The Language of Evasion: How Semantic Similarity Between Questions and Answers Predicts Stock Returns

We examine executive responsiveness during earnings calls and its impact on stock returns. Using large language model embeddings, we measure semantic similarity between analyst questions and executive responses, capturing direct answers versus deflection. Executives who provide semantically aligned responses generate 3.9% annual alpha (t = 3.41), robust to sentiment, firm characteristics, and market factors. Human validation on 1,642 Q&A pairs shows low-similarity responses are rated evasive 67% of the time versus 22% for high-similarity responses (r = 0.360, p < 0.001; Cohen’s d = 1.01).