Evasive Answers in Financial Q\&A: Earnings Calls vs. FOMC Press Conferences

Khaled AlNuaimi, Gautier Marti, Alexis Marchal, Mohamad Alansari, Andreas Henschel

Keywords: Earnings Calls, Evasive Answer Detection, FOMC

TL;DR: We introduce evasive answer detection in financial Q&A, showing it is a distinct, measurable phenomenon that impacts transparency and financial NLP tasks

Abstract: Question--answer (Q&A) sessions in earnings calls and central bank press conferences provide high-stakes, unscripted insights into firms and the macroeconomy. Executives often respond evasively by avoiding, reframing, or obscuring answers, which limits transparency and biases downstream NLP tasks such as sentiment, risk, and event prediction. We introduce the task of evasive answer detection in financial Q&A and propose a three-level taxonomy grounded in pragmatics and psychology. Using annotated transcripts from earnings calls and FOMC press conferences, we show that lightweight features including hedges, verbosity, tense shifts, and semantic alignment capture robust signals of evasiveness. Our baselines demonstrate that evasiveness is linguistically and semantically distinct from sentiment and veracity, supporting its treatment as a standalone problem. This work establishes a foundation for benchmarks and models that incorporate evasiveness cues into financial NLP pipelines, market surveillance, and transparency assessment.

Submission Number: 51