RealFin: How Well Do LLMs Reason About Finance When Users Leave Things Unsaid?
Abstract
Reliable financial reasoning requires knowing not only how to answer, but also when an answer cannot be justified.In real financial practice, problems often rely on implicit assumptions that are taken for granted rather than stated explicitly, causing problems to appear solvable while lacking enough information for a definite answer.We introduce RE-ALFIN, a bilingual benchmark that evaluates financial reasoning by systematically removing essential premises from exam-style questions while keeping them linguistically plausible.Based on this, we evaluate models under three formulations that test answering, recognizing missing information, and rejecting unjustified options, and find consistent performance drops when key conditions are absent.Generalpurpose models tend to over-commit and guess, while most finance-specialized models fail to clearly identify missing premises.These results highlight a critical gap in current evaluations and show that reliable financial models must know when a question should not be answered.The dataset and code are available at https://github.com/xnlp-lab/RealFin.A 5-year EUR government bond is expected to experience a 25 bp yield decline over the next 6 months.Which statement about the rolldown return is correct? A. The rolldown return equals the difference between the price of the 5-year bond and that of a 4.5-year bond at the lower yield-to-maturity.B. The rolldown return consists of the 5-year bond's basis point value mul-tiplied by the expected 25 bp yield-to-maturity change over the next six months. C. The rolldown return will be negative if the 5-year bond has a zero coupon and is trading at a premium.A 5-year EUR government bond is expected to experience a yield decline over the next 6 months.The bond is a zero-coupon bond currently trading at a premium.Over the holding period, the bond's total return is expected to be close to zero.The most likely magnitude of the yields-to-maturity to decline by ?bps A. 10 B. 20 C. 25 D. 30 AOption A correctly describes this calculation.Option B incorrectly uses BPV, which relates to parallel shifts.Option C is incorrect as premium/discount status doesn't determine rolldown return direction.Conf: 95
Fields of Study & Subjects
Spreadsheets and End-User Computing
Personal Information Management and User Behavior
FinTech
Crowdfunding
Digital Finance
Publication details
- Journal / source
- Findings of the Association for Computational Linguistics: ACL 2026
- Publisher
- Association for Computational Linguistics
- Pages
- 25050–25080
- Publication date
- 1 January 2026