Logo FinanceMath

Knowledge-Intensive Math Reasoning in Finance Domains

Yilun Zhao1, Hongjun Liu2, Yitao Long3, Rui Zhang4, Chen Zhao2,3, Arman Cohan1,5,
1Yale University, 2NYU Shanghai, 3New York University
4Penn State University, 5Allen Institute for AI
ACL 2024 Oral
Paper Code

🤗

 Dataset

🔮

 Visualize

🏆

 Leaderboard
data-overview

Overview of FinanceMath.

Introduction

Large Language Models (LLMs) have demonstrated impressive problem-solving capabilities across various domains, but their ability to handle knowledge-intensive mathematical reasoning in specialized fields remains understudied. To address this gap, we introduce FinanceMath, a benchmark designed to evaluate LLMs' performance in knowledge-intensive mathematical reasoning within the finance domain. FinanceMath comprises 1200 examples covering a wide range of finance subareas, with 40.2% of problems requiring interpretation of tabular data. Each problem is accompanied by expert-annotated, Python-formatted solutions.

📢 Update (July 2026): the FinanceMath test set annotations (Python solutions and answers) are now publicly released. You can evaluate on the test set locally — see the GitHub repository for the data and evaluation scripts.

FinanceMath Dataset

Visualization

BibTeX

@misc{zhao2024financemath,
      title={FinanceMath: Knowledge-Intensive Math Reasoning in Finance Domains}, 
      author={Yilun Zhao and Hongjun Liu and Yitao Long and Rui Zhang and Chen Zhao and Arman Cohan},
      year={2024},
      eprint={2311.09797},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2311.09797}, 
}