FinCoT: Grounding Chain-of-Thought in Expert Financial Reasoning
arXiv link
Date: 9th July 2025
Key Points
- Improves LLM reasoning in finance applications via prompt engineering
- Structured CoT prompting:
- Outlines the reasoning steps the agent should take. Finance workflows are fairly well known. Telling the model to following
these workflows improves model performance.
- Obtained by using DeepResearch on finance methods and then o3 to outline the steps required to solve each problem.
This is checked by a human in the loop.
- Benefit: more auditable