FN-Agents: Analysis of Exchange Rate Volatility Prediction Based on Multi-Agent Systems

Exchange rate prediction is challenging. While large language models (LLMs) offer new approaches, their lack of domain-specific fine-tuning hinders effective assessment of factors influencing exchange rates. To address this, we propose FN-Agents—a framework integrating technologies such as vector databases, Retrieval-Augmented Generation(RAG), and self-reflection—to enable LLMs to better utilize external resources for exchange rate tasks. FN-Agents assists LLMs in acquiring expert knowledge, learning through self-reflection, building fine-tuned datasets, and ultimately generating interpretable core feature sets autonomously. Experiments demonstrate that FN-Agents excels in feature selection and enhances the predictive accuracy of mainstream models.