This paper investigates the application of multi-agent large language models (LLMs) in traditional finance (TradFi) and decentralized finance (DeFi), with a focus on addressing challenges such as inefficiencies, security vulnerabilities, and regulatory requirements. We propose a conceptual framework for implementing multi-agent LLMs, which combines the advanced reasoning and language capabilities of LLMs with the collaborative nature of multi-agent systems. Through detailed case studies, we demonstrate the effectiveness of this framework in portfolio management, fraud detection, smart contract optimization, and regulatory compliance. Empirical data from these case studies show significant improvements in performance metrics such as accuracy, speed, and cost efficiency compared to traditional approaches. For instance, in portfolio management, multi-agent LLMs achieved a 3.8% increase in average returns and a 0.6 improvement in the Sharpe ratio. In DeFi, smart contract optimization reduced vulnerabilities by 66%, while decentralized lending protocols saw a 17% increase in liquidity utilization. This paper concludes with recommendations on how to improve effectiveness of LLMs and outlines future research directions, including the integration of quantum computing and the development of explainable AI techniques.
Paper
Full text
Multi-Agent Large Language Models for Traditional Finance and Decentralized Finance
Semantic Scholar · 2025
Abstract
This paper investigates the application of multi-agent large language models (LLMs) in traditional finance (TradFi) and decentralized finance (DeFi), with a focus on addressing challenges such as inefficiencies, security vulnerabilities, and regulatory requirements. We propose a conceptual framework for implementing multi-agent LLMs, which combines the advanced reasoning and language capabilities of LLMs with the collaborative nature of multi-agent systems. Through detailed case studies, we demonstrate the effectiveness of this framework in portfolio management, fraud detection, smart contract optimization, and regulatory compliance. Empirical data from these case studies show significant improvements in performance metrics such as accuracy, speed, and cost efficiency compared to traditional approaches. For instance, in portfolio management, multi-agent LLMs achieved a 3.8% increase in average returns and a 0.6 improvement in the Sharpe ratio. In DeFi, smart contract optimization reduced vulnerabilities by 66%, while decentralized lending protocols saw a 17% increase in liquidity utilization. This paper concludes with recommendations on how to improve effectiveness of LLMs and outlines future research directions, including the integration of quantum computing and the development of explainable AI techniques.