Abstract
Significant progress has been made in automated problem-solving usingsocieties of agents powered by large language models (LLMs). In finance,efforts have largely focused on single-agent systems handling specific tasks ormulti-agent frameworks independently gathering data. However, multi-agentsystems' potential to replicate real-world trading firms' collaborativedynamics remains underexplored. TradingAgents proposes a novel stock tradingframework inspired by trading firms, featuring LLM-powered agents inspecialized roles such as fundamental analysts, sentiment analysts, technicalanalysts, and traders with varied risk profiles. The framework includes Bulland Bear researcher agents assessing market conditions, a risk management teammonitoring exposure, and traders synthesizing insights from debates andhistorical data to make informed decisions. By simulating a dynamic,collaborative trading environment, this framework aims to improve tradingperformance. Detailed architecture and extensive experiments reveal itssuperiority over baseline models, with notable improvements in cumulativereturns, Sharpe ratio, and maximum drawdown, highlighting the potential ofmulti-agent LLM frameworks in financial trading. More details on TradingAgentsare available at https://TradingAgents-AI.github.io.