ATLAS: Multi-Agent LLM Framework for Adaptive Trading
ATLAS (Adaptive Trading with LLM AgentS) represents a comprehensive multi-agent system designed to amalgamate structured data from markets, news, and corporate fundamentals, thereby enhancing trading decision-making. The primary trading agent functions within an action space that is aware of orders, ensuring that its outputs translate into actionable market orders instead of mere abstract signals. Utilizing Adaptive-OPRO, an innovative prompt-optimization method, the agent can adjust its prompts in real-time based on stochastic feedback during trading. This framework effectively tackles issues like delayed rewards masked by market noise, integrates diverse information sources, and aligns model outputs with actionable market activities.
Key facts
- ATLAS stands for Adaptive Trading with LLM AgentS
- It is a multi-agent framework for financial trading
- Integrates information from markets, news, and corporate fundamentals
- Central trading agent uses an order-aware action space
- Adaptive-OPRO is a novel prompt-optimization technique used
- Addresses delayed rewards, heterogeneous information synthesis, and action execution gaps
- Published on arXiv with ID 2510.15949
- Announce type is replace-cross
Entities
Institutions
- arXiv