ATLAS: Adaptive Trading with LLM AgentS Through Dynamic Prompt Optimization and Multi-Agent Coordination

TL;DR

The paper addresses the challenge of deploying LLMs for autonomous trading under delayed and noisy rewards by introducing ATLAS, a framework that integrates market intelligence, a central execution agent, and a dynamic prompt optimizer (Adaptive-OPRO). ATLAS links reasoning to executable orders through an order-level decision interface and employs windowed ROI-based prompt updates to continuously improve performance. Across regime-specific equity studies and multiple model families, Adaptive-OPRO consistently outperforms fixed prompts, while reflection-based tuning fails to deliver systematic gains. The work provides a practical, auditable architecture for sequential decision-making with LLMs in high-stakes markets, highlighting the value of structured inputs and explicit execution channels for reliable deployment.

Abstract

Large language models show promise for financial decision-making, yet deploying them as autonomous trading agents raises fundamental challenges: how to adapt instructions when rewards arrive late and obscured by market noise, how to synthesize heterogeneous information streams into coherent decisions, and how to bridge the gap between model outputs and executable market actions. We present ATLAS (Adaptive Trading with LLM AgentS), a unified multi-agent framework that integrates structured information from markets, news, and corporate fundamentals to support robust trading decisions. Within ATLAS, the central trading agent operates in an order-aware action space, ensuring that outputs correspond to executable market orders rather than abstract signals. The agent can incorporate feedback while trading using Adaptive-OPRO, a novel prompt-optimization technique that dynamically adapts the prompt by incorporating real-time, stochastic feedback, leading to increasing performance over time. Across regime-specific equity studies and multiple LLM families, Adaptive-OPRO consistently outperforms fixed prompts, while reflection-based feedback fails to provide systematic gains.

Figures (8)

• Figure 1: ATLAS Framework Overview. The Central Trading Agent submits orders to the Trading Execution Engine via prompts shaped by three specialized analysts and the proposed Adaptive-OPRO optimization technique.
• Figure 2: ROI across three assets using Adaptive-OPRO.
• Figure 3: Daily vs weekly reflection mechanism performance comparison across models and assets, showing ROI percentages (solid = daily, striped = weekly).
• Figure 4: Header and trader identity modifications between iteration 4 and iteration 5, showing title changes and mission statement refinements. Lines in red with a leading "-" and lines in green with a leading "+" indicate deletions and additions, respectively, proposed by Adaptive-OPRO.
• Figure 5: Structural reorganization consolidating sections into a unified PORTFOLIO & CONSTRAINTS section. Lines in red with a leading "-" and lines in green with a leading "+" indicate deletions and additions, respectively, proposed by Adaptive-OPRO.