Llm agent capabilities reveal gaps beyond financial market outcomes
LiveMACE: Process-Aware Evaluation of LLM Agent Capabilities in Evolving Markets
Artificial IntelligenceComputation and Language
Summary
Measuring AI agents only by their results can hide how they actually work. The authors created LiveMACEBench, a way to test AI agents using real, changing financial markets that show how agents make decisions over time. They found that similar financial outcomes can come from very different ways of using tools, memory, and teamwork. This means looking at results alone misses important differences in AI abilities. Their approach helps spot where AI agents struggle so improvements can be more targeted.
What this means in practice
- •For financial technology teams: Use live market data to test and improve how AI agents apply tools and memory in trading environments beyond just profit results.
- •For multi-agent system developers: Diagnose specific strengths and weaknesses in AI teamwork and rule-following by analyzing detailed decision traces in evolving conditions.
Authors
Jun Zhao, Leiming Fu, Yanbo Wen, Yiding Wang, Xuantong Liu, Yang Shu, Yuyang Lu, Xuanran Xing, Jingqi Tong, Hao Xu, Qi Zhang, Xuanjing Huang
Abstract
Evaluating agents by outcomes alone can obscure the capabilities that produce them. This problem is especially pronounced in evolving environments, where outcomes reflect a closed-loop interaction between agent behavior and changing external conditions. We introduce LiveMACEBench, a process-aware benchmark that uses live financial markets as a naturally evolving testbed for persistent LLM agents. Five frontier LLMs operate along continuous trajectories under matched Tool Use, Persistent Memory, Rule Following, and Multi-Agent Collaboration configurations. We evaluate them through both realized outcomes and mechanism-specific diagnostics derived from complete decision traces. Across 30 days of live evaluation, we find a pronounced outcome-capability gap: realized returns often diverge from capability-specific measurements, and similar outcomes can arise from markedly different patterns of mechanism use. Trace-level diagnostics further expose distinct bottlenecks across capabilities, demonstrating that mechanism access, effective mechanism use, and downstream performance are not interchangeable measures of agent capability. LiveMACEBench makes this distinction measurable, turning live markets from a performance leaderboard into a diagnostic environment for agent capability