LLM Agents in a Double Auction Converge Slower Than Humans or Not at All

arXiv 2609.02580·medium signalXBlueskyLinkedInCopy link

Replicating seminal economic experiments with LLM agents substituted for human subjects, the authors place agents in a double auction — a widely used market mechanism — and test whether it still delivers efficient allocation, framing this as a novel alignment dimension: compatibility with a mechanism designed for humans. Markets populated by LLM agents showed slower convergence toward equilibrium or none at all, giving less efficient allocations than human-populated markets, with substantial heterogeneity across model families and market roles. A lexical analysis of the chain-of-thought traces found the decision to execute a trade rather than keep incrementally adjusting price coincides with a shift from strategic language toward urgency; the testing framework is publicly released.