Title: Assessing the Efficacy of Matching Mechanisms in LLM-Agent Markets

Abstract:

This research investigates the operational viability of conventional matching mechanisms within markets driven by Large Language Model (LLM) agents, which act as delegated decision-makers in allocation processes. The study contrasts decentralized markets reliant on free negotiation against centralized markets governed by mechanism-based rules, encompassing several prominent mechanisms. In controlled one-to-one matching scenarios, mechanism-driven markets consistently demonstrate superior performance in both stability and efficiency compared to free negotiation.

Furthermore, the findings reveal that LLM agents exhibit significantly higher rates of truthful preference reporting than human participants in comparable Deferred Acceptance (DA) and Evenly Adjusted Deferred Acceptance (EADA) environments. However, the alignment between truth-telling and formal strategy-proofness is not universal across all mechanisms. Notably, the Top Trading Cycles (TTC) mechanism, although strategy-proof, does not consistently induce higher levels of truthfulness than EADA. These outcomes indicate that while matching theory offers valuable insights for designing institutions in LLM-agent markets, it remains an incomplete framework for such applications.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC