LLM Agents in Tradable Credit Schemes: Market-Network Dynamics under Threshold-Based Eligibility

Abstract

This study develops a large language model (LLM) based behavioural simulation framework to examine how tradable credit schemes perform when car use depends on a qualification threshold in daily operation. The framework combines traveller agents with LLM based decision policies, repeated intraday credit trading, and MFD based network loading in a repeated daily setting. Using a 5/8 credit regime, we show that the system approaches a stable operating state, but realised car use remains below the theoretical upper bound implied by perfect reallocation. Agent level tracing and follow up interviews indicate that threshold inefficiency is sustained by option value holding and optimistic threshold chasing, while car access is redistributed unevenly across traveller groups. These findings provide a mechanism based understanding of daily scheme performance and offer a basis for policy design and future testing of credit endowments, eligibility thresholds, and market rules.