TraderTalk: An LLM Behavioural ABM applied to Simulating Human Bilateral Trading Interactions

TL;DR

The paper addresses modeling nuanced human bilateral trading interactions in markets with sparse public data by integrating a general-purpose LLM into an ABM, forming TraderTalk. It demonstrates that a GABM with LLM-based negotiation can produce realistic trade intentions and execution patterns in a stylised government-bond market without heavy domain tuning. The approach emphasizes prompt design and coordinated agent interactions to maintain realism while avoiding overfitting and rapid model obsolescence. The work suggests that LLM-augmented ABMs can be a practical tool for policymakers, regulators, and researchers to explore bilateral liquidity and negotiation dynamics in OTC financial markets.

Abstract

We introduce a novel hybrid approach that augments Agent-Based Models (ABMs) with behaviors generated by Large Language Models (LLMs) to simulate human trading interactions. We call our model TraderTalk. Leveraging LLMs trained on extensive human-authored text, we capture detailed and nuanced representations of bilateral conversations in financial trading. Applying this Generative Agent-Based Model (GABM) to government bond markets, we replicate trading decisions between two stylised virtual humans. Our method addresses both structural challenges, such as coordinating turn-taking between realistic LLM-based agents, and design challenges, including the interpretation of LLM outputs by the agent model. By exploring prompt design opportunistically rather than systematically, we enhance the realism of agent interactions without exhaustive overfitting or model reliance. Our approach successfully replicates trade-to-order volume ratios observed in related asset markets, demonstrating the potential of LLM-augmented ABMs in financial simulations

Figures (1)

• Figure 1: Model Architectures