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

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is: how to simulate human behavior in bilateral financial transactions by combining large - language models (LLMs) with agent - based models (ABMs). Specifically, the paper aims to: 1. **Enhance the sense of realism in simulation**: By using LLMs trained on a large amount of human texts, capture and simulate the complex bilateral dialogues in financial transactions, thereby improving the authenticity of the simulation. 2. **Address structural and design challenges**: Solve the challenges encountered in coordinating the interactions between LLMs - based agents (such as taking turns to speak) and in interpreting the outputs of LLMs. 3. **Verify the application of LLMs in complex decision - making**: Explore whether LLMs can accurately simulate the decision - making process of humans in bilateral asset transactions, especially in the government bond market. ### Specific research questions The paper explores the above - mentioned goals through two research questions: - **Research question 1 (RQ1)**: Can LLMs make realistic and appropriate responses in bilateral trading interactions? - Through simple scenario tests, evaluate whether LLMs can correctly reason and choose not to trade (i.e., "no trade") given natural - language trading intentions, in order to avoid complex numerical calculations. - **Research question 2 (RQ2)**: Can generative agent - based models (GABMs) make trading decisions in a realistic manner? - In more complex scenarios, by passing specific agent information to LLMs and allowing LLMs to participate in the decision - making process, evaluate whether they can generate reasonable simulated dialogues according to trading intentions and execution situations. ### Main contributions - **Propose the TraderTalk model**: A novel generative agent - based model for simulating real - world bilateral financial trading interactions. - **Empirical analysis**: Through multiple simulation experiments, verify the reasoning ability and decision - making accuracy of LLMs at different levels of complexity. - **Enhanced realism**: Demonstrate that LLMs - enhanced ABMs can, to a certain extent, simulate the trading decision - making process under sparse data conditions in the real world, providing meaningful behavioral simulations in the bilateral trading environment. ### Conclusion The research shows that although LLMs may be misinterpreted or make mistakes in some cases, on the whole, they can significantly improve the authenticity and rationality of the simulation, laying the foundation for future multi - agent and multi - market simulation research.

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

TradingAgents: Multi-Agents LLM Financial Trading Framework

Reproducing and Extending Experiments in Behavioral Strategy with Large Language Models

An Experimental Study of Competitive Market Behavior Through LLMs

Modelling Opaque Bilateral Market Dynamics in Financial Trading: Insights from a Multi-Agent Simulation Study

Simulating Human Strategic Behavior: Comparing Single and Multi-agent LLMs

Simulating Financial Market via Large Language Model based Agents

Learning and Calibrating Heterogeneous Bounded Rational Market Behaviour with Multi-Agent Reinforcement Learning

GLEE: A Unified Framework and Benchmark for Language-based Economic Environments

Neural Stochastic Agent-Based Limit Order Book Simulation: A Hybrid Methodology

Systematic Biases in LLM Simulations of Debates

Smart Agent-Based Modeling: On the Use of Large Language Models in Computer Simulations

LLM-driven Imitation of Subrational Behavior : Illusion or Reality?

Agent-Based Modelling Meets Generative AI in Social Network Simulations

LLM economicus? Mapping the Behavioral Biases of LLMs via Utility Theory

MetaAgents: Simulating Interactions of Human Behaviors for LLM-based Task-oriented Coordination via Collaborative Generative Agents

How Well Can LLMs Negotiate? NegotiationArena Platform and Analysis

Many learning agents interacting with an agent-based market model

TradingGPT: Multi-Agent System with Layered Memory and Distinct Characters for Enhanced Financial Trading Performance

Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?

Interacting Large Language Model Agents. Interpretable Models and Social Learning