TwinMarket: A Scalable Behavioral and Social Simulation for Financial Markets

TL;DR

This work tackles the challenge of modeling social emergence in financial markets by introducing TwinMarket, a scalable framework that combines LLM-driven agents with a Belief-Desire-Intention cognitive architecture and a dynamic social network to simulate investor behavior. The approach unites micro-level cognition and macro-level information diffusion within an order-driven market, grounded in real-world data sources. Through micro- and macro-level validations and extensive ablations, the authors show that the framework reproduces key financial stylized facts (fat tails, leverage effect, volatility clustering) and population-level phenomena (wealth inequality, opinion leadership) while enabling scalable experiments up to large agent populations. The results offer a high-fidelity synthetic data generator and a versatile platform for investigating behavioral finance theories and social emergence in complex markets.

Abstract

The study of social emergence has long been a central focus in social science. Traditional modeling approaches, such as rule-based Agent-Based Models (ABMs), struggle to capture the diversity and complexity of human behavior, particularly the irrational factors emphasized in behavioral economics. Recently, large language model (LLM) agents have gained traction as simulation tools for modeling human behavior in social science and role-playing applications. Studies suggest that LLMs can account for cognitive biases, emotional fluctuations, and other non-rational influences, enabling more realistic simulations of socio-economic dynamics. In this work, we introduce TwinMarket, a novel multi-agent framework that leverages LLMs to simulate socio-economic systems. Specifically, we examine how individual behaviors, through interactions and feedback mechanisms, give rise to collective dynamics and emergent phenomena. Through experiments in a simulated stock market environment, we demonstrate how individual actions can trigger group behaviors, leading to emergent outcomes such as financial bubbles and recessions. Our approach provides valuable insights into the complex interplay between individual decision-making and collective socio-economic patterns.

Figures (37)

• Figure 1: Overview of TwinMarket environment where each user has a unique persona within the social network, interacts with the environment in real-time, and influences it through their actions. This framework enables the study of emergent social phenomena.
• Figure 2: The simulation workflow driven by the Belief-Desire-Intention framework. The micro-level simulation focuses on modeling individual user behavior, while the macro-level simulation addresses the dynamics of the social media platform and trading system.
• Figure 3: Chord diagram of user dynamic social network.
• Figure 4: Information propagation in network.
• Figure 5: Rising Gini coefficient, indicating widening wealth inequality.