Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Shifting Power: Leveraging LLMs to Simulate Human Aversion in ABMs of Bilateral Financial Exchanges, A bond market study

Alicia Vidler, Toby Walsh

TL;DR

This work tackles the difficulty of modeling opaque OTC bilateral bond markets by proposing TRIBE, a generative ABM that embeds LLM-driven, human-like client decision-making. The approach enables rich client agency and variability in trading intent, revealing that even modest aversion signals can halt trading, while human-like variability can shift power toward clients and trigger systemic instability. Through three experiments in an Australian government bond setting, TRIBE demonstrates both the feasibility of combining LLMs with ABMs and the sensitivity of decentralized markets to agent heterogeneity and prompt design. The findings carry practical implications for market design and regulation, illustrating how dynamic client behavior can reshape liquidity and stability in OTC markets and offering a platform for regulatory experimentation and policy testing.

Abstract

Bilateral markets, such as those for government bonds, involve decentralized and opaque transactions between market makers (MMs) and clients, posing significant challenges for traditional modeling approaches. To address these complexities, we introduce TRIBE an agent-based model augmented with a large language model (LLM) to simulate human-like decision-making in trading environments. TRIBE leverages publicly available data and stylized facts to capture realistic trading dynamics, integrating human biases like risk aversion and ambiguity sensitivity into the decision-making processes of agents. Our research yields three key contributions: first, we demonstrate that integrating LLMs into agent-based models to enhance client agency is feasible and enriches the simulation of agent behaviors in complex markets; second, we find that even slight trade aversion encoded within the LLM leads to a complete cessation of trading activity, highlighting the sensitivity of market dynamics to agents' risk profiles; third, we show that incorporating human-like variability shifts power dynamics towards clients and can disproportionately affect the entire system, often resulting in systemic agent collapse across simulations. These findings underscore the emergent properties that arise when introducing stochastic, human-like decision processes, revealing new system behaviors that enhance the realism and complexity of artificial societies.

Shifting Power: Leveraging LLMs to Simulate Human Aversion in ABMs of Bilateral Financial Exchanges, A bond market study

TL;DR

This work tackles the difficulty of modeling opaque OTC bilateral bond markets by proposing TRIBE, a generative ABM that embeds LLM-driven, human-like client decision-making. The approach enables rich client agency and variability in trading intent, revealing that even modest aversion signals can halt trading, while human-like variability can shift power toward clients and trigger systemic instability. Through three experiments in an Australian government bond setting, TRIBE demonstrates both the feasibility of combining LLMs with ABMs and the sensitivity of decentralized markets to agent heterogeneity and prompt design. The findings carry practical implications for market design and regulation, illustrating how dynamic client behavior can reshape liquidity and stability in OTC markets and offering a platform for regulatory experimentation and policy testing.

Abstract

Bilateral markets, such as those for government bonds, involve decentralized and opaque transactions between market makers (MMs) and clients, posing significant challenges for traditional modeling approaches. To address these complexities, we introduce TRIBE an agent-based model augmented with a large language model (LLM) to simulate human-like decision-making in trading environments. TRIBE leverages publicly available data and stylized facts to capture realistic trading dynamics, integrating human biases like risk aversion and ambiguity sensitivity into the decision-making processes of agents. Our research yields three key contributions: first, we demonstrate that integrating LLMs into agent-based models to enhance client agency is feasible and enriches the simulation of agent behaviors in complex markets; second, we find that even slight trade aversion encoded within the LLM leads to a complete cessation of trading activity, highlighting the sensitivity of market dynamics to agents' risk profiles; third, we show that incorporating human-like variability shifts power dynamics towards clients and can disproportionately affect the entire system, often resulting in systemic agent collapse across simulations. These findings underscore the emergent properties that arise when introducing stochastic, human-like decision processes, revealing new system behaviors that enhance the realism and complexity of artificial societies.

Paper Structure

This paper contains 12 sections, 4 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Literature and Recent work
  3. Agent-based models
  4. Large Language Models (LLMs)
  5. Human Trading behaviors: Aversions and Delays
  6. TRIBE: Artificial society and trading marketplace model
  7. The Experiment and Challenge
  8. Experiment 1: TRIBE Traditional ABM (benchmark)
  9. Experiment 2: TRIBE Incorporating (any) Aversion in Prompt with LLMs
  10. Experiment 3: TRIBE(LLM) - Success of prompting for Timeliness and the emergent Shift in Society Dynamics
  11. Future Work
  12. Conclusion

Figures (3)

  • Figure 1: Tribe Architecture
  • Figure 2: APRA data: Balance in $ Billion Australian dollars
  • Figure 3: Yes/no Ratio through time vs Yes/no Rolling 10 Requests

Theorems & Definitions (2)

  • definition 1
  • definition 2