Scholarlink

<p>In stock markets, prices often respond to specific anchors such as past peaks adjust only gradually to supply–demand imbalances, rather than reflecting fundamentals immediately. These phenomena suggest path dependence, where investor decisions are influenced not only by current market conditions but also by the trajectory of prices leading up to the present. One plausible driver of such path dependence is human behavioral biases, which cause aggregate market outcomes to deviate systematically from rational equilibrium. The challenge in formalizing these biases is that they are inherently context-dependent: their manifestations vary not only with current market conditions but also with factors such as individual trading history or the environment in which the investor operates. To capture the context-dependent nature of behavioral biases and to investigate how such context dependence contributes to the emergence of path-dependent price dynamics in stock markets, we adopt a large language model- (LLM-)augmented agent-based modeling approach. Our analysis proceeds in two stages. Micro analysis: we run controlled trading experiments to test LLMs’ context-dependent behavioral biases. Macro analysis: we introduce the LLM-based agents in an artificial market and conduct multi-agent simulations. These experiments revealed that (1) LLMs’ behavioral biases are context-dependent similar to humans and (2) inclusion of LLMs into artificial market simulations enables the reproduction of path-dependent anomalies in stock prices that conventional agent models had previously failed to capture. Together, these results demonstrate how LLM-based agents can advance constructive modeling of market dynamics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

LLM agents reveal how human bias shapes path-dependent market dynamics

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月19日

查看全文
  • Ryuji Hashimoto,
  • Takehiro Takayanagi,
  • Masahiro Suzuki,
  • Kiyoshi Izumi

摘要

In stock markets, prices often respond to specific anchors such as past peaks adjust only gradually to supply–demand imbalances, rather than reflecting fundamentals immediately. These phenomena suggest path dependence, where investor decisions are influenced not only by current market conditions but also by the trajectory of prices leading up to the present. One plausible driver of such path dependence is human behavioral biases, which cause aggregate market outcomes to deviate systematically from rational equilibrium. The challenge in formalizing these biases is that they are inherently context-dependent: their manifestations vary not only with current market conditions but also with factors such as individual trading history or the environment in which the investor operates. To capture the context-dependent nature of behavioral biases and to investigate how such context dependence contributes to the emergence of path-dependent price dynamics in stock markets, we adopt a large language model- (LLM-)augmented agent-based modeling approach. Our analysis proceeds in two stages. Micro analysis: we run controlled trading experiments to test LLMs’ context-dependent behavioral biases. Macro analysis: we introduce the LLM-based agents in an artificial market and conduct multi-agent simulations. These experiments revealed that (1) LLMs’ behavioral biases are context-dependent similar to humans and (2) inclusion of LLMs into artificial market simulations enables the reproduction of path-dependent anomalies in stock prices that conventional agent models had previously failed to capture. Together, these results demonstrate how LLM-based agents can advance constructive modeling of market dynamics.