In real-world stock markets, certain chart patterns—such as price declines near historical highs—cannot be fully explained by fundamentals alone. These anomalies suggest path dependence in price formation, where investor decisions are shaped not only by current market conditions but also by preceding price trajectories. In behavioral finance, such path dependence has been attributed to loss aversion, anchored to personal reference points like purchase prices or past peaks. However, incorporating these subtle behavioral biases into traditional agent-based market simulations has been challenging. To address this, we propose the Fundamental-Chartist-LLM-Agent (FCLAgent), which leverages large language models (LLMs) to emulate human-like trading decisions. In this framework, LLMs determine buy/sell intentions based on individual contexts, while order price and volume are generated by standard rule-based mechanisms. Simulation results demonstrate that FCLAgents successfully reproduce path-dependent patterns that conventional agents fail to capture.