This paper presents an agent-based stock market simulation where trading behavior is driven by exogenous sentiment processes rather than strategic optimization. The model incorporates three key sentiments: buy/sell dis-balance (ψ), volatility (σ with calm and jump states), and trading intensity (ρ). Agents draw their individual sentiments from Gaussian distributions centered on these common processes, influencing their order direction, limit price, and participation frequency. The stock exchange uses an order book with a matching engine to determine equilibrium prices.

Simulations with 1000 agents over up to 100,000 steps show that the stock price mean-reverts around the ratio of total cash to shares, with a slight downward bias from the absolute value in limit price draws. The wealth distribution of the top 25% agents converges to a Pareto distribution with exponent a ≈ 4, regardless of initial allocation (identical, uniform, normal, or Pareto). Introducing jump volatility (calm σ=0.05, breaking σ=0.2) produces fat-tailed log-returns, matching empirical features of stocks like TSLA and FB. A shift in buy/sell sentiment from neutral to bearish (ψ from 0 to -0.3) causes a rapid price decline of about 26%, while heterogeneous sentiments across subgroups lead to more gradual adjustments and differential wealth dynamics. The study highlights that sentiment-driven models can replicate key market phenomena without assuming rational expectations or strategic agents.