Multi-Scale Sentiment and Market Microstructure — An Agent-Based Framework for Cryptocurrency Markets

This paper presents a methodological framework for agent-based modeling (ABM) of cryptocurrency market microstructure with real-time sentiment integration. The proposed architecture combines three novel components: (1) a Monte Carlo Dropout sentiment analyzer providing uncertainty-aware sentiment scores from social media streams, (2) a heterogeneous agent population including market makers, informed traders, noise traders, and arbitrageurs with sentiment-responsive behavior rules, and (3) a dynamic factor model (DFM) for regime detection enabling adaptive agent parameterization.