Multi-scale sentiment analysis framework for agent-based modeling of cryptocurrency market microstructure. Blends institutional macro signals (ASRI framework) with retail micro signals (CryptoBERT with MC Dropout). In stylized simulation, multi-scale blending reduces volatility from 877% to 5.8% (p=0.013) and spreads from 147 to 3.5 bps (p<0.001) compared to single-source sentiment. Regime-adaptive weighting adjusts macro/micro blend based on detected market conditions.
@misc{farzulla2025sentimentabm,
author = {Farzulla, Murad},
title = {The Extremity Premium: Sentiment Regimes and Adverse Selection in Cryptocurrency Markets},
year = {2025},
howpublished = {Farzulla Research Working Paper DAI-2510},
doi = {10.5281/zenodo.17989810},
url = {https://farzulla.org/papers/sentiment-abm}
}