The Impact of AI on the Fear & Greed Index on Limit Order Book Microstructure and Execution Costs: An Agent-Based Simulation of Crypto Markets

Abstract

This conference paper reports the core findings of an agent-based limit order book (LOB) simulation designed to examine whether the Fear & Greed Index (FGI) changes crypto-market microstructure and execution quality. The NetLogo model combines explicit buy and sell order books, a price-priority matching mechanism, mixed Poisson/Hawkes-like order arrivals, heavy-tail order sizes, and rule-based cancellations. Sentiment is introduced parsimoniously: the probability that the next order is a buy order equals FGI/100. In a 50,000-trade experiment, higher FGI regimes produce a strong directional shift in order flow, with buy-order share rising from 25.64% in the 20-35 FGI bin to 73.05% in the 65-80 bin. Average order-book imbalance simultaneously moves from -0.5175 to 0.4424. Quoted spread changes only modestly across sentiment regimes, but effective spread increases from 3.0029 to 3.9329, indicating materially worse execution costs under greed-dominated conditions. The model also reproduces important stylized facts, including fat-tailed return distributions and persistent volatility clustering in mid-price returns. These results suggest that sentiment primarily influences market quality through directional order-flow pressure liquidity depletion rather than through a simple widening of top-of-book quotes.