Informational Herding and Conditional Performance of AI-Driven ETFs: A Comparative Analysis with Institutional Investors

While AI is often presumed immune to human behavioral biases like herding, existing literature lacks insights into whether AI-driven funds exhibit herding, their nature, and how they perform under different market conditions—a critical gap for asset management. To address this, we use a modified Lakonishok, Shleifer, and Vishny (LSV) model on CRSP database data to empirically compare the investment patterns of the AI-Powered Equity ETF (AIEQ) with institutional investors. Our key findings are: (1) AI-driven funds do engage in herding, but this is not irrational imitation—instead, it is “informational herding” driven by a 29% average commonality in stock holdings with human managers; (2) a critical “AI gap” emerges in bear markets: AI funds significantly underperform institutional investors due to strategic inflexibility (failing to adapt herding strategies to market downturns). Economically, these findings help investors avoid bear-market risks from AI funds and guide regulators/asset managers in optimizing AI design and deployment; academically, they provide novel evidence on the nature and conditional effectiveness of AI herding, enriching behavioral finance, and AI-in-finance literature.