Financial Stability Implications of Generative AI: Taming the Animal Spirits

TL;DR

The paper investigates how generative AI affects financial stability by conducting controlled lab experiments with LLM-based AI traders and comparing them to human professionals in a classic herd-behavior framework. It uses a Cipriani-style design with Bayesian price updating to quantify rationality, information cascades, and the conditions under which herd behavior is optimal. Key findings show AI agents are more rational and display far fewer cascades than humans, suggesting AI adoption could dampen asset-price bubbles, but when explicitly guided to maximize profits, AI can herd optimally, with nuanced implications for stability. The results highlight both potential stabilizing benefits of AI-driven rationality and the importance of cautious, targeted AI deployment and regulation to manage emergent, incentive-driven herding dynamics in financial markets.

Abstract

This paper investigates the impact of the adoption of generative AI on financial stability. We conduct laboratory-style experiments using large language models to replicate classic studies on herd behavior in trading decisions. Our results show that AI agents make more rational decisions than humans, relying predominantly on private information over market trends. Increased reliance on AI-powered trading advice could therefore potentially lead to fewer asset price bubbles arising from animal spirits that trade by following the herd. However, exploring variations in the experimental settings reveals that AI agents can be induced to herd optimally when explicitly guided to make profit-maximizing decisions. While optimal herding improves market discipline, this behavior still carries potential implications for financial stability. In other experimental variations, we show that AI agents are not purely algorithmic, but have inherited some elements of human conditioning and bias.

Figures (13)

• Figure 1: Herding behavior and financial stability The diagram shows how herding can lead to a financial stability event both when optimal and suboptimal. While suboptimal herding is the greatest concern from a financial stability perspective, optimal herding can build up financial vulnerabilities as well.
• Figure 1: System prompt This prompt describes the instructions of the experiment, which is given to the LLMs through their system prompt.
• Figure 2: Flow diagrams of experiments The figure shows diagrams of the order of events for each session of the experiments under (a) Treatment I (without event uncertainty), (b) Treatment II (with event uncertainty), and (c) Treatment III (without price updating). The experiment is an adoption of cipriani2009herdcipriani2005herd and is based on the avery1998multidimensional model.
• Figure 2: System prompt(continued)
• Figure 3: Flow diagrams of experiments(continued)