Ex Machina: Financial Stability in the Age of Artificial Intelligence

Kartik Anand, Sophia Kazinnik, Agnese Leonello and Ettore Panetti

No 20681, CEPR Discussion Papers from Centre for Economic Policy Research

Abstract: We study financial decision making by reinforcement learning algorithms and large language models (LLMs) in a mutual fund redemption model with strategic complementarities and multiple equilibria. Both depart from the efficient equilibrium and trigger financial fragility, but through distinct mechanisms. Reinforcement learning (Q-learning) algorithms redeem excessively because of a learning bias, which we characterize formally and show can produce non-Nash equilibrium outcomes. LLMs fail to coordinate because they hold different beliefs about others' decisions, which we derive from their chain-of-thought reasoning. Our results show that the design of AI agents can shape collective outcomes and, ultimately, financial stability.

Keywords: Coordination games; Financial stability; Q-learning; Large Language Models; Artificial intelligence; Strategic uncertainty (search for similar items in EconPapers)
JEL-codes: C63 G01 G23 (search for similar items in EconPapers)
Date: 2025-09
New Economics Papers: this item is included in nep-cmp and nep-rmg
References: Add references at CitEc
Citations:

Downloads: (external link)
https://cepr.org/publications/DP20681 (application/pdf)

Related works:
Working Paper: Ex Machina: financial stability in the age of artificial intelligence (2026)
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:cpr:ceprdp:20681

Ordering information: This working paper can be ordered from
https://cepr.org/publications/DP20681

Access Statistics for this paper

More papers in CEPR Discussion Papers from Centre for Economic Policy Research 33 Great Sutton Street, London EC1V 0DX, UK.
Bibliographic data for series maintained by CEPR ().