← Back
Evidence source 5098Spot Checked

Ex Machina: Financial Stability in the Age of Artificial Intelligence

papers.ssrn.com2025-10-07Paper
Executive summary

A new study compares Q-learning algorithms and large language models (LLMs) as AI agents in mutual fund redemption scenarios. Q-learning agents coordinate but trigger early, excessive redemptions, increasing financial fragility. LLMs use reasoning and context, reducing unnecessary redemptions and improving stability, but struggle with coordination due to diverse beliefs. The research uses simulations to replace human investors with AI, revealing that AI design can shape financial outcomes. The authors call for more real-world data and policy attention.

What it examines

This paper studies how different types of artificial intelligence (AI) agents, specifically Q-learning algorithms and large language models (LLMs), affect financial stability in mutual fund redemption scenarios. Using simulations, it explores how these AI agents make decisions under strategic uncertainty and multiple possible outcomes.

What it concludes

The study finds Q-learning agents coordinate but can increase financial fragility, while LLMs improve stability but reduce coordination. These results suggest that the design of AI systems can impact financial stability, highlighting the need for careful regulation and further research on AI behavior in finance and other strategic settings.

Extracted from this source

Evidence objects

Evidence 395086% extraction confidence
Researchers simulated mutual fund redemptions using Q-learning and large language model (LLM) AI agents, revealing how different AI designs can dramatically impact financial stability under uncertainty and asset illiquidity.

key_findings bullet 1 · key_findings · validation V0

Evidence 395186% extraction confidence
Q-learning agents coordinated actions but triggered excessive early redemptions, increasing system fragility and crisis risk. In contrast, LLM agents used reasoning to reduce unnecessary redemptions, though struggled with coordination due to diverse beliefs.

key_findings bullet 2 · key_findings · validation V0

Evidence 395286% extraction confidence
The study highlights that AI system design shapes economic outcomes in unexpected ways, urging policymakers to rethink technologys role in finance. Authors call for more real-world data to validate these simulation-based findings.

key_findings bullet 3 · key_findings · validation V0

Evidence 395386% extraction confidence
This paper introduces a novel simulation-based framework comparing Q-learning and LLM agents in coordination games with multiple equilibria, uniquely analyzing their impact on financial stability during mutual fund redemptions. Its original focus on agentic AI, belief heterogeneity, and financial fragility offers compelling insights for both academic research and practical risk management applications.

key_findings bullet 4 · key_findings · validation V0

Raw abstract and provenance

This paper develops a simulation-based framework to study how AI agents behave in a mutual-fund redemption game with strategic complementarities and multiple

Source row: 747 · abstract type: snippet