Jörg Osterrieder
Generative artificial intelligence—encompassing large language models, generative adversarial networks, and diffusion models—is transforming financial services across modeling, trading, risk management, and compliance. This paper develops a Micro-Meso-Macro conceptual framework identifying sixteen channels—twelve risk and four benefit—through which GenAI affects systemic risk in finance. At the micro level, channels address model risk amplification, operational dependency, competitive divergence, and automation bias. At the meso level, channels capture algorithmic monoculture, market microstructure transformation, information ecosystem distortion, and interconnectedness. At the macro level, channels encompass procyclicality amplification, regulatory arbitrage acceleration, too-connected-to-fail AI infrastructure, and cross-border regulatory fragmentation. Four benefit channels identify conditional innovation pathways. We derive sixteen testable propositions with identification strategies and falsification conditions.
Keywords: Generative AI, systemic risk, financial stability, large language models, conceptual framework, financial regulation, model risk, algorithmic monoculture
JEL: G01, G18, G21, G28, O33