AI Personas predict Fed rate moves 3 quarters ahead

Researchers Hayden Helm and Andrew Dassori introduced a novel AI-driven index to predict U.S. Federal Open Market Committee (FOMC) rate decisions. Their method constructs a digital 'persona' for each FOMC member using a retrieval-augmented generative system trained on nearly 25,000 chunks of publicly available data. Each persona's generated content is nearly indistinguishable from real statements (detectability score 0.23 vs. 0.15 floor), and the personas accurately identify their real-world counterparts (8× chance). The index captures members' monetary-policy stances (hawk–dove ordering) with Kendall's τ=0.63, outperforming simple retrieval-only representations.

For the 2022–2025 period, the Persona-based Rate Action Index tracked the actual rate cycle with Kendall's τ=0.68 (p<10⁻⁶) and led the federal funds target rate by approximately three quarters. A simple classifier built from the index achieves 0.69 accuracy against a 0.47 base rate. This is the first demonstration of capturing time-varying group behavior through a collection of AI personas, offering a powerful new tool for economic forecasting and multiagent system analysis.