AI Persona Index Predicts Fed Rate Decisions
A new research paper proposes an index for predicting U.S. Federal Open Market Committee (FOMC) decisions on federal funds target rate changes. The index is based on how a collection of AI personas, each representing a committee member, responds to current market conditions. The researchers collected nearly 25,000 retrievable chunks from publicly available data, partitioned into per-member corpora, and used each as a retrieval database for a generative system called a "persona." The personas were evaluated on identifiability and detectability, showing highly attributable behavior (average member-conditional recall 8 times chance) and generated content nearly indistinguishable from held-out real content (detectability score 0.23 against a 0.15 floor). The study presents evidence that query-conditioned representations of the personas capture relevant information for rate decisions.
Key facts
- The index predicts FOMC decisions to hike, hold, or cut the federal funds target rate.
- The dataset consists of nearly 25,000 retrievable chunks from publicly available data.
- Data is partitioned into per-member corpora for each FOMC member persona.
- Average member-conditional recall is 8 times chance.
- Detectability score is 0.23 against a 0.15 floor.
- The research is published on arXiv with ID 2607.26545.
- The approach uses generative AI personas to model committee member behavior.
- Query-conditioned representations of personas capture market conditions.
Entities
Institutions
- Federal Open Market Committee
- arXiv
Locations
- United States