Deciphering Federal Reserve Communication via Text Analysis of Alternative FOMC Statements
Taeyoung Doh,
Dongho Song and
Shu-Kuei Yang
American Economic Journal: Macroeconomics, 2026, vol. 18, issue 3, 379-412
Abstract:
We propose a text-based measure of monetary policy stance that models FOMC statements as convex combinations of dovish and hawkish alternatives, providing a tractable representation of the Committee's position along the policy spectrum. Leveraging staff-drafted alternative statements, we fine-tune a pretrained language model to capture both quantitative precision and semantic tone. Stance is defined as the product of tone and novelty, and decomposed into expected and surprise components using high-frequency financial data. Surprises arise from shifts in tone relative to expectations or from statement novelty. Our framework enables counterfactuals showing how alternative communication could have moved markets.
JEL-codes: D83 D84 E31 E32 E43 E52 E58 (search for similar items in EconPapers)
Date: 2026
References: Add references at CitEc
Citations:
Downloads: (external link)
https://www.aeaweb.org/doi/10.1257/mac.20220312 (application/pdf)
https://doi.org/10.3886/E238704V1 (text/html)
https://www.aeaweb.org/articles/materials/25600 (application/pdf)
https://www.aeaweb.org/articles/materials/25601 (application/zip)
Access to full text is restricted to AEA members and institutional subscribers.
Related works:
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:aea:aejmac:v:18:y:2026:i:3:p:379-412
Ordering information: This journal article can be ordered from
https://www.aeaweb.org/journals/subscriptions
DOI: 10.1257/mac.20220312
Access Statistics for this article
American Economic Journal: Macroeconomics is currently edited by Simon Gilchrist
More articles in American Economic Journal: Macroeconomics from American Economic Association Contact information at EDIRC.
Bibliographic data for series maintained by Michael P. Albert ().