Financial Market Reactions to the Novelty of Information in FOMC Minutes
Abstract
We study how financial markets respond to the incremental information conveyed by FOMC minutes. Using paragraph-level embeddings, we score each paragraph by its semantic distance to the closest paragraph in previously released public communications. Aggregating these paragraph-level scores yields two indices: overall novelty, which captures the degree of novelty, and novelty tilt, which captures the composition of novelty, i.e., whether new content is concentrated in the staff review or the committee discussion sections. In high-frequency event study regressions around minutes releases, overall novelty is associated primarily with the magnitude, rather than the sign, of asset-price responses, whereas novelty tilt is informative about the direction of repricing. Methodologically, we develop a disclosure-based framework for quantifying novelty in FOMC minutes over time.
Keywords
Artificial intelligence; Central bank communication; High-frequency event study; FOMC minutes; Textual novelty