arbiter
JournalsAccount

Decoding central bank communications with large language models

Kairan Chen et al.

Journal of International Financial Markets, Institutions and Money2026https://doi.org/10.1016/j.intfin.2026.102325article
AJG 3ABDC A
Weight
0.50

Abstract

• We develop a way to measure the “Procedural Linguistic Minutes Information Shocks”. • The shocks quantify the extra linguistic information in the Minutes beyond the Statements. • We identify the causal impact of the shocks on expectations about interest rate. This paper examines the effects of Fed announcements on market expectations about interest rates from a linguistic perspective. A simple framework using large language models is developed to measure the proposed ‘Procedural Linguistic Minutes Information Shocks’ (PLMIS). These shocks represent the additional linguistic information in the Minutes beyond the Statements for the same FOMC meetings. This paper investigates the causal impact of the constructed shocks on market expectations, using 1-minute bar data on U.S. Treasury futures and a high-frequency event study approach. The key finding is that the PLMIS conditional on dovish text has a positive causal effect on price changes within the 30-minute event window. This overall positive impact can be decomposed into two opposing effects, each driven by a distinct type of topical content: ‘views on the recent economy’ and ‘forward guidance’. The main findings remain robust when the prompts are rephrased, a narrower event window is used, and an equal-weighted index is applied.

Open via your library →

Cite this paper

https://doi.org/https://doi.org/10.1016/j.intfin.2026.102325

Or copy a formatted citation

@article{kairan2026,
  title        = {{Decoding central bank communications with large language models}},
  author       = {Kairan Chen et al.},
  journal      = {Journal of International Financial Markets, Institutions and Money},
  year         = {2026},
  doi          = {https://doi.org/https://doi.org/10.1016/j.intfin.2026.102325},
}

Paste directly into BibTeX, Zotero, or your reference manager.

Flag this paper

Decoding central bank communications with large language models

Flags are reviewed by the Arbiter methodology team within 5 business days.

Evidence weight

0.50

Balanced mode · F 0.40 / M 0.15 / V 0.05 / R 0.40

F · citation impact0.50 × 0.4 = 0.20
M · momentum0.50 × 0.15 = 0.07
V · venue signal0.50 × 0.05 = 0.03
R · text relevance †0.50 × 0.4 = 0.20

† Text relevance is estimated at 0.50 on the detail page — for your query’s actual relevance score, open this paper from a search result.