Data notice and terms of use
This site presents an analysis of central bank communication. Please read the following before citing, reusing, or drawing conclusions from anything shown here.
How to cite
Every figure, table, and data file taken from this site must cite the working paper below. This applies to reuse in papers, presentations, briefings, and press material, whether the material is reproduced as shown or redrawn from the downloaded data.
Silva, T. C., Moriya, K., & Veyrune, R. (2025). From Text to Quantified Insights: A Large-Scale LLM Analysis of Central Bank Communication. IMF Working Paper No. 2025/109. International Monetary Fund, Washington, DC.
@techreport{silva2025text,
author = {Silva, Thiago Christiano and Moriya, Kei and Veyrune, Romain},
title = {From Text to Quantified Insights: A Large-Scale LLM Analysis
of Central Bank Communication},
institution = {International Monetary Fund},
type = {IMF Working Paper},
number = {2025/109},
year = {2025},
address = {Washington, DC},
url = {https://www.imf.org/en/Publications/WP/Issues/2025/05/30/From-Text-to-Quantified-Insights-A-Large-Scale-LLM-Analysis-of-Central-Bank-Communication-566953}
}Where the figure depends on the data behind it, quote the build details below alongside the citation. View the paper
Where the documents come from
Every document analyzed here was collected from the public website of the central bank that published it. Nothing is confidential, and nothing was obtained under an agreement with a central bank.
The documents remain the property of the central banks that wrote them. This site shows measurements taken over them, together with short extracts for context; it is not a document archive and does not republish the originals.
The collection is assembled automatically and at scale, and that process is not exact. Read what is shown as indicative of the record as collected, and check anything material against the central bank's own publication before relying on it.
The labels are produced by a model
The topic, tone, audience and time-orientation attached to each sentence are predictions from a fine-tuned large language model, not judgments made by an economist. Any single sentence, document or small group of documents may be mislabeled.
The language of each document is also detected automatically and is likewise imperfect, particularly for documents that mix languages or that were converted from scanned images.
Document text is read from files of varying quality, including scans. Reading them introduces errors of its own, and those errors carry into every measure built on the text.
Coverage is uneven, and absence is not evidence
The collection is not exhaustive. Coverage differs across central banks, document types and years, for reasons that include what each central bank publishes, what it keeps online, and what could be retrieved and read.
A gap in a chart therefore means the collection holds nothing for that period — not that the central bank published nothing. Conclusions of the form "this central bank stopped writing about X" cannot be drawn from this site alone.
Figures may change
Figures are computed from a fixed build of the data, stamped below. They can change without notice for two independent reasons.
Methods change: classification, language detection, text extraction and the definitions of individual measures are improved over time, and an improvement can move a published number.
The underlying collection changes: documents are added, re-retrieved, re-extracted or corrected, which restates the figures computed over them.
A figure taken from this site today may therefore not reproduce later. If you cite one, cite it together with the build date shown below.
Third-party data
Some views place communication measures beside market and macroeconomic series supplied by third parties under license. Those series are shown for comparison only. Where a license does not permit redistribution, the series is displayed but its values are not available for download, and the view says so.
Names, borders and territorial status
The designations used and the presentation of material on maps and in tables do not imply the expression of any opinion whatsoever on the part of the IMF concerning the legal status of any central bank, territory, city or area, or of its authorities, or concerning the delimitation of its frontiers or boundaries.
Entities are grouped for analytical convenience. Groupings follow standard IMF reference classifications and are not an official designation of any country's status.
Status of this site
The views expressed here are those of the authors and do not necessarily represent the views of the IMF, its Executive Board, or IMF management. Nothing here is an official IMF position, policy advice, or investment advice.
The material is provided as it is, without warranty of any kind, and is intended for research and analytical use.