Mapping the space of central bankers' ideas

Summary

Focus

We use machine learning techniques to impose mathematical structure on almost 20,000 central bank speeches from over 100 central banks spanning four decades. We plot numerical representations of speeches (ie embeddings) as clusters in two- or three-dimensional charts to understand how monetary policy ideas evolve and spread globally. In this representation, similar ideas appear close together in the vector space that embeds the speeches. The resulting output creates a visual "space of economic ideas" showing what central banks are talking about and how their communication priorities change over time.

Contribution

Central bank speeches are a critical tool for central bank communication. Yet understanding of global patterns in the large body of speeches has been limited by difficulty in finding a comprehensive way to compare and cluster the speeches that preserves their semantic significance. Our work demonstrates how modern artificial intelligence (AI) tools can convert the collection of thousands of speeches into visual maps that researchers can explore and analyse. Our maps show how language and meaning relate to policy episodes, economic conditions and specific central banks. We demonstrate practical methods for exploring central bank communications visually across different institutions, topics and time periods.

Findings

Central bank speeches cluster mainly by region and specific institution, reflecting local economic conditions and pressing new challenges as they emerge. Financial stability appears in the centre of the visual map, while issues that have arisen more recently, such as digital forms of cash and climate change, emerge at the edges and grow over time. Speech clusters correspond to major economic events, with hawkish sentiment prevailing more during the surge in inflation in 2022. Our approach shows that central bank speeches are shaped by emerging priorities.

---

Abstract

This paper explores the landscape of economic ideas as revealed in the machine learning embedding of a comprehensive dataset of central bank speeches. This dataset, maintained by the BIS, encompasses 19,742 speeches delivered by almost 1,000 officials from over 100 central banks over a period spanning three decades, from 1997 to 2025. As well as topic analysis of speeches at any moment in time, the evolution of the topics over time provides insights into how the focus of central bank thinking has been shaped by shifting policy challenges since 1997. Parsing the embedding both through topics and through time provides rich insights into how economic ideas have taken shape through communication practices of central banks worldwide. To demonstrate its utility, we have conducted a series of analyses that map the global landscape of monetary policy discourse. Furthermore, we construct a quantitative framework-referred to as the "space of central bankers' ideas"-which uncovers institutional patterns and highlights shifts in policy approaches over time.

JEL classification: E52, E58, C55, C38

Keywords: central bank communication, central bank speeches, AI, topic modeling, embeddings

The views expressed in this publication are those of the authors and not necessarily those of the BIS.