[Paper Review] Evolution of topics in central bank speech communication

Magnus Hansson|arXiv (Cornell University)|Sep 21, 2021

Computational and Text Analysis Methods4 citations

TL;DR

This paper uses Dynamic Topic Models (DTM) to analyze central bank speeches from 1997–2020, revealing that central banks discuss a broad range of topics beyond traditional monetary policy. The study finds strong, persistent topic trends driven by narrative effects, not fully explained by financial controls, suggesting central bank communication operates through story-based narratives that shape market expectations over time.

ABSTRACT

This paper studies the content of central bank speech communication from 1997 through 2020 and asks the following questions: (i) What global topics do central banks talk about? (ii) How do these topics evolve over time? I turn to natural language processing, and more specifically Dynamic Topic Models, to answer these questions. The analysis consists of an aggregate study of nine major central banks and a case study of the Federal Reserve, which allows for region specific control variables. I show that: (i) Central banks address a broad range of topics. (ii) The topics are well captured by Dynamic Topic Models. (iii) The global topics exhibit strong and significant autoregressive properties not easily explained by financial control variables.

Motivation & Objective

• To identify the global topics discussed in central bank speeches from 1997 to 2020.
• To examine how these topics evolve over time and whether they exhibit persistence.
• To assess whether topic persistence is driven by financial fundamentals or narrative effects.
• To evaluate the effectiveness of Dynamic Topic Models in capturing central bank communication content.
• To explore whether central bank communication functions as a form of narrative economics, shaping expectations through story-based discourse.

Proposed method

• The study applies Dynamic Topic Models (DTM) to a corpus of 119 to 423 central bank speeches annually from nine major central banks between 1997 and 2020.
• DTM estimates time-varying topic distributions by modeling word probabilities in speeches as dynamic processes over time.
• The model is trained on a combined corpus of all nine central banks to ensure topic coherence and generalizability.
• A case study focuses on the Federal Reserve using its speeches alone, incorporating regional financial control variables such as the VIX, S&P 500 returns, and 1-year Treasury yields.
• Autoregressive (AR(1)) coefficients are estimated for topic probabilities to measure persistence, both with and without control variables.
• Topic coherence scores and manual inspection of representative documents validate the interpretability and quality of the extracted topics.

Experimental results

Research questions

• RQ1 What global topics do central banks discuss in their public speeches between 1997 and 2020?
• RQ2 How do these topics evolve and persist over time?
• RQ3 To what extent is topic persistence explained by financial and macroeconomic control variables?
• RQ4 Are the observed topic trends consistent with narrative-based communication, as suggested by narrative economics?
• RQ5 Can Dynamic Topic Models effectively capture the evolving content of central bank communication in a meaningful and interpretable way?

Key findings

• Central banks discuss a broad range of topics beyond traditional monetary policy, including financial stability, trade, and central bank digital currency (CBDC).
• Dynamic Topic Models effectively capture the content of central bank speeches, with high topic coherence and strong alignment with representative documents.
• Topic probabilities exhibit strong and statistically significant autoregressive properties (AR(1) coefficients > 0.5 for many topics), indicating persistent communication patterns.
• Even after controlling for regional financial variables such as the VIX, S&P 500 returns, and 1-year Treasury yields, topic persistence remains significant, suggesting narrative effects beyond fundamentals.
• The topic on financial system and trade show full explanation by controls, but other topics (e.g., inflation, macroeconomic conditions) remain persistent, indicating narrative-driven communication.
• The evolution of word probabilities within topics aligns with epidemiological models of narrative spread, supporting the theory that central bank communication functions as story-based economic narratives.