WP – 2026-10-23 – Severin Joho And Enzo Rossi: The Evolution Of Frequency, Language Complexity And Policy Narrative: Benchmarking Python-based NLP And ChatGPT On A Century Of SNB Rhetoric

TL;DR

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

A Swiss National Bank working paper by Severin Joho and Enzo Rossi analyzes 1,063 speeches by senior SNB officials delivered from 1928 through 2024. It reports that speeches became more frequent, shorter and more accessible over time, while inflation gained prominence in policy discussion during the flexible-exchange-rate era.

The Swiss National Bank has published a working paper examining how its officials’ public language changed across 1,063 speeches delivered from 1928 to 2024. Authors Severin Joho and Enzo Rossi report that speeches became more frequent, shorter and more accessible over time, while inflation moved closer to the center of policy discussion in the era of flexible exchange rates.

The study describes a newly compiled dataset of speeches by high-ranking SNB officials. It analyzes changes in speech frequency, length, readability and themes, and compares two approaches to text analysis: conventional Python-based natural-language processing and generative AI using ChatGPT.

According to the paper, the annual number of speeches followed an upward trajectory from the mid-1950s, with a more pronounced rise after the move to flexible exchange rates. The researchers also identify a broad shift toward shorter speeches and language they characterize as increasingly accessible from the 1960s onward.

The paper reports a change in subject emphasis as well. In the flexible-exchange-rate era, inflation became more prominent in SNB rhetoric, while cyclical economic development and unemployment received less emphasis. These are findings about the content and form of speeches; the supplied summary does not establish that the changes themselves caused shifts in policy or public understanding.

How SNB Communication Changed

Central bank speeches are one way officials explain economic conditions and policy priorities. A long-run record of how often officials speak, how they write and which subjects they stress can help researchers study the evolution of institutional communication alongside changes in the monetary environment.

The findings offer a historical reference for readers following the SNB: the paper links the rise in inflation’s prominence in speeches to the flexible-exchange-rate era, while documenting less emphasis on cyclical development and unemployment. That does not by itself show how any particular speech affected markets, policy decisions or audiences. The value of the study is its account of patterns across a century, rather than a claim about the impact of individual statements.

Its comparison of text-analysis methods also matters beyond this dataset. Researchers increasingly use generative AI to interpret large collections of documents. The authors present ChatGPT and conventional Python methods as complementary tools, while drawing attention to methodological pitfalls and questions that still need investigation.

A Century of SNB Speeches

The dataset covers speeches delivered from 1928 through 2024, giving the authors a record spanning changes in Switzerland’s monetary-policy setting and in public communication. The paper identifies the mid-1950s as the start of a sustained increase in annual speech frequency and says the increase accelerated after the transition to flexible exchange rates.

The researchers examine not only topic frequency but also speech length and readability. Their approach pairs deterministic text-processing methods written in Python with a generative language model, ChatGPT. The paper’s stated aim includes drawing methodological lessons from that comparison, rather than relying on an AI system alone to characterize a historical archive.

“The annual number of speeches has followed an upward trajectory since the mid-1950s, accelerating notably after the transition to flexible exchange rates.”

— Severin Joho and Enzo Rossi, authors of the SNB working paper

Limits of the Speech Analysis

The available summary does not give numerical estimates for the size of the changes in speech frequency, length or readability, nor does it specify how those measures were calculated. It also does not describe the detailed prompts, validation procedures or error rates used in the ChatGPT comparison. Readers therefore cannot assess from the summary alone how strongly individual results depend on analytical choices.

The reported relationships do not establish why officials changed their communication or whether shorter, more accessible speeches improved public understanding. The paper flags algorithmic pitfalls and open research questions, but the supplied material does not list those questions in detail. Its dataset ends in 2024, so it cannot show whether the trends continued in 2025 or 2026.

Questions for Future Text Research

The working paper points to further research on the use of large language models to analyze central-bank communication. Its authors call attention to the need to address algorithmic pitfalls and identify open methodological questions, while treating AI-based analysis as a complement to established text-processing frameworks.

No follow-up study, new dataset or timetable is specified in the supplied report. The immediate next step for readers assessing the findings is to consult the full paper for its methods, definitions and results, including how the two analytical approaches were benchmarked. Future work could also extend the archive beyond 2024, but the source does not confirm that such an update is planned.

Key Questions

What did the SNB working paper examine?

It analyzed 1,063 speeches by high-ranking Swiss National Bank officials from 1928 to 2024, covering frequency, length, readability and thematic content.

What changes did the authors report?

The authors say speeches became more frequent, shorter and more accessible over time. They also report that inflation gained prominence in the flexible-exchange-rate era, while cyclical economic development and unemployment received less emphasis.

How did the researchers use ChatGPT?

They benchmarked ChatGPT against traditional Python-based text analysis as part of their examination of the speeches. The paper’s stated methodological lesson is that the approaches can complement each other, while AI analysis has pitfalls and unresolved questions.

Does the study show that SNB speeches changed policy?

No such causal conclusion is stated in the supplied summary. The paper reports patterns in speech frequency, readability and topics; it does not establish that those changes caused policy decisions or changed public understanding.

What remains unknown from the report summary?

The summary does not provide detailed measurements, the full validation procedures for ChatGPT, or the size of each reported change. It also does not say whether the speech trends continued after the dataset’s final year, 2024.

Source: primary