Swiss central bank speeches grow shorter as inflation takes priority
Severin Joho and Enzo Rossi find shorter, more readable speeches and a growing emphasis on inflation in Swiss National Bank (SNB) rhetoric. Their October 2026 working paper analyzes 1,063 German-language speeches from 1928 to 2024.
More speeches, fewer words
Speech frequency rose from the mid-1950s and accelerated after Switzerland adopted floating exchange rates in 1973.
Annual addresses peaked at 27 in 1987, 2007 and 2009 before declining to 12 in 2024.
Average speech length fell from roughly 3,500 words in earlier decades to 2,500 in the 1980s and below 2,000 after the mid-1990s.
German-adapted indices show improved readability, particularly during the 1980s, but recent speeches still require upper-secondary-level literacy.
Inflation displaced broader economic development as the dominant theme, reaching 80 percent of filtered thematic references after the pandemic.
Explicit references to unemployment and the output gap were virtually absent.
The authors link these patterns to the evolution of the monetary policy framework.
ChatGPT counts do not converge
The authors benchmark Python-based natural language processing against ChatGPT using identical cleaned texts.
Python records a mean speech length of 2,681 words against ChatGPT's 2,342; their word-count correlation is 0.966. Most readability measures closely agree, but mean Vienna Nonfiction Formula scores differ at 11.54 versus 10.11, with a correlation of 0.728. ChatGPT sometimes misapplied formulas or produced batch outliers.
The authors recommend combining deterministic calculations with LLM-assisted cleaning and thematic interpretation.
Readability is not comprehension
The paper distinguishes simpler syntax from demonstrated public understanding.
Its historical patterns are descriptive, with causal mechanisms left for further research.
The authors also require independent validation of LLM-generated numerical outputs rather than treating them as reproducible calculations.