Does CBRT communication affect inflation expectations? Let's ask ChatGPT

This study examines whether the verbal communication of the Central Bank of the Republic of Türkiye (CBRT) affects the inflation expectations of economic agents in Türkiye. Leveraging a large language model (i.e. ChatGPT), the analysis quantifies the verbal surprise tightness embedded in the press releases on interest rates of CBRT between January 2015 and December 2025, a period of persistently high and volatile inflation, conditional on contemporaneous economic variables and policy rates. Then, the estimated measure of surprise is incorporated into an empirical framework to assess its impact on the inflation expectations of different economic agents while controlling for relevant macroeconomic and financial variables. The results yield four main findings. First, the behavior of economic agents is in line with the demand-side view, in which the communicated tighter policies are anticipated to decrease the inflation expectations. Second, the effect of central bank communication in Türkiye is shaped by the surrounding policy context. When the gap between the inflation rate and the target widens, its influence is often limited or reversed; however, it becomes more consistently disinflationary when it operates within a credible monetary tightening process. Third, the effects vary considerably across groups, with the transmission of communication emerging first among market participants, then among firms, and only later among households. Fourth, communication is not effective in isolation. Its impact is strongest when it is supported by coherent policy actions and accompanied by realized disinflation.