This study is dedicated to the adaptation of the DistilBERT architecture to better interpret
the sentiment embedded within monetary policy speeches. While the DistilBERT model has
demonstrated its efficacy across various domains, the unique lexicon and subtleties present in
central banking communications necessitate specialized tuning. This study involved the meticulous
annotation of speech, report, and press release data from the European Central Bank (ECB) and
subsequent model fine-tuning. Results underscore the model’s adeptness at discerning sentiment
in this specialized context, offering a valuable tool for researchers and analysts examining the
intersection of sentiment and monetary policy.
