Presenter: Michele Lenza (ECB and CEPR)
Co-authors: Douglas Araujo (BIS) Nikola Bokan (ECB), Fabio Alberto Comazzi (ESM)
Title: Word2Prices: Embedding central bank communications for inflation prediction
Abstract: The embeddings of the European Central Bank's introductory statements at monetary policy press conferences significantly improve out-of-sample core inflation forecasts multiple quarters ahead, even when the embeddings are learned from relatively simple Word2Vec models. Other common textual analysis techniques, such as dictionary-based metrics or sentiment metrics do not obtain the same results. Embeddings learned from more sophisticated, pre-trained large language models improve the forecasting performance, although in these cases the exercise can only be considered to be purely out-of-sample after each models' training data cutoff date. This work contributes to the literature by documenting a novel and simple approach to leverage central bank texts for policy-relevant forecasting.