Publications

Existing research on monetary policy communication has mainly focused on the textual content of press-conference transcripts, while studies on voice sentiments remain limited. We advance the relevant literature by leveraging advanced audio analysis techniques, a large language model and high-frequency financial market data at the one-second frequency, to provide new insights into the changing communication features during ECB press conferences. We evaluate ECB communication across different topics and sample periods, and assess their impacts on euro area financial markets between 2012 and 2024. Voice analysis reveals that our model distinguishes valence, arousal and dominance features with relative accuracy. The regression results indicate that communication regarding the broader economic outlook and inflation triggers stronger and more widespread market responses than the topic of monetary policy measures. At the same time, the text sentiments and arousal-valence feature of voice generate a much higher number of significant effects compared to the voice dominance characteristic. These findings illustrate the importance of audio analysis in explaining monetary policy communication messages and in internalizing the nuanced central bank communication with its heterogeneous impact on financial markets.

Keywords: ECB, euro area, audio analysis, monetary policy, communication, sentiment analysis, financial markets.

JEL codes: C80, E43, E44, E58, G14.