Sentiment and Policy Transmission: Deep Learning Analysis of European Central Bank Communication

Abstract

This study examines European Central Bank (ECB) communication sentiment using transformer-based natural language processing on 791 speeches. Applying Vector Autoregression modeling and Explainable AI techniques, we analyzed how sentiment affects key economic indicators across four ECB presidential regimes. Results showed sentiment functions as a monetary policy transmission mechanism with heterogeneous effects across different variables: Consumer Confidence and Harmonized Index of Consumer Prices responded immediately to sentiment shocks, while Industrial Production and Unemployment Rate showed delayed but persistent effects. Negative communication during downturns correlates with deviation from rational inflation expectations. These findings demonstrate ECB communication holds significant influence over markets, expanding the central bank's policy toolkit beyond traditional monetary instruments.

Keywords: Sentiment Analysis, Forward Guidance, Natural Language Processing, Deep Learning, Explainable Ai.