Do central bank speeches help predict monetary policy?

Sammanfattning: This paper studies the information contained in central bank speeches, focusing on predicting future monetary policy. Departing from previous papers which use mainly dictionary-based methods, we employ the deep transfer learning technique to process the text data. We find that central bank speeches can be used to predict policy rate change over the next three months (with an accuracy of 71.4%). In this paper, we also discuss the importance of each step in our suggested model workflow.