Shifts in ECB communication: a text mining approach.

This thesis investigates how fundamental changes to communication, made by the European Central Bank (ECB) during the press conference following monetary policy decision, affect stock market volatility. First, the ECB press conferences are dissected into topics using Latent Dirichlet Allocation (LDA), an unsupervised generative model for text. Then turning points in ECB communication are captured using the estimated topic probabilities. The proposed approach does not rely on subjective interpretation of topical content. The thesis finds that the topics surge and die out over time, revealing communication patterns that match the ECB monetary policy stance. Furthermore, the content of the ECB press conference is informative for the market, consistent with the previous literature. Market uncertainty increases if the ECB switches to a different communication regime. The main revisions to communication on the monetary analysis and the economic analysis are perceived to be of high importance, whereas the Q&A session does not convey incremental information.