A Time Series Analysis of Emotional Loading in Central Bank Statements

Proceedings of the Second Workshop on Economics and Natural Language Processing, pages 16–21 Hong Kong, November 4. c⃝2019 Association for Computational Linguistics 16 A Time Series Analysis of Emotional Loading in Central Bank Statements Sven Buechel♣ Simon Junker♦ Thore Schlaak♦ Claus Michelsen♦ Udo Hahn♣ ♣ Jena University Language & Information Engineering (JULIE) Lab Friedrich-Schiller-Universität Jena Fürstengraben 27, D-07743 Jena, Germany https://julielab.de/ ♦ German Institute for Economic Research (DIW) Mohrenstraße 58, D-10117 Berlin, Germany https://www.diw.de Abstract We examine the affective content of central bank press statements using emotion analysis. Our focus is on two major international play- ers, the European Central Bank (ECB) and the US Federal Reserve Bank (Fed), covering a time span from 1998 through 2019. We reveal characteristic patterns in the emotional dimen- sions of valence, arousal, and dominance and find—despite the commonly established atti- tude that emotional wording in central bank communication should be avoided—a correla- tion between the state of the economy and par- ticularly the dominance dimension in the press releases under scrutiny and, overall, an impact of the president in office. 1 Introduction Central Bank (henceforth, CB) communication has become increasingly important in the past 20 years for the world economy (Blinder et al., 2008). Until the mid-1990s, there was consensus that cen- tral bankers should remain more or less silent and, if urged to make official statements, should try to hide their personal believes and assessments. This code of conduct changed fundamentally in re- cent years. Especially in times of unconventional monetary policy, central bankers are now trying to communicate proactively to economic agents, to give forward guidance and, thereby, try to increase the effectiveness of monetary policy (Lucca and Trebbi, 2009). This has led to a fast growing eco- nomic literature about the content, type and timing of CB communications and the observable effects on the economy (e.g. Ehrmann and Fratzscher, 2007b,a). CB communication and the reactions it causes are, in essence, verbally encoded—both in terms of official statements being issued as well as their assessment by other economic players and infor- mation gate-keepers (e.g., journalists, lobbyists). Hence, more recent empirical work tries to in- corporate NLP methods into economic analyses, e.g., using topic modeling (e.g. Kawamura et al., 2019) or information theory-based scores (Lucca and Trebbi, 2009). However, these analyses are based on the assumption that statements by CBs are free from emotions and contain factual infor- mation only. But it is quite unlikely that even experienced communicators can fully hide their emotions in such a way that they cannot be traced by analytic means. Hence, NLP methods might help reveal latent emotional loadings in CB communiqués. Yet, if emotions can be identified, what is their added value for the interpretation of CB communi- cation? In this paper, we intend to gather prelim- inary evidence that once emotional traces can be unlocked from CB communication, this additional information might help to better understand purely quantitative time series data signalling economic development congruent with emotional moves in CB press releases. Regarding NLP, …

Content truncated to 500 words