“Monetary policy is 98 percent talk”*

The Fed press conferences: analyzing the tone

In this paper, Gorodnichenko et al. develop a deep learning model to detect emotions embedded in press conferences after the Federal Open Market Committee (FOMC) meetings in order to examine the influence of these emotions on financial markets. In particular, they look at FOMC press conferences from April 2011–June 2019 and split the audio segments corresponding to the response of the speaker to each question raised during the event. The split audios are then run through a machine learning model, which is trained to recognize emotions from voice variations. Answers are then rated as: positive (happy or pleasantly surprised emotions), negative (sad or angry emotions), and neutral. Finally, they look at how the tone of any given press conference affects a series of financial variables. They find:

[M]aking the voice tone more positive by one standard deviation could raise S&P 500 returns by approximately 75 basis points. This order of magnitude is comparable to what one can observe after a one standard deviation shock to forward guidance. In other words, the voice component can generate economically significant effects on the stock market. We also find that inflation expectations and exchange rates respond to variations in voice tone, e.g., a more positive tone leads to a decrease in expected inflation. At the same time, the evidence for the bond market is more mixed in our sample. These results suggest that policy communication is more nuanced than reading and posting prepared remarks and speeches. It appears that a certain level of acting skill may be helpful for ensuring that the public receives the policy message fully and accurately.

An ETF that tracks the S&P 500 index (SPY), Federal funds rate (FFR), forward guidance (FG), and asset purchase (AP) shocks.

The Voice of Monetary Policy
Authors: Yuriy Gorodnichenko, Tho Pham, Oleksandr Talavera
From: University of California-Berkeley, University of Reading, University of Birmingham

The ECB press conferences: analyzing the terms used

In this paper, Gnan et al. examine transcripts of the ECB President’s press conference statements to extract the ECB’s stance towards five topics: ‘rate guidance’, ‘economic activity’, ‘inflation’, ‘financial & monetary conditions’, and ‘fiscal policy’. They use machine learning to measure the central bank tone and whether the ECB indicates policy easing or tightening. They use the recently established Euro Area Monetary Policy event-study Database which allows for a consistent
construction of different types of monetary policy shocks from asset price data. They find:

• Shocks to the short end of the yield curve are associated with changes in the ECB’s ‘rate guidance’-stance, the medium term with ‘economic activity’, and the long end with ‘financial & monetary conditions’.
• The euro appreciates (relative to USD, GBP, and JPY) when the ECB’s topic stance towards ‘financial & monetary conditions’ becomes unexpectedly positive;
• The differential response of sovereign bonds in core and periphery countries to ECB announcements is related to the ECB’s stance towards ‘fiscal policy’.

Deciphering Monetary Policy Shocks
Authors: Phillipp Gnan, Maximilian Schleritzko, Maik Schmeling, Christian Wagner
From: Vienna University of Economics and Business, Goethe University Frankfurt