Discussion of "The Voice of Monetary Policy"
By Yuriy Gorodnichenko, Tho Pham, and Oleksandr Talavera
Advanced analytics: new methods and applications for macroeconomic policy
Bank of England, European Central Bank, and K ing’s College London
Jonathan Benchimol (Bank of Israel)
Novemb er 3, 2021
This presen tation do es not necessarily re‡ect the views of the Bank of Israel.
1 / 8
IThis paper proposes another communication channel for
central bank chairmen.
INot only is the content important, but also the voice tone in
monetary policy communication.
IVoice tone moves …nancial markets, even when controlling for
CB actions or press conference content.
IWhy is this an important area of research?
IAbility to assess and quantify all of the relevant information
embedded in central bank communication
=)key to the e¤ective implementation of monetary policy?
IFive comments for a very cool paper.
2 / 8
1. Units of Observation
IWhy do we not see an immediate reaction?
IIs more “high-frequency” analysis possible in this setting?
IFor example, Gorodnichenko and Weber (2016) use
high-frequency identi…cation of monetary shocks on the
volatility of stock returns.
ICurti and Kazinnik (2020) use minute-level data to analyze
Fed Chair’s facial expressions on investor sentiment.
IThis might be the more di¢ cult thing to implement.
IMehrabian’s communication rule cannot apply to the stock
market unless you assume:
Istock markets to perceive the chairman’s voice like humans.
(Kind of) behavioral market e¢ ciency?
Ipeople process this information (communication content)
conditional on the chairman’s voice.
IETFs are unbiased.
IUnclear which and how many market players listened to the
=)unclear how much the voice e¤ectively contributed.
3 / 8
IStrong assumption: The chairman has only …ve kinds of
IMissing emotions: annoyed, exasperated, and fear.
INot easy to …x, but worth a try.
IBeing worried may not lead to being sad for some chairmen
but may be perceived as sad for others.
IOthers may not be in the set of emotions or simply display a
IStrong assumption: a particular emotion is perceived in the
“correct” way by everyone.
IRAVDESS dataset is unbalanced.
4 / 8
IWhy emotions? Because...
Iof the monetary policy?
Ia colleague was disappointing right before the speech?
Ithe chairman or a family member is sick?
=)The exhibited emotion may not be related to monetary policy.
=)Control with the chairman’s voice before becoming chairman.
IAs far as the Q&A is considered, the results may depend on
the journalists, not only the chairman
IOne journalists’characteristic is to provoke the chairman.
ISome are sympathetic.
IGovernors are trained/coached to manage their behavior
(physical, voice, tone) without necessarily conveying the
Itruth of the text content
Ieconomic situation they think about.
IExperience (as a chairman or previous ones) shapes the
management of the voice and so the results.
Ie.g., Benchimol, El-Shagi and Saadon (2021).
5 / 8
4. Text Sentiment
IEach text-mining sentiment or score is di¤erent according to
the dictionary used.
IOne dictionary may control better than another
Ie.g., Benchimol, Kazinnik, and Saadon (2021).
IA more robust approach to capturing text sentiment is needed.
IDictionary-based approach (hawkishness/dovishness) might
miss a lot of nuances present in the text.
INew addition compared to the previous versions of the paper:
in conjunction with the dovish/hawkish dictionary,
transformers (e.g., BERT) are used to capture the text tone,
and whatever is remaining is coming from voice variation.
IVoice for soft information vs. voice for hard information.
ITrain the learning one various/di¤erent speeches.
6 / 8
5. Policy Implications
IDiscussions related to policy implications of the …ndings are
IWhat should the take-aways for the policymakers be?
IDo these price movements su¢ ciently signi…cant to change
the central bank/chairman’s behaviors?
IWould you suggest an alternative way to communicate?
7 / 8
IThank you for your attention.
8 / 8