Strategic Control of Facial Expressions by the Fed Chair

View on arXiv ← Back to list

Authors: Hunter Ng

Published: 2024-10-26 16:16:11+00:00

Comment: 20 pages main text, 30 pages of tables and figures and appendix

AI Summary

This paper investigates whether Federal Reserve Chairs strategically control facial expressions during FOMC press conferences and how these nonverbal cues impact financial markets. It finds that facial expressions are a distinct public signal, differentially interpreted across Chairs, and that Fed Chairs do not strategically control them despite influencing market reactions. The study also suggests investors process these cues using a dual-processing, finite-state Markov memory model.

Abstract

This article investigates whether the Federal Reserve Chair strategically controls facial expressions during FOMC press conferences and how these nonverbal cues affect financial markets. I use facial recognition technology on videos of press conferences from April 2011 to December 2020 to quantify changes in the Chair's nonverbal signals. Results show that facial expressions serve as a separate public signal, distinct from verbal content. Using deepfakes, I find that the same facial expressions expressed by different Fed Chairs are interpreted differentially. As their tenure increases, negative expressions become more frequent, eliciting adverse market reactions. Furthermore, the markets interpretation of these expressions evolves over time, suggesting that investors process facial cues with dual-processing finite-state Markov memory. In line with the Fed's goals of transparency and non-volatility, I find that Fed Chairs do not strategically control their expressions.

Key findings
Facial expressions serve as a complex, distinct public signal from verbal content, with the same expressions interpreted differently across Fed Chairs. Fed Chairs do not strategically control their facial expressions, as negative expressions increase with tenure and elicit adverse market reactions, contradicting goals of transparency and non-volatility. Investors process these nonverbal cues using a dual-processing, finite-state Markov model, where reactions diminish over time but are recalibrated by recent events like congressional testimonies.
Approach
The study quantifies Fed Chair facial expressions using facial recognition technology (DeepFace with VGG-Face model) on FOMC press conference videos and analyzes their impact on minute-level financial market data. It employs deepfakes (generated with DeepFaceLabs) to demonstrate differential interpretation of expressions across Chairs and uses NLP tools (OpenAI Whisper, FinBERT, spaCy) for verbal content control. Relationships are analyzed using fixed-effects OLS regressions.
Datasets
FOMC press conference videos (from Federal Reserve's YouTube channel, April 2011-December 2020), minute-level financial market data (SPDR S&P 500 (SPY), CBOE Volatility Index (VIX), Euro-to-USD exchange rate (EUR), Japanese Yen-to-USD exchange rate (JPY)). FinBERT was pre-trained on the Financial phrase-Bank dataset, and VGG16 on ImageNet.
Model(s)
UNKNOWN
Author countries
USA
← Previous
Next →