The Voice of Monetary Policy

Abstract

We develop a deep learning model to detect emotions embedded in press conferences after the Federal Open Market Committee meetings and examine the influence of the detected emotions on financial markets. We find that, after controlling for the Federal Reserve’s actions and the sentiment in policy texts, a positive tone in the voices of Federal Reserve chairs leads to significant increases in share prices. Other financial variables also respond to vocal cues from the chairs. Hence, how policy messages are communicated can move the financial market. Our results provide implications for improving the effectiveness of central bank communications. (JEL D83, E31, E44, E52, E58, F31, G14)

Machine Learning for Economics Research: When What and How?

“…Similarly, article [6] published at JME, they use a large news corpus and ML algorithms to investigate the role played by the media in the expectations formation process of households, and in [7] published in JoE, uses the twitter data and ML model to measure inflation expectation. Likewise, artciel [8] published in AER, the authors use satellite data to measure GDP growth at the sub and supranational regions, in [9] also published in AER, the authors employed a computer vision algorithm that measures the perceived safety of streetscapes, and how strongly it is correlated with population density and household income, and in [10] published in AER, the authors use deep learning to detect emotions embedded in press conferences after the Federal Open Market Committee meeting and examine the influence of the detected emotions on financial markets.…”

“…The use of deep learning models for NLP is evolving rapidly, and various large language models (LLMs) could be used to process text data; However, transformer models [21] are proven to be more useful to efficiently extract useful information from the textual data [1]. For instance, in [10], they use transformer to detect emotions embedded in press conferences after the Federal Open Market Committee meeting for sentiment analysis. Moreover, almost all large general-purpose LLMs, including GPT-3 and chatGPT, are trained using Generative Pre-trained Transformer [22].…”

More than words: Fed Chairs’ communication during congressional testimonies

“…Our paper is closely related to two recent papers that analyze the effects of Fed Chairs' emotional cues on financial markets. Gorodnichenko et al (2021) (n.d.) along several new dimensions. While these papers study communications during FOMC press conferences, we study Fed Chair's congressional testimonies, which offer several advantages.…”

Can central banks be heard over the sound of gunfire?

“…To our knowledge, this is the first study to use artificial intelligence (i.e., Chat‐GPT) to analyze and classify central bank announcements. Moreover, we examine the case of a developing country during a war, whereas the existing literature has typically used dictionary‐based models (Brzeszczynski et al, 2017; Fiser & Horvath, 2010; Gardner et al, 2022) or has employed large pre‐trained language models (LLMs) (Doh et al, 2020; Gorodnichenko et al, 2023) to quantify the sentiment of central bank communications, almost always with a focus on the developed economies in times of peace.…”

“…For instance, Rosa (2011) has investigated the effect of the Federal Reserve's decisions and statements made in relation to U.S. stock market indices and found that the latter can have a greater impact. Moreover, Gorodnichenko et al (2023) and Hayo and Zahner (2023) have demonstrated that sentiment conveyed in central bank announcements, and even the voice of the speaker, can influence financial markets. However, most previous studies conducted on central bank communications have focused on Western Economies that typically operate in low-volatility environments.…”