Banca de QUALIFICAÇÃO: Alexandre Henrique Lucchetti

Uma banca de QUALIFICAÇÃO de DOUTORADO foi cadastrada pelo programa.
STUDENT : Alexandre Henrique Lucchetti
DATE: 07/02/2024
TIME: 09:30
LOCAL: Vídeo Conferência
TITLE:

Natural Language Processing in Economics and Finance: Literature Review and Applications to Monetary Policy Communication

KEY WORDS:

Natural Language Processing, Monetary Policy Communication, Monetary Policy Coordination, Yield Curve, Sentiment Analysis

PAGES: 103
BIG AREA: Ciências Sociais Aplicadas
AREA: Economia
SUMMARY:

This work consists of three papers on Natural Language Processing (NLP) applied to economics and finance. The first paper surveys the main methods, requisites and applications of NLP in economics and finance, guided by the following research questions: what is NLP currently capable of and how it can contribute to economic research? What are the methods and techniques to achieve such contributions? What kind of data is needed and suitable for these methods' implementation? And what can we expect to achieve with NLP in economics and finance? These four questions outline the structure of the paper, designed to help shorten the path economic researchers have to trail to get introduced to this field, with topics that cover: (i) useful NLP tasks to economics; (ii) NLP models; (iii) economic and financial textual data for NLP; and (iv) NLP applications in economics and finance. We further contribute by resorting to bibliometric tools to help us visualize the literature map of this field, also providing valuable insights to our survey and the remainder of our work. We finally indicate that there is much room to apply natural language processing to economic issues, but alert that, more than ever, researchers must be careful not to stray away from questions motivated by hypotheses closely tied to economic theories. With the insights from the first paper, the second and third papers propose novel applications of the discussed techniques to monetary policy communication. The second paper provides a forward-looking measure of how central banks implicitly coordinate their actions, accounting for similarities on what policymakers weigh on the economic outlook, the instruments they rely on and the forward-guidance they communicate, all extracted from public manifestations in the form of speeches' transcripts. In order to do that, we resort to the central bankers' speeches database made available by the Bank for International Settlements and build a network of similarities that connects central banks whose policymakers' present speech similarity, adapting for this context the method proposed by Cajueiro et al. (2021). Our results show that our network successfully captures the long-term global importance of central banks overseeing the G10 currencies. Additionally, we show that the financial crisis of 2007-2008 halted a long-term trend of coordination reduction, reverting to coordination increase. The same pattern was also detected years later, during the Covid-19 Pandemic. These latter results suggest that central banks tend to increase coordination in times of turbulence, in line with the evidence from the literature (Blanchard; Ostry; Ghosh, 2013). Finally, the third paper proposes a framework for estimating expectation-embedded multidimensional sentiment from monetary policy communication, combining economic fundamentals and state-of-the-art deep learning neural networks. In order to do that, we first resort to the Litterman e Scheinkman (1991) yield curve decomposition model --- with its level, slope and curvature factors accounting for the three dimensions of our approach --- to estimate market participants' expectation shifts as priced along the entire term structure of interest rates in response to monetary policy communication. From there, we propose a two-stage estimation framework to make sure we incorporate all the information regarding monetary policy communication known by economic agents. We do that by first estimating the communication state as known by agents from the Brazilian monetary authority statement, released on the same day as its rate decision. The second stage combines the communication estimated from the first stage with new communication from monetary meeting minutes, released six days after the meeting, to predict how market participants will price the new information in the yield curve's factors. This framework shows excellent out-of-sample performance not only in its baseline setup, which predicts the direction of each factor's movement, but also in a more sophisticated setup, in which we also allow the model to infer the magnitude of such movements.

COMMITTEE MEMBERS:
Presidente - 1642911 - DANIEL OLIVEIRA CAJUEIRO
Externo à Instituição - JOSE ANGELO COSTA DO AMOR DIVINO - UCB
Interno - 1229140 - ROBERTO DE GOES ELLERY JUNIOR