Ensaios em macroeconomia aplicada: choques petrolíferos e previsão

• Ano de defesa: 2022
• Autor(a) principal: Medeiros, Rennan Kertlly de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Economia; Programa de Pós-Graduação em Economia; UFPB
• Palavras-chave em Português: Macroeconomia; Choques do petróleo; Sentimento textual; Projeção local; Previsão fiscal; Arrecadação; Modelos de machine learning; Elastic Net; Textual sentiment; Oil shocks; Local projection; CNPQ::CIENCIAS SOCIAIS APLICADAS::ECONOMIA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/27532

Chapter 1 - Effects of oil market sentiment on macroeconomic variables This essay aims to evaluate the effects of oil price shocks on macroeconomic variables, for the economies of the United States and Brazil. We develop a variable that measures the volatility of oil prices, from a textual sentiment analysis. We evaluate oil price shocks using the Local Projection method. Our results suggest that changes in oil prices cause larger impacts on the US economy, compared to the effects on the Brazilian economy. The responses of the US and Brazilian variables were similar when using the sentiment indicator or the VIX volatility index. Finally, we find that decreasing the frequency of the variables, together with changing the method, does not change the response trajectories of the macroeconomic variables. Chapter 2 - Fiscal forecasting strategies: an empirical study for the Brazilian economy This article aims to forecasting Brazil’s federal tax revenues from different machine learning methods, for different samples. To forecasting the variable of interest, 34 explanatory variables were used. The estimation methodology used is divided into three learning categories: shrinkage, ensemble and factor. The results suggest that the Elastic Net model has the highest accuracy for monthly forecasts using 20% and 30% of the test sample, for short periods. For forecasts of accumulated periods, the LASSO model has high performance. Finally, we verify that the Bagging model is limited with an increase in forecast periods and, above all, with a decrease in the frequency of variables. Keywords: Textual sentiment; Oil Shocks; Local Projection.