Ensaios sobre seleção de variáveis e impulso-resposta em estruturas a termo de spreads de crédito soberanos

• Ano de defesa: 2022
• Autor(a) principal: Candido, Guilherme Amaral
• Orientador(a): Fernandes, Marcelo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Modelagem de estrutura a termo; Spreads de crédito; Decomposição de Nelson Siegel; Seleção de variáveis; Choques estruturais
• Palavras-chave em Inglês: Term structure modeling; Credit spreads; Nelson Siegel decomposition; Selection of variables; Structural shocks
• Link de acesso: https://hdl.handle.net/10438/32906

This thesis seeks to identify the determinants of variations in the term structures of Latin American offshore sovereign debt spreads, differentiating between global and idiosyncratic movements. In particular, I seek to answer what are the determinants of these term structures and what are the expected effects of structural shocks. In the first chapter, I analyze different models, with Nelson Siegel latent factors driven by macro-finance state variables, to describe the offshore term structure of the Brazilian sovereign credit spreads, from 2004 to 2020. The models rely on combinations of different indicators, which I select using three methods: the sorting by the explanatory capacity of each one in individual regressions; Lasso selection; and genetic algorithm. Models that combine external risk appetite variables with idiosyncratic variables in Brazil offer the best explanatory power. In the second chapter, I extend the discussion to identify the effects of structural shocks on macro variables in the offshore term structures of other Latin American sovereign issuers, including: Chile, Peru, Colombia and Mexico. This time I extract credit spreads from Credit Default Swaps (CDS) prices, in order to ensure data uniformity and compatibility. For comparison sake, I revisit the Brazilian term structure as well. Among the term structure drivers, I consider three common external and three individual local indicators, and identify shocks using structural VAR models and Local Projections. In addition, I compare the results with narrative shocks. I find that Latin American term structures are strongly driven by external factors, responding in similar ways to structural shocks, irrespectively of the identification method that I use.