Prevendo o Consumo Agregado das Famílias a Partir de Modelos VEC, VAR e Regressões MIDAS

• Ano de defesa: 2023
• Autor(a) principal: Arruda, Daniel Ferreira
• Orientador(a): Ribeiro, Marcel Bertini
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Consumo agregado; Previsão; Correção de erros; Nowcasting; MIDAS; U-MIDAS
• Palavras-chave em Inglês: Aggregate consumption; Forecasting; Error correction
• Link de acesso: https://hdl.handle.net/10438/33311

This paper aims to evaluate the predictability of the Brazilian household aggregate consumption data in different time horizons. For the short and medium term, VEC and VAR models are estimated to capture the dynamics between aggregate household consumption and its main determinants. For the very short term (nowcasting), Mixed Data Sampling (MIDAS) regressions are estimated using amplified retail sales, consumer confidence and the IBC-Br, and varying the number of months known to the regressors within the forecasted quarter in each specification. Results indicate that the error correction models bring a slight improvement in the predictive performance in relation to the random walk, especially when including household credit growth or debt service. In addition, the MIDAS regressions show a good capability to predict consumption with regressors known within the quarter, again outperforming the random walk, and it is possible to state that the restricted MIDAS performs better than the unrestricted version (U-MIDAS).