Previsão da série temporal da inflação usando modelos orientados por score dinâmico e métodos de aprimoramento de acurácia preditiva: o caso do Brasil.

• Autor(a) principal: Castro, Carlos Henrique Dias Cordeiro de
• Data de Publicação: 2023
• Outros Autores: [e-mail]
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UERJ
• Texto Completo: http://www.bdtd.uerj.br/handle/1/20904
• topic: Inflation forecast; Generalized Autoregressive Score; FOCUS Bulletin; Inflação – Brasil; Previsão econômica – Brasil; Modelos econométricos; Economia – Brasil; Previsão de inflação; Boletim FOCUS; CIENCIAS SOCIAIS APLICADAS::ECONOMIA

This thesis analyzes monthly forecasts of Brazilian inflation measured by the IPCA. These forecasts are obtained using models based on the GAS (Generalized Autoregressive Score) approach, which was developed by Creal et al. (2008) and Harvey and Chakravarty (2008). We compare these GAS-based forecasts with those published in the Boletim Focus, along with other competing models commonly used in the literature. GAS models fall into the category of observation-driven models, where the parameters vary over time based on a mechanism involving past parameter values and past observations via the score function. Score-based models are an attractive method for forecasting and have been successfully applied in various empirical studies related to economic and financial variables. The initial comparative results regarding point predictions, stability, and predictive density indicate that GAS models, particularly those using Student's t-distribution, outperform competing models. However, it's worth noting that GAS models tend to have slightly higher mean prediction errors than the Focus Bulletin projections, a finding supported by statistical significance tests. To address this, we employ several empirical strategies to enhance forecast accuracy. Three strategies were adopted: (i) incorporating high frequency information to forecasts estimated by individual models via the MIDAS (Mixed Data Sampling) regression model by Ghysels et al. (2004); (ii) the incorporation of inflationary expectations via the Black and Litterman model (Black and Litterman, 1990); (iii) Finally, based on a confidence set, a wide range of combination methods were applied to the forecast components. The results showed that all these three steps were important to gain forecasting performance. Some results found by de Castro and Aiube (2022) ratify the results of the present work. First, the GAS models can be competitive in relation to the Boletim Focus projections and other traditional models in the literature for forecasting Brazilian inflation. Second, the GAS models with Student's t-distribution and heteroskedastic produced forecast estimates with better performance in terms of point forecast, stability and predictive density. Third, the combination of models was able to gene.