MATHEUS ALVES PEREIRA DOS SANTOS
Data: 26/02/2024, 11h
Sala Zoom: link do Google meet: https://meet.google.com/ckr-vuwz-skk.
Orientador: Davi Michel Valladão | PUC-Rio
Resumo:
The study of time series plays a crucial role in the decision-making process, resulting in the proposition of numerous methodologies over time for this purpose. In this context, score-driven models stand out as a flexible and interpretable approach. However, due to the considerable number of parameters, the estimation process of these models tends to be complex. In order to address this complexity, this work aims to evaluate how the use of modern optimization techniques impacts the final performance of the model. In addition to simplifying the parameter estimation process, this paradigm shift allows for the incorporation of new techniques, such as robust optimization, in the model formulation, which has the potential to enhance its performance. The SDSS.jl package, which enables the fitting and forecasting of Scoredriven models based on unobserved components using modern optimization techniques, emerges as one of the main contributions of this work. By using monthly data on the electric load of the Brazilian system and competition series, it was possible to highlight its good performance even during periods of regime change in the data, thanks to the application of robust techniques, and compare its performance with implementations already available in the literature