Estratégia metodológica inovadora na determinação do impacto de fatores subjetivos em modelos de previsão

• Ano de defesa: 2024
• Autor(a) principal: Senna, Viviane de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Engenharia de Produção; UFSM; Programa de Pós-Graduação em Engenharia de Produção; Centro de Tecnologia
• Palavras-chave em Português: Séries temporais; Variáveis qualitativas; Variáveis dummy; Modelos de previsão; Time series; Qualitative variables; Model; Forecast; CNPQ::ENGENHARIAS::ENGENHARIA DE PRODUCAO
• Link de acesso: http://repositorio.ufsm.br/handle/1/31625

Traditional time series models consider only quantitative observations regarding a certain aspect to be explained. However, there are situations that occur in certain scenarios that can generate interference in observations, but cannot always be quantified, as they are qualitative variables. The objective of this study was to develop a methodological strategy capable of capturing the effects of qualitative variables and optimizing the integration of traditional models. The methodology adopted to do so was the Design Science Research Methodology – DSRM, which aims, through a rigorous process of designing projects to solve problems, to evaluate project results and communicate the conclusions obtained. The quantitative variables selected to apply the strategy were indices of stock exchanges located on all continents, cryptocurrencies, and the qualitative variables were the frequencies of wildfires that occurred in the countries where the exchanges are located. ARIMA models and extensions for all series were adjusted, such as ARIMAX-GARCH. These models were improved by applying the strategy of inserting qualitative variables in the Box Jenkins methodology, as exogenous dummy types “0” or “1”. The observations were divided into quartiles and defined as dummies “1” in all frequencies positioned above the third quartile. The dummies inserted in more complex models, such as ARIMAX-GARCH, are statistically significant.