Análise de múltiplos pontos de mudança em modelos normal multivariados

• Ano de defesa: 2017
• Autor(a) principal: Leonardo Brandão Freitas do Nascimento
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: multivariados; matriz de variância e covariância; Análise de múltiplos pontos; Estatística; Estatistica; Verossimilhança (Estatistica); Verossimilhança (Estatística)
• Link de acesso: http://hdl.handle.net/1843/ICED-ANLRMM

In this work, it is proposed an extension of the Product Partition Model for the identification of multiple change points, over time, in the vector of mean and the variance and covariance matrix of a data sequence with Multivariate normal distribution. Conjugates prior distributions were used to estimate the vector of means and the matrix of variance and covariance over time. In addition, it is proposed to carry out a comparison of each parameter sequentially. For this purpose, Higher density intervals (HPD intervals) for the difference of parameters at successive instants of time were used. To evaluate the model, some simulated scenarios were considered and an application in financial data, more specifically an analysis of the impact of the United Kingdom's exit from the European Union.