Utilize este identificador para referenciar este registo: http://hdl.handle.net/10362/2638
Título: The portuguese pharmaceutial market in the near future - A time series exploration approach
Autor: Baptista, Maria Helena Miranda Flores
Orientador: Mendes, Jorge Morais
Palavras-chave: SSA
Time series
Forecasting
Pharmaceutical market
Séries temporais
Previsão
Mercado farmacêutico
Data de Defesa: 24-Nov-2008
Relatório da Série N.º: Mestrado em Estatística e Gestão de Informação;TEGI0222
Resumo: Using a novel exploratory technique for time series analysis, Single Spectrum Analysis (SSA), it was possible to develop a comprehensive study of the Portuguese pharmaceutical market. In the introductory chapter this technique is described in detail, for the decomposition step, homogeneity structure testing and forecasting. A bibliography review was conducted on the technique. To the best of our knowledge this was the first time that SSA was applied to any pharmaceutical market, so it was not possible to compare results with other published work. A detailed explanation on the Portuguese pharmaceutical market is provided in order to allow comprehensiveness of the work. The Portuguese pharmaceutical market is divided in 15 classes, which aggregates all drugs sold in the country. The technique was applied to those 15 time series plus the “Total Market” time series. Applying SSA, time series were decomposed in the respective components, which can be described as trend, cyclical movements and seasonality. The structure of all time series was tested for homogeneity. With those steps concluded, a monthly forecast, for the years 2008 and 2009 (with the respective confidence bounds) were produced for all the 16 time series. As a complex methodology, decisions need to be taken in several steps of the study. Even if not all possible choices are presented in the work, lengthy analyses were done to reach the best possible results. In fact, choosing between possible window lengths, Singular Value Decomposition (SVD) approaches, and eigentriples to be grouped together is sometimes more an “art” than a science; experience and previous knowledge of the actual phenomena can and should help. For confidentiality reasons the raw data is not provided in this work, but both forecast values and confidence bounds are presented.
Descrição: Dissertação apresentada como requisito parcial para obtenção do grau de Mestre em Estatística e Gestão de Informação
URI: http://hdl.handle.net/10362/2638
Aparece nas colecções:NIMS - Dissertações de Mestrado em Estatística e Gestão da Informação

Ficheiros deste registo:
Ficheiro Descrição TamanhoFormato 
TEGI0222.pdf638,6 kBAdobe PDFVer/Abrir


FacebookTwitterDeliciousLinkedInDiggGoogle BookmarksMySpace
Formato BibTex MendeleyEndnote 

Todos os registos no repositório estão protegidos por leis de copyright, com todos os direitos reservados.