Analise Bayesiana em series temporais com erros exponenciais

The main purpose of this thesis is the application of Bayesian methodology to the analysis of time series models with exponential noise. Since the exact Bayesian analysis proved to be very complex, even for the simplest models, it was decided to search for methods that could provide good approximations to the exact solution, so that Bayesian inference and prediction for the models could be performed in a more systematic way. The models considered are very special cases of non linear time series models. The first two chapter of the thesis are dedicated to this review, while the other chapters are concerned with the main theme of the thesis. Chapter 1 deals with a systematic review of the current development of the theory underlying the non linear time series models. Here processes are considered to have a general white noise, although sometimes, for convenience, references are made to particular situations where the white noise is Gaussian. Chapter 2 considers time series models with an autoregressive structure, whose errors are either mixtures of exponentials or have exponential distribution..Available from Fundacao para a Ciencia e a Tecnologia, Servico de Informacao e Documentacao, Av. D. Carlos I, 126, 1249-074 Lisboa, Portugal / FCT - Fundação para o Ciência e a TecnologiaSIGLEPTPortuga