Dynamic Bayesian Models for Continuous Proportions
Helio S. Migon, Cibele Q. daSilva, Joao B. Pereira
of Statistical Methods Department, Federal University of Rio de Janeiro, Rio de Janeiro, RJ, Brazil

This paper introduces a new methodology to model time series of continuous proportions. This class of models allows the joint estimation of trend, seasonality and the effect of covariates. Also monitoring and interventions can be take into account. An extension of the class of dynamic generalized models is proposed.The inference explore conjugacy and use linear Bayes estimation as a first order approximations. The observations are assumed conditionally independent and Dirichlet distributed. A link function is introduced to relate the natural parameters of the Dirichlet family with the states describing the time series components. Artificial generated time series data and also real vectors of continuous proportions will be considered in order to illustrate the proposed methodology.

Keywords: Bayes linear estimation; Laplace approximation; Dirichlet model

Biography: Helio S. Migon, Professor of Statistics at Universidade Federal do Rio de Janeiro, Brazil. I have got the PhD degree from University of Warwick, Uk in the 80's. I am publishing papers in different statistical journals and supervising many MSc and PhD students eversince.