Gabriel Huerta and Raquel Prado
12/31/2004 09:00 AM
Applied Mathematics & Statistics
We present structured prior modeling for multiple time series focusing on latent component structure for a collection of autoregressive processes. Similar to the univariate case, the state-space representation of these vector processes implies that each univariate time series can be decomposed into simple underlying components. Such components may have a common structure across the series that define the vector process. Additionally, this approach allows the consideration of uncertainty on the number of latent processes across the multiple series and consequently, it handles model order uncertainty in the vector autoregressive framework. Posterior inference and implementation are developed via customized MCMC methods. Issues related to inference and exploration of the posterior distribution are discussed. Illustrative data analyses are presented.
