|
Abstract:
|
The notion that seasonal patterns are related across multiple time series has been considered for nearly half a century. Nevertheless, until recently, most approaches to signal extraction (seasonal adjustment) have been developed in the univariate context and ignore the possibility of between series dependence. Building off of recent developments, we propose a Bayesian approach to multivariate signal extraction that allows us to calculate a model-based seasonal adjustment. To reduce the dimension of the parameter space we utilize recent advances in Bayesian variable selection. Our approach is illustrated using series from the U.S. Census Bureau.
|
