|
Activity Number:
|
414
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Wednesday, August 5, 2009 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Bayesian Statistical Science
|
| Abstract - #303483 |
|
Title:
|
Particle Learning in Autoregressive Models with Structured Priors
|
|
Author(s):
|
Raquel Prado*+ and Hedibert F. Lopes
|
|
Companies:
|
University of California, Santa Cruz and The University of Chicago
|
|
Address:
|
Baskin School of Engineering MS:SOE2, UCSC, Santa Cruz, CA, 95064,
|
|
Keywords:
|
autoregressions ; state-space models ; structured priors ; particle learning ; sequential Monte Carlo
|
|
Abstract:
|
We consider sequential parameter learning and filtering in state-space models where the states or the observations follow an autoregressive process with structured priors. Following an approach similar to that presented in Huerta and West (1999), priors are built on the reciprocal roots of the AR characteristic polynomial. Models with structured priors on the reciprocal roots appropriately incorporate scientifically meaningful information in applied scenarios. However, standard methods for posterior inference based on MCMC are not computationally efficient in settings where online estimation and filtering are required. We propose an algorithm based on the particle learning approach of Carvalho, Johannes, Lopes and Polson (2008). We compare the performance of such algorithm to alternative particle methods in simulated and real data sets.
|
