|
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 - #303362 |
|
Title:
|
Parameter Estimation for General State-Space Models: A Review
|
|
Author(s):
|
Jonathan R. Stroud*+
|
|
Companies:
|
The George Washington University
|
|
Address:
|
2140 Pennsylvania Ave., NW, Washington, DC, 20052,
|
|
Keywords:
|
particle filter ; Bayesian ; maximum likelihood ; stochastic volatility ; jumps ; spatio-temporal
|
|
Abstract:
|
The particle filter is a powerful method for sequential state estimation in general state-space models. However, parameter estimation within the particle filter remains an unsolved problem. This talk compares the existing approaches for sequential parameter estimation in state-space models. We consider maximum likelihood and Bayesian approaches, and compare the generality and computational efficiency of these algorithms. Using a stochastic volatility jump-diffusion model and a high-dimensional spatio-temporal model, we compare the strengths and weaknesses of each approach. We conclude with some new ideas for parameter learning algorithms that scale to higher-dimensional systems.
|
