|
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 - #303770 |
|
Title:
|
Sequential Monte Carlo Methods for Long Memory Stochastic Volatility Models
|
|
Author(s):
|
Christian Macaro*+ and Hedibert F. Lopes
|
|
Companies:
|
Duke University and The University of Chicago
|
|
Address:
|
Department of Statistical Science, Durham, NC, 27708-0251,
|
|
Keywords:
|
Stochastic volatility ; Long Memory ; Bayesian Inference ; Sequential Monte Carlo
|
|
Abstract:
|
Stochastic volatility (SV) models can predict heteroscedasticity of financial returns in a flexible way. The estimation procedures focus on the implementation of Sequential Monte Carlo (SMC) techniques. The availability of financial high frequency data and the advances in the analysis of long memory, has shifted the attention towards the estimation of SV models together with long memory. The infinite Markovian representation implied by long range dependence, prevents the implementation of existing estimation methods. In this work, a SMC scheme, based on an augmented representation, is proposed. The infinite order expansion is truncated and a new latent component is introduced. The new characterization is suitable to state and parameter learning. An application emphasizes the benefits of this proposal when dealing with real time estimation of financial high frequency data.
|
