TY - JOUR

T1 - Stochastic Model Specification Search for Gaussian and Partial Non-Gaussian State Space Models

AU - Frühwirth-Schnatter, Sylvia

AU - Wagner, Helga

PY - 2010

Y1 - 2010

N2 - State space models are a widely used tool in time series analysis to deal with processes which gradually change over time. Model specification however is a diffcult task as one has to decide which components to include in the model and to specify whether these are fixed or stochastic. In the Bayesian approach, model selection relies on the posterior probabilities of a model given the data. These can be determined for each model separately by using Bayes' rule, which requires estimation of the marginal likelihoods by some numerical methods. The modern approach to Bayesian model selection is to apply model space MCMC methods by sampling jointly model indicators and parameters, as e.g. in the stochastic variable selection approach (George and McCulloch, 1997), which is usually applied to model selection for regression models. In this talk we show that a stochastic model search MCMC method is feasible for Gaussian as well as non-Gaussian time series data (binary data, multinomial data, count data) that chooses appropriate components in a structural time series model and decides, if these components are deterministic or stochastic. For non-Gaussian state space models the stochastic model search MCMC methods makes use of auxiliary mixture sampling developed in Frühwirth-Schnatter and Wagner (2006) for count data and in Frühwirth- Schnatter and Frühwirth (2007) for binary and multinomial data.

AB - State space models are a widely used tool in time series analysis to deal with processes which gradually change over time. Model specification however is a diffcult task as one has to decide which components to include in the model and to specify whether these are fixed or stochastic. In the Bayesian approach, model selection relies on the posterior probabilities of a model given the data. These can be determined for each model separately by using Bayes' rule, which requires estimation of the marginal likelihoods by some numerical methods. The modern approach to Bayesian model selection is to apply model space MCMC methods by sampling jointly model indicators and parameters, as e.g. in the stochastic variable selection approach (George and McCulloch, 1997), which is usually applied to model selection for regression models. In this talk we show that a stochastic model search MCMC method is feasible for Gaussian as well as non-Gaussian time series data (binary data, multinomial data, count data) that chooses appropriate components in a structural time series model and decides, if these components are deterministic or stochastic. For non-Gaussian state space models the stochastic model search MCMC methods makes use of auxiliary mixture sampling developed in Frühwirth-Schnatter and Wagner (2006) for count data and in Frühwirth- Schnatter and Frühwirth (2007) for binary and multinomial data.

UR - http://www.osg.or.at/download/news/293/ARSStatistica07_Wagner_Abstract.pdf

U2 - 10.1016/j.jeconom.2009.07.003

DO - 10.1016/j.jeconom.2009.07.003

M3 - Journal article

SN - 0304-4076

VL - 154

SP - 85

EP - 100

JO - Journal of Econometrics

JF - Journal of Econometrics

ER -