Deviance information criterion (DIC) is a widely used information criterion for Bayesianmodel comparison. In this paper a rigorous decision-theoretic justiÖcation of DIC is providedfor models without latent variables or incidental parameters. For models with latentvariables, however, it is shown that the data augmentation technique undermines the theoreticalunderpinnings of DIC, although it facilitates parameter estimation via Markovchain Monte Carlo (MCMC) simulation. Data augmentation invalidates the standardasymptotic arguments and conventional estimators of latent variables may be inconsistent.In this paper, a robust form of DIC, denoted as RDIC, is advocated for Bayesiancomparison of latent variable models. RDIC is shown to be a good approximation toDIC without data augmentation. While the later quantity is di¢ cult to compute, theexpectation ñ maximization (EM) algorithm facilitates the computation of RDIC whenthe MCMC output is available. Moreover, RDIC is robust to nonlinear transformations oflatent variables and distributional representations of model speciÖcation. The proposedapproach is applied to several popular models in economics and Önance.
Singapore Management University
City or Country
LI, Yong; YU, Jun; and ZENG, Tao.
Robust bayesian model Selection. (2016). Research Collection School Of Economics.
Available at: http://ink.library.smu.edu.sg/soe_research/2112
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.