Robust Bayesian Analysis of Shifting Normal Sequence: Sensitivity to Prior
by
Ashok K. Bansal
University of Delhi

Shift point inference problem occurs when mechanism underlying sequence of observations abruptly changes. Essentially, there are two problems associated with a changing model. First, one will want to detect a change, and second, assuming that a change has occurred, be able to estimate it as well the other parameters of the model.

Bayesian approaches is often used to avoid analytic and numerical difficulties. Broemelerg and Tsurume (1987) provided solutions using normal-gamma priors for pre- and post-change mean and precision of the normal sequence. A valid criticism of the Bayesian methods is that the chosen conjugate prior may not be appropriate. It is, therefore, important to examine sensitivity of posterior inferences to potential misspecification of the prior distribution. Very little has appeared in regard to robustness and changing parametric models.

We shall discuss effects of non-normality in the post-change prior distribution of the unknown normal mean and also those of moderately non-gamma priors for changing precision of the normal sequence on posterior odds used for detection of the change also on the Bayes estimates of the change point.

Date received: October 6, 2002

Copyright © 2002 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cais-33.