Robust Bayesian Variable Selection in Finite Mixture Regression Model with an Application to Financial Crisis Data

Abstract

A Bayesian variable selection approach for finite mixture regression model is proposed which is able to simultaneously accommodate model uncertainty population heterogeneity and outlier effect Variable selection is mainly accomplished through the idea of data augmentation and special spike and slab prior specification and model inference is based on MCMC output The proposed method is further applied to analyze the global financial crises data Under two-subpopulation setting some important covariates for each group are found as well as several countries that are possible outliers

Date of Award	2015 Jul 23
Original language	English
Supervisor	Kuo-Jung Lee (Supervisor)