TY - CHAP TI - Bayesian regularization in latent variable models through shrinkage priors AB - Bayesian methods are nowadays widely accepted among statisticians and econometricians, mostly because Bayesian inference in combination with MCMC methods allows great flexibility in modelling complex data. However, many researchers feel uneasy about choosing priors, because they want to be as objective as possible. In the hope to let the data speak themselves they try to identify non-informative priors. The present talk gives some reasons why it is often advantageous to use carefully selected informative priors instead of non-informative ones, in particular in cases where the likelihood function is highly non-regular. Examples include Bayesian inference for finite mixture models when the number of components is unknown and variable selection problems in latent variable models. AF - CFE 2011, 5th CSDA International Conference on Computational and Financial Econometrics PP - Senate House, University of London UR - http://www.cfe-csda.org/cfe11/London2011BoA.pdf PY - 2011-12-01 AU - Frühwirth-Schnatter, Sylvia ER -