Bayesian infinite factor models with non-Gaussian factors
File(s)
Published version
Author(s)
Grushanina, Margarita
Type
Conference Paper
Abstract
Bayesian factor models represent a very popular tool in the analysis of high-dimensional datasets. The cumbersome task of determining the number of factors has in recent years been addressed in literature by employing nonparametric models for the automatic inference on the number of factors. However, factors are usually assumed to be normally distributed. In reality, this assumption may prove to be too restrictive. Here, the factor model with automatic inference on the number of factors is extended to the non-Gaussian case. We relax the assumption of normality by employing a Laplace prior on factors. Two types of shrinkage priors are considered: the multiplicative gamma process prior and the cumulative shrinkage process, based on a sequence of spike-and-slab-distributions. An estimator of the covariance matrix, used to bound the prior on the idiosyncratic variances away from zero, is adapted to the non-Gaussian case. The models are tested both on simulated data sets as well as on a Eurozone countries’ inflation rates data set.
Date Issued
2021-08-08
Date Acceptance
2021-08-01
Citation
JSM Proceedings, International Society for Bayesian Analysis (ISBA) Section, 2021, pp.396-415
Publisher
American Statistical Association
Start Page
396
End Page
415
Journal / Book Title
JSM Proceedings, International Society for Bayesian Analysis (ISBA) Section
Copyright Statement
© 2021 American Statistical Association.
Identifier
https://ww2.amstat.org/meetings/proceedings/2021/data/assets/pdf/1913713.pdf
Source
Joint Statistical Meetings
Subjects
Factor analysis
multiplicative gamma process
Laplace prior
non-Gaussian factors
adaptive Gibbs sampling
shrinkage
spike-and-slab prior
Publication Status
Published
Start Date
2021-08-08
Finish Date
2021-08-12
Coverage Spatial
Virtual