Invited Discussion of ‘Sparse Bayesian factor analysis when the number of factors is unknown’
Published in Bayesian Analysis, 2024
Recommended citation: Canale, A., Schiavon, L, Stolf, F. (In press) Invited Discussion of ‘Sparse Bayesian factor analysis when the number of factors is unknown’ by Frühwirth-Schnatter S., Hosszejni D., and Freitas Lopes H. in Bayesian Analysis.
Invited Discussion of ‘Sparse Bayesian factor analysis when the number of factors is unknown’
in: Bayesian Analysis, in press.
Citation: Canale, A., Schiavon, L, Stolf, F. (In press) Invited Discussion of ‘Sparse Bayesian factor analysis when the number of factors is unknown’ by Frühwirth-Schnatter S., Hosszejni D., and Freitas Lopes H. in Bayesian Analysis
Abstract: Coupling identifiability through the unordered generalized lower triangular (UGLT) structures discussed in Frühwirth-Schnatter et al. (2023) with an efficient Markov chain Monte Carlo (MCMC) algorithm enables straightforward uncertainty quantification for all unknowns, including the factor loading’s sparse structure and dimension, thereby facilitating interpretation—often the ultimate goal of latent factor methods. In this discussion, we aim to highlight how the proposed approach can also be advantageous when exogenous information about the variables is available and integrated into the inferential process, specifically in inferring the sparsity structure of the factor loading matrix.