Template-Type: ReDIF-Paper 1.0

Author-Name: Sergei Seleznev
Author-Email: SeleznevSM@cbr.ru
Author-Workplace-Name: Bank of Russia, Russian Federation





Title: Truncated priors for tempered hierarchical Dirichlet process vector autoregression



Abstract: We construct priors for the tempered hierarchical Dirichlet process vector autoregression
model (tHDP-VAR) that in practice do not lead to explosive forecasting dynamics.
Additionally, we show that tHDP-VAR and its variational Bayesian approximation with
heuristics demonstrate competitive or even better forecasting performance on US and
Russian datasets.


Length: 37 pages
Creation-Date: 2019-10
Revision-Date:
Publication-Status:
File-URL: http://cbr.ru/Content/Document/File/87576/wp-47_e.pdf
File-Format: Application/pdf
File-Function:
Number:wps47
Classification-JEL: C11, C32, C53, E37.
Keywords: Bayesian nonparametrics, forecasting, hierarchical Dirichlet process, infinite hidden Markov model.
Handle:RePEc:bkr:wpaper:wps47