Title: Monitoring economic linkage with a semi-parametric VAR model
Authors: Hong Wang - Monash University (Australia) [presenting]
Catherine Forbes - Monash University (Australia)
Bonsoo Koo - Monash University (Australia)
Abstract: An extension to the Global Vector AutoRegressive (Global-VAR) model for economic variables observed with mixed frequency is proposed. The modelling framework employs the notion that certain variables observed only at low frequency (e.g. quarterly) have latent (unobserved) high frequency (e.g. monthly) values that are ``missing''. Such variables are nevertheless related to several high frequency observed variables through a regression framework. Bayesian analysis of the model, which is specified in hierarchical form, is amenable to the use of Markov chain Monte Carlo methods. Two cases are considered, where the latent variable evolves linearly, and where the latent variable evolves nonlinearly, with the latter specified non-parametrically using a Gaussian Process prior. The proposed modeling framework is applied to progressively to ``nowcast'' (i.e. predict the present state of) the missing values and their corresponding quarterly aggregates. Data available from 36 countries, and spanning the period from 1979:Q2 to 2013:Q4, are explored, with the nowcast distributions resulting from the two approaches evaluated over an (expanding) out-of-sample period that includes the Global Financial Crisis.