A Mixed Frequency Bayesian Vector Autoregression Model for the Australian Economy

Speaker: Kelly Trinh
Affiliation: Data61, CSIRO

Abstract

Australian macroeconomic data are typically released on either a monthly or quarterly frequency. However, investigations into the effects of monetary policy shocks with vector autoregressions (VARs) are usually conducted on the coarsest frequency data, i.e., quarterly. Policymakers are therefore missing out on potentially important information regarding the transmission of these shocks on higher frequency indicators. In light of this, we propose a mixed frequency vector autoregression (MFVAR) model for backcasting, nowcasting, and forecasting Australian macroeconomic indicators. The model provides monthly estimates of CPI and GDP from January 1991 to December 2023, which have historically been released only at a coarser quarterly frequency. In an in-sample analysis, we demonstrate that these higher frequency indexes offer credible advantages over existing methods. The MFVAR also provides competitive point and density forecasts of four key macroeconomic indicators --- CPI, GDP, the cash rate, and the unemployment rate --- compared to a quarterly VAR model, while also providing higher frequency monthly forecasts of all variables.            

Biography:

Kelly Trinh is a research scientist at Data61, CSIRO. Her research focuses on Bayesian statistics, time series, and spatial-temporal analysis. Recently, she has expanded her methodological focus to include hybrid modelling integrated with neural networks. Her work has broad applications across diverse fields, including economics, health, natural hazards, and agriculture.