Title: Nowcasting the Austrian economy with mixed-frequency VAR models
Authors: Ines Fortin - Institute for Advanced Studies (Austria) [presenting]
Jaroslava Hlouskova - Institute for Advanced Studies (Austria)
Abstract: Having information on GDP more timely than provided by national statistical offices, usually released with a one-quarter lag, is often desirable. This is why economists have started to build models exploiting data available at higher frequencies. These data are used to nowcast the present, in the most efficient way possible and without a loss of information. The related models explicitly rely on data at different frequencies. We nowcast Austrian GDP, investment, consumption, and exports/imports. Current research suggests that the use of large models with parameter shrinkage should provide good nowcast accuracy. However, this also requires the continuous management of big datasets which are updated at different points in time and goes along with substantial computational time. We rather build small models with few high-frequency variables, which nowcast as well as possible, applying a mixed-frequency VAR model. As benchmark models, we consider the traditional (quarterly) VAR model as well as univariate AR models and MIDAS regressions. We also provide a measure of uncertainty surrounding our nowcasts.