| Popis: |
The paper compares the nowcasting quality of a range of models of Russian GDP using high-frequency data. The models compared are MIDAS in different specifications, including models with regularization and dimensionality reduction using principal components and Mixed-Frequency Bayesian VAR with Minnesota prior. Indices corresponding with GDP by production components are used as explanatory variables. Nowcasts of MFBVAR models are shown to have higher accuracy then obtained by any type of MIDAS models on different test time periods. We also analyze dynamics of nowcasting errors of models and calculate monthly estimate of GDP growth rate that can be obtained with MFBVAR models. |
