Working Paper: CEPR ID: DP17800
Authors: Juan Antolin-Diaz; Thomas Drechsel; Ivan Petrella
Abstract: A key question for households, firms, and policy makers is: how is the economy doing now? This paper develops a Bayesian dynamic factor model that allows for nonlinearities, heterogeneous lead-lag patterns and fat tails in macroeconomic data. Explicitly modeling these features changes the way that different indicators contribute to the real-time assessment of the state of the economy, and substantially improves the out-of-sample performance of this class of models. In a formal evaluation, our nowcasting framework beats benchmark econometric models and professional forecasters at predicting US GDP growth in real time.
Keywords: nowcasting; dynamic factor models; real-time data
JEL Codes: E32; E23; O47; C32; E01
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Heterogeneous dynamics and fat tails (C69) | Nowcasting economic activity (E27) |
| Modeling heterogeneous dynamics (C69) | Contribution of economic indicators to nowcasting (E27) |
| Modeling fat tails (C46) | Contribution of economic indicators to nowcasting (E27) |
| Hard indicators (C43) | Weighting in nowcasting process (C51) |
| Hard indicators (C43) | Accurate nowcasts of GDP growth (E37) |
| Bayesian DFM (C11) | Forecasting accuracy compared to benchmarks (C53) |
| Nowcasting process improvements (C53) | GDP growth nowcasts (O49) |