A0890
Title: Nowcasting GDP in Real Time:A Machine Learning Approach with Mixed Frequency Data
Authors: Giacomo Caterini - Italian Parliamentary Budget Office (Italy) [presenting]
Cecilia Frale - Italian Parliamentary Budget Office (Italy)
Libero Monteforte - Bank of Italy (Italy)
Abstract: Advances in data availability and storage have provided forecasters with an increasingly rich set of economic indicators. Given the delayed and uncertain release of official GDP figures, nowcasting models have emerged to estimate current-quarter activity in real time. Our findings show that machine learning models generally dominate during normal periods when updated indicators are available, while Dynamic Factor model performs better when timely data are scarce. Surprisingly, during the COVID-19 pandemic, DFM_MF adapted much more quickly than ML. Additionally, the set of predictive economic variables has evolved in post-pandemic period, with ML proving valuable in identifying the most relevant and timely indicators over time.
