| dc.description.abstract | Timely economic indicators are crucial for effective macroeconomic decision-making. In particular, this
applies during crises when economic shifts can be large and costly. In this thesis, we produce weekly
Gross Domestic Product (GDP) growth estimates for 22 large Western economies using a neural network
ensemble and a unique combination of macroeconomic data and text variables derived from 686 390 news
articles. Our contribution to the literature on GDP growth nowcasting is investigating whether comprehensive
information about the news coverage combined with macroeconomic indicators can be leveraged to improve
nowcasting. Specifically, we combine sentiment analysis with macroeconomic variables and news topic
analysis through zero-shot classification. We find that our model effectively captures the GDP growth trend
in most countries during normal times and excels during crises, such as the Global Financial Crisis and the
COVID-19 pandemic. Analyzing Shapley values revealed that the ensemble model identifies interaction
effects and non-linearities, which enable it to capture sharp shifts during crises. | en_US |
