.png){kind=icon}
.png){kind=icon}
Return
Sparse Warcasting
.png){kind=icon}
PRE
AI
DOI：10.1111/sjpe.70058.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Forecasting economic activity during institutional collapse requires nowcasts derived exclusively from alternative data sources. Such sources are abundant yet theoretically unanchored and potentially weakly informative. This study examines whether sparse supervised dimension reduction extracts reliable signals in a context rich in data but poor in statistics. Applying sparse Partial Least Squares to nowcast Ukrainian GDP during the 2022 invasion using only Google search categories, the methodology achieves lower nowcast errors than unsupervised Principal Component Regression. Geographic disaggregation amplifies gains: capital city search data systematically outperforms national aggregates across GDP components, consistent with information centralization in economic centers during existential threats.
Keywords:
embedded time series
Google categories
machine learning
nowcasting
partial least squares
sparsity
Journal
S
IF:
1.2
Papers:
40
Citations:
764