There has been a recent push to improve timeliness of official statistics to enable users to conduct analysis of conditions and emerging trends in near real-time. This need became particularly important during the COVID-19 pandemic to allow government to make evidence-based decisions to respond to current economic conditions. One promising avenue for timeliness gains is commonly referred to as nowcasting, and consists of applying model-based methods to predict values for recent reference periods for which traditional estimates are not yet available. In this context, any model that can incorporate auxiliary information can be considered. This session will contrast a number of potential models including machine learning models, traditional time series models, and other complex automated models designed specifically for this context, and highlight differences and similarities among the methods as well as practical aspects of their implementation.

Submissions

A latent weekly GDP indicator for Germany