64th ISI World Statistics Congress - Ottawa, Canada

64th ISI World Statistics Congress - Ottawa, Canada

On the Path to Real-Time Economic Indicators

Conference

64th ISI World Statistics Congress - Ottawa, Canada

Format: IPS Abstract

Session: IPS 171 - Advancing Timeliness of Official Statistics through Model-based Nowcasting

Wednesday 19 July 10 a.m. - noon (Canada/Eastern)

Abstract

The COVID-19 pandemic has highlighted the need for faster and more accurate economic indicators. With this context, we present experimental work on two use cases that helps push towards developing practical approaches for nowcasting economic indicators in the setting of a National Statistical Office. Initial attempts at modelling targeted the aggregate level and involved applying a dynamic factor model to model monthly national GDP with high-frequency datasets such as Google Trends and Yahoo Finance, complemented with publicly available official statistics. We found that, while high-frequency data can be useful for detecting shocks, they are not reliable enough in stable states to produce advance indicators with the necessary precision. Loosening constraints on timeliness and data availability, a second attempt at modelling involved leveraging early respondent data and working closely with subject matter experts to create a predictive model (not dissimilar to a time-series based imputation method) for electricity generation at the unit level. With considerations for explainability and robustness, we included simpler modelling approaches to predict electricity generation. Results obtained were promising with current work now looking at where else the unit-level modelling approach could be applied and how it might generalise to a broader nowcasting framework. In conclusion, while real-time nowcasting is a complex challenge, the aim is to make practical strides in this direction through experimentation with new data and methods. This work highlights the importance of balancing speed and accuracy in economic indicators and offers practical insights on developing nowcasting models, especially for those working within a National Statistical Office.