Aysmmetric Uncertainty: Nowcasting Using Skewness in Real-time Data


Aysmmetric Uncertainty: Nowcasting Using Skewness in Real-time Data

By Paul Labonne

Nowadays there is an abundance of statistical data

Rhodes, E.C. (1937) ‘The Construction of an Index of Business Activity’, Journal of the Royal Statistical Society, 100(1), pp. 18–66. https://doi.org/10.2307/2980281

You might have heard or read something along those lines recently. But so did people in 1937. This is how British economist Edmund Cecil Rhodes starts his article “The construction of an index of business activity”. He goes on by suggesting a formal statistical approach for summarising the economic series published by The Economist into a single index representing underlying business activity.

The idea that macroeconomic data are driven by an unobserved common factor representing economic activity is now at the centre of economic nowcasting – forecasting the recent past, present and near future. Nowcasting models rely on common factors to exploit real-time data – economic series related to GDP growth but available more frequently and rapidly. Each release of real-time data, for instance unemployment figures, are used to update the common factor which then yields a new nowcast of GDP growth.

The recent large economic fluctuations in the wake of the pandemic, however, have highlighted important limitations of the current generation of nowcasting models. The New York Fed, for instance, suspended the publication of the nowcasts based on its dynamic factor model. Nowcasting models have failed notably in providing a reliable measure of uncertainty attached to their predictions. Estimating and communicating statistical uncertainty is an important research agenda at ESCoE.

In this new ESCoE discussion paper, I extend the idea behind common factors to prediction uncertainty. Currently real-time data are used mainly to retrieve a point prediction, a specific number representing the nowcast. But I show that real-time data also contain information useful for retrieving the uncertainty attached to this prediction, information which can be summarised through common factors. To investigate the benefit of extending common factors to nowcasting uncertainty, I apply the method to US GDP growth. Real-time macroeconomic data are extracted from Fred-MD which provides historical data vintages, thus facilitating both real-time analysis and replication.  The results show that two novel empirical features can be extracted from real-time data: a common factor characterising general forecasting uncertainty and another one tracking macroeconomic risk. Exploiting these common factors for nowcasting GDP growth yields more reliable measures of nowcasting uncertainty and more precise predictions when macroeconomic uncertainty is at its peak.

 Read the full ESCoE Discussion Paper here.

ESCoE blogs are published to further debate.  Any views expressed are solely those of the author(s) and so cannot be taken to represent those of the ESCoE, its partner institutions or the Office for National Statistics.

About the authors

Paul Labonne

Research Projects

Related publications