By Ana Galvão
The Office for National Statistics (ONS) publishes the first estimate of quarterly real GDP for the United Kingdom (UK) 40 days after the end of the reference quarter. Our nowcasting system aims to anticipate the ONS quarterly GDP release by using monthly series that are published with a shorter delay than the GDP data. Nowcasting seeks to provide accurate estimates of current economic conditions even before the ONS publishes quarterly GDP data.
The nowcasting system we propose in our ESCoE Discussion Paper ‘Real-time Probabilistic Nowcasts of UK Quarterly GDP Growth using a Mixed-Frequency Bottom-up Approach‘ has three main features.
The first of these is that forecasts are computed with a menu-driven econometric package (Eviews) and can be updated fast as new data arrives. Our choice of employing Mixed-Data Sampling (MIDAS) regressions to compute forecasts explains this feature. These methods were first applied to macroeconomic forecasting by Clements and Galvao (2008). MIDAS regressions accommodate the fact that the target variable (GDP growth) is sampled quarterly. Still, candidate predictors, such as the index of services and consumer confidence, are sampled monthly.
The second relevant feature of our system is that we apply a bottom-up strategy. We first compute forecasts for a set of GDP components, and then in a later step we combine them to obtain GDP growth forecasts. We consider expenditure (aggregate consumption, investment and trade balance) and output (services, production and construction) components. Forecasts for these quarterly components are computed using MIDAS regressions with predictors selected from a set of 10 monthly candidates. We show that by combining GDP components we improve forecasting accuracy.
The third feature of our system is that it delivers not only point forecasts, but also the uncertainty around nowcasts using predictive intervals. We show that well-calibrated 90% predictive intervals are only achieved if we allow the forecasting uncertainty to be time-varying. The main reason for this is that forecasting uncertainty increases during the recession period (2008-2009) until 2012, but then declines fast.
To evaluate the proposed nowcasting system, only data actually available in real-time were employed to estimate and compute forecasts for the first release of quarterly UK GDP growth. Our empirical results suggest that the automated nowcasting system is able to deliver accurate forecasts, even in comparison with professional macroeconomic forecasts.
A full version of this Discussion Paper is available here.
Ana Galvão is an ESCoE Research Associate and Professor of Economic Modelling and Forecasting at Warwick Business School, University of Warwick.
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.
Friday, May 01, 2020