Measuring GDP at Different Publication Horizons


Measuring GDP at Different Publication Horizons



Jump to


Many users of economic statistics want to understand the state of the economy in real time. This project investigates how data sources of different frequencies can be combined to obtain real-time predictive intervals for the first release of quarterly GDP growth. We demonstrate how ‘nowcasts’ of GDP can be improved by supplementing standard macroeconomic indicators with big data, and by exploiting methodological advances to extend standard nowcasting models. We also explore leading and real-time indicators of turning points in economic growth.


There is a need for early, reliable estimates of GDP. For many years, ONS met this demand by producing a first estimate of quarterly GDP growth, based only on the output approach, twenty-five days after the end of the reference quarter. This first release, which was based on limited survey information only covering output, has since August 2018 been replaced with an estimate published forty days after the end of the reference quarter, incorporating additional information on the GDP expenditure components.

While ONS has taken a step forward by producing more soundly based first estimates of GDP growth, many users of economic statistics would also like to know what is happening in the economy in real time. We investigated whether reliable nowcasts of GDP growth can be constructed based on real-time data sources at different frequencies. In particular, the project investigated the reliability of nowcasts produced at horizons of one, two and three months before ONS’s first estimate is published. In addition to traditional point forecasts, we produced predictive intervals of the type popularised by Bank of England Inflation Report fan charts, providing a guide to the reliability of the nowcasts.

We also considered how unstructured big data sources can be incorporated in nowcasts of economic activity and how modelling pervasive features of macroeconomic data, including outliers and seasonality, can improve real-time assessment of economic activity.

Turning points refer to when the economy moves from one phase of the economic cycle to the next. When this information is available, decision making can be adjusted more appropriately in response to changing economic conditions. We considered potential data sources for identifying these.


The main element of the project used a bottom-up, mixed-frequency approach to compute accurate nowcasts for the first ONS releases of UK quarterly growth. Our results suggested that this nowcasting system can deliver well-calibrated short-term forecasts by combining forecasts of GDP components from both the expenditure and the output approaches. One of the main advantages of this system is that it is technically easy to implement in real time by professional economists.

We suggested a process by which big unstructured data, unsuitable for time series analysis, can be translated into structured data that can then be used in standard econometric models to produce nowcasts and forecasts. In a separate exercise we extended the dynamic factor model – the workhorse model used for nowcasting economic activity – to account for several key features of macroeconomic time series that are typically treated as nuisance parameters. We put these modelling innovations to the test on UK and US data in a comprehensive evaluation exercise made possible by cloud computing.

We provided evidence of the impact of GDP data revisions on turning point dating and surveyed leading statistical agencies and economic institutions on the use and performance of real-time turning point indicators.


The nowcasts developed in this project are competitive in comparison to those produced by professional forecasters such as the Bank of England and NIESR. They point to the importance of modelling past changes in the volatility of forecasting errors to achieve well-calibrated predicted intervals. Other modelling innovations proposed in this project are also capable of delivering substantial improvements in real-time point and interval nowcasts of GDP.

Using a basic selection of keywords in Google Trends, we were able to improve nowcasts of key UK macroeconomic variables, suggesting that the use of ‘high quality’ big data available in statistical agencies, government departments and central banks should further improve the accuracy of real-time measures of economic activity. Qualitative business and consumer surveys remain the most popular sources of data for turning point indicators, on the basis that they tend to lead official statistics.


Our nowcasting tools offer a way for ONS and other statistical agencies to further develop their early estimates of GDP and predictive intervals. The ONS Data Science Campus now publishes faster indicators of economic activity. Our methods can be used to evaluate the usefulness of these in nowcasting and forecasting GDP. The National Institute of Economic and Social Research publishes a widely referenced monthly GDP Tracker that provides a benchmark for assessing the economic news in the monthly GDP data release. Some of the new approaches developed in this project are being incorporated into this Tracker.

The methods developed in this project have benefitted from feedback from ONS, the Bank of England, central banks in Europe and the US and from engagement with business users via the Economists and Strategy Council and The Conference Board.

External Project Papers

Lenoël, C., Young, G. (2020) “Real-time turning point indicators: Review of current international practicesONS Economic Review, Apr 2020, Office for National Statistics


Project partners

Related publications