Ana Beatriz Galvão and Rory MacQueen
The Office for National Statistics (ONS) publishes the first estimate of aggregate investment and consumption for the United Kingdom 40 days after the end of the reference quarter. Because of this publication delay, economists have developed nowcasting models that rely on economic indicators available in real-time.
Even with the 40-day publication delay, the ONS computes the first estimate of these macroeconomic aggregates with incomplete survey data. Regular data revisions are then published to improve the accuracy of the initial estimates by exploiting a richer information set. In this paper by Ana Galvão and Rory MacQueen, a sizeable gap is found between the first estimate of investment (gross fixed capital formation) and its revised and improved ONS estimates based on additional survey data content.
The ESCoE Discussion Paper investigates whether a nowcasting model can improve over the ONS’s first estimate of consumption and investment growth if the aim is to predict the values in their latest release.
First, it finds that nowcasting models based on real-time weekly and monthly indicators (including the ONS’ monthly sectoral GVA growth series) cannot predict the latest estimates more accurately than the ONS’ first estimate. That said, some classes of economic predictors can assist in enhancing early estimates under some circumstances – such as during periods of turbulence. Monthly sectoral growth data and real-time data on job vacancies can offer useful insights into investment and consumption changes during the COVID-19 pandemic.
Second, this work considers several predictors in a mixed data sampling (MIDAS) model to nowcast quarterly investment and consumption growth and to assess their performance (accuracy) The accuracy of models that combine forecasts of single indicator models is assessed in terms of bias and root mean squared forecast error (RMSFE). They are benchmarked against a simple autoregressive model (AR model), which forecasts future values based exclusively on past values of the same variable. Values less than one indicate superior performance, values greater than one indicate worse performance.
The first indicator group considered is the 20 monthly sectoral components of the gross value added (GVA). These include sectors such as mining, agriculture, and property. The second group includes traditional monthly survey indicators, such as ONS Retail Sales, the GfK Consumer Confidence indicator, the OECD Business Confidence indicator, and CBI Retail Orders. The third uses the ONS diffusion indices created from HMRC VAT data. The final group contains monthly times series for on line advertisement and card spending. These new ONS real-time indicator are linked to the time series of vacancies and retail sales.
An inspection of the relative performance of groups of indicators in Figure 1 indicates that neither monthly GVA nor traditional survey indicators improve significantly on the autoregressive model over the decade preceding the pandemic. If, instead, the two years of the COVID-19 period are evaluated as presented in Figure 2, we find that nowcasting models improve more frequently over the benchmark, even though the accuracy of the models is in general worse due to the pandemic volatility. GVA components, VAT diffusion indices and real-time indicators now significantly improve on the autoregressive benchmark.
Overall, nowcasts using the MIDAS model and real-time indicators are not more accurate than the ONS’s first estimate for consumption and investment growth. Our research indicates, however, that new sources of high-frequency real-time data may improve the accuracy of timely forecasts. A further forthcoming ESCoE paper will investigate how input-output table information can be used to nowcast these macroeconomic aggregates using timely sectoral data.
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.
