Large-scale administrative data held by HM Revenue and Customs (HMRC) might be used to enhance the accuracy of early estimates of GDP and its components. In practice, incorporating these sources into the production of the National Accounts involves addressing a range of statistical issues. In this project we investigated methods for making use of companies’ Value Added Tax (VAT) returns in the calculation of early GDP estimates. We then investigated the stability of our modelling framework as more data become available, to inform a the use of these data for nowcasting.
Administrative data such as companies’ VAT records provide potential source material for producing National Accounts. Output figures are currently compiled from turnover data collected by a monthly survey of businesses, the Monthly Business Survey (MBS). This enumerates all large firms and a sample of small firms. VAT returns, made by nearly all firms with an annual turnover larger than £83,000, similarly include data on firm turnover. This project has shown how the VAT returns for small firms can be used to produce turnover figures which can be used in the calculation of GDP. We consider if the model provides consistent results as more data gradually become available, and, consequently, if the model is stable enough to be implemented in ONS production processes.
In response to the coronavirus pandemic, we also considered whether early vintages of the VAT data were good predictors of the path that could be derived from later vintages of VAT data. If they were, then the data could be used for nowcasting as well as backward-looking estimates. Using a two-stage process, we showed the early estimates were fairly stable and could be used for nowcasts, providing sufficient care was taken in the cleaning and use of the data.
We obtained vintages of VAT data from ONS for small firms, industry by industry, with the attribution issues resolved as far as possible and grossed up to represent the whole of each industry. ONS also provided MBS data for small and large firms separately by industry. We used the turnover data for large firms to inform the temporal disaggregation and nowcasting of the data for small firms. Initially we explored least-squares methods of temporal disaggregation. These, however, delivered very erratic estimates. As an alternative, we developed a state-space model to handle jointly the issues of temporal disaggregation and seasonal adjustment. We estimated this model in partnership with one for large firms in the MBS as a set of seemingly unrelated time series equations, with correlations in the disturbances estimated in the VAT and MBS models. We accounted for measurement errors in the VAT figures by devising methods to deal with the difficult task of separating the irregular changes in the seasonally adjusted figures from these measurement errors. We also examined the stability of our modelling framework in a nowcasting context. By estimating the state space model over an extended sample, it is possible to establish the stability of both its parameters and of the interpolands that it generates.
Our contributions are both theoretical and empirical. First, we developed a series of methods to allow use of complex administrative data for economic statistics, which account for the unusual timing and periodicity of the data, and large skewed measurement errors. We demonstrate that suitable methods are feasible to make use of VAT data and show promise for implementation in statistical production. We tested a range of cleaning methods and found that score-based approaches generally did better than discarding extreme observations.
Second, we produced a range of results on the differences and similarities between industry output estimates derived from the Monthly Business Survey (MBS) and VAT data. The estimates from VAT data were similar in aggregate, albeit with some differences historically. However, we also showed that the VAT-based nowcasts could provide a timely indication of economic recessions.
Our research has shown how VAT returns for small companies can be used to improve the information content of early GDP estimates. The methods that we have developed provide a means by which ONS can make full and efficient use of the VAT data accruing in real time as a basis for producing output data.
Labonne, P. (2022) ‘Aysmmetric Uncertainty: Nowcasting Using Skewness in Real-time Data’ ESCoE Discussion Paper Series ESCoE DP 2022-23, 13 October 2022
Labonne, P. and Weale, M. (2022) ‘Nowcasting in the presence of large measurement errors and revisions‘ ESCoE Discussion Paper Series ESCoE DP 2022-05, 9 March 2022
Labonne, P. (2021) ‘Nowcasting in the Presence of Large Measurement Errors and Revisions’ ESCoE Conference on Economic Measurement 2021 Poster Exhibition, 11-13 May 2021. Poster Presentation.
Labonne, P. and Weale, M. (2020) “Temporal disaggregation of overlapping noisy quarterly data: estimation of monthly output from UK value-added tax data” Journal of the Royal Statistical Society Statistics in Society Series A, Vol 183, Issue 3 Wiley. https://doi.org/10.1111/rssa.12568
Labonne, P. and Weale, M. (2019) “Nowcasting GDP from VAT returns in the UK” Workshop on Time Series Methods for Official Statistics, 26-27 September 2019 Paris, The Seasonal Adjustment Centre of Excellence (SACE), in collaboration with Eurostat and OECD.