Communicating uncertainty in GDP estimates

Graph picture

The UK economy has been affected by some unprecedented events in recent years, with the effects of the COVID-19 pandemic followed immediately by the huge rise in energy costs due to the war in the Ukraine. For the ONS, these have increased the challenge of producing economic statistics that are both fast and accurate. Dramatic changes in the underlying data tend to bring added uncertainty to our initial estimates. As Craig McLaren explains, we are exploring ways to better portray this in our statistical outputs.   

GDP is a developing picture  

There is always a balance between timeliness and data quality. We produce quick estimates as these are important for policymakers. However, the earliest estimates of GDP are always subject to revision as later data become available. Sometimes a slightly different picture of the whole economy can be revealed. In times of upheaval, as we saw in the sharp downturn and then recovery of the ‘peak COVID’ years of 2020-22, subsequent revisions have also been larger than usual. 

There will always be some level of uncertainty in the estimation of official statistics, as it can take up to three years to collect all the available information about economic activity in any quarter. This is because we rely on complex information from households and businesses which can take time to collect and process in the detail we need, particularly from our large comprehensive annual surveys. We have, though, well-practiced approaches to confront and interpret the different data sources which in turn enable us to present a coherent picture of the economy.  

Current practice  

We already regularly publish articles highlighting the scale of revisions, what drove them and how they have changed over time. Our latest findings show our revisions performance has improved over time and that revisions are usually small and unbiased – that means that GDP is no more likely to revised up than down – and revisions usually tend not to be significant in size.  

Despite their importance, national statistical institutes such as the ONS generally do not explicitly show these levels of uncertainty for GDP. In the UK, we will sometimes refer to it if the compilation of estimates of GDP has been particularly challenging and so there might be higher-than-normal uncertainty. This was the case over the 2020 to 2021 period. To help explain the position we have already published a range of information on the conceptual and practical challenges in producing estimates of GDP to communicate that uncertainty.  

Immediate changes and next steps  

As the National Statistician Sir Ian Diamond discussed in his speech at the Royal Statistical Society, we are fully committed to the continuous improvement of the measurement of our economic statistics. We are working to expedite the GDP production cycle so that more data are available to us earlier.  

We are also improving clarity in the presentation of GDP estimates. Recently we invited the Office for Statistics Regulation to review our approach to revisions. Its report finds that while the ONS’s approach to revisions is “appropriate and well managed” it says that the ONS could also “improve public understanding of uncertainty in early estimates of GDP without reducing trust in statistics”. 

As part of that commitment, our December 2023 GDP publication released today highlights how the three approaches to measuring GDP – production (output), income, and expenditure compare. This reflects how the GDP compilation process naturally confronts our available data content as we compare the various sources and assess what each is telling us. They don’t always show exactly the same picture. We are also making it more explicit about where and how we have applied adjustments to our data to produce our central estimate, as well as highlighting the scale of past revisions. Where uncertainty exists in our estimates we have also made that much clearer.  

Over the coming months we will enhance this work by including more information on the data content of each release and explore enhanced use of visualisations.  

The development of our approach is informed by input from key partners including the independent Economic Statistics Centre of Excellence and further enhancements are being actively explored.  

Shining a light on uncertainty where it exists will help users understand the uncertainty inherent in measuring GDP, with the result of better anticipating future data revisions. We will look to do that over the coming months. We welcome feedback on this topic by emailing 

Craig McLaren

Craig McLaren, Head of National Accounts