GDP revisions in Blue Book: 2026

• We maintain a transparent approach in showing revisions to UK gross domestic product (GDP) estimates by publishing real-time estimates of GDP, alongside our data vintages, and by providing a timely assessment of these revisions.
• There is a mean revision between the first estimate and the final estimate of positive 0.1 percentage points over the period Quarter 1 (Jan to Mar) 1961 to Quarter 4 (Oct to Dec) 2023; for the period Quarter 1 2020 to Quarter 4 2023, there is a mean revision of positive 0.3 percentage points, which includes the impact of the uncertainty around the coronavirus (COVID-19) pandemic.
• When we look at the sign of quarterly GDP growth, of the 175 times where the growth in the first estimate was positive, 168 of these (96%) have remained positive in the final estimate, while 42 of the 64 (66%) instances where the growth in the first estimate was negative have remained negative.
• For the period Quarter 1 1997 to Quarter 2 (Apr to June) 2026, the mean absolute revision to the quarterly change in volume GDP in Blue Book 2026 is 0.04 percentage points.
• For the period between May 2018 and December 2023, the mean revision to the first estimate of the three-month change in GDP is positive 0.26 percentage points.
• Revisions to quarterly business investment growth were not statistically significant over the period Quarter 1 2022 to Quarter 4 2025.

Compiling gross domestic product (GDP) is a complicated task – not only for the UK, but for all national statistical institutes across the world.

In the UK, the value of the economy in 2025 is estimated to be £3.0 trillion; this estimate is compiled from over 363 data sources in the annual production round, with a time series that goes back to 1948 annually and 1955 quarterly. In the UK, we produce a first estimate of quarterly GDP six weeks after that quarter ends, alongside our higher-frequency monthly GDP estimates, which are also available around 40 days after the month ends.

There is a trade-off between the timeliness and accuracy of estimates of GDP. As additional information becomes available, we have a more complete picture of economic activity in that period. This compilation cycle for the national accounts estimates, which includes these data revisions and is explained in the National Accounts Revisions Policy, can take up to three years and naturally leads to revisions in our estimates of GDP.

We also introduce major methodological improvements during the annual Blue Book process to ensure that changes are implemented in a consistent and coordinated way. These changes can take place beyond the three-year production cycle. Our Why GDP figures are revised article and our recent Managing the trade-off between timeliness and accuracy in producing our best estimates of GDP blog give more details.

We can analyse revisions to our GDP estimates using three internationally recognised approaches, as outlined in the Organisation for Economic Co-operation and Development's (OECD's) Revisions in quarterly GDP of OECD and key partner countries update (PDF, 5.2MB).

These are:

• the mean revision (MR), which shows whether there is a systematic tendency for initial estimates to be revised upwards or downwards – the MRs for quarterly GDP are in Table 1
• the mean absolute revision (MAR), which measures the absolute size of revisions so that upward revisions are not offset by downward revisions of the same magnitude
• the relative mean absolute revision (RMAR), which incorporates the size of the initial estimate in relation to the revision, as a large revision on a very large initial growth rate can have less of an impact than a large revision to a very small initial growth rate

This article analyses the revisions to our quarterly GDP estimates that were published in our Blue Book 2026 publication in October 2026.

From September 2025, we reverted our main commentary on our monthly GDP bulletin to focus on the three-month on three-month movements, as it was in the pre-coronavirus (COVID-19) pandemic period. This presentational change to our monthly GDP release was accompanied by a blog post on GDP: Getting the best insight each month, which provided some background and context for our users.

We continue to share commentary around the data content, uncertainty and revisions of our short-term GDP estimates in our regular statistical bulletins and articles, and seek opportunities to improve this on an ongoing basis.

We are exploring how we can further assist users to interpret movements in our main economic statistics by developing a smoothed seasonally adjusted measure of monthly GDP, which is planned to be published before the end of 2026. An example of the application of smoothed estimates is available from the Australian Bureau of Statistics article Are you being misled by the seasons?

Figure 1 shows the mean absolute revisions (MAR) to estimates of the quarterly change in real gross domestic product (GDP) because of the annual national accounts (Blue Book) process.

This looks at the revisions in each Blue Book and relates to the period from Quarter 1 (Jan to Mar) 1997 to the latest quarter available in that publication that was open to revision. It is one indicator of the trade-off between timeliness and accuracy, and also captures methodological improvements that come through our annual Blue Books.

The main methodological and data improvements in Blue Book 2026 are the incorporation of the Annual Survey of Goods and Services (ASGS), which gives us more information on products that are produced by the services industries, and changes to our estimates of actual and imputed rentals. For quarterly changes in volume GDP, the MAR for Blue Book 2026 was 0.04 percentage points.

Another indicator of the extent of revisions in a Blue Book is to look at how the level of volume gross domestic product (GDP) evolves over time. Revisions to the change in volume GDP will also be reflected in how the level has changed. This can give a more complete understanding of the impact of revisions in any one Blue Book, which might not always be captured in looking at a mean revision. This can also be understood if we look at the vintages of volume GDP, which captures the tendency for how a first estimate is revised.

Figure 2 shows the real-time estimates of the level of volume GDP that have been published in every quarterly national accounts from June 2008 to September 2026.

To enable comparability, this has been indexed so that we can see how the economy has evolved since Quarter 1 (Jan to Mar) 2008. It shows how revisions have unfolded in real time over this period, including the known additional uncertainty through the global financial crisis and coronavirus (COVID-19) pandemic, while also capturing the methodological improvements as part of each Blue Book. It reflects the iterative process of revisions, which provides additional framing around the impact of Blue Book 2026 on the level of volume GDP. This underpins how we communicate revisions in real time as part of our efforts to be transparent around uncertainty.

Figure 2 also shows the latest Blue Book 2026 estimates, where there was a mean revision of 0.0 percentage points to the quarterly change in volume GDP. This is why we recommend looking at the additional impact of revisions in each Blue Book, which can then be compared over time to improve the public understanding of how these estimates have been revised in real time.

Figure 2: Real-time quarterly snapshots show the iterative process of revisions

Real-time estimates of quarterly GDP growth, UK, referenced to Quarter 1 2008 = 100

There will be some periods where revisions are larger, where this likely reflects higher levels of uncertainty in early estimates (for example, the coronavirus (COVID-19) pandemic) and/or where methodological and data changes have had more pronounced impacts in specific years. This can help to show what drives larger revisions to gross domestic product (GDP).

Two examples of years that have seen larger annual GDP revisions over a number of subsequent Blue Books are 2012 and 2017. In this section, we explore how the level of GDP has been revised up in these years because of methodological and data improvements across subsequent Blue Books.

Figure 3 shows the real-time evolution of the annual volume change in GDP in 2012. There was a first estimate for 2012 annual GDP growth of 0.2% published in February 2013, which was then subject to some data revisions in those early estimates. However, there were then a series of Blue Book revisions that came in over the next few years, which led to this being revised up from the first estimate.

For example, Blue Book 2014 involved the first supply and use table (SUT) revisions back to 2012, which also included the implementation of the new international frameworks in the European System of Accounts (ESA) 2010. This included the capitalisation of research and development (R&D), where it had previously been recorded as being only intermediate consumption, which by definition, raised the level of GDP. As a consequence of this "boundary" change, this led to a revision for 2012 annual GDP growth from 0.3% to 0.7%, as part of Blue Book 2014, published in September 2014.

There was then a further upward revision to the annual growth of 2012 in Blue Book 2015 from 0.7% to 1.2%, which included the impacts of remaining gross national income (GNI) changes being implemented. This included changes to the measurement of the UK "exhaustiveness" provisions, which reflected improvements around the recording of undercoverage of the incomes accruing to small businesses and income concealed by businesses and households through the evasion of taxes.

The impacts of both the Blue Book 2014 and 2015 changes were described in detail in our Revisions to GDP and components article. Figure 3 shows that the main revision occurs within the initial three years of the first estimate, but then over the last 10 years, revisions tend to be smaller.

The implementation of Blue Book 2026, specifically the Annual Survey of Goods and Services, led to a revision to this historical data and now reflects a latest estimate of 1.7% for annual GDP in 2012. This is a consequence of a change in methods to bring in an entirely new data source that shed greater light on goods production in the service sector.

Figure 3: Larger revisions to estimates of 2012 annual GDP growth are seen in the initial updates

Annual GDP growth for 2012 by data vintage, UK, from February 2013 to September 2026

Figure 4 shows how estimates of the change in volume GDP have evolved for 2017, where methodological revisions have also been the driver to upward revisions over time. However, the revisions profile shows that these have tended to come much later in the process. For example, Figure 4 shows that there was some stability in the range of estimates in the first few years after the first estimate was published, including the initial SUT balancing processes in 2019 and 2020.

However, there have been a series of methodological revisions over the last six years, which have had a particular impact for the change in 2017. For instance, the introduction of the Financial Services Survey (FSS) in 2021 led to a revision from 1.7% to 2.1%. Another example is in Blue Book 2023 where the revision was brought about through:

• the enhancement in the quality of price deflators, including new methods to account for changes in the quality of computer hardware

• expanding the use of Services Producer Price Indices (SPPI) in the national accounts

• introducing new weighting methods for market output deflators

• bringing in method improvements to trade in services travel deflators

There were further revisions in Blue Book 2025, including changes to improve survey estimates of R&D and align them with information from administrative tax records. The latest estimate in Blue Book 2026 of 3.3% also reflected new improvements to how we record actual and imputed rental changes.

Figure 4: Annual 2017 GDP growth saw upward revisions across a number of Blue Books in the 2020s

Annual GDP growth for 2017 by data vintage, UK, from February 2018 to September 2026

These case studies show how revisions not only reflect data revisions within the production cycle, but also methodological improvements that can take place beyond the three-year period of the first estimate.

Some of the revisions have reflected changes in the production boundary of GDP (increasing the scope of measured activity), while some have captured an effort to improve our measurement of GDP. It is important to be aware of the impact of these revisions to GDP, including understanding why there might be a tendency to revise up the level of GDP over time. We are committed to transparency in these communications, including learning from this iterative process of revisions.

We can use our first estimate of quarterly gross domestic product (GDP) to compare with one that is published later. Our focus is on those estimates published one quarter (three months) and three years (36 months) after the first estimate of quarterly GDP, reflecting the three-year production cycle of UK GDP.

Table 1 shows there is a zero-mean revision (MR) at t plus 3 months over the period Quarter 2 (Apr to June) 1961 to Quarter 1 (Jan to Mar) 2026, which means that there is no tendency for the first estimate to be revised up or down. The size of revisions has not been constant over time.

Table 1 shows that the MR is a little higher when comparing the first published quarterly estimate with the estimate published three years later. This analysis continues to show that while initial revisions (t plus 3 months) are not statistically significant, those revisions that tend to reflect the SUTs balancing process (t plus 36 months) are marginally statistically significant over the full span.

The quality of early quarterly estimates has improved substantially over time (Figure 5). There have been improvements to the measurement of GDP, with improved data coverage from both surveys and administrative data sources, although the challenges of measurement during the coronavirus (COVID-19) pandemic period have proven to be an exception.

Figure 5 shows that the MAR is larger at the three-year horizon for each of the last six decades. As additional information becomes available, the revision increases between t plus 3 months and t plus 36 months, although there are other revision points on the way to t plus 36 months. These include the first annual dataset in the quarterly national accounts round in March each year, and the first supply and use table (SUT) balance at t plus 24 months.

One likely explanation is that these revisions reflect the impact of balancing our full range of information in a SUT framework for the first time, and methodological improvements, which cannot be anticipated at the time of the first estimate. However, the MAR is smaller in recent decades, although again with the partial exception of the pandemic period.

We have included a separate revisions analysis for the period Quarter 1 (Jan to Mar) 2020 to Quarter 1 2026 on a t plus 3-month basis, as there are now over six years' worth of data making up this data point, while the t plus 36-month analysis now has four years of data.

Revisions compare a first and a later estimate of the change in GDP, where the “final” estimate is a proxy of the true value. It would be expected that the mean revision tends to zero as you get closer to that final estimate, as you get a cleaner signal of this true value. This will be true by construct if you take a “final” estimate as being that of a specific vintage.

Figure 6 shows how the estimate of GDP converges on average towards the final estimate through the production cycle, where we look at these revisions over different vintages of the data. Across all time periods from Quarter 1 (Jan to Mar) 1962 onwards, the mean revision to the first estimate is 0.13 percentage points. As the data sources behind our estimates mature, there is a tendency for that revision to fall.

For example, the mean revision between the estimate published a year later (m plus 12) and this “final” estimate is 0.08 percentage points. It will then be zero by the time of the estimate that is published three years later, which is taken to be the “final” one.

We also look at this profile of revisions in more recent times to improve our understanding of how these revision properties have performed of late. If we look at how vintages of GDP converge to the final estimate from Quarter 1 2010 onwards, it shows a much quicker convergence to the “final” estimate, reflecting the improvements to the early estimates of quarterly GDP in recent decades.

Figure 6: Estimates of quarterly GDP will typically converge to their "final" estimate through the three-year production cycle

Mean revision: t+36m – (t+i)m, Quarter 1 (Jan to Mar) 1962 to Quarter 4 (Oct to Dec) 2023, quarterly GDP growth, UK, percentage points

Notes:

1. The latest data point is the period ending Quarter 4 (Oct to Dec) 2023 for the t plus 36 months data.
2. We do not have a t plus 36 months estimate yet for Quarter 3 (July to Sept) 2023 and Quarter 4 2023. However, our National Accounts Revisions Policy is such that these current estimates will not be revised before we get to the t plus 36 months stage, so this has been inferred for these purposes. These have gone through two annual supply and use tables balancing processes, which is the basis of the "final" estimate being chosen.

One focus of the Organisation for Economic Co-operation and Development (OECD) in communicating uncertainty is the extent to which revisions might change the sign of the change in GDP, which can be particularly important from a real-time communications perspective.

Table 2 shows the distribution of the 248 quarterly observations (Quarter 1 (Jan to Mar) 1962 to Quarter 4 (Oct to Dec) 2023) over the same period, where the revision has not changed our first estimate of whether GDP is expanding or contracting in almost 85% of these observations. However, approximately 11% of these observations have seen a change in the sign of the first estimate, particularly in the mid-1960s and early 1990s.

Another way to look at this is that 168 of the 175 times (96%) where the growth in the first estimate has been positive, they have remained positive in the final estimate, while 42 of the 64 (66%) instances where the growth in the first estimate was negative have remained negative. This might imply that a negative growth for the first estimate is more likely to change sign than a positive growth for the first estimate.

This might be one indicator of higher uncertainty if the economy is contracting in real time. However, this is also conditional on the distribution of these growth rates for the first estimates. First estimates that are closer to no growth (for example, 0.0%) are more likely to change sign if there is a revision, where the mean revision over this entire sample is positive 0.13 percentage points. Our Communicating the UK Economic Cycle methodology article explains some of the principles in communicating uncertainty in real time.

As part of our commitment to transparency of the revision performance of gross domestic product (GDP), we also publish revisions triangles for monthly gross domestic product (MGDP). We have provided some preliminary analysis of these revision properties, noting that there is only a limited sample for these real-time vintages, but also that much of this period covers the coronavirus (COVID-19) pandemic. These revisions should not be considered as indicative of what is expected to be the typical revision performance in the future, but it can still be useful in understanding the uncertainty in real-time estimates of the monthly change in GDP so far.

For example, from the period May 2018 to August 2025, we find a mean revision between the first estimate and the one published a year later of positive 0.1 percentage points. As our sample increases, we will move to a “final” estimate that is the one that is published three years later to improve comparability.

Figure 7 shows the real-time snapshots of the level of GDP in every MGDP release from January 2022 to August 2026, with data points starting from January 2022 to provide some further indication of data uncertainty as the UK recovered from the pandemic.

There are level changes in the 2022 and 2023 periods reflecting the impact of previous Blue Book revisions, in which there was a benchmarking to new fully balanced levels of GDP. This was a period where there were known higher levels of uncertainty as part of the recovery from the pandemic, which had been communicated at the time. This dispersion has narrowed for early estimates through more recent years, which might provide some indication of lower data uncertainty. However, we recognise there have been fewer vintages produced for this period.

Figure 7: The levels of monthly GDP through 2022 to 2026

Real-time estimates of the levels of monthly GDP, UK, January 2022 to August 2026, indexed to January 2022 = 100

Figure 8 shows how much the mean revisions to the three-month on the previous three-monthly change in GDP compared with the “final” estimate changes by the vintage of GDP between May 2018 and December 2023. We compare the revision through the three-year production cycle, with an overall mean revision to the first estimate of positive 0.26 percentage points compared with three years later. We see there is some convergence to the final estimate over the first 18 months.

Figure 8: Within the Monthly GDP dataset, the three-month on previous three-month movement converged on the final estimate over the first 18 months

Mean revision: t+36m – (t+i)m, May 2018 to December 2023, three-month on previous three-month GDP growth, UK, percentage points

Notes:

1. The latest data point is the period ending October to December 2023 for the t plus 36 months data.
2. We do not have a t plus 36 months estimate yet for the periods from September to December 2023. However, our National Accounts Revisions Policy is such that these current estimates will not be revised before we get to the t plus 36 months stage, so this has been inferred for these purposes. These have gone through two annual supply and use tables balancing processes, which is the basis of the "final" estimate being chosen.

Table 3 shows the distribution of the 68 monthly growth rate observations over the period of May 2018 to December 2023. It shows the extent to which a revision between the first and final estimate 36 months later has led to a change in the sign of the monthly growth rate. In 76% of these observations there has not been a change, while for 24% of the observations there has been a change of sign, including to or from zero growth.

For completeness, we have undertaken the same analysis for the three-month on previous three-month GDP growth rates where 50 out of 68 (74%) of the initial estimates retain the same sign after three years.

Business investment is published separately as a breakdown alongside quarterly GDP. It is a closely watched indicator of investment activity. Following the publishing of our full business investment revisions article last year, we have included a shortened analysis within this article.

Estimates of business investment were updated in Blue Book 2026 because of revised and newly available data, methodological improvements and the annual review of our seasonal adjustment models. More information on the revisions can be found in Section 3 of our Business investment in the UK: April to June 2026 revised results bulletin published in September 2026, consistent with Blue Book 2026.

We analysed revisions to quarterly business investment growth using three internationally recognised measures:

• the mean revision (MR)

• mean absolute revision (MAR)

• relative mean absolute revision (RMAR)

The MR indicates whether initial estimates tend to be revised upwards or downwards. The MAR measures the average absolute size of revisions, so upward and downward revisions do not offset one another. The RMAR assesses the size of revisions relative to the initial estimate, helping to distinguish a revision that is large relative to a small initial growth rate from one applied to a much larger initial movement.

We also used a t-test to assess whether the MR was statistically different from zero. A result that is not statistically significant means that the test does not provide sufficient evidence of a systematic upward or downward revision within the period assessed; it does not mean that no revisions occurred.

Results for 2022 to 2025 are presented at the three-month (t plus 3 month) revision horizon. The earliest period for which revisions have changed since Blue Book 2025 is Quarter 2 (July to Sept) 2022, at the t plus 36 months vintage. As each annual test is based on four quarterly observations, the findings should be interpreted with caution.

From 2020 onwards, revisions to quarterly business investment growth were larger than those seen over the longer historical period. At the three-month (t plus 3 month) horizon, the MR was 0.8 percentage points and the MAR was 1.8 percentage points. At the 36-month (t plus 36 month) horizon, the MR increased to 1.6 percentage points and the MAR increased to 3.9 percentage points. This indicates that revisions continued to accumulate as additional data sources became available and estimates were refined over time.

The t-statistics for the 2020 onwards period were 1.2 at both the t plus 3 month and t plus 36 month horizons. Neither result was statistically significant under the test used. Although the MRs were positive, indicating that later estimates were, on average, higher than the initial estimates, the analysis does not provide sufficient evidence that these upward revisions were systematic. The results therefore suggest that revisions occurred in both directions and that the average revision was not significantly different from zero.

Compared with earlier periods, the 2020 onwards data show larger average revisions, particularly when measured against the final estimate at t plus 36 months. This likely reflects the increased uncertainty surrounding measurement of business activity during and following the coronavirus (COVID-19) pandemic as well as the research and development improvement brought in during Blue Book 2025. However, despite the larger magnitude of revisions, the statistical tests do not provide evidence of a consistent upward or downward bias in the initial estimates of quarterly business investment growth over this period.

Figure 9 shows the MARs to estimates of the quarterly change in business investment as part of implementing our annual national accounts (Blue Book).

Across the entire time series, the average mean absolute revision to business investment was 1.37 percentage points. Revisions were substantially higher in 2013 and 2014 than in other years, while revisions since 2015 have generally been lower and closer to the long-term average. More information on the revisions in Blue Book 2013 and Blue Book 2014 can be found in Section 4 of our Business investment in the UK revisions in Blue Book: 2025 article.

Gross domestic product (GDP)

A measure of the economic activity produced by a country. There are three approaches used to measure GDP:

• the output approach
• the expenditure approach
• the income approach

A more detailed glossary of terms used within the national accounts is available.

More information about strengths and limitations of national accounts data used in the Blue Book can be found in our GDP quality and methodology information (QMI).