Author Name(s): Jones J. and Walker G., ONS
This article is part of the series of articles launched on 14 September (Guled, 2012), which intend to keep users informed of statistical continuous improvements being carried out across the Office for National Statistics (ONS). This is the second article in the series and reports on current continuous improvement areas relating to the compilation and publication of gross domestic product (GDP). It pulls together into a single action plan, a number of distinct activities that are or will be taking place, which progress will be reported on every six months.
As part of the ONS commitment to being the place where people come first for trusted statistics, ONS continuously reviews and improves data sources, methods and systems. This also assists users in their understanding of the quality of the statistics, their fitness for purpose and their relevance. This article covers a range of continuous improvement work being undertaken in relation to producing estimates of quarterly and annual GDP.
2. Brief background to the compilation of gross domestic product
GDP is an integral part of the United Kingdom’s (UK) national accounts and provides a measure of the total economic activity in a region. GDP is often referred to as one of the main 'summary indicators' of economic activity with GDP growth being the growth between two consecutive quarters.
2.1 GDP approaches
In the UK three different theoretical approaches are used in the estimation of GDP: output, expenditure and income.
2.1.1 GDP output
GDP from the output (O) or production approach - GDP(O) measures the sum of the value added created through the production of goods and services within the economy (our production or output as an economy). This approach provides the first estimate of quarterly GDP and can be used to show the different contributions of industries (for example, agriculture) within the economy.
2.1.2 GDP expenditure
GDP from the expenditure (E) approach - GDP(E) measures the total value of final expenditures on all finished goods and services produced within the economy; this will be by consumers, non-profit institutions and government. It includes gross capital formation, and exports of goods and services. It does not include imports of goods and services.
2.1.3 GDP income
GDP from the income (I) approach - GDP(I) measures the total income earned by individuals and businesses in the production of goods and services within the economy. The estimates provided breakdown this income into, for example, income earned by businesses (corporations), employees and the self employed.
2.1.4 Compilation of GDP
Quarterly estimates of GDP are 'gross' because the value of the capital assets actually worn away (the 'capital consumption') during the productive process has not been subtracted. However, in the annual national accounts process capital consumption is included.
GDP is produced using the 1995 European System of Accounts (ESA95) concepts and definitions and the derived estimates are published in both nominal terms (GDP at market prices) and as gross value added (GVA). In the quarterly estimates of GDP we assume that changes in gross value added (an approximation of GDP) move in line with output. The three GDP approaches are calculated by:
using data from a variety of sources including surveys, secondary sources and forecasts
applying industry weights derived from the annual national accounts supply use process, which includes an adjustment for the under-coverage of unregistered self-employed people and businesses
removing the effects of price changes by deflating the data using price indices
seasonally adjusting the data
GDP is therefore a complex estimate to produce and is often difficult for people to understand how it is produced. To assist in understanding GDP data sources, a list is included in the article by Walker et al, published on 30 May 2012.
Table 1: Examples of GDP data sources
|GDP Measure||Measurement of:||Examples of source data||Examples of deflators|
|Output||Output of UK businesses||ONS Monthly Business Survey of 44,000 businesses across the retail, construction, production, manufacturing and some of the service sectors||Consumer Prices Index; Producer Price Indices; Services Producer Price Indices|
|Agricultural data supplied by Department for Environment, Food and Rural Affairs (Defra)|
|Secondary sources for some of the service sectors|
|Expenditure||Household expenditure||ONS Living Costs and Food Survey||Consumer Prices Index|
|Business expenditure||ONS Capital Expenditure Survey|
|ONS Quarterly Stocks Survey|
|Government expenditure||HM Treasury|
|Income||Household income||HM Revenue and Customs (HMRC) income data|
|Business income||ONS Quarterly Profits Inquiry and HMRC aggregated corporations tax data|
|Government income||HM Treasury data|
2.2 The publication of GDP
Estimates for GDP are produced both on a calendar year and quarterly basis, with publication dates announced well in advance. Annual estimates are published in late summer as part of the UK National Accounts. Quarterly estimates, which are updated with more information as it becomes available each month, are published on the following basis:
2.2.1 Quarterly estimates of GDP
In the first month following the relevant quarter the preliminary estimate of GDP is published 3.5 weeks after the end of the quarter. This is based on information on output and provides the first estimate of growth in GDP. The timing of the preliminary estimate is based on a trade-off between accuracy and timeliness, with a strong user requirement for this preliminary official estimate. At this point in the cycle the GDP (O) estimate is based on approximately 44 per cent of data.
In the second month following the relevant quarter the second estimate of GDP is published. This is based on information from all three approaches and provides information on the level of GDP as well as the growth in GDP. At this point in the cycle the GDP estimate is based on approximately 83 per cent of data.
In the third month following the relevant quarter the third estimate of GDP is published in the UK National Accounts. At this point in the cycle the GDP estimate is based on approximately 92 per cent of data.
2.2.2 Annual estimates of GDP
In the annual national accounts process the quarterly data are ‘benchmarked’ to annual data. For example, quarterly household income will be ‘benchmarked’ to annual household income data from HMRC. The different sources of data for the different components are brought together using the supply and use tables (SUTs) framework. Compiling the tables reconciles levels not growth rates. The supply and use framework is used to balance supply and demand for products at a detailed level (112 products in the UK). Additionally, data by 112 industry groups are confronted and balanced to give definitive estimates of GDP for all years covered. Further information on UK SUTs can be found in Mahajan (2006) (110.2 Kb Pdf) .
In the UK, SUTs are compiled for annual data with the first balance published around 18 months after the end of the year. So, in the 2012 Blue Book dataset published on 28 June, 2010 data were balanced for the first time. The second balance is usually based on ‘final’ data, except for any subsequent methodological or sources changes.
3. Reviewing and improving GDP data sources and methods
Given the variety of monthly, quarterly and annual data sources and the methods used to produce quarterly and annual GDP it is essential that we continuously review and work to improve how GDP is estimated.
The sections below provide an overview of areas either already being reviewed, to assess if they can be improved or awaiting review. The review areas can be broadly categorised into five main headings:
GDP compilation methods
addressing GNI reservations
improving dissemination and user understanding
A summary of the review areas is provided in Annex A.
3.1 Source data
3.1.1 Index of Services industry reviews
The Index of Services (IoS) accounts for 77 per cent of GVA in the UK economy in 2009. The IoS was launched as an experimental statistic in December 2000 and designated as a National Statistic in April 2007 with the aid of a structured IoS industry review programme between 2002 and 2009. In March 2012, the IoS industry reviews programme was re-launched with three work streams:
work stream one, will concentrate on the financial sector (industries 64, 65 and 66), which has undergone significant change in recent years
work stream two, will concentrate on public service dominated industries (such as industries 84, 85, 86, 87 and 88). Industry 86M (human health: market sector) will be reviewed first
work stream three, will concentrate on industries that in the previous IoS review programme were identified as requiring further developments. Work will commence with a review of industry 68 (real estate), including both 68.1-2 (on a non-fee or contract basis) and 68.3 (on a fee or contract basis)
Further information on the IoS re-launched review programme can be found in Stephens (2012) (249.1 Kb Pdf) . During the review process we are looking to establish some industry user groups. If you would like to be involved in any of these groups please contact email@example.com
3.1.2 Construction survey review
In 2008, the quarterly construction output survey was transferred to ONS from the Department for Business Innovation and Skills (BIS). From a GDP perspective, there is a need for monthly estimates of construction output, which prior to the transfer of the survey to ONS had been fulfilled by using construction forecasts produced by Experian; the data being collected monthly from a panel of 800 construction businesses. On a quarterly basis, the montly Experian forecasts were replaced with quarterly BIS construction survey estimates.
Following the transfer of the quarterly construction output survey, a methodological review was carried out on the quarterly construction survey with the objective of changing its’ frequency to monthly and improving the methods used. Following the review changes were made to the sample design and estimation, questionnaire and imputation methods (further information on the survey methods used for the construction survey can be found in the Guidance and Methodology section (500.7 Kb Pdf) . These changes have now been in place for over two and a half years and in accordance with principles in the UK Code of Practice for Official Statistics it was agreed that a post implementation methodological review should be carried out that would include seasonally adjusting the monthly construction series.
The construction survey review commenced in July 2012 and is currently focussed on the following areas: sample design and estimation, data collection and deflators; work to seasonally adjust the monthly data will commence once there are three full years of data (January 2013).
Sample design and estimation review
The sample design and estimation review has initially focussed on assessing whether any improvement could be made to the accuracy of either the estimate of total output or to components variable estimates by using alternative sample allocations but retaining the current sampling frame & sample size. The analysis has shown that any gains would be very marginal. It should be noted at this point that the construction survey selects 8,000 construction businesses and that the construction industry population is dominated by medium and small businesses and changes rapidly with businesses and self-employed people entering and leaving the construction market place.
The sample design and estimation review has found that:
for estimates of total output, turnover performs better than employment as an auxiliary variable. Turnover should therefore continue to be used as an auxiliary in the ratio estimation
some of the regional estimates for repair and maintenance have large coefficients of variation, which indicates that the regional estimates are very variable. It is not practical to carry out an allocation reducing the variance of these estimates as region cannot easily be used for stratification
it is not possible to improve on the current employment sizebands without adding extra strata
to a certain extent, lowering the band 5 turnover threshold (which currently fully enumerates businesses with turnover of £60 million and over, and employing 10-99 employees) would improve the accuracy of total output estimates, however, the impact on burden and cost should also be considered
we could reduce the sample size of the smallest businesses (compensating by sampling more medium-sized businesses) with a minimal impact on the accuracy of estimates
In December 2012, the full results of the sample design and estimation review will be published.
Following the review, further research will now be undertaken to look at the optimum construction survey design if no constraints were in place e.g., money and business burden. It is anticipated that this latter approach will provide an indication of whether any further enhancements can be made.
Data collection review
The data collection review has focussed on three areas: (1) the questions and questionnaire; (2) data validation; and (3) response chasing. Further information on these three areas are below:
questions and questionnaire - the European Short Term Statistics Regulation specifically states that construction should be measured in ‘output’ terms which means, for example, that sub-contracted work should be reported by the business undertaking the sub-contracted work. Periodically, checks to ensure that output is being reported have been carried out with construction survey respondents. As part of this review, further checks are being carried out with respondents and the questions and questionnaire are being reviewed to see if any further improvements can be made, for example to the instructions on the questionnaire
data validation – validation rules are being reviewed along with the process used to validate data with respondents
response chasing – respondents to the construction survey, especially for medium and small businesses, are typically out at work during the day and response chasing calls are often received when they are working for example, ‘up scaffolding’. The review is considering how this issue can be improved so contact can be made at more convenient times
Construction volume estimates are compiled by removing the effects of price changes. This is carried out using construction prices indices (deflators) from the Building Cost Information Service (BCIS) of the Royal Institute of Chartered Surveyors (RICS).
For each of the published construction components, ONS receives deflators on a quarterly basis from BCIS, who are currently contracted by the Department for Business Innovation and Skills (BIS) to provide this information. As part of ONS's emphasis on continuous improvement, the methodology used to construct the deflators is being reviewed.
Seasonally adjustment of monthly construction data
In January 2013, three years of monthly construction output data will have been collected. In seasonal adjustment terms this is still regarded as a ‘young’ data series and although it will be possible to produce a seasonally adjusted times series from these data it is generally recommended, from a quality perspective, to have at least seven years of data before undertaking seasonal adjustments (Eurostat, 2009). It is anticipated that we will be able to produce monthly seasonally adjusted estimates in early 2013. However, it is likely that the ‘young’ monthly seasonally adjusted construction series will be less stable and more prone to revisions as each new period is added, and therefore will be kept under constant review.
3.1.3 Coverage of small businesses
The samples for business surveys conducted by the Office for National Statistics are generally selected from the Inter-Departmental Business Register, which includes all businesses registered for Value Added Tax (VAT) and/or Pay As You Earn (PAYE); unregistered businesses are therefore not included in the business population used to select the survey samples.
In compiling GDP the under-coverage of unregistered businesses is corrected during the annual national accounts supply and use process by up rating Annual Business Survey (ABS) sales, purchase, and hence GVA data by industry. For example, for the 2012 annual national accounts the latest balanced (SUTs) year was 2010 and included in the SUTs process was an average adjustment to the ABS sales and purchase data of +1.4 per cent for unregistered businesses. The ABS data covers around 70 per cent of the UK economy; once combined with data from administrative and other sources, the adjustments amounted to an increase of 0.9 per cent (£11.5bn) in total UK GVA. The annual current price weights from the supply and use process (including the under-coverage adjustments) are then used in the quarterly GDP process. It is therefore imperative that the under-coverage adjustment is regularly reviewed using the most appropriate data sources.
With increasing number of start-up businesses there is additional concern about under-coverage in the GDP estimates.
In terms of quarterly current price GDP levels, the under-coverage of small businesses would have to equate to at least a £350 million change in levels between two consecutive quarters to produce a 0.1 per cent impact on quarterly GDP. To ensure that the under-coverage adjustment is accounting correctly for these small businesses, ONS is now in the process of working with other government departments and major internet platforms to investigate how the under-coverage adjustment can be optimised.
3.2 GDP compilation methods
GDP has not always been calculated and presented as it currently is. Prior to September 1989, the first estimate of GDP (labelled as provisional) was published around 11 weeks after the end of the quarter. Four different estimates of growth were published, relating to each of the three components and an "average" measure which was a simple arithmetic calculation giving equal weight to each component. This average did not take account of the relative quality of the source data. This approach culminated in some major criticism of the quality of GDP estimates in an annex to the 1988 Autumn Statement,
This criticism resulted in the 1989 Scrutiny Review examining Government Economic Statistics. This review, known as the Pickford review, noted a number of issues around the then method for calculating GDP and made a number of recommendations that have been developed further over subsequent years into the system that we use now.
In particular, it recognised that, given the data sources in the UK, the short term estimate of GDP growth should be based on the output approach and that, in the longer term, the level of GDP should be determined by confronting the various sources using annual supply and use tables as a framework.
Some 25 years after the publication of the Pickford Review, it seems appropriate that the GDP compilation methods should be reviewed to ensure that they are still fit for purpose. This will be undertaken in conjunction with the future vision for quarterly supply use balancing and annual constant price balancing. The review will also assess the quality of sources for the components of GDP and prioritise areas for improvement.
3.3 Addressing ‘GNI reservations’
Each year in September, the UK, in line with all other European Member States provides Eurostat with an estimate of gross national income. This estimate is used to calculate a large part of the UK contribution to the EU budget. The detail of this calculation is explained on the Europa website.
As part of the GNI Regulation, Eurostat verify the numbers produced by each Member State and put ‘reservations’ on the estimates for a particular year, or number of years, if there are concerns over the methodology used to compile the estimates. There are currently 10 reservations on the estimates for the UK covering the period 2002 to 2010.
Two of these reservations were addressed in the GNI numbers submitted in September 2012. The remaining eight need to be addressed by September 2014 and it is planned to do so over the next two years. A list of the reservations was recently published in the Guidance and Methodolgy area of the ONS website.
3.4 Assessing quality
3.4.1 Revision analysis
One aspect of quality which is of particular concern to users is the extent to which GDP estimates are revised. ONS provides a comprehensive database of successive estimates of GDP to allow users to do their own revisions analyses but also produce occasional articles summarising the position. An enhancement to the current analysis would be to analyse past revisions according to whether they arose from new data, new methods or changes to international definitions and frameworks. It is the first of these that most accurately give a guide to possible future revisions. ONS is investigating the feasibility of including this dimension in the analysis of future revisions.
3.4.2 Quality information
To aid interpretation of the quality of GDP estimates ONS provides quality information within the statistical bulletin; quality information is also included within the statistical bulletins of related releases, for example, the Index of Production and Retail Sales. Supporting quality information can also be found in the GDP Quality and Methodology Information document which will continue to be reviewed and updated on an annual basis.
3.5 Improving dissemination and user understanding
3.5.1 Improving dissemination
To understand how users’ are using and what they think of existing ONS statistical products i.e. statistical bulletins, time series dataset, podcasts and quality reports, six user focus groups were held in October 2012. The focus groups included the following types of users:
other Government Departments
The focus groups provided a very useful insight into how users currently use ONS statistical products and their views on the utility of these products. The focus groups discussions are currently being analysed and a report will be published later this year. The results from the focus group discussions will be used to inform further developments to existing ONS statistical products e.g. presentation, commentary and supporting information, and possible new products.
3.5.2 User understanding
In section 2.1.4 it was noted that ‘GDP is therefore a complex estimate to produce and is often difficult for people to understand how it is produced’. With this in mind, ONS has been considering what can be done to help people understand how GDP is produced and the quality of the GDP estimates. To date, a number or articles have been published, for example:
Balancing the three approaches to measuring gross domestic product (Lee, 2012)
Why is GDP revised? (Walker et al. 2012 (269.9 Kb Pdf) )
Measuring coherence between official estimates of economic activity and external indicators (Alhassan et al. 2012)
Attention has also been given to producing supporting articles that can be released to accompany the GDP estimates. To date, these have included:
Special events in quarter two 2012 (Jones et al. 2012), which included information on how the quarter two 2012 preliminary estimate of GDP was compiled including details of the assumptions made to calculate the estimate. The aim of the article was to help users interpret affected statistical outputs that included May and June 2012, which might have been affected by the change in bank holidays due to the Diamond Jubilee and the poor weather.
A revamp of the monthly economic review that is produced to accompany the quarterly estimates of GDP.
Further work will continue in this area and any suggestions for how ONS can help users understand GDP and the quality of the GDP estimates can be emailed to: firstname.lastname@example.org
4. Concluding comments
To ensure that the most appropriate data sources and methods are used to produce the estimates of quarterly and annual GDP a number of review and improvement activities are currently being carried out and planned by ONS; these include improvements to how the estimates are disseminated. It is anticipated that progress will be reported on a six monthly basis.
Annex A: Summary of current and planned GDP review and improvement areas
|Index of Services||Industry reviews including data sources and methods.||In progress. Is a rolling review programme|
- Sample and estimation
- Data collection
- Monthly seasonal adjustment
|Coverage of small unregistered businesses||To continually improve the annual under-coverage adjustment for small unregistered businesses||In progress|
GDP compilation methods
|GDP compilation methods||Focus||Status|
|Compilation methods||GDP compilation methods will be reviewed to ensure that they are still fit for purpose. To be undertaken in conjunction with the future vision for quarterly supply use balancing and annual constant price balancing. The review will also assess the quality of sources for the components of GDP and prioritise areas for improvement.||To be started|
|GDP compilation methods||Focus||Status|
|GNI reservations||Address the remaining eight GNI reservations by September 2014.||In progress|
|Assessing quality||Investigate the feasibility of including information on why revisions occurred i.e. from new data, new methods or change to international definitions and frameworks.||In progress|
|Continue to review and update the Quality and Methodology Information document on an annual basis.||In progress|
Improving dissemination and user understanding
|Dissemination and user understanding||Focus||Status|
|Dissemination of statistics||To assess users’ needs for improving existing statistical products and for new statistical products.||In progress|
|User understanding||Continue to look for opportunities to release supporting information to aid user understanding of GDP.||In progress|
Jacqui Jones, +44 (0)1633 455643, email@example.com, Business Indicators and Balance of Payments, ONS.