Analysis of Revisions to Gross Domestic Product

Australian gross domestic product (GDP) measures the value and volume of goods and services produced in Australia. The Australian Bureau of Statistics (ABS) publishes estimates of GDP based on the best information available at the time. These estimates may be revised in subsequent publications. This paper describes the causes and timing of revisions to GDP growth rates, and analyses revisions to reference periods between 2010–11 and 2017–18.

In this paper, revisions refer to differences between estimates for the same concept and reference period in the sequence of national accounts publications.

The ABS revises GDP estimates to improve their accuracy – the degree to which the data correctly describe the desired concept. Revisions reflect trade-offs between accuracy and other aspects of statistical quality described in the ABS Data Quality Framework such as:

• the delay between the reference period and publication – an aspect of timeliness
• the degree of comparability between estimates across the time series – an aspect of coherence.

The ABS publishes an initial estimate of GDP two months after the reference quarter. This short delay strikes a balance between the accuracy of GDP data sources, which become more complete over time, and the timeliness of estimates to support economic analysis and decision making. As more complete and detailed data becomes available over the next few years, the ABS progressively revises GDP estimates to improve their accuracy.

GDP and its components are published as time series, allowing users to measure growth over time. The quality of these growth rates depends on how consistently GDP is measured across the time series. From time to time, updates to economic concepts, data sources or measurement methods enable the ABS to measure GDP more accurately. To maintain coherence, substantial updates are reflected across as much of the time series as relevant, causing revisions to estimates for previous reference periods.

Although revisions are intended to improve the accuracy of GDP estimates, they reduce reliability – the proximity between initial and later estimates. This revisions analysis provides information on the reliability of initial GDP estimates, which may be of interest to economic analysts who make use of them.

For more detailed discussion of the quality of GDP and other national accounts statistics, see Information Paper: Quality Dimensions of the Australian National Accounts and chapter 24 of Australian System of National Accounts: Concepts, Sources and Methods.

Tracing the GDP compilation process helps to illustrate sources of revisions. For more information, see Australian System of National Accounts: Concepts, Sources and Methods.

An estimate of GDP for the latest quarter first appears in the Australian National Accounts just over 2 months after the quarter ends. Quarterly growth in components of GDP are measured using the growth of relevant quarterly indicator series, often from other ABS statistical products, that measure related economic concepts. These indicator series are subject to revisions, which tend to be largest in the quarter after the reference period.

Seasonally adjusted estimates for each quarter are produced by removing seasonal factors from components that display regular seasonal variation. These factors are re-estimated as the seasonal pattern evolves over time, causing revisions to seasonally adjusted quarterly growth. These revisions tend to be largest in the year after the estimates are first published but may affect the entire seasonally adjusted time series.

The Australian National Accounts also contains annual financial year estimates. Estimates for the latest financial year are first released in the June issue and are subsequently published in the Australian System of National Accounts, 4 to 5 months after the end of the financial year. In this publication:

• the 3 previous financial years are routinely open for revisions from more definitive estimates of levels (benchmarks) for each component of GDP
• the weighting pattern used to aggregate annual volume growth in the latest year is updated to reflect the previous year’s benchmark, volume measures are rescaled (re-referenced) and annual deflators are incorporated
• any substantial updates to economic concepts, data sources and measurement methods are introduced, and all relevant benchmarks are updated accordingly.

Together these updates revise annual growth estimates for the latest year and previous 3 financial years, or longer if there are substantial changes to concepts, sources or methods. The largest revisions tend to occur in the previous financial year due to the introduction of benchmarks.

The September issue of the Australian National Accounts is released soon after the Australian System of National Accounts. In this issue, quarterly estimates are updated to align with the most recent benchmarks:

• a method known as benchmarking forces the 4 quarterly levels for each financial year to sum to the corresponding benchmarks, while keeping quarterly growth as consistent with the indicator as possible
• the weighting pattern used to aggregate volume growth is updated for consistency with the previous year’s benchmarks.

A starting point for measuring revisions is to compare the time series that appear in successive publications. The table below contains quarterly growth rates of GDP for June 2019 onwards, as published in Australian National Accounts issues between June 2019 and June 2021. Each column contains the growth rate for a different reference period, and rows contain the time series from successive publications. A table of this form is referred to as a revisions triangle; comparing values within a column reveals how a reference period’s initial growth rate has been revised in subsequent publications.

Revision triangles for key National Accounts aggregates can be made available on request.

This analysis covers revisions to quarterly and annual growth rates between their initial estimates and the publication 3 years later, at which point they are close to final.

The analysis of quarterly growth covers revisions to reference periods between September 2010 and June 2018. The graph below presents estimates of quarterly growth in GDP for each reference period in the analysis span, as first published and as published 3 years later in the Australian National Accounts, together with the revisions between these publications (in percentage points). Quarterly revisions are analysed in seasonally adjusted terms.

The analysis of annual growth covers revisions to reference periods between 2010–11 and 2017–18. The graph below presents estimates of annual growth in GDP for each reference period in the analysis span, as first published in the June issue of the Australian National Accounts and as published in the Australian System of National Accounts just over 3 years later, together with the revisions between these publications (in percentage points). Annual revisions are analysed in original terms.

Revisions have been analysed for the three measures of GDP (expenditure, income and production), and the major components of each measure. Most measures and components are analysed in volume terms, with the following exceptions:

• Components of the income measure are only measured in current price terms, so revisions are analysed in current prices only.
• The level of the changes in inventories component measures growth in the value of inventory levels in dollars. Revisions are analysed in current prices only, to prevent the annual re-referencing of volume levels from being interpreted as revisions to growth.

McKenzie and Gamba (2008) describe a range of statistics for summarising revisions to GDP.

This analysis focuses on the average size of revisions, as measured by the mean absolute revision between initial estimates and estimates published 3 years later.

We also examine whether the average revision tends to be close to zero, or whether there is evidence of a systematic direction to the revisions, by

• measuring the mean revision between initial estimates and estimates published 3 years later
• testing whether the mean revision is significantly different from zero at a 95% confidence level, using robust standard errors as suggested by Di Fonzo (2005).

This test indicates whether, for example, growth rates tend to be systematically revised upwards or downwards over the three years after they are first published.

The graphs and tables below present revisions to related GDP measures and components side by side. Where one component has smaller revisions than another, this indicates there has been less scope to improve its initial estimates; these have been more reliable, but not necessarily more accurate.

On average, the size of revisions to initial estimates of quarterly GDP growth was 0.19 percentage points after three years. Revisions to the income and expenditure measures of GDP, GDP(I) and GDP(E), were larger than revisions to the production measure, GDP(P). On average, revisions increased initial estimates for all measures, but were not significantly different from zero.

On average, the size of revisions to initial estimates of annual GDP growth was 0.24 percentage points after three years. Again, revisions to GDP(I) and GDP(E) were larger than revisions to GDP(P). Mean revisions were not significantly different from zero for any of these measures.

The size of revisions varied substantially across the 7 expenditure components examined, which incorporate data from a range of sources. Revisions were largest for public gross fixed capital formation, and smallest for household final consumption expenditure, the largest component of GDP(E). The mean revision to quarterly growth in private sector gross fixed capital formation was significantly different from zero, suggesting a systematic tendency for initial estimates to be revised upwards.

Amongst the 4 income components examined, revisions were largest for gross mixed income, and smallest for compensation of employees, the largest component of GDP(I). However, the mean revision to annual growth in compensation of employees was significantly different from zero, suggesting a systematic tendency for initial estimates to be revised upwards.

Revisions were measured for 21 components of GDP(P), each of which made up less than 11% of the total in 2019–20. The component with the largest revisions was gross value added (GVA) by the Agriculture, forestry and fishing industry division. Several components had mean revisions to quarterly and/or annual growth that were significantly different from zero.

The ABS regularly monitors whether the accuracy of GDP measures can be improved through changes to:

• data sources – for instance, whether alternative data sources could be used improve the accuracy of initial estimates
• methods – for instance, whether improvements can be made to models used to combine data sources or extrapolate components that are not directly measured
• compilation practices – for instance, whether practices for editing anomalies tend to identify and resolve more errors in one direction than the other.

Revisions analysis reveals how the accuracy of initial estimates is improved when further information is available. Components that have exhibited large or significant revisions are good candidates for investigating whether accuracy can be improved in the first place. Further analysis of when revisions tend to arise, or which subcomponents are responsible, may help to isolate the main factors that contribute to them. Unfortunately, it may not be possible to make improvements in every case, especially if more accurate or timely data is difficult to obtain.

Many countries revise official GDP estimates as more information becomes available.

The Organisation for Economic Co-operation and Development (OECD) maintains a database of revisions to GDP and other main economic indicators for member countries, which can be used to compare revisions. It is important to keep in mind that revisions do not measure the accuracy of initial estimates, but rather how much the accuracy has been improved after 3 years.

Each country has its own schedule for publishing estimates for the latest reference quarter:

• Some countries publish both preliminary estimates of GDP, around 30 or 45 days after the end of the quarter, and more detailed or accurate estimates around 2 to 3 months after the end of the quarter.
• Others, including Australia, publish a single set of estimates, most commonly around 2 months after the end of the quarter.

For consistency, we measure and compare revisions to quarterly growth rates for OECD members between the earliest estimate 3 months or more after the end of the reference quarter, and the estimate dated 39 months after the end of the reference quarter. The analysis covers reference quarters between September 2010 and June 2018. We exclude any reference quarter for which a country’s first estimate in the database is dated more than six months after the quarter ends, as these estimates may already incorporate revisions. This excludes all reference quarters for Colombia and Costa Rica, and reduces the length of the analysis span for Greece, Latvia and Lithuania.

On average, revisions to Australia’s GDP have been smaller than the OECD median.

Di Fonzo, T (2005) ‘The OECD Project on Revisions Analysis: First Elements for Discussion’, Organisation for Economic Co-operation and Development (OECD), accessed 5 August 2021.

McKenzie, R and Gamba, M (2008) ‘Interpreting the results of Revision Analyses: Recommended Summary Statistics’, OECD, accessed 5 August 2021.

OECD (2021) Revisions Analysis Dataset, Infra-annual Economic Indicators [data set], OECD, accessed 14 October 2021.