1301.0 - Year Book Australia, 2012
ARCHIVED ISSUE Released at 11:30 AM (CANBERRA TIME) 24/05/2012
Page tools:
 Print Page
| |||
|
Statistics contained in the Year Book are the most recent available at the time of preparation. In many cases, the ABS website and the websites of other organisations provide access to more recent data. Each Year Book table or graph and the bibliography at the end of each chapter provides hyperlinks to the most up to date data release where available.
Document Selection These documents will be presented in a new window.
|
||
ARTICLE – WHICH IS THE BEST SHORT-TERM MEASURE OF GROSS DOMESTIC PRODUCT?
This article discusses the decision to make GDP(A) Australia's official estimate of GDP in volume terms. The ABS currently publishes four measures of GDP (in chain volume terms) in the Australian National Accounts. These are the:
The publication of GDP(A) followed research indicating that it was, on balance, the most satisfactory indicator of short term growth (Aspden, 1990). Quarterly estimates of GDP(A) were first published in the June quarter 1990 issue of Australian National Accounts: Gross Product, Employment and Hours Worked (5222.0). Prior to GDP(A) becoming the headline volume measure in December quarter 1991, GDP(I) was the official measure of GDP and a statistical discrepancy between GDP(I) and GDP(E) was shown on the expenditure side. GDP(I) remained the official measure for current price GDP until the implementation of the 1993 System of National Accounts in the September quarter 1998. Following the final issue of 5222.0 in the September quarter 1992, GDP(A) remained the official measure as published in Australian National Accounts: National Income, Expenditure and Product (5206.0).
WHY THE GDP MEASURES DIFFER
In principle, the three measures of GDP should give the same result, but in practice they differ due to limitations of data sources. These limitations include sampling error, reporting error, incomplete coverage in the numerous individual data sources, and variations in the timing of recording of transactions.
Differences between the three alternative measures of GDP are an inevitable consequence of the derivation of measures from combinations of independent data sources. Improvements in the quality of data sources and the use of data confrontation may reduce the magnitude of differences between the series, but such statistical discrepancies can never be eliminated entirely. In Australia's case, a statistical discrepancy was first identified in the national accounts with the compilation of quarterly estimates of GDP(E) and GDP(I) in the 1960s. With the publication of quarterly estimates of GDP(P) for the first time in 1988, it became possible to calculate differences between all three measures and to publish GDP(A).
As seen in graph S30.1, the quarterly movements in the three measures of GDP do not exhibit a uniform pattern in relation to each other across time. The discrepancies in all three measures are offsetting at various times, meaning that there is no single component of GDP that is responsible for the statistical discrepancy.
OVERSEAS PRACTICES
Australia is one of only a few OECD member countries for which independent quarterly measures of GDP(E) and GDP(I) are produced, and is one of only three countries that publish these and an independent estimate of GDP(P) (the other two being Canada and the United Kingdom). The United States of America produces two independent measures, GDP(E) and GDP(I), while New Zealand produces independent estimates of GDP(E) and GDP(P).
ANALYSING THE GDP MEASURES
The short-term fluctuations in a seasonally adjusted time series are commonly referred to as its irregular component. The irregular components of the GDP measures can be considered to have two parts: measurement error and an actual irregular. To give the clearest possible picture of what is really happening in the economy, it is desirable to minimise measurement error.
A way of filtering out the measurement error is to apply a moving average to seasonally adjusted estimates to obtain a trend estimate (graph S30.2). However, the moving averages used to derive trend estimates cannot distinguish between measurement error and actual irregulars, and both types of movement are smoothed out. For example, the downturn resulting from the global financial crisis, where the main impact on GDP occurred in the December quarter 2008, is spread over a number of neighbouring quarters in the trend estimates of GDP. Another problem that leads to revisions occurs at the end of a trend series. Trend estimates are updated as more observations become available for the moving average. This is because weights of the observations at the end of series change as more observations are added.
Another way of obtaining a smoother measure of GDP is to average the three seasonally adjusted chain volume measures of GDP to obtain GDP(A). Positive and negative measurement errors in each of the independent measures of GDP are cancelled out to some extent in GDP(A). If each independent measure of GDP had the same degree of measurement error and the measurement errors were uncorrelated with each other, then the variance of the measurement error of the average would be a third of that of each individual measure. Unlike a trend estimate, only the measurement error is smoothed out in GDP(A). Also, unlike trend estimates, GDP(A) is subject to revision only when the seasonally adjusted estimates of the independent GDP measures are revised. Thus, there is a prima facie case for considering whether GDP(A) is a useful additional measure of GDP.
QUALITY ASSESSMENT AND RESULTS
There are three principal criteria available to assess the short-term movements of the seasonally adjusted estimates of the three measures of GDP and their average. These are timeliness, accuracy and reliability. They are defined as follows:
Timeliness
There is often a trade-off between timeliness versus accuracy and detail. All four measures of GDP are released approximately two months after the reference period for Australia. Around half of the 34 OECD countries release their national accounts with a similar lag to Australia. The International Monetary Fund (IMF) Special Data Dissemination Standard (SDDS) requires that countries release data for GDP and major expenditure aggregates/or by productive sectors no later than 90 days after the reference quarter.
Accuracy
The accuracy of the GDP estimates cannot be assessed definitively since the true value of GDP is unknown. However, two quantitative comparisons of accuracy are available. The first is to determine which measures of GDP are least variable in terms of quarter-to-quarter movements and the second is to determine which measure of GDP is a leading indicator in economic growth and recession cycles.
Variability
A comparison of the variability of quarter-to-quarter movements is a useful quantitative comparison since low variability indicates less erratic data. If one measure of GDP is less variable than another, it does not always mean that it is better. The lower variability could have been achieved due to source data limitations, where some components may have been imputed using relatively smooth indicators. Nevertheless, with all else being equal, a smooth measure of GDP implies that it is subject to less measurement error than a more erratic measure. Table S30.3 presents results of a variability test that show the average of quarterly movements without regard to sign from September quarter 1974 to December quarter 2010.
movement in seasonally adjusted estimate
change in trend(b)
of previous period trend
(b) Trend is calculated using a 7-term Henderson moving average.
Source: Australian National Accounts: National Income, Expenditure and Product, Dec 2010 (5206.0).
In line with the results obtained by Aspden (1990), table S30.3 shows that GDP(A) is the smoothest measure of GDP. It has the lowest average variation in the seasonally adjusted series and equal lowest variation in the trend series. GDP(A) also has the lowest value of variation in the irregular component, being slightly ahead of GDP(P). This indicates that GDP(A) and GDP(P) quarter-to-quarter movements remain better indicators of underlying trend than are the movements in GDP(I) and GDP(E). Aspden (1990) also examined the distribution of the magnitude of quarterly movements without regard to sign as they were in the March quarter 1990.
The earlier analysis discussed why GDP(P) should be less variable than either GDP(I) or GDP(E). The main factor noted was related to timing issues. Timing issues arise when transactions occurring in a particular quarter are inadvertently allocated to a neighbouring quarter. Timing problems are likely to be more acute for GDP(I) and GDP(E) than for GDP(P).
Timing discrepancies can occur in the compilation of GDP(E) through lags in tracing transactions through the economy. On the arrival of an imported good into Australia, it will be debited to imports and credited to wholesale stocks. Then, when it is sold to a retailer, it will be debited to wholesale stocks and credited to retail stocks. When it is sold to the final purchaser, it will be debited to retail stocks and credited to one of the expenditure aggregates (either final consumption expenditure or gross fixed capital formation). For each transaction, the credit and debit entries should have the same value and be recorded in the same quarter, otherwise a measurement error will occur. The good itself makes no contribution to GDP, but the chain of services provided in delivering it to its final purchaser does.
Similar potential timing problems occur with the compilation of GDP(P) as it is defined to be gross output less intermediate input. Timing inconsistencies in recording output and intermediate input would lead to measurement error in value added. However, the method of double deflation – deriving volume estimates of value added for an industry by subtracting a volume estimate of intermediate input from a volume estimate of gross output – is used only to derive quarterly volume estimates of gross farm value added. For all the other industries, quarterly volume estimates are derived by extrapolating value added estimates by a volume indicator. It is assumed that the gross value added grows at the same rate as the volume measure of output (gross output method), but changes in the ratio of value added to output can cause a discrepancy. Apart from agriculture, the potential short-term timing issues are avoided, although it can be argued that the use of extrapolative methods in non-farm production can be inferior when examining longer-term movements.
The original article noted that GDP(I) had an advantage over GDP(E) as an indicator of recent short-term movements since it was less reliant on the ABS business register. There is a continuous flow of business births, deaths and re-organisations and there are inevitable lags between these events and their notification to the ABS to enable the register to be updated. While, as a proportion of GDP, the effect of this may be small, it can be significant in terms of GDP growth, particularly when the economy is in an expansion phase. However, more recent estimates of GDP(I) are more heavily dependent on the business register. Private sector compensation of employees, gross operating surplus and gross mixed income, which comprise more than 70% of GDP(I), now use indicators derived from the Quarterly Business Indicator Survey (QBIS). This survey is based on the business register, while the proportion of GDP(E) using the business register remains at about 40%. This relates mainly to estimates of capital formation and the retail component of household final consumption. The continuing advantage of GDP(I) over GDP(E) could lie in improvements to the updating of the business register since 1990.
Sampling and non-sampling errors can occur within the source data used to compile the various measures of GDP. Sampling error exists because a sample and not the entire population is being observed, while non-sampling errors relate to other issues such as question framing and accuracy of reporting by respondents.
Aspden (1990) noted that the quality of the volume indicators for a number of service industries in gross value added and GDP(P) were weak. Hours worked was used as a proxy for output for some service industries. Recently there has been much greater use made of output measures from QBIS for estimating volume estimates for service industries. This has potentially improved the quality of the GDP(P) volume measure.
Leads and lags
A variable is said to lead another if it consistently changes before the other variable. Leading indicators provide information that may help predict the turning points of GDP at either the commencement or the end of a business cycle. Similarly, a lag indicator would consistently show changes in its cycle after other indicators have changed.
Aspden (1990) used two methods to examine the possibility of leads and lags:
In phase shift analysis, the attributes of component waves were correlated across the three independent measures of GDP in terms of their duration of wavelength, height of amplitude and their timing or phase. The results indicated that the attributes of the three independent measures coincided and there was no significant difference in most cycles from September quarter 1974 to June quarter 1990.
From the phase shift and growth cycle turning point comparisons, it was found that the three independent measures of GDP are correlated most of the time. However, GDP(P) showed less irregular movement in the growth cycle than the other measures as also indicated by the tests of variability.
Turning point analysis on three recent economic cycles is presented below in table S30.4. The results show that GDP(P) led GDP(A) by one quarter in the cycle peak of March quarter 2007, but GDP(A) led GDP(E) and GDP(P) by one quarter at the troughs in March quarter 1991 and December quarter 2008, respectively.
Reliability
A desirable property of a statistical measure is reliability, that is, a low susceptibility to revision. However, the desire for reliability needs to be qualified by the need for accuracy as well. The ABS policy has been to revise published estimates as better source data become available. However, there have been some changes to revisions policy since the introduction of supply and use tables.
Aspden (1990) focused on the revision characteristics of GDP(E), GDP(I) and an average of these two measures. GDP(P) and GDP(A) could not be included in this analysis because of the short time that the GDP(P) series had been available. The mean revision and mean absolute revision of quarterly percentage movements from initial to latest observations were examined for each series. The results showed that GDP(E) had both a greater mean revision and mean absolute revision than GDP(I). This was attributed to the greater reliance of GDP(E) on the business register.
An updated revisions analysis was published in Information Paper: Quality Dimensions of the Australian National Accounts, 2007 (5216.0.55.002). This utilised quarterly data from March quarter 1988 to June quarter 2002. Table S30.5 below shows mean revision and mean absolute revision for the percentage movement from initial release to three years later for all three measures of GDP and their average. The results for the mean revision support the original findings, with GDP(I) being slightly better than GDP(E). However, a notable result was that GDP(A) had the lowest mean absolute revision of all measures.
to Dec 2002
to Dec 1992
to Dec 1997
to Dec 2002
to Dec 2002
to Dec 1992
to Dec 1997
to Dec 2002
The information paper compared the correlation of the three independent measures against GDP(A) over the 75 quarters. GDP(P) was found to be the most highly correlated measure against GDP(A) with a coefficient of 0.87, while GDP(E) and GDP(I) had coefficients of 0.77 and 0.78, respectively. The GDP(E), GDP(I) and GDP(P) measures moved in the same direction in 80% of quarters. In 75% of quarters, the disparity in growth rates between the strongest and weakest components was greater than 0.5 percentage points and in 90% of quarters, it was greater than 0.3 percentage points. If only the two most consistent components at each point in time are taken into account, the coherence improves as only 30% of quarters had a difference greater than 0.3 percentage points.
CONCLUSION
On the basis of variability, GDP(A) is the best performing measure. GDP(A) is marginally less irregular than GDP(P) and is appreciably less irregular than GDP(I) or GDP(E). GDP(A) encompasses all the data measures, which appear to display partially offsetting measurement errors at various times. GDP(P) is less subject to timing problems than either GDP(E) or GDP(I).
A comparison of growth-cycle turning points and phase-shift analysis reveals that there is little difference between the different measures of GDP, as growth cycles coincide in most cases and no one measure of GDP consistently leads the others.
In examining mean revisions, GDP(I) is slightly more reliable than the other measures. However, GDP(A) is superior in terms of the mean absolute revision.
The available data indicate that GDP(A) is the better indicator of short-term movements, given its low variability and high reliability. This analysis supports the use of GDP(A) as the official measure of GDP.
BIBLIOGRAPHY
Aspden, C 1990, "Which is the best short-term measure of Gross Domestic Product?" originally published in Australian National Accounts: National Income and Expenditure, June Quarter 1990, (5206.0), pp. 57–65
Information Paper: Quality Dimensions of The Australian National Accounts, 2007, (5216.0.55.002), Chapter 5 "Accuracy of the national Accounts”, pp 21–31
Australian National Accounts: Gross Product, Employment and Hours Worked, (5222.0)
Australian National Accounts: National Income, Expenditure and Product, Dec 2010, (5206.0)