• Summary
• Downloads
• Explanatory Notes
• Related Information
• Past Releases

REVISIONS HISTORY OF THE QUARTERLY AUSTRALIAN NATIONAL ACCOUNTS

1. Introduction

The statistical measures presented below are consistent with those presented by the OECD and other countries, although there are many possible variations such as the time periods chosen for analysis, use of value data or percentage change data, and the extent and depth of analysis.

The two key revisions measures used are as follows:

• Mean revision - measures the extent to which, on average, initial estimates are either higher or lower than more 'mature' or the latest vintages. A mean revision different from zero indicates a tendency for initial estimates to either understate or overstate later estimates in the study period. Whether this represents a systematic bias in revisions is the subject of statistical significance testing. The significance test used in this study is the Heteroskedasticity Autocorrelation Consistent (HAC) t-statistic by Newey and West (1987).
• Mean absolute revision - measures the 'spread' of initial estimates around the corresponding revised estimates. Revisions are measured without regard to sign.

The choice of whether to use the latest available estimate or a 'mature' estimate of a set interval as the reference point for calculating revisions is not clear cut. A set interval such as two or three years after the initial estimate has the advantage that it allows a consistent comparisons of revisions performance over time. It can be used to answer the question - are the national accounts becoming more or less reliable? Use of the latest available estimate allows revisions statistics to be calculated for a more recent time frame, but could present a misleading picture of the reliability of the series if revisions to observations that have not been through most of the processing cycle are included with those that have matured more fully. A further disadvantage with using latest available estimates is that the results are more subject to influence by occasional conceptual changes over time.

The study presented here derives revisions as the difference between the initial estimate and the estimate three years later (Y3). By this time the estimates have been subject to two rounds of the supply-use balancing process which incorporates final data sources for most series. This time interval is also consistent with the revisions data presented by the OECD and some other statistics agencies. The revisions analysis is undertaken on quarter on previous quarter percentage change estimates.

Analysis has been conducted on seasonally adjusted, chain volume growth rates, except for income-based GDP and components which are only available in current price terms. These are the series most used by economic policy advisers and analysts. Revisions in such estimates can stem from revisions in a large number of original current price components of an aggregate as well as the price deflators and seasonal factors used to create the estimates. It is a major task to fully examine the reasons behind revisions and it has not been attempted here. Future research will attempt to throw further light on this.

Initial estimates included in the study are June quarter 1988 through to December quarter 2002. At the time the data were drawn for the study (early 2006), this was the latest quarter that had gone through the standard three years of revisions (Y3 is December quarter 2005). Where the 'latest' available vintage is used as the point of comparison with initial estimates it is also taken as the data available as at the December quarter 2005. Initial and revised estimates of GDP growth for more recent quarters (up to December quarter 2006) can be seen in the data triangle above, although they are not included in the results presented below.

Revisions statistics for three 5 yearly sub-periods are also shown:

Period 1 - March quarter 1988 to December quarter 1992
Period 2 - March quarter 1993 to December quarter 1997
Period 3 - March quarter 1998 to December quarter 2002.

2. Results

Summary

The main findings from an analysis of revisions to initial estimates of quarterly growth for the 59 quarters between March quarter 1988 and December quarter 2002 are as follows:

• Initial estimates of quarterly GDP growth were revised upwards on average by 0.1 percentage points which can be compared with its average growth rate of 0.81%. The expenditure, production and income variants also tended to be revised upwards, although the mean results for the series below that level were in both directions.
• When subjected to a probabilistic test of statistical significance, mean revisions to initial estimates of GDP, GDP(P) and private gross fixed capital formation were found to be biased downwards (initial estimates understate growth), but in each of these cases the earliest sub-period was the main contributor to the result. No other series in the study were found to be biased.
• The spread of initial estimates of quarterly GDP growth as measured by the average absolute revision was 0.37 percentage points, which can be compared with the mean absolute growth rate in GDP of 0.90%. Average absolute revisions to the expenditure, production and income variants were slightly higher and for some of the component series significantly higher, but this has to be seen in the context of the higher volatility of some of the components.
• Although average revisions to GDP are not insignificant compared to its growth rate, one-third of revisions were between plus and minus 0.0 and 0.2 percentage points and initial estimates successfully indicated that the economy was growing above or below the long term trend growth rate 75% of the time.
• Initial estimates performed satisfactorily in identifying the most significant peak and trough in the early 1990s.
• OECD research shows that the magnitude of revisions to Australia's GDP growth is in the middle for the 18 OECD countries included in their study.
• Some component series have been revised substantially. Components that should receive the highest priority for future work are private and public gross fixed capital formation, change in inventories and government final consumption expenditure. These series are prone to revision and are large expenditure-side contributors to growth in GDP.

It should also be borne in mind that the ABS continues to make improvements to the data sources and methods used in the national accounts and therefore revisions results for past periods are not necessarily indicative of the characteristics of future revisions.

Gross domestic product (GDP)

The initial and revised quarterly growth rates for each quarter in the study period together with the magnitude of the revision (in percentage points) are shown in the following graph. The table below presents data for average revisions for the whole period and each of the three 5 yearly sub periods, together with other related data.

GDP, Initial estimate versus estimate three years later
	Full	Period 1	Period 2	Period 3
	1988:2 to 2002:4	1988:2 to 1992:4	1993:1 to 1997:4	1998:1 to 2002:4
Number of observations	59	19	20	20
Mean growth rate	0.81	0.48	1.05	0.90
Mean absolute growth rate (%)	0.90	0.69	1.05	0.95
Mean revision (Y3-Initial) (% points)	0.10	0.14	0.10	0.05
Median revision (Y3-Initial) (% points)	0.08	0.08	0.11	0.02
Mean revision statistically significant?	Yes	No	No	No
Proportion of positive revisions (%)	56	53	60	55
Mean absolute revision (Y3-Initial)	0.37	0.45	0.35	0.31
Initial estimates with incorrect sign	3	3	-	-
- nil or rounded to zero (including null cells)

The mean absolute revision to initial estimates was 0.37 percentage points compared with the mean absolute growth rate of 0.90 per cent. One-third of the revisions were no greater than 0.2 percentage points and just under three-quarters of revisions were no greater than 0.5 percentage points. The maximum revision to an initial estimate was 1.05 percentage points. When the sub-period data in the table is considered, the size of mean absolute revisions has declined.

The direction of growth of initial estimates was reversed 3 times out of the 59 quarters in the study period (December quarter 1989, December quarter 1990 and September quarter 1991). The fact that the sign reversals occurred around and during the 1990-91 recession period is consistent with the observation that economic measurement is typically more difficult at these times. Also, growth rates are also often close to zero, making a sign reversal more likely. Turning points and recession periods are of particular interest to economic analysts and policy advisors, and it is even more important that economic information is reliable at these times.

The graph below highlights various vintages of GDP data around the 1990-91 recession period. The initial estimates successfully identified the turning points but indicated a deeper decline in the early period and a stronger temporary recovery in the middle period of the recession than the revised estimates. It should be noted that the ABS changed its preferred headline measure of GDP during this period. Up until September quarter 1991 GDP(I) was the headline measure, although the preferred average measure (GDP(A)) was available with a lag from the June quarter 1990 and became the headline from the December quarter 1991 issue.

When forming a view about the economy, it could be expected that users of economic data will not only consider the estimates for any one quarter or any one economic data item in isolation. The ABS warns that, 'given the qualifications regarding the accuracy and reliability of the quarterly national accounts ... trend estimates provide the best guide to the underlying movements, and are more suitable than either the seasonally adjusted or original data for most business decisions and policy advice' (cat. no. 5206.0, Explanatory Notes). The revisions data described here also indicate that too much emphasis should not be put on initial estimates in isolation.

The data included in the analysis cuts-off in December quarter 2002 in order to present revisions data based on a consistent three years of revision throughout. The table below provides an update for the more recent period. It shows both the initial estimates of GDP growth and the estimates as published in the December 2006 issue of the quarterly national accounts. On the evidence so far, nearly all of the quarters since March quarter 2003 have been revised upwards. This observation has to be used with caution because the revisions cycle for much of the period is still incomplete. No revisions statistics are presented.

GDP is derived in Australia as the average of the conceptually equivalent expenditure (GDP(E)), production (GDP(P)) and income (GDP(I)) components. The chain volume measure of GDP(I) is the current price measure deflated by the GDP(E) implicit price deflator. The table below shows the summary revisions statistics for these three top level components alongside GDP itself. The mean revision statistics shows that the initial estimates of each of the three components also have a tendency to slightly understate growth, although none pass a statistical test of significance.

GDP and its top level components, Initial estimate versus estimate three years later
	Mean Revision (% points)				Mean Absolute Revision (% points)
	Full	Period 1	Period 2	Period 3	Full	Period 1	Period 2	Period 3	Number of Sign Changes
GDP	0.10	0.14	0.10	0.05	0.37	0.45	0.35	0.31	3
GDP(E)	0.09	0.05	0.10	0.10	0.60	0.89	0.44	0.50	8
GDP(P)	0.16	0.26	0.13	0.10	0.44	0.55	0.40	0.37	6
GDP(I)	0.07	0.10	0.10	0.03	0.51	0.44	0.45	0.64	6

However, as a first step, the correlation between revisions in GDP and the top level components can be presented. The scatter plots below show that revisions to GDP(E) are more highly correlated to revisions in GDP than either GDP(P) or GDP(I). The correlation coefficients are 0.83, 0.51 and 0.60 respectively. The upper left and lower right quadrants show revisions that were in the opposite direction to revisions in GDP. Given the equal weighting of the three components in GDP this would be consistent with revisions in GDP(E) being larger or that revisions to GDP(P) and GDP(I) tend to go in opposite direction or a combination of both. Some further observations about the contribution of the expenditure components to revisions in GDP are made in the section on expenditure-based GDP below.

Expenditure-based GDP and components, Initial estimate versus estimate three years later
	Mean Revision				Mean Absolute Revision
	Full 59 obs	Period 1 19 obs	Period 2 20 obs	Period 3 20 obs	Full 59 obs	Period 1 19 obs	Period 2 20 obs	Period 3 20 obs	Mean growth rate	Mean absolute growth rate
	% points	% points	% points	% points	% points	% points	% points	% points	%	%
GDP(E)	0.09	0.05	0.10	0.10	0.60	0.89	0.44	0.50	0.79	0.93
Household final consumption expenditure	-0.01	0.04	0.04	-0.10	0.31	0.34	0.24	0.34	0.86	0.89
Government final consumption expenditure	-0.26	-0.69	-0.17	0.06	1.98	2.27	2.05	1.62	0.72	1.24
Private gross fixed capital formation	0.88	1.20	0.81	0.65	2.20	1.74	1.56	3.26	1.26	3.71
Public gross fixed capital formation	-0.22	-1.02	-0.81	1.12	7.38	6.29	4.95	10.85	1.66	10.42
Changes in inventories	. .	. .	. .	. .	. .	. .	. .	. .	. .	. .
Exports of goods & services	0.04	-	-0.05	0.17	1.48	2.36	1.32	0.79	1.42	2.42
less Imports of goods & services	-0.21	-0.13	-0.35	-0.16	1.15	1.68	0.79	1.02	1.88	2.98
. . not applicable
- nil or rounded to zero (including null cells)

The substantial variation in the magnitude of revisions between components has to be seen in the context of the volatility of each series. It could normally be expected that initial estimates of growth in a volatile series would be more subject to revision than a smoother series, and the table shows this to be the case. Users of the national accounts are likely to have a different level of tolerance to revisions depending on the series volatility. All else being equal, a given percentage point revision in a volatile series will be of less concern than the same level of revision in a relatively smooth series.

The scatter plots above showed that revisions to GDP are more highly correlated to revisions in GDP(E) than either the product or income measures. Some indication of the relative contribution of expenditure side components to revisions in quarterly GDP can be provided by weighting the component revisions. The table provides two sets of weights. The first is the components average contribution to the level of GDP over the study period. Household final consumption expenditure obviously receives the highest weight, contributing on average 58.1% of GDP, and revisions to change in inventories receives the lowest, contributing on average only 0.3% of GDP. The second set of weights is the mean absolute contribution of the component to growth in GDP. Although private gross fixed capital formation contributes less than half that of household final consumption expenditure to the level of GDP, its mean contribution (without regard to sign) to growth in quarterly GDP is greater. Change in inventories contributes as much to growth as household final consumption expenditure, although its average contribution to the level of GDP was only 0.3%. This reflects the extreme quarter to quarter volatility of change in inventories. The expenditure side statistical discrepancy was also a significant contributor to growth.

Expenditure components, Weighted revisions
	Mean Absolute Revision	Mean Contribution to GDP	Weighted mean absolute revision to GDP	Mean absolute contribution to growth in GDP	Weighted mean absolute revision to growth in GDP
	% points	%	% points	% points	% points
Household final consumption expenditure	0.31	58.20	0.18	0.52	0.17
Government final consumption expenditure	1.98	18.80	0.37	0.24	0.38
Private gross fixed capital formation	2.20	17.60	0.52	0.66	0.39
Public gross fixed capital formation	7.38	3.90	0.29	0.31	0.22
Changes in inventories	. .	0.30	. .	0.52	. .
Exports of goods & services	1.48	16.30	0.24	0.40	0.24
less Imports of goods & services	1.15	-15.10	0.17	0.40	0.15
Statistical discrepancy (E)	. .	-0.10	. .	0.42	. .
GDP	0.37	100.00	0.37	0.90	. .
. . not applicable

Whichever set of weights are applied to the revisions for each component, the weighted results indicate that public and private gross fixed capital formation and government final consumption expenditure are key drivers of revisions in GDP from the expenditure side, although the revisions in components, including the statistical discrepancy, will be largely offsetting. Change in inventories, is the equal second largest contributor to growth in GDP on the expenditure side. Although it is not possible to calculate a weighted mean revision figure in percentage points for comparison in the table, data provided later in this section shows it is heavily revised and should be grouped with private and public gross fixed capital formation as deserving of priority attention in future revisions analysis work.

GDP(E)

The mean revision to GDP(E) was 0.09 percentage points, with the proportion of positive revisions being 59%. The mean absolute revision was 0.60 percentage points, with a maximum revision of 2.2 percentage points. The first sub-period was particularly subject to revisions. There were 8 cases (14%) where initial estimates of GDP(E) gave an erroneous indication of the direction of growth.

The relatively low level of revision to HFCE reflects in part the weight given to the monthly Retail Business Survey in the national accounts estimation process. It is regarded as the mostly accurate source of information for the goods component of HFCE and the estimates of that part of GDP will tend to remain unrevised in the supply and use balancing process. The services component of HFCE is more likely to be subject to revision as later sources become available.

Significant revisions were made to private GFCF data prior to June quarter 1998 as a result of changes to the assets boundary with the implementation of SNA93. The scope of assets was widened to include computer software, mineral exploration and entertainment and artistic originals and the changes were backcast to the full length of the series. This will have had some impact on the revisions statistics prior to that date, but the use of the Y3 estimate as the point of reference for revisions will have mitigated the effect on the results.

In 23% of cases, the initial estimates gave an erroneous indications of the direction of growth. In a number of cases, these resulted from relatively minor revisions to a volatile series.

In 30% of cases, the initial estimates gave an erroneous indication of the direction of growth.

Public GFCF is a particularly volatile series, and as for private GFCF, revision statistics should be assessed in that context. Nevertheless, it appears that initial estimates for this series have been a relatively poor indicator of growth.

As sensible percentage change data cannot be calculated for change in inventories, the graph below shows the magnitude of revisions in current dollar values. Despite the difficulties in interpreting current price data over time, there is no indication of a strong tendency for initial estimates to be either under or overstated, although the number of quarters that were revised upwards was slightly higher (59%). However, the series is subject to substantial revision, and in 25% of instances the initial estimates had their direction of change reversed.

Change in inventories is a difficult series to estimate because it is the difference between estimates of opening and closing inventory stock levels. Small revisions to data for stock levels can have significant impacts on the difference, and an even bigger impact on the second difference which feeds through to growth in GDP(E). There is a further difficulty arising from the seasonal adjustment of crops - forward estimates have to be prepared of the value of the harvest in the current year, and depending on actual weather conditions and prices there may be significant revisions to the earlier estimates when the crops come in. The difference between the forecast and the actual crop will be reflected in the revisions data for seasonally adjusted change in inventories.

No revisions data are presented for in the industry components of GDP(P). Industry data are yet to be included on the real-time data base of national accounts data.

The mean revisions to initial estimates of GDP(I) chain volume measure was 0.07 percentage points and on 53% of quarters the revision was in a positive direction indicating a slight tendency for estimates to be revised upwards over the study period. The mean absolute revision was 0.51 percentage points. There were 6 cases (8%) where the sign of the initial growth rate was reversed.

GDP(I) and components, Initial estimate versus estimate three years later - Current prices
	Mean Revision Period				Mean Absolute Revision Period				Contribu- tion to GDP(I)(a)	Mean growth rate	Mean absolute growth rate
	Full	1	2	3	Full	1	2	3	Full	Full	Full
	% pts	% pts	% pts	% pts	% pts	% pts	% pts	% pts	%	%	%
GDP(I)	0.06	0.21	-0.04	0.01	0.52	0.47	0.55	0.54	100.00	1.48	1.56
Compensation of employees	0.12	0.34	0.07	-0.04	0.60	0.57	0.64	0.59	47.60	1.51	1.60
Gross mixed income(b)	. .	. .	. .	0.68	. .	. .	. .	2.11	8.20	1.87	2.78
Gross operating surplus(b)	. .	. .	. .	0.04	. .	. .	. .	1.52	32.80	1.50	1.94
Gross operating surplus - Private non-financial corporations	-0.22	-0.49	-0.20	-	2.47	2.37	2.32	2.70	18.00	1.78	3.08
Taxes less subsidies on production and imports	0.04	0.60	-0.18	-0.27	2.28	1.33	2.59	2.86	11.00	1.59	2.37
. . not applicable
- nil or rounded to zero (including null cells)
(a) As at December 2005.
(b) Revisions statistics based on 18 observations (1998:3 to 2002:4).

GDP(I)

The mean revision to initial estimates of GDP(I) was 0.06 percentage points, and the proportion of individual quarters which were revised upwards was 64%. Mean absolute revision was 0.52 percentage points. Only on one occasion was the direction of growth revised.

The mean revision to initial estimates of GMI was 0.68 percentage points, with 61% of revisions being in a positive direction. The mean revision result was affected by the particularly large upward revision in September quarter 2001, which if removed reduces the mean revision to 0.35 percentage points. The mean absolute revision was 2.11 percentage points. The direction of initial growth was reversed on 33% of occasions. These results are drawn from 18 observations.

GOS - Private non-financial enterprises

GOS - private non-financial corporations contributed 18.0% of GDP(I) in the December quarter 2005. In order to provide a longer time period of data, revisions to initial estimates of private non-financial corporations GOS is shown in the graph below. That component was not affected by the SNA93 changes and is available for the full length of the study period. The mean revision to initial estimates was -0.22 percentage points with the proportion of positive revisions being 47%. The mean absolute revision was 2.47 percentage points. The direction of initial growth was reversed on 19% of occasions.