Revisions in relation to the national accounts may be defined as the differences between a published sequence of estimates for a given reference period for the same national accounting variable over a particular period of time. Revisions arise from the progressive incorporation of more up to date data, re-weighting of chain volume series and reassessment of seasonal factors, and from time-to-time the introduction of new accounting concepts and improved data sources and methods.
Revisions analysis is concerned with the reliability of initial (or subsequent) estimates rather than the accuracy of estimates. Accuracy is always the main focus of statistical agencies. However, as mentioned earlier in this paper, a standard accuracy measure, such as the standard error of GDP, has been found impossible to produce in practice because of the complicated compiling process and aggregation structure of the national accounts. For example, a Eurostat (1999) task force concluded that:
"even under the most optimistic assumption that confidence intervals can be provided for all basic sources, objective error margins for national accounts aggregates appear to be out of reach. The problems posed by identifying, measuring and aggregating all errors that remain after adjustments are made at the various compilation stages seem insurmountable."
Reliability refers to the ability of the successive vintages of national accounts estimates to present a consistent, reliable picture of the economy as the estimates are revised to incorporate increasingly comprehensive and improved source data. Therefore, reliability is a relative measure rather than an absolute measure like a standard error. Revisions analysis is often employed as a diagnostic tool to indicate possible sources of unreliability, but the results should never to be used as an adjustment factor in compiling or interpreting the national accounts.
Revisions should not be see as synonymous with error. Revisions are an inevitable consequence of the national accounts process, reflecting both the complexity of economic measurement and the need to provide economic policy advisers and other users with initial estimates that are timely in order to maximise their use in analysis of current economic conditions. While the trade-off between timeliness and accuracy/reliability is a major reasons for revisions to initial estimates, delaying the release of data by a year or more would not mean that revisions are no longer required. There are several reasons for this:
- Unlike in business accounting, where the aim is to record accurately the transactions and financial position of a particular business entity in isolation, the national accounts process is aimed at bringing together sets of accounts that are coherent for the whole nation. It is in effect a quadruple accounting system, and achieving coherence between the transactions of economic agents is an iterative process proceeding over a number of years.
- Some of the most complete data are only available every few years (such as the Census of Population and Housing).
- To make the national accounts more useful for economic analysis the data are subject to transformations to produce seasonally adjusted and trend estimates. The ABS uses X-11 methodology to estimate seasonal factors. This is essentially a model based process that relies on seasonal patterns established over a number of years and the estimation of forward (projecting) factors. As new quarters are added and previous quarters are revised the seasonal patterns and trend-cycle component (hereafter referred to as trend) are re-estimated and depending on specifications of the model can effect the estimates for a number of years. These estimations are subject to revision, particularly in real time as a result of so-called end-point problems that are endemic to filters used to estimate seasonal factors and trend.
- Where possible, improvements to data sources and estimation methods, as well as occasional changes to national accounting standards are backcast to earlier periods.
ABS and international data quality assessment frameworks include revisions history as one of the indicators of quality, linking it with the accuracy element. Users need to be aware of the potential for revisions, and to satisfy themselves that initial estimates provide an acceptable indication of later estimates that are based on more complete data that has been subjected to coherence checks within the national accounts system. As part of an overall quality assessment program, national accounts compilers are encouraged to regularly conduct an analysis of revisions in order to inform users and to help guide the continuous quality improvement process within statistics agencies.
The ABS has published studies of revisions in the past, as have a number of other national statistics agencies. However, more recently, the OECD have observed that most producers of economic statistics do not quantify revisions to their data. This has motivated them to establish the freely available 'Main Economic Indicators Original Release Data and Revisions Data Base' for a limited number of key economic statistics together with an analytical spreadsheet. This is designed to assist compilers of statistics and analysts to derive a standard set of revisions measures using either OECD data or their own data. This should also facilitate international comparisons of revisions.
The remainder of this section describes: the national accounts compilation process and revisions policy; the development of an infrastructure for the regular analysis of revisions; and presents the results of a recent study into revisions to the quarterly national accounts.