3.1. Another aspect of the quality of balance of payments statistics is their reliability. Reliability has been defined earlier as the proximity of a preliminary estimate of an aggregate for a period to the final estimate for that period. Attention therefore is focused on the extent to which an initially published estimate is revised as it is subsequently republished. The extent of such revisions is a measure of the reliability of the statistics. From an analytic perspective, users would like to know how much reliance to place on an estimate of an aggregate - particularly the initial estimate. If it is likely to be revised, how much is that revision likely to be and in which direction? Are decisions made on the basis of the initial estimates likely to prove ill-founded in the light of later revisions? The purpose of the following discussion is to answer questions of this type and to extend the picture of quality built up so far by providing the reader with an assessment of the impact of revisions.
3.2. As noted earlier, in the case of balance of payments statistics, estimates are always subject to revision and therefore, in principle, are never final. In practice however, estimates do settle down after a period and thereafter are usually only subject to change if an improved methodology or conceptual change is introduced. The length of this settling down period varies from item to item but, in the following analyses, generally estimates for periods which have undergone at least one year's revision (i.e. 12 monthly, 4 quarterly or 1 annual revision cycle(s)) are included. For annual data, however, it has been assumed that the December quarter 1994 issue of the quarterly publication Balance of Payments, Australia (Cat. no. 5302.0) captures results for the 1993-94 reference year from most annual source surveys and therefore constitutes a reasonable approximation of an annual revisions cycle for the purpose of these analyses.
3.3. The current account estimates used in the analyses are consistent with those published in both the January 1995 issue of the monthly publication Cat. no. 5301.0 and the December quarter 1994 issue of the quarterly publication Cat. no. 5302.0; and the capital account estimates are consistent with the December quarter 1994 issue of the quarterly publication Cat. no. 5302.0. Estimates in these issues are taken to be the final or latest estimates for all observation periods under examination. This means that the most recent periods for which data are included in analyses of revisions after one year from initial estimation are the month of December 1993, the December quarter 1993, and the year 1993-94 (where one year revisions are simulated to coincide with the availability of annual survey results). For other analyses data through the month and quarter of June 1994 are included.
3.4. Therefore the observation periods (i.e. months, quarters and years in respect of which statistics are provided) in these analyses are either the 96 months from January 1986 to December 1993, or 102 months to June 1994; either the 32 quarters from March 1986 to December 1993, or 34 quarters to June 1994; and the 13 years from 198182 to 1993-94. The aggregates examined are based on those published in Table 1 of the monthly and quarterly balance of payments publications (Cat. nos 5301.0 and 5302.0, respectively). The annual capital account data by direction of investment, and the annual net series for services, income, transfers and the balance on goods and services are presented from 1985-86.
3.5. A number of general points need to be borne in mind when considering the analyses of reliability of balance of payments statistics presented in this paper. First, the analyses inevitably reflect past experience and this does not necessarily give any indication of the likely behaviour of the statistics in the immediate future. For example, any recent action to improve reliability will not be fully reflected. Second, the findings for aggregate or net series should be treated with particular caution as they will reflect the varying impacts of revision on component series. Third, substantial change in the volume and size of transactions for an aggregate occurring over a relatively short time will make the effect on the reliability of the series difficult to predict. Fourth, in these analyses the assumption is made that the latest estimate is always a better approximation of the notional true value than any earlier estimate.
3.6. Finally, if there has been significant change in the concept being measured in an item, comparison of the statistics before and after the change will not be valid unless steps are taken to remove the effects of these changes from the time series. However, while accuracy may improve as improved methodologies and new concepts are employed to better approximate economic reality, from the perspective of the user faced with significant and frequent revisions arising from these changes, the reliability of the series under analysis is impaired. In this study it has only been possible to exclude the effects of one significant change in concept (the move to recording merchandise exports primarily on a shipping date basis). The analyses therefore generally reflect what the users observe in the reliability performance of the published series. Cautionary notes are provided where the significant shifts in estimation levels reflecting conceptual/methodological change need to be borne in mind.