|Page tools: Print Page Print All|
3.44. The previous subsection of this paper looked at the bias and dispersion of initial and revised estimates, when compared to the final or latest estimate, over a significant period of revision (24 months, 12 quarters and 6 years for monthly, quarterly and annual estimates respectively). The reference in the analysis to the final estimates provided a perspective on the shifts in estimation levels derived from conceptual/methodological changes as well as mapping the process of regular revision through at least one annual collection cycle.
3.47. To illustrate how the data in Table 6 can be used, consider income debits. The median initial monthly estimate of income debits is $1,578 million. The median revision after one year to initial estimates was an increase of $54 million, while the median-scaled one-year revision to initial estimates was 3.9%. The pattern for quarterly and annual estimates is similar. The median initial quarterly estimate was $4,837 million, the median one year revision was $155 million and the median-scaled revision was 2.8%. The corresponding annual estimates were $14,584 million, $397 million and 2.3%.
3.48. The direction in two or three years of revisions may also be calculated. However, for purposes of this paper, the advantage of using a shorter revision period to allow more recent data to be included in the analysis, and to reflect a common focus on the immediate past in balance of payments analyses, outweighed the advantages of presenting the analysis of a longer revisions process.
3.49. Table 7 (monthly statistics) and Table 8 (quarterly statistics) analyse the direction of revisions in the same way as Table 3.1 but with statistics included for the following sub-periods of the overall period covered in Table 6.
3.50. Comparing patterns of directions of revisions between the different sub-periods can highlight trends in the quality of the statistics that are not apparent in the all periods total. Although annual statistics can be broken down in the same way, they are not presented here because the small number of observations for each sub-period means that any conclusions drawn from comparing data between sub-periods are likely to be of questionable validity.
3.51. The picture that Table 6 presents is largely consistent with that shown in Appendix 1. Most median revisions to gross statistics are positive or a small negative value - indicating a generally positive bias to revisions to offset the mainly negative biases present in initial estimates shown in Appendix 1. Bias present in revisions to net aggregates may be positive or negative depending on whether revisions to the credit or debit components of the net item are larger.
3.52. Looking at current account items shown in Tables 6, 7 and 8 it can be seen that, on average, revisions to merchandise exports and imports are very small both in dollar terms and in scaled form. This is true for all periodicities and all sub-periods. Amongst the other current account components, median revisions to unrequited transfers are small both absolutely and proportionally for all periodicities of statistics and all subperiods.
3.53. Services are a little less reliable with median proportional revisions lying between -0.5% for quarterly credit estimates and 2.7% for monthly credit estimates. Interestingly, the all-period proportional revisions to monthly services debits are lower than the quarterly revisions. This reflects delays in obtaining information from several annual surveys with the result that one-year monthly revisions for some component series often exclude survey data for the reference period. The revisions to the quarterly series would capture some of the revisions sourced from these surveys. New quarterly data sources have been developed in recent years (particularly new quarterly data collections within the ABS Survey of International Trade in Services) which means both that the one-year revisions analysis for monthly series now captures survey data revisions to extrapolations (which are still required for up to six months ahead of the survey results becoming available), and that the quarterly series are only extrapolated one quarter in advance, reducing the scale of one-year revisions to these series.
3.54. Income estimates appear the least reliable of the current account items. The median one-year revisions performance for the annual credits series (5.3%) reflects the major revisions necessary to extrapolations for reinvested earnings on investment abroad when annual survey results become available. Since 1988-89, extrapolations have varied from between about half the final result to eight times the final result. Because the one-year revisions to monthly series only capture survey data for the January to June period, the monthly series (0.4% revision) appears more reliable than either the quarterly counterpart (-0.9%) that captures more of the survey data in the first year of revision, or the annual series.
3.55. For income debits the median proportional revisions to both monthly and quarterly estimates improve in the second subperiod (January 1990 to December 1993), reflecting on average improved coverage of income debits transactions within the first year of revision.
3.56. The net effect of all revisions to the current account components are reflected in revisions to the balance on current account. For all periodicities, median revisions to this balance show a bias of between 2.0% (annual estimates) and 4.2% (quarterly estimates). While the performance between sub-periods is more volatile, in general median proportional revisions to both monthly and quarterly estimates are smaller in the more recent sub-periods.
3.57. Turning to the capital account aggregates shown in Tables 6, 7 and 8, it can be seen that proportional and absolute median revisions are generally larger than those applied to aggregates in the current account. Because monthly capital account statistics only include net official sector transactions, only limited comparisons can be made with quarterly and annual statistics. Within monthly statistics, the size of revisions to general government sector transactions has approximately doubled from sub-period 1 (median revision of $57 million or 2.8%) to sub-period 2 (median revision of $94 million or 5.6%). This deterioration in reliability largely reflects the increased significance of, and revisions to, data reported to ABS in respect of State government borrowing. Estimates for the other component of the official sector the Reserve Bank are completely reliable over both sub-periods.
3.58. General government sector transactions are dominated by transactions classified to FIA. While the median revision to the FIA component for all quarterly periods combined is only $108 million, the scaled one-year revisions for the two sub-periods are 14.4% and -8.9%. The annual median one-year revision is $625 million (6.7%).
3.59. Median one-year revisions to transactions of the non-official sector are large both absolutely and proportionally. Median revisions to initial annual estimates of FIA ($2,561 million or 19.7%) are considerably larger than revisions to the initial quarterly estimates ($394 million or 11.2%). This reflects the different pattern of revisions between the quarterly and annual estimation cycles that arises because comprehensive survey data are generally only available about six months after the end of the reference year. In particular, the one-year revision to each annual estimate includes results from the annual collections within the ABS Survey of Foreign Investment for that year. One-year revisions to the March and June quarters' estimates, will also pick up annual survey revisions to those periods. However, for the September quarter one-year revision, no annual survey results are available; and for some years for the December quarter one-year revision, only interim annual survey results are available. It should be noted that the quarterly scaled revision in the second sub-period is very much larger (17.3%) than in the first sub-period (7.8%) reflecting improvement over time in capturing annual survey data more quickly.
3.60. However, users need to bear in mind the limitations of using median estimates when assessing the reliability of net series such as foreign investment in the non-official sector. For example, while the median one-year revision to quarterly observations for this series is only $394 million, the mean revision in absolute terms is in excess of $1 billion.
3.61. Median one-year revisions to initial quarterly estimates of Australian investment abroad by the non-official sector ($590 million or 45.4%) are, as expected, more biased than the equivalent revisions to annual estimates ($924 million or 13.6%). Revisions to the quarterly estimates are large in both sub-periods shown in Table 3.3, although there has been some decline in the relative size of the median revision in the more recent period.