Contents >>
Appendix 1. Bias in balance of payments statistics
1. Graphs in this appendix show the pattern of revisions through which various balance of payments aggregates pass from their initial estimates (denoted as O on the horizontal axis) through various revision points (sequentially numbered on the horizontal axis) to final estimates (denoted as the unitary line at ratio 1.00 on the vertical axis).
2. In the case of monthly statistics, initial estimates refer to the first published estimates for any month in the reference month issue of the ABS balance of payments publication (5301.0); similarly, initial quarterly estimates refer to the first published quarterly estimates appearing for that reference quarter in 5302.0; and for annual series, initial estimates refer to the first published estimates for a full financial year appearing in the June quarter issue of 5302.0 for that year. The final estimates are the latest available estimates and in this analysis represent data consistent with the January 1995 issue of 5301.0 and the December quarter 1994 issue of 5302.0.
3. Graphs are presented for monthly, quarterly and annual statistics. For monthly statistics, the initial estimate and revisions to it over the next 2 years (24 monthly revisions) are plotted against the latest estimate; for quarterly statistics, the initial estimate and revisions to it over the next 3 years (12 quarterly revisions) are plotted against the latest estimate; and for annual statistics, the initial estimate and revisions to it over the next 6 years (6 annual revisions) are plotted against the latest estimate.
4. The observation periods used for the graphs in this appendix are the 102 months from January 1986 to June 1994; the 34 quarters from March quarter 1986 to June quarter 1994; and the 13 years from 198182 to 199394 (other than for capital account component series by direction of investment, for which the annual periods commence in 198586). The chosen final observations mean that, for every annual observation period there is at least one revision; for quarterly observations there are at least 2 revisions and for monthly series there are at least 7 revisions. For example, this means that in the monthly graphs there will be 102 observations contributing to representations for the initial estimate through to the seventh revision, but with only 85 observations being available to contribute to representations at the twentyfourth revision. In the annual series, sixthround revisions will be analysed based on only 8 observations for the current account and balance on capital account. For all the components of the capital account, there is a maximum of 9 annual observations (for initial estimates) and only 3 observations for sixth round revisions.
5. Observations plotted for each aggregate are ratios of preliminary estimates (including the initial estimate) to the latest estimate after various numbers of revisions. For example, a preliminary estimate of 180 for which the latest estimate is 200 will have a ratio value of 0.90. The latest estimate (with a ratio value of 1.00) is shown as a horizontal line in the middle of each graph.
6. Median ratio values, at each revision cycle, are shown as a heavy black curve. The method used to calculate the median ratio of, say, monthly exports at revision number 3 is, first, to find, for each month, the ratio of the estimate of exports after the third revision to the latest estimate for that month. The ratios so calculated are then ranked in ascending order and the middle value selected as the median.
7. Also shown on these graphs, as hatched lines, are the upper boundaries of the first and third quartiles of ratio observations (i.e. the lines corresponding to 25% and 75% of ratio observations when they are ranked in order). The area between these two lines therefore portrays the behaviour of the middle 50% of ratios or, alternatively, the 25% of ratios immediately above the median and the 25% immediately below.
8. Because ratios are being graphed in this appendix, all balance of payments series that are compiled on a gross basis (either as credits or debits) or which are net series that have carried the same sign for the entire period under analysis (such as the balance on current account), will generate ratios that are positive, and which will generally be within a limited span. However, care is needed in interpreting the graphs compiled for net series that change sign. For such series (all the capital account series), it is possible that initial estimates and some subsequent revisions will have a different sign to the final estimate, producing a negative ratio; or be proportionately very much larger than a final estimate that is close to zero, producing a very large ratio; or, for initial estimates close to zero, be a very small fraction of a final estimate, producing a ratio close to zero. Much wider scales are used to present the capital account ratios to capture the generally wider fluctuations in revisions history, and the scales widen from monthly to quarterly and then annual series where time period aggregates may generate much greater proportional revision for net series. However, occurrences of sign turnaround and movements from or to estimates close to zero cannot be readily accommodated on scales that would allow the graphs to meaningfully display the remaining ratios, particularly for the medians. Where such occurrences account for less than half the ratios lying above and/or below the median that are calculated for a revision point within the period of observation for any series, graphing the median ratios and the first and third quartiles will not involve representing these extreme ratio values. However, for annual series where relatively few observations are available for each revision point, the occurrence of extreme ratio values is evident for some net capital account series. In Graph A46, revisions to three of the nine initial estimates have changed the sign of the net series, preventing presentation of the first quartile plot point for initial estimates. In Graph A52, the turnarounds in the sign of the balancing item resulting from revisions (for example, in the case of initial estimates, the sign changes in 4 out of 13 observations) and revisions that are many times the size of the final estimate (again in the case of initial estimates, in 4 out of 13 observations the ratios were above 2.00) preclude meaningful presentation of both the first and third quartile ratios for most revision points.
9. Reviewing the graphs in this appendix for the balance on current account provides an example of the interpretations that can be made from this presentation. Graph A11 shows that the median initial estimate and the first three median revisions of the monthly balance on current account are very close approximations of the final median estimate. However, subsequent revisions over the following 12 to 15 months, on average, overstate the deficit, which remains overstated through the subsequent revision points that are shown. The first and third quartiles are roughly evenly distributed around the median estimate, and narrow over the revision period under analysis.
10. The quarterly series for the balance on current account (see Graph A25) displays a pattern similar to the monthly series. The median initial estimate is close to the median final estimate but moves away with the first quarterly revision and continues to worsen the overstatement of the deficit through to the fifth revision point (equivalent to the deterioration seen up to the fifteenth monthly revision point). However, by extending the revision period to three years compared with the two years shown for monthly estimates, the quarterly graph throws further light on developments with revisions to the estimates. Median estimates after the eighth revision point are all closer to the final median estimate than that eighth revision point estimate.
11. Graph A44 shows the annual series for the balance on current account. The median initial estimate shown here understates the final estimate (a result not observed in the quarterly series) and by more than was observed for the monthly series. The smaller number of initial observations used (only nine) compared with the monthly and quarterly analysis contributes to this outcome (there were five understatements and four overstatements of the deficit), as does the circumstance that understatement in only one or two quarterly estimates can result in an annual estimate being understated. However, consistent with the monthly and quarterly series, the annual revisions raise the median deficit estimate, which overstates the deficit even after six annual revision cycles.
Graphs A.1 to A.52
The attached file contains a printable copy of pages 73 to 85 of the paper publication containing graphs A.1 to A.52.
Download this entire file for viewing
and printing in Adobe Acrobat format

File size: 810 Kbytes 