Introduction
3.86. In addition to examining the bias of estimates and the direction and magnitude of revisions to initial estimates, it is useful to consider whether all revisions are worthwhile in the sense that they improve the accuracy of an estimate. An indication of this is given in the graphs included in Appendix 1 where the value of an aggregate, at various intervals after its initial estimate, is related to the latest estimate of that aggregate (the latest estimate is assumed to be the best available approximation of the true value). Table 11 and Table 12 provide complementary information by quantifying the proportion of revisions which reduces the error in monthly (Table 11) and quarterly (Table 12) estimates. Error in this context is measured as the difference between the latest estimate of an aggregate and a preliminary estimate produced one year or less after the initial estimate. As with Table 10, the analysis in Table 11 and Table 12 provides information on counts of revisions rather than monetary values and therefore is limited in that a revision that, say, halves an error has the same weighting as a minor amendment that alters the error only marginally.

11 PROPORTION OF REVISIONS TO MONTHLY ESTIMATES THAT REDUCES ERROR RELATIVE TO LATEST ESTIMATE: JANUARY 1986 TO JUNE 1994 (a)  
    Withinfirstyear revisions (b)  Oneyear revisions (c)  
    Reduce
error  Increase
error
(d)  Proportion that
reduces error  Reduce
error  Increase error
(d)  Proportion that
reduces error  
 Aggregate  no.  no.  %  no.  no.  %  
  
 Current account        
  Merchandise exports  373  160  70  90  4  96  
  Merchandise imports  181  103  64  82  7  92  
  Balance on merchandise trade  376  180  68  86  10  90  
  Services credits  337  293  54  59  37  62  
  Services debits  327  301  52  57  39  59  
  Income credits  358  324  53  68  29  72  
  Income debits  452  270  63  78  19  81  
  Unrequited transfers credits  160  19  89  79  2  98  
  Unrequited transfers debits  125  73  63  57  25  70  
  Balance on current account  529  432  55  77  19  80  

Capital account        
  Official sector transactions  244  208  54  59  29  62  
   General government transactions  242  207  54  60  28  63  
   Reserve Bank transactions  16  2  89  14    100  

(a) The latest estimate is that consistent with the January 1995 issue of 5301.0.
(b) Counts of successive monthly revisions to an estimate, occurring within the first 12 months after the initial estimate, according to whether each revision moves the revised estimate closer to, or further away from, the latest estimate of that aggregate.
(c) Counts of the difference between the initial estimate of an aggregate and the estimate after 12 months classified according to whether the revision moves the revised estimate closer to, or further away, from the latest estimate for that aggregate.
(d) Includes revisions that ‘overshoot' i.e. they worsen the error but in the opposite direction.  
3.87. In the first three columns of the tables, revisions to an estimate occurring within the first year after the original estimate are analysed according to whether they move the estimate closer to or further away from the latest estimate. The third column shows the percentage that reduces error. The last three columns are similar except that they analyse the accumulated net revision to the initial estimate after one year in relation to the latest estimate.
Do monthly revisions reduce error?
3.88. Looking at the monthly data in Table 11, it can be seen that the proportion of revisions that reduces error within the first year (that is moves an estimate closer to its latest value) varies between items. Services debits and income credits were the worst items, with only 52% and 53%, respectively, of revisions moving the estimates closer to their latest values. The corollary to this is that, for each of those items, about half the revisions made during the first year in fact worsened the accuracy of the estimate. General government capital transactions and services credits performed little better, each with only 54% of revisions in the right direction. Examination of Graphs A.5, A.7 and A.8 confirms this picture for income credits and services where it can be seen that the median estimate during the first year both rises and falls.
3.89. Table 11 shows both that there are relatively few revisions to monthly Reserve Bank capital transactions made within the first year, and that these almost always reduce error. It also shows that the cumulative revisions after one year always reduce error, correcting the two withinyear revisions that moved the estimate in the wrong direction. This supports the evidence examined earlier which suggested that estimates of this item are very reliable.
3.90. For all items, the majority of cumulative revisions after one year reduced error, although the proportion was low for services credits (62%) and for services debits (59%). This suggests that for those items substantial revisions remained to be applied after the end of the first year. The cumulative effect of revisions to current account items constitute the revisions to the balance on current account which, over the course of the first year of estimation, generally reduced error (80% of oneyear revisions reduced error). The pattern for oneyear revisions to official sector capital transactions shows 69% of revisions reduced error.
Do quarterly revisions reduce errors?
3.91. Table 12 provides a similar analysis for quarterly statistics. The unrequited transfers debits series has the lowest proportion of revisions within the first year in the right direction (52%). As with monthly statistics, the quarterly services and income items also have low proportions of revisions within the first year that reduce error relative to the latest estimate. In the case of services credits, just over half the revisions within the first year reduce error (54%) and less than half of the oneyear cumulative revisions reduce error (43%). This pattern is evident in the Appendix 1 graph for services credits (Graph A.21) which suggests minor negative revisions follow many of the more significant positive revisions towards the final estimates.

12: PROPORTION OF REVISIONS TO QUARTERLY ESTIMATES THAT REDUCES ERROR RELATIVE TO LATEST ESTIMATE: MARCH QUARTER 1986 TO JUNE QUARTER 1994 (a)  
     Within first year revisions (b)  One year revisions (c)  
     Reduce error  Increase error (d)  Proportion that reduces error  Reduce error  Increase error (d)  Proportion that reduces error  
 Aggregate  no.  no.  %  no.  no.  %  
  
 Current account        
  Merchandise exports  81  22  79  29  2  94  
  Merchandise imports  48  19  72  25  3  89  
  Balance on merchandise trade  79  28  74  30  1  97  
  Services credits  66  57  54  13  17  43  
  Services debits  66  55  55  20  11  65  
  Income credits  77  53  59  22  9  71  
  Income debits  76  47  62  25  6  81  
  Unrequited transfers credits  79  46  63  28  3  90  
  Unrequited transfers debits  43  6  88  25  2  93  
  Balance on goods and services  32  29  52  18  7  72  
  Balance on current account  71  58  55  19  12  61  

Capital account        
  Official sector transactions  60  39  61  24  6  80  
   General government transactions  60  39  61  24  6  80  
    Foreign investment in Australia  59  30  66  25  4  87  
    Australian investment abroad  37  25  60  14  11  56  
   Reserve Bank transactions  9  4  69  4    100  
    Foreign investment in Australia  1  1  50        
    Australian investment abroad  8  3  73  4    100  
   Nonofficial sector transactions  74  50  60  19  12  61  
    Foreign investment in Australia  70  47  60  24  6  80  
    Australian investment abroad  76  44  63  27  4  87  
  Balance on capital account  70  51  61  12  13  48  

(a) The latest estimate is that consistent with the December quarter 1995 issue of 5302.0
(b) Counts of successive monthly revisions to an estimate, occurring within the first 12 months after the initial estimate, according to whether each revision moves the revised estimate closer to, or further away from, the latest estimate of that aggregate
(c) Counts of the difference between the initial estimate of an aggregate and the estimate after 12 months classified according to whether the revision moves the revised estimate closer to, or further away, from the latest estimate for that aggregate.
(d) Includes revisions that ‘overshoot' i.e. they worsen the error but in the opposite direction.  
Follow us on...
Like us on Facebook Follow us on Twitter Add the ABS on Google+ ABS RSS feed Subscribe to ABS updates