Introduction
3.92. Because of the preliminary nature of statistics for the latest months or quarters shown in balance of payments publications, care is needed in interpreting period-on-period (month-on-month or quarter-on-quarter) changes in the magnitude of balance of payments aggregates for the more recent periods. Nonetheless, it is useful to examine the extent to which the direction of an initial movement in a balance of payments aggregate is subsequently modified by revisions. For example, if a month-on-month (or quarter-on-quarter) change in an aggregate initially suggests a turning point, or a point of inflection, in the time series for that aggregate, how often is this pattern subsequently altered by revisions?
3.93. Scatter graphs included in Appendix 3 are intended to shed light on this question. These graphs compare the initial and latest figures for the change in an aggregate from the previous month or quarter. For each of the periods covered in this analysis, the initial estimate of the movement is plotted against the latest estimate of the movement for the same period. Initial estimates are plotted on the horizontal axis and the latest estimates on the vertical axis. Graphs of annual statistics are not included here for two reasons. First, because the small number of observations available may give a misleading impression; and, second, because initial annual estimates are rarely used as the basis for assessing changes in macro-economic conditions.
3.94. A monthly or quarterly movement which has remained unchanged since initially published will be plotted on an imaginary 45° line running from the bottom left corner to the top right corner of the graph and passing through the origin. Points that lie off this line indicate periods for which the latest estimate of the period-on-period change differs from the initial estimate of that change. Examination of the proportions of points lying above or below the 45° line can also give insights into the bias of the estimates. Those points below the line correspond to movement estimates that have decreased since the initial estimate, that is, the initial estimate of movement was overstated (positively biased). On the other hand, points above the 45° line represent initial estimates that understated the latest estimate of movement (negatively biased). It can also be seen in the graphs that points lying to the right of the vertical zero axis represent observations for which the initial and latest estimates of period-on-period movement are positive while points to the left of that axis correspond to estimates of change that are negative. Similarly, points lying above the horizontal zero axis represent observations for which the latest estimates of the period-on-period movement are positive while points below that axis correspond to estimates of change that are negative.
3.95. In cases where the initial period-on-period change does suggest a significant change in trend, the question then arises as to how much subsequent revision can be considered to make the information conveyed by the initial change misleading? In other words, how often is the informed user of the statistics likely to draw the wrong conclusions about macro-economic conditions when the user's judgements are based on initial estimates? For the purposes of this analysis, subjective judgements have been made about the size of revisions that are considered acceptable in that they would be unlikely to invalidate any broad decisions taken on the evidence of the initially published period-on-period change. These judgements have been expressed in terms of bounds around the imaginary 45° line and are included on the graphs as dotted lines. The size of revision that is considered acceptable obviously varies from item to item depending principally upon the absolute size of the aggregate involved. For example, bounds of ±$300 million are used to define the acceptance region for the monthly balance on current account item while this acceptance region is increased to ±$700 million for the quarterly balance on current account item.
3.96. An important point to bear in mind when considering these graphs is that estimates are expressed in current dollars and therefore no account is taken of the impact of inflation or changes in the absolute size of an aggregate over the period under review. For a gross aggregate that has grown progressively bigger over the period (as most have), the impact of a variation of a fixed size will be greater the further back in the time series a particular period lies. It should also be noted that the scale is not the same on all graphs.
3.97. Points on a graph that lie in either the top left or bottom right quadrants indicate a change in sign for the periodon-period movement between the initial and latest estimate. For example, these quadrants include cases where an initial improvement in the deficit on current account was subsequently revised to indicate a worsening of the deficit, or vice versa. Other things being equal, these cases might be considered to be the most likely to mislead the user and should be of most concern
Monthly revisions and estimates of movement
3.98. Looking first at monthly statistics (see Appendix 3 Graphs C.1 to C.14), it can be seen that the graphs for the merchandise trade items (Graphs C.1 and C.2) suggest very reliable statistics. With few exceptions, observations are clustered tightly along the imaginary 45° line, indicating that the month-on-month movement has changed little between the initial estimate and the latest estimate for the same month. Observations are also clustered approximately evenly around the 45° line - indicating a relative absence of bias in the estimates. There is a slight predominance of points lying to the right of the vertical axis - indicating more positive than negative movements (consistent with a gradually increasing value of traded goods). For both merchandise exports and imports items there are few observations lying outside the acceptance region - although there are more for exports than for imports - and none can be considered an extreme outlier.
3.99. Graphs C.3 and C.4 respectively show the pattern for the non-merchandise credit and non-merchandise debit items of the current account. The fact that observations are more closely clustered around the centre of Graph C.3 than Graph C.4 indicates that month-on-month changes in non-merchandise credits are generally smaller than those for non-merchandise debits. This is consistent with the typical value of these aggregates - Table 3.1 shows that the median value of monthly estimates for non-merchandise debits is more than 75% larger than the equivalent credit item. Month-on-month changes to non-merchandise debits are not only larger than changes to credits but the spread of observations above and below the 45° line indicates that the estimates of change are also less reliable. The preponderance of positive observations above the 45° line and negative observations below the 45° line (about two-thirds of the points are in these categories) also suggests that the initial estimates of month-on-month movement for non-merchandise debits tend to be understated (i.e. there is a negative bias in the estimates).
3.100. An examination of the graphs for the current account aggregates that contribute to the non-merchandise items show that income debits (Graph C.6) is principally responsible for the variability of estimates of month-on-month movement for total non-merchandise debits. The graphs also show that the negative bias in initial estimates of change evident in total non-merchandise debits is principally due to the negative bias in services debits (Graph C.8). Income debits also has a number of observation points lying outside the acceptance region of which, importantly, six lie either in the top left or bottom right quadrants indicating month-on-month movements for which the direction of change has been reversed between the initial and latest estimate.
3.101. Amongst the non-merchandise credit items, services credits (Graph C.7) appears less reliable than either income or unrequited transfers credit. There is some evidence of negative bias in the initial estimates of movement for services credits but it is not marked and of the points lying outside the acceptance region only two are in the top left or bottom right quadrants.
3.102. Graph C.11 shows the pattern for month-on-month movements for the balance on current account. This summarises movements in the component items of the current account. It indicates a relatively reliable series of observations without obvious bias or extreme points lying outside the acceptance region.
3.103. Within the monthly capital account items (Graphs C.12, C.13 and Graph C.1), the graph for the month-on-month changes in net transactions of the Reserve Bank shows an unbiased and very reliable series with a wide band of monthly movements ranging from about +$3,100 million to $2,600 million. Month-on-month changes in net capital transactions of the general government sector span a band of changes that is about twice as variable - ranging from approximately +$5,800 million to -$5,800 million - but with no clear evidence of bias in the initial estimates. Of the 15 observation points outside the acceptance region, only two are in the top left or bottom right quadrants and none can be considered to be extreme outliers.
Quarterly revisions and estimates of movement
3.104. Turning to quarterly statistics (see Appendix 3, Graphs C.15 to C.33), the graphs for the merchandise trade items indicate relatively reliable estimates of quarter-on-quarter change, with little evidence of bias and no observations lying outside the acceptance regions within which revisions are unlikely to confuse a user of the statistics. The picture for non-merchandise credits and non-merchandise debits is different. Both show a marked predominance of positive quarter-on-quarter changes (consistent with steady increases over time in the value of the items that make up these aggregates). There is evidence of negative bias in the initial estimates of quarter-on-quarter change in non-merchandise credits, but the debit aggregate shows no bias in this respect. The wide spread of observations above and below the 45° line indicate considerable variability in the estimates of change for both aggregates, but particularly debits.
3.105. For all current account components, a clear majority of observations indicate positive estimates of quarter-on-quarter change. Income debits (Graph C.20) contributes most to the variability in non-merchandise debits estimates of quarter-on-quarter movement and also accounts for most of the outlying observations. In spite of this variability, the graph for the balance on current account (Graph C.25) suggests that when all the changes are added together they result in initial estimates of movement that are relatively reliable, unbiased, and without extreme outliers. In other words, an analyst, using initial estimates of quarter-on-quarter movement in the balance on current account as evidence in assessing trends in broad macro-economic conditions, is unlikely to have been misled in the light of subsequent revisions to that movement.
3.106. With the exception of the Reserve Bank (for which estimates of movement are very reliable), components of the capital account show relatively large variability between the initial and latest estimates of movement. Sizeable revisions to quarteronquarter movements are evident for FIA transactions of the non-official sector (Graph C.30) and FIA transactions of the general government sector (Graph C.26). While there are few observations lying outside the acceptance regions, an analyst would need to exercise some caution in using the initial estimates of quarteronquarter movements in capital transactions of the general government and non-official sectors.