3.107. This section of the paper has reviewed the reliability of balance of payments statistics. The analysis has concentrated on the process of revisions to the statistics. In considering the revisions process, evidence of bias and dispersion has been assembled and differences between periodicities of the statistics and over time have also been identified.

3.108. The general conclusion is that initial estimates of most items are negatively biased (i.e. understate the latest estimate). This implies that generally positive revisions are required in later estimates. The pattern is similar for all periodicities - monthly, quarterly and annual statistics. Merchandise trade items and the capital transactions of the Reserve Bank are exceptions that are revised little and are generally free of bias.

3.109. Of the non-merchandise components of the current account, initial estimates of unrequited transfers are generally revised the least - in proportional terms and within the first year of revision by an average of less than 1% for all periodicities. Initial estimates of services credits and debits are less reliable and analysis of the revisions history for services credits shows significant upward revisions occurring some years after the initial estimate due to methodological improvements.

3.110. In terms of the average value of revisions, income debits are the least reliable of the current account components for all periodicities. This is also true in proportional terms except for annual estimates where revisions to income credits are particularly high, reflecting the revisions experience with reinvested earnings estimates.

3.111. For all periodicities, average oneyear revisions to the balance on current account are negative (i.e. revisions make the deficit on the current account larger). Subsequent revisions bring the estimate back down to a value close to the initial estimate. It is also interesting that in proportional terms, average one-year revisions to quarterly estimates of the balance on current account (4.2%) are larger than the revisions to both monthly (2.3%) and annual (2.0%) estimates of this balance.

3.112. Bias of capital account items is larger and more erratic than current account items. As noted, transactions of the Reserve Bank are virtually unbiased. Transactions of the general government sector are dominated by foreign borrowings and initial estimates of these are consistently negatively biased, with annual statistics experiencing the largest average revisions in proportional terms (6.7%). Initial estimates of non-official sector transactions are also strongly negatively biased (for both FIA and AIA) in quarterly and annual statistics.

3.113. Initial estimates of the current account deficit have tended to be relatively close to the final estimate, and initial estimates of movements in this balance have been relatively reliable and unbiased. On the other hand, the initial measures of net capital transactions are strongly negatively biased (implying an understatement of the surpluses) and the behaviour of revisions is erratic. This means that the initial estimates of the balancing item and subsequent sizeable revisions to it largely reflect deficiencies in measuring net capital account transactions.