3. Data


To calculate confidence intervals, we compare the Budget forecasts from the 1998-99 Budget onwards with current outcomes to generate historical forecasting errors.8 Outcomes for nominal and real GDP are from the latest quarterly national accounts release for the June quarter 2013.

For the fiscal aggregates (receipts, payments and the underlying cash balance), there are some specific issues to bear in mind. First, the fiscal impact of all policy decisions made after a forecast was published have been added back to the relevant fiscal aggregate forecast to eliminate this potential source of error from the analysis. As a result, the root mean square errors for receipts, payments and the underlying cash balance do not reflect variations caused by subsequent policy changes. However changes in public debt interest as a result of policy decisions have not been removed from the data set.

Second, there was a change in the Budget reporting standard from cash to accrual in the 1999-2000 Budget. This change does not have a material impact on the underlying cash series. However, for many years in the sample, the Budget Papers published the effect of policy decisions only in accrual terms. As a result, we have constructed a composite series of policy decisions from records of cash and accrual measures, which has been used to adjust the relevant fiscal series.

Third, the potential for actual spending to exceed payment forecasts is taken into account through the Budget process. This is done through a provision made in the Contingency Reserve called the Conservative Bias Allowance (CBA). The CBA is an allowance for the tendency for expenses estimates of existing Government policy to be revised upwards in the forward years. This is of particular importance for demand driven programs where precise cost estimates are difficult to forecast. The allowance is calculated as a percentage of accrual expenditure and is unwound at each estimate update until it is removed completely when the year becomes the budget year. The reduction in the percentage over time reflects the fact that program estimates are progressively updated, thereby decreasing the bias. As such, the incorporation of the CBA into the budget updates is used to reduce bias and improve the accuracy of payment forecasts. (For further detail on the CBA see page 6–63 of the 2013–14 Australian Government Budget, Budget Strategy and Outlook.)

Fourth, abstracting from policy changes to construct confidence intervals around the fiscal variables does not encapsulate some cases where parameter variations have more in common with decisions of government, particularly concerning payment forecasts. For example, specific decisions to re-profile spending (due to changes in timing of projects) are captured as parameter variations. Similarly, the Natural Disaster Relief and Recovery Arrangements enable all new spending decisions relating to unprecedented natural disasters to be captured as parameter variations. We have abstracted only from variations that have been reported for budgeting purposes as policy decisions, and have therefore included errors that in reality are not driven by parameter variations in isolation. Further, in line with longstanding convention and due to inherent difficulties in forecasting future impacts of natural disasters, estimates of the impact of disasters are not included beyond the Budget year. The forecast errors that arise as a result of this treatment have not been removed from the data.

Fifth, the underlying cash balance and receipts series have been constructed from each published Budget update and then payments have been derived as a residual. Note that underlying cash balance data exclude Future Fund earnings, while receipts data do not. To ensure the residual series (payments) is consistent with the reporting of payments in budget documentation, Future Fund earnings have been removed from the residual data.

Finally, the incorporation of GST receipts and payments in the 2008-09 Budget Papers saw significant changes to the Government's balance sheet as this was not considered a Commonwealth tax in Budgets prior to that year. Consequently, GST data have been removed from receipts and payments data to abstract from any forecasting error associated with this change.

Further details on the data can be found in Appendix A.

8 In this analysis, the latest available estimate of the outcome is used as the measured outcome. Another approach could be to use a vintage of data close to when the forecasts were released. For example, Tulip and Wallace (2012) use GDP as reported in the fourth-published estimate (that is, released around four quarters after the event). Using later data vintages lessens issues around incomplete incorporation of source data but may increase issues from changing data definitions. However the use of forecast growth rates rather than levels in this paper should lessen the problems associated with changing data definitions (see footnote 12).