
Australian Government Budget documents present forecasts for key economic and fiscal variables. Estimates of uncertainty around such forecasts can help convey to readers a better appreciation of the risks associated with the economic and fiscal outlook. For example, they can help inform readers about how likely it is that outcomes will be close to the forecasts. This can be useful for informing government policy and public discourse more generally. For example, confidence intervals can highlight the amount of adjustment required to meet a budget target should particular risks materialise, throwing into sharper relief the trade-offs that government may face.
Estimates of forecast uncertainty can also improve the credibility and transparency of the forecasting process, key topics in the Review of Treasury Macroeconomic and Revenue Forecasting (Treasury Forecasting Review) (Treasury, 2012). Explicit estimates of uncertainty can aid in making clear that point forecasts may turn out to be incorrect and that forecasts may be more usefully considered as a range rather than a point estimate. Being explicit about inherent uncertainties may lead to fewer misunderstandings about the forecasts and what they represent.
A number of fiscal agencies provide estimates of uncertainty around their forecasts. This includes the US Congressional Budget Office (CBO), which publishes a fan chart of probabilities around its projections for the budget balance (as a share of GDP). The UK Office for Budget Responsibility (OBR), in its Economic and Fiscal Outlook, publishes measures of uncertainty around its central projections for real GDP growth, public sector net borrowing (as a share of GDP) and cyclically-adjusted budget balance (as a share of GDP). The New Zealand Treasury reports confidence intervals around its revenue forecasts and cyclically-adjusted budget balance forecasts as part of New Zealand Budget papers (New Zealand Treasury, 2013).3 For Australia, the 2013 Pre-election Economic and Fiscal Outlook (PEFO) also presented confidence intervals around key aggregates; the first time this has been done.4 Table 1 describes some measures of forecast uncertainty provided by a range of fiscal institutions.
Table 1: Some measures of uncertainty published by fiscal institutions
| Institution | Variables reported | Measure of uncertainty | Method of construction |
|---|---|---|---|
| Congressional Budget Office (US) | Budget balance. | 10, 20, 30, …, 90 per cent confidence intervals. | Based on historical forecast errors, assuming normally distributed errors. |
| Office for Budget Responsibility (UK) | Real GDP growth, public sector net borrowing and cyclically-adjusted budget balance. | 20, 40, 60, 80 per cent confidence intervals. | Based on historical forecast errors, assuming modified normal distribution. |
| NZ Treasury | Core crown tax revenue and cyclically-adjusted balance. | 20, 40, 60, 80 per cent confidence intervals. | Based on historical forecast errors, assuming normally distributed errors. |
Note: The approach adopted below uses root mean square errors of past forecast errors to construct confidence intervals in line with the NZ Treasury. More detailed information about what is reported can be found in the original sources.
Sources: CBO (2007a), CBO (2007b), OBR (2012), OBR (2013), New Zealand Treasury (2013), Parkyn (2010).
Publishing measures of uncertainty around economic forecasts is also common practice among central banks around the world. The Reserve Bank of Australia (RBA), the US Federal Reserve, the European Central Bank (ECB) and others report measures of uncertainty around their GDP and inflation forecasts (see Table 2 below). In the Australian context, confidence intervals around RBA forecasts have been published in an RBA discussion paper (Tulip and Wallace, 2012) and have been subsequently reported in the RBA Statement on Monetary Policy.
Table 2: Some measures of uncertainty published by monetary institutions
| Institution | Variables reported | Measure of uncertainty | Method of construction |
|---|---|---|---|
| Reserve Bank of Australia | Real GDP growth and underlying CPI inflation. Unemployment rate and CPI inflation also reported in discussion paper. | 70 and 90 per cent confidence intervals. | Based on historical forecast errors, assuming symmetric errors. |
| Bank of Canada | CPI inflation and core inflation. | 50 and 90 per cent confidence intervals. | Combination of historical forecast errors and model errors, assuming normally distributed errors. |
| Bank of England (UK) | CPI inflation and level and growth of real GDP. | 30, 60 and 90 per cent confidence intervals in some charts. Detail varies over time and within publication. | Based on forecast errors, modified for judgement. |
| Bank of Japan | GDP growth and CPI excluding fresh food. | Range and confidence intervals. | Based on forecast distributions of individual Policy Board members. |
| European Central Bank | Real GDP growth and its components and Harmonised Index of Consumer Prices inflation. | Range. | Twice the past mean absolute projection errors, with outliers excluded. |
| Federal Reserve (US) | Real GDP growth, unemployment rate and inflation. | Root mean square errors. Also participants' range of forecasts and judgement of risk and uncertainty. | Root mean square errors based on historical forecast errors. Also presents views of FOMC participants. |
| Sveriges Riksbank (Sweden) | Real GDP growth, CPI and core CPI inflation and re-purchase rate. | 50, 75 and 90 per cent confidence intervals. | Based on past forecast errors, assuming normally distributed errors. |
| Norges Bank (Norway) | Policy interest rate, output gap and CPI and core CPI inflation. | 30, 50, 70 and 90 per cent confidence intervals. | Based on model, assuming a normal distribution, constrained by zero lower bound for the policy interest rate. |
Note: The approach adopted below uses root mean square errors of past forecast errors to construct confidence intervals in line with the Riksbank. More detailed information about what is reported can be found in the original sources.
Sources: Bank of Canada (2009), Bank of Canada (2013), Bank of England (2007), Bank of England (2013), Bank of Japan (2008), Bank of Japan (2013), ECB (2009), ECB (2013), Federal Open Market Committee (FOMC) (2012), Kjellberg and Villani (2010), Norges Bank (2013), Sveriges Riksbank (2007), Tulip and Wallace (2012).
In addition, some institutions, like the UK Office for Budget Responsibility and the European Central Bank, provide a comparison of forecasts from a number of institutions, making differences in views between different forecasters more transparent. This potentially serves as a basis both for a better appreciation of the difficulties in forecasting and for discourse about why forecasts differ.
This paper presents estimates of uncertainty around key Budget parameters, similar to those developed by other institutions. After discussing the theory behind our approach in section 2 and data issues in section 3, we use recent forecast errors to construct confidence intervals around the 2013–14 Budget forecasts, presented in section 4. These confidence intervals highlight the important point that there has always existed a range of plausible alternative outcomes around any given point estimate.
3 In New Zealand, the Budget forecasts in the Economic and Fiscal Update are supplied to the Minister of Finance by the New Zealand Treasury, while the Australian Budget forecasts are produced by the Australian Government after receiving the advice of the Australian Treasury and Department of Finance.
4 The Charter of Budget Honesty Act 1998 requires the Secretaries to the Treasury and the Department of Finance to release a PEFO prior to an election.