Forecasting Performance

Macroeconomic forecasting performance

The Government's macroeconomic forecasts are prepared using a range of modelling techniques including macroeconometric models, spreadsheet analysis and accounting frameworks. These are supplemented by survey data, business liaison, professional opinion and judgment.

Forecasts are subject to inherent uncertainties. Generally, these uncertainties tend to increase as the forecast horizon lengthens. Forecast errors (the difference between forecasts and outcomes) can arise for a range of reasons — for example, differences between the assumed path of key variables and outcomes, as well as changes in the relationships between different parts of the economy.

Confidence intervals seek to illustrate that there is a range of plausible outcomes around any forecast. Confidence intervals are based on observed historical patterns of forecast errors. They are a guide to the degree of uncertainty around a forecast and can span a wide range of outcomes.

Real GDP forecasts

Real GDP forecasts in the Budget are based on a number of key assumptions including the exchange rate, interest rates and commodity prices. The forecasts also incorporate judgments about how developments in one part of the Australian economy affect other parts and how the domestic economy is affected by events in the international economy. The accuracy of the forecasts depends on the extent to which the assumptions and judgments underpinning them prove to be correct — and also the reliability of the economic relationships embodied in the macroeconomic models used to produce them.

For example, a lower exchange rate than assumed would be expected to result in higher than forecast growth in Australia's export volumes, including in tourism and manufacturing. At the same time, import prices would be higher, resulting in lower growth in import volumes. Overall, this would lead to a larger contribution from net exports to economic growth, although there would be some mitigating effect on real GDP from the impact of higher import prices on real household income.

Forecast errors for real GDP can also result from unexpected shifts in the pace or nature of economic activity during the forecast period. A faster than forecast pick‑up in Australia's economic growth in 2016‑17 could be driven by stronger consumer spending, underpinned by faster than forecast growth in employment, as activity gathers pace in the economy's labour‑intensive service sectors. Alternatively, faster economic growth could be driven by stronger than expected major trading partner growth, which could boost exports and, in turn, stimulate incomes and demand throughout the economy.

Over the past 20 years, Treasury's forecasts of real GDP growth have exhibited little evidence of bias, and accuracy has generally remained within a range of ½ to one percentage point (Chart 1). While forecasts of real GDP growth were less accurate in the years during and immediately after the global financial crisis (GFC), forecast errors seem to have since returned to the usual range.

National Accounts data are not yet available for the whole of 2015‑16. Information to date suggests that real GDP growth is evolving broadly in line with last year's Budget forecast, however there are offsetting results at the component level. Stronger growth in dwelling investment and exports are expected broadly to balance higher import growth and softer than expected business investment, particularly in the economy's non‑mining sectors. Other components of GDP, including consumption and mining investment, have so far evolved broadly as forecast in the 2015‑16 Budget.

Chart 2 shows that the average annualised growth rate in real GDP in the two years to 2016‑17 is expected to be around 2½ per cent, with the 70 per cent confidence interval ranging from 1¾ to 3½ per cent. In other words, if forecast errors are similar to those made over recent years, there is a 70 per cent probability that the growth rate will lie in this range.

Nominal GDP forecasts

Compared with real GDP forecasts, nominal GDP forecasts are subject to additional sources of uncertainty from the evolution of domestic prices and wages, and world prices for commodities.

Over the past decade, nominal GDP forecast errors have reflected the difficulties in predicting movements in global commodity prices (Chart 3). Faster than expected or assumed declines in the prices of key commodities in recent years — particularly for iron ore — have meant that nominal GDP was overestimated.

In 2015‑16, the outcome for nominal GDP growth is expected to be lower than forecast in last year's Budget. This primarily reflects weaker than expected wages and domestic price inflation, which has contributed to lower than forecast outcomes for the GDP deflator.

The confidence intervals around nominal GDP forecasts are wider than those around the real GDP forecasts, reflecting both the uncertainty over the outlook for real GDP and the added uncertainty about the outlook for domestic prices and commodity prices. Average annualised growth in nominal GDP in the two years to 2016‑17 is expected to be around 3½ per cent, with the 70 per cent confidence interval ranging from 2 to 4¾ per cent (Chart 4).

Fiscal forecasting performance

The fiscal estimates contained in the Budget are based on economic forecasts and projections as well as estimates of the impact of Government spending and revenue measures. Changes to the economic forecasts and projections underlying the estimates — for example, inflation, profits, wages growth, population and unemployment — will affect forecasts for receipts and payments. As such, this will have a direct impact on the profile of the underlying cash balance and government debt. Even small movements in economic forecasts and projections or outcomes that differ from the forecasts and projections can result in large changes to the budget aggregates, for example, decreasing payments or increasing receipts with flow‑on effects to the underlying cash balance.

Receipts

The Government's tax receipts estimates are generally prepared using a 'base plus growth' methodology. The last known outcome (2014‑15 for the 2016‑17 Budget) is used as the base to which estimated growth rates are applied, resulting in tax receipts estimates for the current and future years.

Most of the indirect heads of revenue, such as GST and fuel excise, are forecast by mapping the growth rate of an appropriate economic parameter directly to the tax growth rate in the relevant head of revenue. A number of income taxes also involve determining whether this tax will be paid in the year the income is earned, such as for pay‑as‑you‑go withholding tax, or in future years, such as for individuals' refunds.

Over the past two decades, receipts forecasts have both under‑ and over‑predicted outcomes (Chart 5).

Generally, there is a strong correlation between the accuracy of the forecasts of nominal GDP and its components and the forecasts for tax receipts. On average, economic forecast errors will be magnified in receipts forecast errors, owing to the progressive nature of personal income tax. Chart 6 plots the forecast errors for nominal non‑farm GDP against the errors for tax receipts excluding capital gains tax (CGT). It shows that where there has been an underestimate of nominal non‑farm GDP growth, tax receipts tend to be underestimated and vice versa.

The forecast for 2015‑16 tax receipts (excluding CGT) in the 2015‑16 Budget is expected to be an over‑estimate of around 2.4 percentage points, compared with an over‑estimate of around 0.8 percentage points for nominal non‑farm GDP growth.

The largest contributor to the expected forecast error in 2015‑16 is from the shortfall in company tax. In 2015‑16, company tax is estimated to be $3.5 billion (5.1 per cent) lower than expected in the 2015‑16 Budget. This is primarily driven by the fall in commodity prices in recent years, lowering profitability in the mining sector.

Another significant contributor to the expected forecast error for 2015‑16 is from gross income tax withholding which is estimated to be $2.6 billion (1.5 per cent) below the forecast of the 2015‑16 Budget as a result of lower‑than‑expected wages growth. Discussions of earlier years' forecast performance can be found in previous budgets.

From 2008‑09, forecast errors in tax receipts have been affected significantly by the economic downturn following the global financial crisis and, in particular, the impact on CGT (Chart 7).

Forecasting CGT is very difficult. Asset price movements above or below the assumption may cause CGT to differ significantly from the forecast. Further, CGT only applies to realised gains, so even if the asset prices are consistent with the assumptions, there may be more or less gains realised than was assumed.

Following the GFC, a large stock of capital losses were carried forward (see Box 2 of Statement 5 of the 2011‑12 Budget), and the utilisation of these losses continues to generate large uncertainties in both the timing and magnitude of the forecasts.

Chart 8 shows confidence intervals around the forecasts for receipts (excluding GST¹ and including Future Fund earnings). Confidence intervals constructed around the receipts forecasts exclude historical variations caused by subsequent policy decisions. These intervals take into account errors caused by parameter and other variations in isolation.

The chart shows that there is always considerable uncertainty around receipts forecasts and that this uncertainty increases as the forecast horizon lengthens. It suggests that in 2016‑17, the width of the 70 per cent confidence interval for the 2016‑17 Budget receipts forecast is approximately 1.8 per cent of GDP ($30 billion) and the 90 per cent confidence interval is approximately 2.9 per cent of GDP ($50 billion).

Payments

The Government's payments estimates are prepared by agencies that comprise the Australian Government general government sector. An assessment of payments forecasting performance is not included in this Statement. However, historical errors have been incorporated in estimated confidence intervals.

Chart 9 shows confidence intervals around payments forecasts (excluding GST). As with receipts estimates, historical policy decisions are excluded², and future policy decisions are out of scope. Payments estimates include the public debt interest impact of policy decisions.³

The chart shows that there is moderate uncertainty around payments forecasts. In 2016‑17, the width of the 70 per cent confidence interval for the 2016‑17 Budget payments forecast is approximately 0.8 per cent of GDP ($15 billion) and the 90 per cent confidence interval is approximately 1.2 per cent of GDP ($20 billion).

Payments outcomes can differ from forecasts for a number of reasons. Demand‑driven programs, such as payments to individuals and some social services, form the bulk of Government expenditure. Forecasts of payments associated with a number of these government programs depend on forecasts of economic conditions. For example, higher than forecast unemployment levels will mean that expenditure on related services, including allowances, will be higher than forecast.

Underlying cash balance

The underlying cash balance estimates are sensitive to the same forecast errors that affect estimates of receipts and payments. Confidence interval analysis shows that there is considerable uncertainty around the underlying cash balance forecasts (Chart 10).

In 2016‑17, the width of the 70 per cent confidence interval for the 2016‑17 Budget underlying cash balance forecast is approximately 2.1 per cent of GDP ($35 billion) and the 90 per cent confidence interval is approximately 3.4 per cent of GDP ($60 billion). In line with receipts forecasts, uncertainty increases over the estimates period.

---

1 GST was not reported as a Commonwealth tax in budget documents prior to the 2008‑09 Budget. As a result, GST data have been removed from historical receipts and payments data to abstract from any error associated with this change in accounting treatment.

2 The allowance for historical policy includes only new policy decisions made at each update. No allowance is made for other decisions, such as assistance for the impact of natural disasters or changes to the timing of projects announced in previous updates. These decisions will contribute to historical forecast errors and therefore increase the size of the confidence intervals around payments.

3 The impacts of past policy decisions on historical public debt interest through time cannot be readily identified or estimated. For this reason, no adjustment has been made to exclude these impacts from the analysis.