The Federal Budget papers contain a number of charts, tables and forecasts. The forecasts are the part most people focus on. They are also the part most likely to be misunderstood.
When the Reserve Bank publishes a forecast, it is trying to predict the path the economy will take. The Bank looks at the data, consults its models, applies its judgement, and prints a number it thinks is most likely to happen. When the budget papers publish a forecast, they are doing something very different. They start with where the economy is today and assume it returns to something like normal over the medium term. The forecast then asks a simple costing question. Under normal economic conditions in the out years, what would revenue and expenses look like?
The distinction matters. The budget forecasts are not predictions. They are an arithmetic exercise built on top of an assumed normal economic equilibrium. The numbers that define that equilibrium, things like the medium-term GDP growth rate, the wage and price assumptions, the unemployment anchor, are framework conventions. They are not Treasury's literal central case for the path the economy will take. But, if the equilibrium is wrong, every number that flows from it is wrong too.
This year's papers contain an equilibrium that does not add up.
The reversion always looks the same
Start with real GDP. The black line is what actually happened. The orange lines are successive Treasury forecasts. Almost every forecast bends back toward 2.5 to 3.5 per cent growth, regardless of where the economy actually is. That end point was higher in the 2010s and it is lower now.
This is the equilibrium speaking. Treasury does not really believe growth will be 2.5 per cent in 2029 and 2030. It believes the speed limit (the speed at which the economy can grow without excessive inflation) is 2.5 per cent under normal conditions. The forecast is the speed limit, not a prediction. Treasury is itself saying the economy's speed limit is lower today than in the past.
The nominal GDP chart shows the same shape from a different angle, and explains why the bottom line in every budget gets revised.
Actuals run well above forecasts almost every year of the past decade. That's not bad forecasting. That's deliberate conservatism on the terms of trade, which feeds straight into company tax receipts.
The terms of trade forecasts do the same thing the GDP forecasts do. Every vintage assumes the growth rate reverts to a mildly negative number, around minus three to minus five per cent, regardless of where the series actually is. When commodity prices are booming the forecast sits well below the actual. When they are collapsing the forecast sits above it. It is not conservatism. It is the same reversion to a fixed assumption, applied to a series that almost never sits at the assumed value.
The revenue consequence is asymmetric, which is why this one is easier to forgive. When the actual comes in above the assumption, as it did through 2021 and 2022, revenue surprises up and the budget bottom line improves. Under-calling a windfall is the politically safer error. So on the terms of trade the reversion machine happens to err in the comfortable direction more often than not. That is luck about which way the series broke, not a considered judgement about commodity risk.
Revenue forecasting was roughly unbiased through the 2010s. The forecast vintages track the actual closely, sometimes slightly high. There is no standing conservatism here.
The expenses side does the opposite
If revenue forecasts lean low in the 2020s, expense forecasts lean low too, but for a different reason. The assumed equilibrium is one in which policy decisions stop being made.
Each forecast vintage shows expenses growing gently along a smooth line. The actual line keeps stepping up as new programs land. NDIS, defence, aged care, and the structural drift in health and social security all appear at their current-policy trajectory, and the actuals then exceed it because new policy keeps expanding them. They also grow faster when program controls prove weaker than anticipated, as NDIS has shown repeatedly.
Post-2020, expenses have come in 10 to 25 per cent above where forecasts placed them three years earlier. The misses on expenses are smaller in percentage terms than the revenue misses, which is why the budget tends to surprise to the upside in the near term and disappoint over time.
Social security and welfare gives the cleanest view.
Every forecast vintage is a less-upward-sloping line relative to the actuals. The forecast assumes slower payment growth. It rarely is.
The other big spending programs are much the same: Health, Education and Defence.
The supply side does not cohere
This is where the equilibrium story starts to fall apart. In the Budget, the medium-term real GDP number sits at 2.5 per cent. That's the speed limit. Speed limits are built from three pieces: how fast the working-age population grows, how many of those people work, and how much each worker produces. Growth in the hours worked by the working age population, the participation rate and labour productivity adds up to growth in potential GDP.
Look at population. The measure that matters for the speed limit is not headline population. It is the civilian population aged 15 and over, because that is the pool the participation rate draws on. The chart shows civilian 15-and-over growth running near 1.8 per cent today, with net overseas migration at around 310,000. The budget assumes migration falls to around 225,000. Since migrants are overwhelmingly working age, that reduction takes roughly 0.4 points off working-age population growth on its own, bringing it down to about 1.5 per cent.
As noted above, migration forecasts step down from the post-pandemic surge to around 225,000 a year by the end of the decade. That's broadly consistent with the pre-pandemic average.
Now look at the labour market.
Participation is predicted to grow a very small amount. The unemployment rate is also expected to fall just a touch. So the labour market does almost nothing in the out years.
Employment growth in the forecasts sits a little above working-age population growth. That wedge is a slightly higher participation rate and a slightly lower unemployment rate. Both are one-off level shifts. Participation cannot rise forever and unemployment cannot fall forever, so over the medium term employment growth has to converge back to working-age population growth, around 1.5 per cent. The speed limit is a medium-term concept, so it is the converged number that matters.
Do the arithmetic on the medium term. Working-age population growth settles at around 1.5 per cent once migration normalises. Participation is flat, unemployment is at the NAIRU, hours per worker are flat, so employment growth converges to that same 1.5 per cent. For GDP to grow at 2.5 per cent, labour productivity has to deliver around 1.0 per cent a year. The problem is that 1.0 is roughly double what we have actually been getting.
Productivity is the only flexible variable left in the system. Population is broadly fixed by demographics and migration policy. The labour market in the forecasts barely moves. Whatever GDP number comes out the end is overwhelmingly determined by what you assume about productivity. So it is worth looking at productivity properly.
The productivity story
Labour productivity is output per hour worked. The blue line below shows what's happened to it since 1980.
The line rises through the 1980s and 1990s, slows in the 2000s, and goes broadly sideways from about 2015. The bump in 2020 to 2022 is an artefact of the pandemic. Low-productivity service jobs in hospitality and retail disappeared during lockdowns, which mechanically lifted the average. As those jobs came back, the average fell again.
Strip out the pandemic and labour productivity has grown by about half a per cent a year over the past decade. Maybe less, depending on where you start the clock. The roughly 1 per cent implied by the budget is the most generous reading of recent history, not a central one.
The orange line is more damning. Multifactor productivity is what you get when you take labour productivity and strip out the contribution from extra capital per worker. It is the closest thing economists have to a measure of pure efficiency: are we getting more output from the same combination of labour and capital? The answer over the past twenty years, outside short cycles, is broadly no. The orange line peaked in 2004 and has been broadly flat since.
This matters because the only reason labour productivity has risen at all since 2015 is that businesses kept adding capital. Each worker has more machines, more software, and more buildings to work with than they did a decade ago. That is what has been doing the work, not any underlying improvement in how efficiently those inputs combine.
So when the budget assumes labour productivity growth of around 1 per cent, it is implicitly assuming either that capital deepening continues at the same pace as the past decade, or that multifactor productivity finally turns. The investment forecasts say capital deepening is slowing. And nothing in twenty years of data suggests multifactor productivity is about to turn.
A more honest central estimate of trend productivity is probably 0.4 to 0.6 per cent. If that is right, then with the population path Treasury has, the coherent speed limit is closer to 1.9 or 2.0 per cent. Not 2.5. Of note, this is the consensus view. The Reserve Bank's February Statement puts potential output growth at around 2 per cent. The Commonwealth Bank independently estimates potential growth at 2.1 per cent and frames it, as I have, as a speed limit. Deloitte Access Economics has 1.9 per cent. The IMF has 2.1. Treasury's 2½ is the outlier.
One possible challenge is that the productivity rebound is policy-driven. The budget's productivity package claims $13 billion in long-run GDP, but Chalmers' speech is explicit that this figure comes from National Competition Policy reforms "underway with states and territories", not from Commonwealth spending. The package leans heavily on state cooperation that has not yet been delivered.
Supporters will argue the reforms leverage above their direct size. Two responses. First, the leverage is already embedded in Treasury’s estimate. The $13 billion figure is not a mechanical accounting saving. It is a modelled GDP impact that already incorporates assumed dynamic effects. Second, the historical examples of genuinely transformative productivity reform, the original National Competition Policy being the canonical case, involved structural changes that removed economy-wide constraints: monopoly utilities, professional restrictions, agricultural marketing boards, public sector trading enterprises. The current package borrows the NCP label but the substance is mostly administrative and regulatory reform: faster approvals, compliance simplification, Digital ID, AI adoption in the public service, and targeted investment incentives. These are real measures, but they are not of the same order of magnitude.
The package may help at the margin. It cannot do the work the equilibrium needs it to do.
The wages and prices side picks the other story
It gets worse. The wages forecast sits at 3.25 to 3.75 per cent.
CPI sits at 2.5 per cent indefinitely (the government's inflation target).
The gap between wages growth and inflation, over sustained periods, is mostly productivity. Labour share movements and terms of trade effects can drive a wedge for a while, but real wages cannot persistently outrun productivity growth. If wages grow 1 per cent faster than prices for a decade, workers' real incomes are rising at 1 per cent a year, and the only way that can be paid for sustainably is if each worker produces 1 per cent more per hour.
So the wages and prices forecasts assume productivity growth of roughly 1 per cent. The GDP and employment forecasts, taken together, also imply around 1 per cent. Neither sits comfortably with the 0.4 to 0.6 the actual data supports. Both halves of the equilibrium independently require trend productivity to roughly double, and nothing in the labour market or investment detail explains where that doubling comes from.
None of these forecasts are talking to each other.
The demand-side detail says nothing has changed
If productivity were genuinely about to rebound, you would expect to see it in the investment numbers. New capital per worker is the main channel through which productivity actually rises. Non-mining investment is the closest the budget papers get to this.
Forecasts park at 2 to 3 per cent growth. That is a capital deepening pace consistent with the productivity track we have had, not the one we would need. Mining investment is essentially flat too.
Household consumption forecasts at 2 to 2.25 per cent are unambitious.
None of the demand-side detail is consistent with the supply-side number you would back out from the wage-price gap. If you believed in the productivity rebound, you would mark up business investment. The forecasts do not.
Public final demand is the giveaway
The clearest sign that the equilibrium is artificial is public final demand. Each forecast vintage assumes government spending growth steps down to around 1.5 to 2 per cent. The actuals run two to three times that.
This is not a mistake. It is a feature of the costing exercise. Treasury cannot forecast policy decisions that have not yet been taken. So in the out years, the spending profile reverts to a passive baseline. Politics then adds spending each year and the actuals drift back above the forecast.
Private final demand does the opposite. Forecasts assume the private sector picks up the slack.
In practice, the private handover rarely arrives on schedule, and the public sector fills the gap. The composition of growth in the budget papers is not the composition we tend to get.
What it adds up to
Put the pieces together and the underlying cash balance projection looks like this.
Deficits stabilising around 1 per cent of GDP. Manageable. The cash balance line is constructed from revenue forecasts that assume conservative commodity prices, expenses forecasts that assume no new policy, GDP forecasts that assume a productivity rebound, and labour market forecasts that assume there isn't one. Each component is defensible in isolation. As a package, they are incoherent.
Net debt then grows steadily.
The forecast line slopes up gently. The net debt line is sensitive to the bond yield assumption. The chart shows debt at market value, so the fall from 2020 to 2023 partly reflects yields rising and revaluing existing bonds down. The forecast trajectory assumes yields broadly hold. If they rise, the immediate effect is to flatter the headline net debt number while raising public debt interest in the budget. For this reason gross debt is also worth keeping an eye on.
Where this leaves us
The reversion to equilibrium framework is fine as a framework for costing purposes. On the terms of trade it happens to err in the politically safe direction, which is forgivable. And Treasury genuinely cannot cost policy decisions that have not been taken.
What is not fine is the equilibrium itself. The package on the supply side is internally inconsistent. The GDP speed limit assumes productivity at around 1 per cent. The actual data, once you strip out the pandemic, says trend productivity is closer to 0.5. Multifactor productivity has been flat for twenty years. The labour market detail assumes no rebound. The wages and prices forecasts independently imply the same 1 per cent. So both halves of the equilibrium need trend labour productivity to double. The investment forecasts assume nothing is changing.
A coherent version of these numbers would do one of two things. It would accept that trend productivity is around 0.5 per cent, mark the speed limit down to about 2.0, and accept that wages cannot sustainably grow much faster than prices. Or it would commit to a genuine rebound, mark business investment up sharply, raise the speed limit, and explain where the rebound is going to come from after twenty years of nothing. Doing both at once is having it both ways.
The deficits in the chart above are calculated on the having-it-both-ways assumption. Half the equilibrium that produces them is doing work the other half denies is happening. That is not a forecast error. It is an equilibrium whose components cannot all be true at once.
No comments:
Post a Comment