Previously I argued the government gave the RBA a flawed framework in 2023. Underneath that framework sits a misreading of one key macroeconomic concept: the NAIRU.
What the NAIRU is
The NAIRU is the non-accelerating inflation rate of unemployment. It is the jobless rate at which inflation holds steady. Push unemployment below it and inflation rises. Let unemployment run above it and inflation falls. Economists write it as u*.
Turn that around. If inflation is holding steady, at whatever level, unemployment should be sitting at u*. For a given labour market, once expectations have caught up, steady inflation at 2 per cent and steady inflation at 6 per cent point to the same jobless rate. Permanently higher inflation does not buy permanently lower unemployment. Over time and across nations, if anything, high inflation is correlated with a higher u*, not a lower one.
It is a structural rate. The way the labour market works sets it: how wages are set, the skills people bring, how well workers match to jobs, the tax and transfer system, and how easily people move between jobs and places. Monetary policy is not on that list. The Reserve Bank sets the cash rate. It does not set u*.
u* is a constraint, not a target. Nobody chooses it. The bank cannot pick it, and the government can shift it only slowly, through the institutions that set it.
The name carries the argument
The word that matters is accelerating. In the traditional expectations-augmented account, when unemployment is below u*, inflation does not settle at a higher level. It climbs. Firms struggle to hire, workers gain bargaining power, and wages and prices rise faster than people expected. That surprise holds unemployment down for a while. But people learn. Wage claims and price setting build in the higher inflation. To keep unemployment below u*, the bank has to keep surprising them, with inflation rising ever faster. That cannot last.
Inflation targeting spread from the early 1990s and anchored expectations. Modern models reflect that, and soften this account. If expectations stay anchored, a hot labour market lifts inflation without sending it into a spiral. But the anchor is exactly what is at risk. Run the labour market hot for long enough and expectations drift. Then you are back in the traditional account.
Many people read the Phillips curve as a long-run trade-off. It is only ever a short-run one. It looks permanent for a reason. Under inflation targeting, expectations stayed anchored, so a little overheating produced a little inflation that soon faded. The curve looked like a stable menu. But that stability was borrowed from the regime. Treat the curve as a permanent menu and you erode the anchor that made it look permanent. Policymakers made that mistake in the 1960s, and the 1970s collected the bill.
In the short run, u* is effectively fixed, and monetary policy should treat it that way. You can borrow jobs from the future for a year or two, at the price of higher inflation. You cannot reliably buy a lower u* with it.
Silent on the level
The NAIRU has no view on the level of inflation. It only says whether inflation is steady, speeding up or slowing down. At u*, inflation heads toward whatever rate people expect. That could be 2.5 per cent or 6 per cent or 10 per cent.
So reaching u* does not, by itself, bring inflation back to target. If expectations sit at 2.5 per cent, inflation gets there. If they have drifted up, it stalls above the RBA's target. Then getting inflation down means unemployment above u* for a while, unless credible policy brings expectations down. That is the credibility route, and it is the cheaper one.
That looks like Australia now. Monthly unemployment near 4.7 per cent in September 2026 is inside my range of u* estimates, so on my numbers the labour market is near balance. Yet trimmed mean inflation has climbed back from near 2.7 per cent in 2025 to around 3.7 per cent. Persistent inflation without a tight labour market is consistent with drifting expectations. If that is the cause, balance alone will not fix it.
Where the 2023 RBA framework goes wrong
The 2023 Review of the Reserve Bank of Australia asked for a more explicit dual mandate. The new Statement on the Conduct of Monetary Policy delivered it: the timeframe for returning inflation to target should, where necessary, balance price stability and full employment. That is a short-run trade. The bank can buy a smaller jobs gap now by bringing inflation down more slowly. But the jobs "saved" are temporary and the price may not be. Inflation that lingers shifts expectations, and shifted expectations are expensive to reverse.
The transition to the more explicit dual mandate was not clean. The Review was commissioned in July 2022, as the tightening began, and reported in April 2023. Its direction was clear well before the Statement was signed that December. Lowe could see the writing on the wall. The bank was weighing jobs more heavily before the words were formalised, so the slow path began under the old framework and was ratified by the new one.
Temporary jobs still have value. The question is what they cost. If expectations hold, a slow return to target is cheap. If they drift, the jobs protected at the start are lost at the end, and more besides. In 2023, a slow return may have looked reasonable. In 2026, with inflation now above target for years, the situation is much more fragile.
Balancing jobs and inflation does not have to mean choosing the slow path. But the pressures lean that way. Job losses are visible now and land on identifiable people. The cost of drifting expectations is diffuse and arrives later. Ministers complain about the first and rarely about the second. So a balancing mandate keeps betting that the anchor holds.
Full employment, properly defined, is u*: the most employment the economy can sustain without accelerating inflation. The bank serves that goal by keeping inflation low and stable, and by keeping unemployment close to u* through the cycle. The Statement keeps the bank's definition close to u*. But it sits beside the government's aspirational definition, where everyone who wants a job can find one without searching too long. And when Treasury puts u* in the low 4s, a given jobless rate looks less inflationary and the case for patience looks stronger. Two definitions, a balancing clause and a wishful estimate of u* all lean the same way. Together they invited the bank to treat u* as softer than it is.
If government wants a lower jobless rate, it has the levers. Skills, training, job matching, wage-setting institutions and the welfare system all shape u*. Those are fiscal and structural levers, and they belong to the government. Asking the central bank to deliver it asks the wrong institution to do it with the wrong tool.
Estimated, but never observed
Nobody observes u*. We infer or estimate it using macroeconomic models. The simplest method treats it as the low-frequency component of the unemployment rate. On average over the business cycle, the unemployment rate (u) should oscillate around the NAIRU. So we can draw a smooth line through the unemployment data, with an HP filter or a LOESS, and call the line u*.
However, a filter on its own is not enough. It assumes unemployment averages out at u* over the window, so it cannot see a long stretch of overheating. Run the labour market hot for a few years and the line bends down to meet it. Adding inflation to the macroeconomic model fixes much of this. A Phillips curve, with wages and other indicators alongside, lets the data pull the estimate back up when the labour market runs hot. The bands still widen at the end of the sample. But the model builds in slow movement, because the institutions behind u* change slowly. Unless something changes how the labour market works, the trend should hold.
My own four specifications, estimated to the June quarter 2026, sit between 4.6 and 4.9 per cent. That spread shows how much the answer depends on the model specification. It is not a confidence interval.
That is the heart of the problem. Inflation is measured every month. u* is inferred, with wide bands and room for honest disagreement. Orphanides and Williams showed that when u* is poorly measured, policy that leans on the unemployment gap inherits the error in the estimate. Policy that responds to inflation, and to changes in unemployment rather than its gap from u*, holds up much better. They also found that underestimating how wrong your estimate might be costs far more than overestimating it, and they used natural-rate misperceptions to help explain the stagflation of the 1970s. The balancing clause puts more weight on the part of the calculation we measure worst. It moved the bank from the easy to identify and fix to the hard to identify and easy to mistake.
A slow anchor beneath the cycle
Over time, unemployment oscillates around u*. Booms push it below. Downturns push it above. The cycle is where monetary policy works.
Over the medium term, u* barely moves. The institutions that set it typically only change slowly: wage-setting arrangements, awards and bargaining, training, the welfare system, the way people find work. Reform takes years to pass and more years to bite. Nonetheless, over a decade or two, u* can shift a long way. Australia's fell substantially after inflation was brought under control in the 1990s. A large shift over two or three years needs an explanation.
That gives a useful test. When an estimate of u* falls sharply over a few years, ask what changed. After COVID, estimates fell into the low 4s, as if the pandemic had reset how the labour market works. I have not seen a convincing account of what that change was. Wage setting, skills, matching and the welfare system looked much as they had in 2019. Zac Gross, whose academic work on the Australian NAIRU goes well beyond mine, does not find the fall either. His u* drifts down through the 2010s to about 4.4 per cent, sits there through 2016 to 2019, then climbs toward 4.9 after the inflation surge. I put it nearer 5 in the late 2010s, a gap inside his uncertainty bands. I tried to replicate his estimates but could not reproduce his line, and the shortfall may well be mine. My own estimates face the same identification problem. The question I would put to his post-COVID rise is the same one: what changed? A u* that climbs with an inflation overshoot may be absorbing changes in expectations, and institutions rarely move that fast.
 |
| Source: Zac Gross, nairu.isaacgross.net |
Immigration is an odd actor here. Unlike wage setting or skills, it can swing hard in a year or two, and Australia went from closed borders to record inflows in quick succession. New arrivals take time to find work, which can lift u* for a while. A larger effect comes if the intake does not match what the labour market needs. Then vacancies and unemployment rise together, and u* moves up for as long as the mismatch persists. That may explain part of the recent rise. A well-matched intake could push the other way. Either way, it is a government lever, not the bank's.
Easier to push up
Monetary policy is better at pushing u* up than down. Once inflation becomes entrenched, bringing it down can take a costly contraction. Long recessions leave people in long-term unemployment, where skills and attachment to work erode. Some never come back. That is how the jobless rate ratcheted higher from the mid-1970s to the early 1990s.
The case for the other direction is weaker. Some argue a tight labour market pulls marginal workers in and keeps them there, nudging u* down. Perhaps, at the margin. But the evidence is thin, and it is not a lever the bank can pull with any confidence. Scarring from recessions is far better documented.
Across countries and decades, high inflation and a high u* tend to travel together. Which drives which is unclear. Friedman argued in 1977 that volatile inflation clouds price signals and makes the labour market work worse. Run the arrow the other way and a high u* tempts governments to push for easier policy, so inflation follows. Or the same rigid institutions, such as the indexed wage system of the 1970s and 80s, raised both. Correlation is not causation, and Australia offers only one long swing up and one long swing down. But none of these stories runs from higher inflation to a lower u*.
None of this makes policy risk one-sided. Hold policy too tight with unemployment above u* and the bank does the damage itself. That was the 2016 to 2019 error Gross and Leigh identified. But on my reading, the cure for that was a symmetric target, not a balancing clause. The asymmetry is in what monetary policy can do to u*, not in which mistakes are possible, and it runs the wrong way for a balancing mandate. Protecting jobs today by going easy on inflation cannot reliably lower u*. It can raise it.
A softer approach to inflation can work while expectations are firmly anchored to a low target. Then a slow return costs little, and the labour market is spared a correction. But everything rests on that anchor. Once expectations start to drift, as they now appear to be, gentleness stops working. It cannot pull expectations back to target. What remains is tough medicine, usually a recession, to force inflation down and re-anchor expectations. That does far more damage to jobs than taking inflation seriously in the first place. The soft path does not avoid the cost. It defers it, enlarges it, and delivers it as the recession we had to have.
No comments:
Post a Comment