Online I meet a lot of post-Keynesian economists who open with "I don’t believe in a NAIRU." Bill Mitchell has made one of the most sustained cases against it. I remain unconvinced. The NAIRU (u*) rests on two ideas. The first is the Phillips curve: when the labour market is tight, inflation rises. That is easy to show in the data, over time within a country and across countries. I showed it for Australia in The New Keynesian Synthesis, Explained. The second, from Friedman and Phelps, is expectations: wage and price setters build the inflation they expect into their decisions. Put the two together and you get the NAIRU, the unemployment rate at which inflation neither speeds up nor slows down, but settles at what people expect.
To reject the NAIRU, you must reject the Phillips curve, reject the role of expectations, or argue that the equilibrium they imply is not economically meaningful. The data make the first two hard to reject. The third is a question of hysteresis. Hysteresis is real. It made the NAIRU hard to read in the 1990s, when unemployment and u* were moving at once. But since then u* has moved slowly, from about 6 per cent in 2004 to just under 5 per cent now, and my four model specifications have converged to within a third of a point. They share data, so agreement is not proof. But that is not how a meaningless equilibrium behaves.
So why does Mitchell reject it? His reasons run deeper than the data. In theory, he holds that employment is set by spending: firms hire to produce what they expect to sell, and there is no unique supply-side unemployment rate to which the economy must eventually return. In practice, he points out that the estimate is latent and moves to fit whatever inflation does, so no outcome can ever count against it. He argues that hysteresis lets u* follow unemployment wherever demand takes it, and that the unemployment rate understates idle labour because it misses the underemployed. Like many post-Keynesians, he sees inflation as a conflict over income shares, settled by bargaining power and institutions more than by slack. And underneath sits a political objection. In Full Employment Abandoned, written with Joan Muysken, the NAIRU is the device that let governments abandon full employment, in the sense that anyone who wants a job can get one. Governments retreated from that commitment after the stagflation of the 1970s, recast a large slice of unemployment as structural, and later handed macro policy to independent central banks. Each objection has some force. None, in my view, is enough.
One counter position is simple: inflation targeting works. Before the RBA adopted it in 1993, Australia had stagflation, unemployment above 10 per cent in two recessions, and repeated boom-bust cycles. After it, inflation settled in or near the band, and the economy ran for almost three decades without a recession until the pandemic. Unemployment fell from 11.2 per cent at the end of 1992 to 4.0 per cent in early 2008. The NAIRU fell with it. Mitchell reads the NAIRU as the device that abandoned full employment. The record suggests the opposite: price stability came first, and lower unemployment followed.
I am a pragmatist about models. I ask three things of one. Does it fit the data over the long run? Does it inform policy? And does it describe the economy rather than prescribe what the economy should be? On those tests, the New Keynesian synthesis wins. Not because it is right in every detail, but because nothing else on offer does the job.
Estimation is the problem, not existence
Nobody observes the NAIRU. It is a latent variable, backed out of the data with a model. The estimates have wide confidence bands. They get revised. They move over time. Critics treat all of this as evidence that the thing does not exist. It is not. It is evidence that the thing is hard to measure. Potential output (Y*) and the neutral real interest rate (r*) face the same problems, and few people deny those exist. Nor is it unfalsifiable. My models let u* move only slowly, so it cannot chase every inflation surprise. And in 2022 the framework faced a test it could have failed. Unemployment fell well below u*, which should add to inflation. Supply shocks were at work too, so it was not a clean test. But in my decomposition, supply led the surge and demand kept it going. Demand added more than a percentage point at the peak and was still adding to inflation long after the supply shocks had faded.
Nor is u* invariant. It is a product of structures within a nation: the wage-setting regime, benefits, bargaining, competition in product markets. Those structures change slowly, so u* moves slowly. Every intervention we make in the economy changes something, and deep recessions leave scars. But a u* that moves is not a u* that is missing.
What is striking is how close the modern NAIRU is to the post-Keynesian theory of inflation. In the Layard, Nickell and Jackman model, the NAIRU comes out of a tug of war. Firms set prices as a mark-up on wages. Workers bargain for wages as a mark-up on prices. If the two sides together want more than the economy produces, they chase each other, and the result is inflation. Unemployment ends the chase by weakening workers’ hand until the claims fit. The Cambridge economist Bob Rowthorn set out the same mechanism in 1977, and his conflict theory of inflation has been a post-Keynesian staple ever since. The disagreement is far narrower than the rhetoric suggests. Both schools describe the same fight. They differ on who should referee it. The mainstream leaves it to the central bank, which settles the claims after the event by letting unemployment weaken workers’ hand. Post-Keynesians want the claims settled in advance, by agreement: an incomes policy like the Accord, coordinated wage bargaining of the Nordic kind, or, in Mitchell’s case, a Job Guarantee wage. Each depends on institutions that can make the deal stick: unions that can deliver their members, and a government that can deliver its side.
A constellation of invisibles
The strength of the New-Keynesian synthesis is not any single estimate. It is a suite of invisibles: r*, u* and Y*. Their levels can move independently. A productivity gain lifts Y* without touching u*, and global saving can shift r* with no bearing on the domestic labour market. But the gaps tend to move together and give the same message. When unemployment sits below u*, output is usually above Y*, and policy needs to sit above r*. That is especially true when inflation is outside the target band. Each draws on partly different data. u* leans on wages, prices and unemployment. Y* leans on capital, hours and productivity. r* leans on output and real rates. When they agree, that is useful. When they disagree, that is informative: something has shifted, and the bank has to work out what.
They also discipline the story. A central bank that says the labour market is tight while holding policy at neutral has to explain why it is not leaning against the tightness. Focusing on any one of them is a false precision trap. Taken together, they turn a dashboard of noisy indicators into a view on whether policy is tight or loose.
DSGE is not the synthesis
Much of the progressive critique is aimed at the DSGE model. DSGE stands for dynamic stochastic general equilibrium. These models build the economy up from the optimising choices of households and firms, often a single representative household, and then hit it with random shocks. They are an easy target. Many estimated DSGE models are fragile. Parameters are weakly identified, so the priors do much of the work. Fit comes from piling on unobserved shocks. Small changes in assumptions produce large changes in results.
But DSGE is one way of expressing the synthesis, not the only way. The RBA does not set rates with a representative agent and an Euler equation. Its main model, MARTIN, is semi-structural, with estimated equations. The Fed's FRB/US works much the same way. My own models are state-space models with slow-moving latent states, not DSGE.
All models simplify. Pointing out a simplification does not by itself identify a weakness. The question is whether the model drops something that matters for the job it is asked to do. Central banks run many models, debate what the differences mean and why they arise, and then take cautious steps. The suite is the framework. Knocking down one model does not touch it.
Expectations explain the missing acceleration
Post-Keynesians point out that inflation has not accelerated for decades when unemployment sat below estimates of the NAIRU. That is true. But the acceleration depended on expectations adjusting to realised inflation. Before central banks adopted inflation targeting, underlying inflation expectations ratcheted up with each burst of inflation, and that is what the accelerationist Phillips curve described. Australia broke that cycle a decade after the United States. Expectations sat around 7 to 8 per cent through the 1980s and only came down in the early-1990s recession. Inflation targeting, from 1993, then anchored them. With expectations held at the target, the curve becomes a level relationship. Running the labour market tight lifts inflation somewhat. It does not set off a spiral. Anchoring is not the only reason the curve flattened. Weaker bargaining power and global competition played a part. But the missing acceleration is what the framework predicts, and the US state-level evidence attributes most of the stability since the 1990s to firmer anchoring, not a steeper curve. Nor is anchoring permanent. If inflation runs above target for long enough, people notice and expectations move. In Australia, short-run expectations already have.
The mainstream absorbed the wins
The critics have won real arguments. Money is created endogenously by bank lending. A government that borrows in its own currency need never default for want of the means to pay, though it can still choose to. Hysteresis is real, even if its size is disputed. Minsky was right about the instability of credit. Conflict over income shares matters for inflation.
The mainstream has taken most of these on board, in its own terms, and without a jot of the post-Keynesian or MMT core. Endogenous money went into the textbooks without changing the policy model. Blanchard derived his r < g argument with standard tools. Financial frictions entered the models after the GFC, though Minsky's endogenous instability remains at the margin. Lorenzoni and Werning have formalised conflict inflation in mainstream terms. What the mainstream rejected was everything that would have required rebuilding the institutions: fiscal policy as the main stabiliser, inflation control through tax, a job guarantee as the anchor.
The MMT core failed its test
In 2020, some MMT advocates said the world would now come to accept their model. Governments ran huge deficits. Central banks bought the bonds. The RBA ran yield curve control. For a while it looked like MMT in practice.
Then came the test. MMT's case is that the binding constraint is inflation, and that fiscal policy handles it when it binds. In 2021 and 2022 it bound. No major government raised taxes to contain it. Central banks raised rates, and the old order reasserted itself almost overnight. MMT advocates can fairly reply that nobody implemented their framework. But it was a political test, not a theoretical one: when inflation bound, would any government raise taxes to fight it? None did. And not just any tax would do. Taxing the rich tends to lower their saving. Taxing the middle and lower end lowers spending. A tax rise that fights inflation has to land where people spend. That is the tax rise no government wants to make, and the one progressives least want to propose.
Central banks were slow. To be fair, the supply shock was huge and unexpectedly so, and they were still assuming slack. Australia was not alone in looking through the early shocks. That was (perhaps) a mainstream misjudgement, not a failure of the framework. Once inflation broadened, central banks moved hard, and for several years expectations largely held, as the framework predicted.
The interest rate is a blunt instrument. But it works because it is in the hands of an apolitical, independent body. A central bank can move rates eight times a year, and it does not face an election. Tax and spending can change between budgets, and automatic stabilisers work all the time. But discretionary fiscal tightening goes through parliament, with long lags and every incentive to delay. The question for MMT is not whether taxes reduce demand. It is whether fiscal adjustments can be made promptly, credibly and consistently enough to anchor expectations. Post-Keynesians, who mostly look to incomes policy instead, would point to the Accord. But the Accord rested on centralised wage fixing and union density around half the workforce. Density is now about an eighth and bargaining is decentralised. Neither school has shown how to turn a mass of noisy data into a timely, credible policy decision.
Underneath it is values
Ask me the core disagreement with Mitchell and I would say it is a values preference. He accepts that inflation constrains employment. His own framework has an inflation-safe ratio of buffer jobs to total employment. He differs on how the constraint arises and how to manage it. For him, full employment is a social right, and inflation is a constraint to be managed with other tools, so that jobs come first. I would accept full employment in his sense, a job for anyone who wants one, if I could be convinced it came without an inflation cost. I have not been.
Why put inflation first? Not because a point of inflation is worse than a lost job. For the household that loses it, unemployment is far worse. But inflation falls on everyone, and hardest on those with the least power to keep up: wage earners whose pay lags prices, and people on fixed incomes. Australian workers took a large real wage cut in 2022 and 2023 and have spent years winning it back. Inflation also blurs the price signals that guide saving and investment, and when it is unexpected, it shifts wealth from savers to borrowers by accident, not design. Nor does it buy jobs. Across countries and decades, high inflation has gone with high unemployment, not low, a point Friedman made the theme of his Nobel lecture. Its costs are spread thin, so they are easy to discount. They are no less real for that.
I share the goal of full employment: the highest employment that can be sustained without destabilising inflation. The NAIRU is its benchmark, and price stability is how you achieve and sustain it. The problem is treating a short-run trade-off as a long-run one. The jobs-inflation trade-off exists only in the short run, and the short run is where it does damage. The lesson of the 1990s is that once inflation is entrenched, restoring price stability can carry large and lasting employment costs. Inflation fell quickly. The labour market took most of the decade to recover. Half-hearted inflation fighting does not protect jobs. It risks postponing the losses and making them bigger. That is why the values argument does not deliver what Mitchell and like-minded colleagues would hope for. Go soft on inflation and you end up with neither the price stability nor the jobs.
That is my problem with the RBA's balanced jobs-inflation mandate. The dual mandate is not the problem. Flexible inflation targeting already weighs activity alongside inflation and recognises a short-run trade-off between them. The problem is the balancing clause when inflation is persistently above target. By putting a near-term rise in unemployment on an equal footing with inflation, it encourages the bank to weigh that cost too heavily. The short-run trade-off becomes a reason to delay restoring price stability. It invites compromise and a narrow path while inflation drifts. The bank cut three times in 2025. The last cut came after quarterly underlying inflation, at an annualised rate, had moved back above the band, and the rate then sat at 3.6 per cent until February. This year the bank has had to raise it by 125 basis points, 50 of them in the past two months. Even at 4.85 per cent, it sits below the 5.5 per cent that a standard textbook Taylor rule, using my estimate of r*, called for in September. The bank risks getting the wrong end of both: higher inflation now and higher unemployment later.
Where we are now
Short-run inflation expectations have climbed back to about 3 per cent, the top of the target band, and trimmed mean inflation is running above them. Unions are asking for higher wages to match. Claims are not settlements, but enterprise agreements run for three or four years, so whatever is settled now will lock in wage growth well into the future. With productivity growth close to zero, wage growth of about 2.5 per cent is consistent with the target unless profit margins compress. Wage growth is running at 3.2 per cent and has sat above that benchmark since 2022. Wages ran below it from 2013 to 2021, so some catch-up was due. But part of that shortfall reflects how the Wage Price Index is built. It prices a fixed set of jobs, so by design it excludes classification creep, the steady drift of workers into higher pay grades. Settlements built on expectations of 3 per cent or more would push wage growth further from where it needs to be.
On Twitter, post-Keynesians still argue for more employment. They resist the evidence that expectations are moving and blame oil and the Middle East. But in the September quarter, fuel rose 7.8 per cent, the second-largest rise of the 87 expenditure classes. It sat in the trimmed tail, so the trimmed mean excluded its direct effect. Some indirect effects, through transport and distribution, will have got through. They cannot explain the breadth of the pressure in the middle of the distribution. Rents, new dwellings, child care, dental services, hairdressing, takeaway food and recreational services all rose between 1.0 and 1.4 per cent in the quarter, an annualised pace of roughly 4 to 6 per cent. Household expectations do track petrol prices, so the oil shock may well have helped push expectations up. That does not change the remedy. What matters is not where the shock started, but whether it has become persistent domestic inflation through wages, margins and expectations. It has.
The labour market has loosened. Unemployment has risen from about 4 per cent at the start of 2025 to 4.7 per cent, which puts it at my estimates of u*. Growth tells the same story. GDP grew 2.1 per cent over the year to June, about my estimate of potential, and the past two quarters ran below it. The gaps agree: the tightness is going. But inflation has not come back to target, and expectations have moved up. Unless productivity recovers or the oil shock unwinds quickly, getting the rest of the way needs unemployment above u*. That is the cost the narrow path deferred. A switch is coming, though. Once expectations have re-anchored, with unemployment above u*, more employment becomes the right goal again. The disagreement is about order, not direction.
They are also arguing for half of their own model. In the conflict-inflation approach, high employment can be sustained if competing claims are reconciled without inflation. That theory says that when combined claims exceed output, inflation persists. Productivity growth could make room, but it is close to zero. Margins could give way, and some critics argue the profit share is high enough to absorb wage claims. But margins can only compress so far, and leaning on them is no reliable way to anchor inflation. That leaves an incomes policy, and there is no institution that could deliver one. Bill Mitchell's answer is the Job Guarantee: government offers a job at a fixed wage to anyone who cannot find one, and that wage anchors the rest. The buffer is meant to grow in downturns and shrink as private employers hire workers away. But a fixed wage stays fixed only until inflation erodes it. Then the pressure to lift it is political, and the anchor moves.
The buffer has its own problem. A Job Guarantee trades the value of the work against its suitability as a buffer. At one end are ongoing services such as aged care, disability support and community care. They produce real value, but withdrawing them when private demand recovers hurts the people who depend on them. Political pressure will favour replacing the workers who leave. At the other end are activities that are easily suspended, the proverbial rock-painting jobs. They use labour, materials, supervision and administration for little return, and the workers doing them could be searching for jobs or training instead. The scheme needs a large stock of work in the middle: productive, scalable, close to where the unemployed live, and able to stop without disrupting anything valuable. Infrastructure maintenance and environmental work come closest. But even they need planning, equipment and supervision, and they cannot be switched on and off quickly. A project abandoned halfway because the private labour market tightened is its own kind of waste. If the jobs are indispensable, they are hard to withdraw. If they are dispensable, their value is doubtful. Government can plainly create jobs. Creating useful jobs that expand and contract with the cycle is much harder than the theory allows.
The Job Guarantee is meant to adjust automatically, without fresh decisions. But keeping it working asks governments to resist constant pressure: to lift the wage, to keep valued services, to replace the workers who leave. Like the fiscal lever, it depends on governments acting against their own incentives. Without a credible anchor, more employment means accommodation. On their own theory, that keeps the conflict going.
Why I still lean New Keynesian
So I lean New Keynesian, though not because the framework is right in every detail. Its stars variables are hard to see, its DSGE versions are fragile, and its critics have won real arguments. I lean on it because it fits the data, it informs policy, and it comes with an anchor that works: an independent central bank that can move quickly, and has. Mitchell’s alternatives may work on paper. Each needs governments to do what they have rarely done: raise taxes to cool a boom, hold a wage down while prices rise, let valued services shrink. Until that changes, the referee stays where it is.
No comments:
Post a Comment