The Phillips curve
The proposed trade-off between unemployment and inflation. It has failed publicly several times and refuses to go away, because something like it keeps being partly right.
The original observation was empirical: in historical data, periods of low unemployment coincided with faster wage growth. The intuition is straightforward — when labour is scarce, employers bid up pay, and higher pay feeds into prices.
The policy conclusion, and its failure
Policymakers read the relationship as a menu: accept somewhat higher inflation, receive lower unemployment. Choose a point on the curve.
Stagflation was not supposed to be possible on the simple reading of the curve. The relationship that had held in the data broke precisely when policy started relying on it — which is a specific instance of a general problem: a statistical regularity exploited for policy often stops holding, because people's behaviour adapts to the policy.
The repair
The expectations-augmented version says inflation depends on expected inflation plus a term for economic slack. Workers do not bargain over nominal pay in a vacuum; they bargain for a real increase, which means they build their inflation expectation into the demand.
Policy can push unemployment below its sustainable rate for a while, because people are surprised by the inflation. Once they adjust their expectations, unemployment returns to where it was and inflation is permanently higher. The trade-off exists only in the interval between the policy and the adjustment — and repeated use shortens that interval, because people learn.
NAIRU
This gave rise to the non-accelerating inflation rate of unemployment: the level of unemployment consistent with stable inflation. Below it, inflation rises; above it, inflation falls.
The concept is useful and the number is not observable. It is estimated from the data, it moves with demographics, technology, institutions and bargaining power, and estimates are revised substantially after the fact. Policy has repeatedly been set against an estimate that later proved wrong in both directions.
The flattening
In the two decades before the 2020s, the relationship appeared to weaken considerably in advanced economies: unemployment fell a long way without inflation responding as expected. Candidate explanations include well-anchored expectations doing exactly the job they were designed to do, globalisation weakening domestic labour's bargaining position, and measurement issues in defining slack.
The subsequent inflation episode was then read by some as the curve reasserting itself in a tight labour market and by others as a straightforward supply shock with little to do with unemployment. That argument is not settled, and anyone claiming it is settled is arguing a position.
What to take from it
- Some relationship between slack and inflation almost certainly exists; its slope is unstable and its timing is unreliable.
- Expectations matter at least as much as slack, which is why central banks talk about credibility constantly.
- Any statement of the form 'unemployment below X means inflation' depends on an unobservable estimate.
- The deeper lesson is about macroeconomics generally: relationships estimated from history can break when policy starts to lean on them.