You Can't Exploit the Phillips Curve Twice

“Class 12,” by Tantri’s own count, and it opens with an unusually large claim for a Sunday-evening Zoom call: “If somebody asks you what economics does — is the real world useless and all — one thing which is definitely… used very, very widely by policy makers, and matters to your life, is this Phillips curve.” He then spends the hour showing how a generation of economists used it wrong, and what it cost the world to find out.

A random finding, and a good deal

The origin story is almost comically modest. “This was a random finding by a professor of LSE, in the 1950s or 1960s” — A.W. Phillips plotted unemployment against inflation and found a clean, negative relationship. Tantri is careful about what that finding actually was, because it’s the mistake nearly everyone makes on first contact: “This is not a causal argument… it’s just a relationship. Causal, we will discuss — that’s why we need theory. That is why we keep making this thing: correlation, not causation.”

What made it more than a curiosity was how well it held up: “This is something that has been put to test by thousands of people, by different types of data — the result holds… otherwise we wouldn’t be discussing in 2026, sitting here in India, something that somebody pointed out in 1960s.” And the shape of it was seductive. At high unemployment, the trade was generous — “it’s sort of flattish. You get a lot of decline in unemployment, with very little increase in inflation… you can print money, create a bit of inflation, and unemployment goes down significantly.” Push further, though, and “the deal becomes rotten after some time… the cost in terms of inflation becomes very, very high, at very low unemployment level. If you become really, really greedy, that is where inflation then really picks up and then goes out of the room.”

Phillips himself never explained why. “He did not theorize — he just said and left it, published a paper, maybe got promoted, and be done with it.” Explaining it fell to Paul Samuelson and Robert Solow, and how they explained it is the whole reason this session matters.

The dial

Samuelson and Solow had the machinery this course already built: the wage-setting and price-setting equations from Session 11, which pin down a natural rate of unemployment where nobody is surprised by prices. Run wage-setting through price-setting and — this is the derivation Session 13 walks through in full — out comes an equation linking inflation to how far unemployment sits from that natural rate. But the theory itself is only half the story; the more consequential half was what Samuelson and Solow concluded a central bank could do with it.

Their reading, in Tantri’s summary, treated the trade-off as a dial a government could simply set: “Central banks have the power to decide on whichever level of unemployment and inflation the economy can be in, and they can go back very easily.” Want low unemployment for a few years? Pay for it in inflation, then reverse the dial: “Let’s say 3 years in a row you want low inflation… 3 years in a row achieve an unemployment rate of 2%. And those 3 years, inflation will be 4%. Fair enough. But the fourth year… you increase interest rate, bring unemployment to 4%, inflation will come back to 2%. There is no permanent problem.” In their model, a low-unemployment binge was a bill you paid only while you were spending — cut it off, and the bill stopped too.

This was, for about two decades, the operating assumption behind a great deal of real policy. It is also, Milton Friedman pointed out, wrong.

Friedman’s objection

Friedman’s argument — delivered in a 1967 address to the American Economic Association, a talk Tantri still assigns as reading — has a name that generalizes far beyond macroeconomics: “This is called the Lucas critique… historical relationships work only so long as you don’t exploit them. The day you start exploiting historical relationships, they cease to work.” The Samuelson-Solow dial assumed that whatever workers and firms expected inflation to be, that expectation would just sit there while the central bank yanked unemployment around underneath it. Friedman’s question was blunt: “Who told you that expectations… is not God-given?”

His alternative was adaptive expectations — next year’s forecast is simply last year’s actual outcome: “Inflation expectation becoming adaptive means this πᵉ is equal to [last year’s π]… if this is constant, it is fine. Now what if this guy starts moving? That means last year’s this becomes this year’s this.” Once expectations move with the very inflation the dial produced, the dial stops being free.

Walk his numbers. Start in a stable 2% world — 2% inflation, unemployment at its 4% natural rate, nobody surprised. Cut rates, push unemployment down to 2%, and — exactly as Samuelson and Solow predicted — that year’s inflation rises to 4%. So far, no dispute. But now expectations have moved: “Even if you take interest rate back to neutral rate… inflation will remain 4%. It is not going to be 2%.” Do it again — cut rates, hold unemployment at 2% a second year — and “inflation this year will be 6%. And not only will be 6%, next year’s inflation expectation will be 6%.” Keep going for a third year and “your inflation will go to 8%.” Same policy, same unemployment target, every single year — and the bill keeps compounding, because each year’s actual inflation becomes next year’s anchor.

Reversing it is the part that actually hurt. Returning unemployment to the 4% natural rate doesn’t reset inflation — it only stabilizes it at whatever level expectations have reached, because at U = Uₙ the model says inflation simply equals its (now-elevated) anchor. To bring an 8% anchor back down, unemployment has to sit above 4% for several rounds running: push it to 6%, and inflation falls to 6% that year — but expectations only fall to 6% too, not to target. Hold rates high again, and it falls to 4%. Do it once more, and you finally reach 2%. “So you have to suffer so much recession… this is called hard landing.” And credibility, once spent, doesn’t come back for free either: “Credibility is not exogenously given… if market sees that you keep doing this every year, this is what happens.” It’s the same mechanism Tantri flagged in Session 9 about Indian state governments quietly borrowing against RBI accommodation — a government that keeps re-exploiting the relationship eventually finds the relationship stops cooperating.

The bill comes due

Friedman’s talk was 1967. The world ran the Samuelson-Solow dial anyway for most of the decade that followed, and “by late 70s, the inflation rate in the US became 13%” — helped along, he’s careful to note, by an actual crude oil shock layered on top (“we’ll discuss how supply shocks cause inflation in the next class”). Reversing it took Paul Volcker, who as Fed chair pushed the funds rate to roughly 14–15% by the early ’80s and delivered exactly the hard landing the model predicts was necessary: “Two years of recession, 1981 and ‘83 — two recessions, back-to-back, in the US.” By the late ’80s, expectations finally came back down to 2%, and the U.S. spent the next two to three decades in what’s remembered as the Great Moderation — the reward, per Tantri, for a central bank that had finally rebuilt the credibility it spent in the ’70s.

The honest coda

The theory has one more twist Tantri won’t paper over. In 2007, crude oil hit roughly $147 a barrel — in today’s terms, “more than $200” — and U.S. inflation barely moved, because “the whole mechanism was, expectation remained well anchored.” By the mid-2010s, inflation had been so quiet for so long that “people had started thinking Phillips curve is dead, Lakshit.” His own honest answer for why: nobody fully knows. One leading explanation is China — an undervalued currency and low wages exporting disinflation to the rest of the world for the better part of a decade, an effect that faded as Chinese wages and its currency both rose. Post-COVID, the anchor broke again: U.S. inflation touched 9%, and even now, with tariffs back in the headlines, it’s “hovering close to 2.7, 2.8%,” with recent Fed research suggesting close to 90% of tariff costs are landing on U.S. consumers rather than being absorbed abroad. “These are not physics equations,” he reminds the room — the mechanism is real, but the exact number always has an error term nobody has fully priced.

Try it

A country has held unemployment 2 points below its natural rate for three straight years, and inflation has ratcheted from 2% to 8% as expectations adapted each year. The new central bank governor wants 2% inflation back. What does Friedman's model say she has to do?

What You Can Now Do

Next time a policymaker promises a “one-time” cost — a growth sugar-rush, a subsidy, a rate cut framed as temporary — ask what happens to the number the year after. If the honest answer is “it becomes the new normal that everyone plans around,” you’re looking at a ratchet, not a dial, and the exit costs more than the entrance did. And when someone declares an old economic relationship dead because it’s gone quiet for a decade, hold two possibilities open at once, the way Tantri does with the Phillips curve: it might genuinely have changed, or the world might just be borrowing against a credibility it hasn’t finished spending yet.

Next: /2026/04/the-course-so-far-medium-run-to-long-run/ — the same curve, formally derived end to end, run live against an actual 2026 supply shock.