US inflation held at 3.4% in August, pushed by a gasoline price surge rather than classic demand overheating — and new central-bank research says the years of "the Phillips curve is dead" commentary were measuring the wrong variable all along. Production cost, not the output gap, is what actually explains inflation. That single substitution changes how this letter reads every inflation print from here.
For most of the last decade, "the old inflation models don't work anymore" has been treated as settled fact in financial commentary. Unemployment fell to historic lows in the 2010s with barely any inflation to show for it; then 2021-2023 delivered a inflation spike with unemployment still low the whole time. Two clean misses from the textbook framework in one decade is a fair reason to distrust it. What gets missed in that story is a body of newer research suggesting the framework was never actually wrong — it was pointed at the wrong input.
US headline CPI rose 3.4% year over year in August 2026, matching July's pace, with the monthly gain accelerating to 0.4% from July's 0.1% — a jump driven primarily by a surge in gasoline prices rather than a broad-based pickup across the consumption basket. Core inflation, which strips out food and energy, came in slightly above forecasts, though the specific figure was not disclosed in the reporting this letter could verify. The Federal Reserve responded on September 16 by raising its benchmark rate 25 basis points to a range of 3.75%-4.00% — the first hike since 2023, explicitly reversing a rate-cutting cycle that had been underway since September 2024 — citing the need for "a timelier return to the Committee's 2 percent" inflation target even while describing the broader economy as "expanding at a solid pace."
A rate hike is a demand-side tool: it works by making borrowing and spending more expensive, cooling the broad economy until price pressure eases. The proximate driver named in this month's CPI report — a gasoline price surge — is a supply-side or energy-market event, not evidence that the broader economy is overheating. If the underlying driver of this inflation print is closer to an energy shock than to excess demand, a rate hike calibrated to cool demand may do comparatively little to address the actual source of the price pressure, while still imposing the borrowing-cost pain that comes with tighter policy. This letter cannot resolve from the available reporting how much of August's inflation was energy-specific versus broader, and neither, on the evidence available, can a simple reading of the headline number alone.
The textbook Phillips curve ties inflation to labour-market slack: tight unemployment should push wages and then prices higher; loose unemployment should do the reverse. Researchers spent much of the 2010s documenting that this relationship had gone "flat" — inflation barely responded to the unemployment rate or the output gap at all, which fed directly into the "these rules are outdated" narrative now common in financial commentary.
A Federal Reserve Bank of New York research note published this February reframes the entire debate. The author's argument, in short: the Phillips curve did not go flat — the *proxy* researchers were using for economic slack did. Using the unemployment rate or the output gap as a stand-in for how much cost pressure firms actually face turns out to be a weak link; substitute real marginal cost — what it actually costs a firm to produce its next unit of output — and the relationship between that cost measure and inflation is steep, not flat, even during the years commentators were calling the framework dead. In manufacturing data used for the study, production-cost fluctuations alone accounted for roughly 70% of observed inflation variation, with a cost-based model's inflation series correlating above 80% with actual producer price inflation.
| Slack Measure Used | Apparent Phillips Curve Shape | Explains Inflation? |
|---|---|---|
| Unemployment rate / output gap (conventional) | Flat | Poorly — the "broken model" story |
| Real marginal cost (new approach) | Steep | ~70% of variation explained |
The lesson is not that inflation is unpredictable or that the old economics was wrong. It is that unemployment was always an indirect, lagging stand-in for the thing that actually drives prices — and a decade of "the model is broken" commentary was really a decade of using the wrong dial to read a system that was working roughly as advertised the whole time.
The cost-based Phillips curve research cited here uses Belgian manufacturing data and a specific methodology; it is a genuine, serious academic contribution, not yet a fully validated real-time policy tool the Fed has adopted as its primary framework. The Fed's own September decision leaned on a fairly conventional demand-cooling rate hike despite citing "elevated" inflation, suggesting policymakers themselves have not yet fully operationalized a marginal-cost-based approach even if the research increasingly supports one. And a single month's gasoline-driven CPI print is not enough evidence to declare the inflation episode either transitory or entrenched in either direction.
"The old inflation rules don't work" is the wrong lesson from the last decade. The more accurate lesson is that the old rules were being tested against the wrong input — unemployment and the output gap — rather than the cost pressures that actually drive prices. For positioning, that argues for favouring pricing-power and cost-structure analysis over headline labour-market watching, and for treating this month's gasoline-driven CPI print as a data point about energy markets specifically, not yet confirmed evidence of the broader demand overheating a rate hike is designed to cool.