For the better part of a decade, macroeconomists have wrestled with an uncomfortable observation: the empirical relationship between unemployment and inflation—the Phillips curve—appeared to have vanished. Unemployment rates fell to multi-decade lows across advanced economies, yet inflation barely stirred. Commentators declared the Phillips curve dead. Central bankers quietly questioned whether their core framework still held.

But declaring the Phillips curve deceased confuses a change in appearance with a change in structure. The relationship between real economic slack and price pressures hasn't disappeared. It has been reshaped—by the very success of monetary policy regimes, by the globalization of factor markets, and by nonlinearities that standard linear estimation was never equipped to detect. The Phillips curve didn't flatline. It was compressed, masked, and pushed into corners of the parameter space where conventional regressions struggle to find it.

Understanding where the Phillips curve is hiding matters enormously for policy design. If the trade-off is genuinely extinct, central banks can pursue indefinitely low unemployment without inflationary consequences—a seductive but dangerous conclusion. If instead the curve is dormant, waiting to reassert itself under specific conditions, then the policy calculus changes dramatically. Recent experience with post-pandemic inflation suggests the latter interpretation deserves far more weight. What follows examines three structural forces that have obscured the Phillips curve and what modern research reveals about its true shape.

The most elegant explanation for the Phillips curve's apparent disappearance is also the most ironic: credible inflation targeting mechanically flattens the observed relationship between slack and inflation. When a central bank successfully anchors inflation expectations near its target, movements in unemployment generate smaller deviations in inflation. The curve doesn't vanish from the structural model—it vanishes from the reduced-form data.

Consider the New Keynesian Phillips curve in its canonical form. Inflation depends on expected future inflation, a measure of real marginal cost (proxied by the output gap or unemployment gap), and a cost-push shock. When expectations are firmly anchored at the target, the expectations term becomes approximately constant. The remaining variation in inflation is driven almost entirely by marginal cost fluctuations and shocks. But here's the subtlety: if monetary policy also responds aggressively to output gaps—as Taylor-type rules prescribe—then the policy reaction itself compresses the variance of the gap. You end up with small movements in both variables, and the econometrician sees a flat scatter plot.

This is what Michael Woodford and others have termed the identification problem of good policy. The structural slope parameter connecting slack to inflation may be unchanged, but the observed correlation collapses because policy is endogenously stabilizing both sides of the relationship. Researchers at the Federal Reserve Bank of Minneapolis have shown that once you properly instrument for this endogeneity—using structural VAR techniques or narrative monetary policy shocks—the Phillips curve slope recovers meaningful magnitudes.

The implications are profound. A policymaker who interprets the flat empirical Phillips curve as permission to run the economy hot indefinitely is committing a Lucas critique error in real time. The flatness is conditional on the policy regime. Change the regime—allow expectations to drift, communicate ambiguously, or delay tightening—and the curve's slope can re-emerge rapidly. The post-2021 inflation episode in multiple advanced economies arguably demonstrated exactly this dynamic.

What looked like a permanent structural change was, in significant part, an artifact of successful policy. The Phillips curve was hiding in plain sight, suppressed by the very framework designed to exploit it. Recognizing this distinction is essential for any central bank calibrating how much slack it can tolerate before inflationary pressures materialize.

Takeaway

A flat observed Phillips curve may reflect the success of monetary policy, not the absence of the trade-off. Confusing a policy-conditional outcome with a structural parameter change is one of the most consequential errors a central bank can make.

The second force obscuring the domestic Phillips curve operates through the increasing integration of goods, services, and labor markets across borders. The global slack hypothesis, advanced prominently by Claudio Borio and colleagues at the Bank for International Settlements, argues that domestic inflation is increasingly influenced by global output gaps rather than purely domestic unemployment conditions. If this is correct, the domestic Phillips curve slope appears flatter not because the structural link between slack and prices has weakened, but because the relevant measure of slack has shifted.

The theoretical mechanism is straightforward. In an open economy with integrated supply chains, the marginal cost facing domestic firms depends on global input prices, foreign wage pressures, and the competitive dynamics of internationally traded goods. A tight domestic labor market may exert upward pressure on wages, but if global supply capacity remains abundant—as it did through much of the 2010s with China's continued integration into manufacturing networks—imported disinflation can offset domestic cost pressures. The domestic unemployment rate becomes a noisier signal of the inflationary pressure actually experienced by price-setting firms.

Empirical evidence on the global slack hypothesis is genuinely mixed, and intellectual honesty requires acknowledging this. Kristin Forbes's influential work finds that global factors—commodity prices, exchange rates, and global output gaps—have become increasingly important determinants of domestic inflation in open economies. However, other researchers, including staff at the European Central Bank, find that the domestic output gap retains explanatory power once you control properly for supply shocks and commodity price movements. The debate remains unresolved.

What seems increasingly clear is that the answer is both. Global and domestic slack are not competing explanations but complementary channels. The appropriate specification likely involves a multi-factor Phillips curve where domestic unemployment, global output gaps, import price dynamics, and sector-specific supply conditions all enter with time-varying coefficients. The flattening of the simple bivariate relationship between domestic unemployment and domestic inflation partially reflects omitted variable bias—we were estimating a misspecified model all along.

For central banks, this complicates the transmission mechanism considerably. Monetary policy primarily operates on domestic demand. If a significant share of inflationary variation is driven by global factors beyond the reach of domestic interest rate adjustments, the Phillips curve that policy can exploit is genuinely flatter than the structural relationship. This doesn't eliminate the trade-off—it contextualizes it within a more complex, multi-dimensional framework that single-equation estimation was never designed to capture.

Takeaway

When we measure the Phillips curve using only domestic unemployment, we may be looking at the wrong thermometer. The relevant slack driving inflation has become partially global, making the domestic relationship appear weaker than the underlying structural link actually is.

Perhaps the most consequential insight from recent research is that the Phillips curve is not a single slope but a convex function—relatively flat when inflation is low and unemployment is near or modestly below its natural rate, but substantially steeper when the economy pushes into extreme overheating or when inflation expectations begin to de-anchor. Linear estimation, which dominated empirical work for decades, averages across these regimes and returns a coefficient that understates the curve's true steepness in the tails.

The theoretical foundations for nonlinearity are well-established. In models with downward nominal wage rigidity, firms resist cutting nominal wages even when slack is abundant, compressing the disinflationary response to recessions. But when demand is strong, wages adjust upward more freely. This asymmetry generates a kink in the Phillips curve around full employment. Similarly, models incorporating state-dependent pricing—where firms adjust prices more frequently in high-inflation environments—predict that the pass-through from marginal cost to prices accelerates as inflation rises. The curve steepens precisely when it matters most.

Empirical work by Harding, Lindé, and Trabandt at the International Monetary Fund has formalized this using piecewise-linear and smooth-threshold Phillips curve specifications. Their results are striking: in the United States, the slope of the Phillips curve is approximately three times larger when unemployment falls more than one percentage point below the natural rate compared to periods of modest slack. The curve that appeared flat during the long expansion of 2010–2019 was simply operating in its flat regime. The inflationary surge of 2021–2023, coinciding with historically tight labor markets, was the curve reasserting itself in its steep regime.

This nonlinearity resolves much of the supposed puzzle. For most of the post-Great Recession period, advanced economies operated with unemployment near or above estimates of the natural rate. Inflation hovered around target—exactly what a flat-in-the-middle Phillips curve would predict. The mistake was extrapolating that flatness into all possible states of the world. When labor markets tightened beyond a critical threshold, the dormant slope activated with force that caught many forecasters off guard.

The policy design implications are asymmetric and urgent. In normal times, central banks have meaningful room to tolerate modest deviations of unemployment below the natural rate without triggering significant inflation. But this room is finite, and the boundary is not visible in real time using standard gap estimates. Once crossed, the inflationary response can be swift and disproportionate. Optimal policy under a nonlinear Phillips curve is therefore more cautious near full employment than a linear specification would suggest—not because the curve is always steep, but because the consequences of being wrong about where the threshold lies are highly asymmetric.

Takeaway

The Phillips curve behaves less like a constant slope and more like a sleeping predator—docile when the economy is near equilibrium, but capable of rapid, disproportionate response when pushed past a threshold that is difficult to observe in real time.

The Phillips curve never died. It was compressed by successful inflation targeting, diluted by global integration, and concealed by the linear lenses through which we insisted on viewing it. Each of these forces is real, but none of them implies the permanent elimination of the inflation-unemployment trade-off.

What modern research reveals is a relationship that is regime-dependent, multidimensional, and nonlinear—far richer and more complex than the simple downward-sloping line that textbooks once promised. The policy framework must evolve accordingly, incorporating state-dependent expectations management, global slack indicators, and threshold-aware reaction functions.

Central banks that internalize these lessons will be better equipped for what comes next. The curve is always there, waiting in the structure of the economy. The question is never whether it exists, but whether we are looking for it in the right places—and whether we are prepared for the moment it wakes up.