What Is the Phillips Curve? Inflation, Unemployment, and Its Limits

The Phillips curve is a family of models and empirical relationships, not one universal line. Its familiar classroom form shows a negative short-run association between inflation and unemployment: when labor markets are tight, firms may compete more for workers, wages and costs may rise faster, and businesses may pass some of those costs into prices. When demand weakens and spare capacity grows, price and wage pressure may ease. That is a possible pattern, not a rule that must hold in every month or country.

The name comes from A. W. Phillips’s 1958 study of the United Kingdom, which documented a relationship between unemployment and the rate of change in money wages over a long historical sample. The original question concerned wage inflation, not every modern measure of consumer-price inflation. Later economists extended the idea to price inflation and added expectations, supply costs, and measures of economic slack. The Federal Reserve’s account of the curve’s history emphasizes the role of resource use and expectations in those later versions. {source:fedKuglerInflationExpectationsPhillips2025}

That distinction matters when reading a chart. A graph of unemployment against wage growth does not measure the same outcome as a graph of unemployment against CPI or PCE price inflation. A model may also use the output gap, vacancies, or a broader resource-use measure rather than the unemployment rate alone. The title “Phillips curve” identifies a family resemblance; it does not specify the data or equation.

One explanation starts with gradual price and wage adjustment. Some employers reset pay or prices only periodically, because renegotiating contracts, updating systems, or communicating a new price takes time. If demand strengthens before every price has adjusted, firms with unchanged prices may meet more orders and hire more workers. Firms that do adjust can raise prices. This combination can produce higher inflation alongside lower unemployment for a time. Federal Reserve descriptions of the framework treat nominal stickiness as one mechanism, not the only explanation. {source:fedPhillipsUnstableCurve2023}

Labor-market slack is broader than the number of people counted as unemployed. An employer may face hiring difficulty, more workers may be changing jobs, hours may be increasing, or productivity may alter how much output wage growth translates into unit labor costs. Those details affect how demand reaches costs and prices. Unemployment is useful partly because it is measured regularly, but it is only a proxy for the pressure on the economy’s productive resources.

The curve does not say that every person who finds a job causes prices to rise. It describes an aggregate relationship that may be relevant to wage setting, business pricing, and forecasts. A firm can absorb a cost increase in its margin, offset it with higher productivity, or pass it to customers; households can change spending; and policymakers can react. Each response changes how a labor-market shift appears in measured inflation.

A simple expectations-augmented version can be written as:

Inflation = expected inflation − β × (unemployment rate − reference unemployment rate) + supply shock

Here, β is a positive coefficient in this sign convention. If unemployment is below the reference rate, the gap in parentheses is negative, so the minus sign adds upward pressure, all else equal. The reference rate is an estimate of the unemployment rate consistent with stable inflation in a model; it is not directly observed and should not be treated as a fixed number known with certainty. Some models instead use the output gap or another measure of slack. Expectations and supply terms also vary by model. {source:frbsfPhillipsCurveBasics2008}

Consider a hypothetical illustration, not an estimated U.S. relationship: expected inflation is 2.0%, the reference unemployment rate is 4.5%, actual unemployment is 3.5%, β is set to 0.4 percentage point of inflation per one percentage-point unemployment gap, and the supply-shock term is zero. The gap is 3.5% − 4.5% = −1.0 percentage point. The slack contribution is −0.4 × (−1.0) = +0.4 percentage point, so the equation gives 2.0% + 0.4% = 2.4% inflation.

Now keep the unemployment gap and β unchanged, but assume expected inflation is 3.0% and a hypothetical supply shock adds 1.2 percentage points. The same equation gives 3.0% + 0.4% + 1.2% = 4.6%. The example shows why an unemployment number alone cannot determine inflation: expectations and other cost or supply forces also matter. The coefficient and inputs are invented solely to demonstrate the arithmetic; they are not a forecast, a fitted parameter, or current data.

A short-run Phillips curve is drawn for a given level of expected inflation and other conditions. If expectations rise, the short-run curve can shift upward: the same unemployment rate may coexist with faster inflation. If expected inflation falls, the curve may move down. The relationship therefore describes combinations under specified assumptions rather than a stable exchange rate between two policy goals. The Federal Reserve’s discussions stress that expectations influence wage demands, business prices, borrowing terms, and purchase timing. {source:fedKuglerInflationExpectationsPhillips2025}

In many standard models with sticky nominal wages or prices, the long-run curve is vertical at a sustainable or “natural” unemployment rate. The intuition is that workers and firms eventually incorporate the ongoing inflation rate into wage and price decisions. Once expectations adjust, a permanently higher inflation rate does not keep unemployment below its sustainable level. This is a model result under assumptions, not a claim that any particular natural rate is precisely known or that long-run debates are closed. {source:fedPhillipsUnstableCurve2023}

“Natural,” “equilibrium,” and “non-accelerating inflation” rates are estimates with uncertainty and can move as labor-market institutions, demographics, matching efficiency, or productivity change. They are not a direct count of available workers and not a target that can be read from one month’s unemployment report. If an analyst treats the reference rate as known, the calculated unemployment gap can look more precise than the underlying evidence warrants.

A supply shock changes production costs or the amount an economy can produce at given prices. A sudden increase in imported energy costs, for example, can raise prices paid by businesses and households while squeezing output and hiring. Inflation may rise even as unemployment rises. That is not a contradiction in the model: it is a shock shifting the short-run relationship, rather than a movement caused only by changing demand along one unchanged curve.

The reverse can happen when supply conditions improve. Lower energy or shipping costs, easing bottlenecks, or stronger productivity can reduce cost pressure while output and hiring continue to grow. Some prices may fall relative to others even if the overall price level still rises. The distinction between a relative-price change and broad inflation is important: one commodity’s price jump is not by itself a complete explanation of the path of an economy-wide index.

The 1970s are often used to illustrate that expectations and supply shocks can change the apparent relationship. Federal Reserve Bank of San Francisco material discussing the historical curves identifies both expectation changes and major supply shocks as possible contributors to shifts, while also describing other explanations for changing data patterns. A historical episode helps motivate the framework; it does not establish that one cause explains every later episode. {source:frbsfPhillipsCurveBasics2008}

A scatterplot of inflation and unemployment often does not trace a neat downward line. The data can combine different expectation regimes, supply shocks, policy reactions, changing labor-market conditions, and different phases of the business cycle. If a central bank responds to inflation with tighter policy, unemployment and inflation may both move in ways that hide the original shock. A simple correlation cannot by itself isolate which force caused which outcome.

Researchers also have to choose which inflation measure, slack variable, lag structure, sample period, and expectations measure to use. The unemployment rate’s estimated sustainable level is uncertain; prices and wages adjust at different speeds; and some models allow the curve’s slope to change when inflation is unusually high or unusually stable. Federal Reserve research reviewing the curve’s slope highlights econometric identification challenges, findings about pre-pandemic flattening, and the possibility of nonlinear inflation dynamics. Those are active measurement questions, not evidence for one fixed coefficient. {source:fedPhillipsSlopeReview2024}

A flatter estimated curve means that a given change in the selected slack measure is associated with a smaller inflation response in that specification. It does not mean inflation cannot change. Expectations, energy and import costs, productivity, and policy can matter more in that sample or model. Research has offered multiple hypotheses for apparent flattening, including anchored expectations, structural changes, measurement error in gaps, effective policy responses to shocks, and nonlinearities. These are candidate explanations to test, not interchangeable facts about every economy. {source:frbsfOriginalPhillipsCurve2021} {source:frbsfPhillipsCurveStressTest2019}

Before comparing two Phillips-curve charts, identify the outcome on the vertical axis. Nominal wage growth is not consumer-price inflation. Unit labor costs also depend on productivity: if output per hour rises, wages can grow faster without unit labor cost rising at the same pace. Headline CPI includes food and energy; core measures exclude some volatile items; and PCE uses a different scope and weighting method. Their measured rates can differ even when they describe the same period.

Then identify the horizontal-axis measure. The unemployment rate counts unemployed people in the labor force; it does not count everyone who wants more hours or has stopped searching. An unemployment gap subtracts an estimated reference rate. An output gap compares actual output with an estimate of potential output. Vacancies or hours worked offer still different views of resource use. These measures can point in different directions, and the unemployment gap can change just because the estimated reference value was revised.

Finally, check whether the chart uses inflation levels or changes in inflation, observed or expected inflation, and a quarterly, annual, or multi-year horizon. A Phillips-curve regression that explains the level of inflation answers a different question from one that models how inflation changes. A coefficient from one country, period, price index, or equation should not be copied into another setting without checking its definition and uncertainty. For related measures, see U-3 and U-6 unemployment rates, CPI, PCE, and the GDP deflator, and nominal versus real GDP.

The Phillips curve can organize questions about how labor-market slack, inflation expectations, supply costs, and pricing behavior may interact. A policymaker or forecaster can use it as one part of a broader model, compare alternative specifications, and ask what happens if expectations or slack estimates change. The Federal Reserve Bank of San Francisco has described both why economists use the framework and why its predictive performance and interpretation remain contested. {source:frbsfPhillipsCurveBasics2008}

It cannot tell you that inflation will always fall when unemployment rises, that a low unemployment rate must trigger a particular inflation rate, or that a central bank can permanently buy lower unemployment with higher inflation. It also does not prove that a supply shock is temporary, that expectations are anchored, or that one measure of slack is correct. Those claims require evidence beyond the curve’s name or a line fitted through historical points.

When someone cites a Phillips curve, ask: Which inflation measure? Which slack measure and reference rate? Are expectations and supply shocks included? What time period and horizon? Is the result a correlation, a model estimate, or a forecast, and how uncertain is it? Those questions turn a familiar diagram into a testable economic claim instead of treating it as a universal policy menu.