Guide

Phillips curve explained

Harbor Manufacturing's macro risk team ran a clean monetarist playbook through 2021–2023: when M2 growth slowed, they trimmed inflation hedges. Services CPI kept printing hot anyway — shelter, healthcare, and wages drove core PCE above 4% while headline unemployment sat near historic lows. The miss was structural: money aggregates ignored labor-market tightness. Rebuilding the sleeve around a wage Phillips curve overlay — unemployment gap, JOLTS quits, and unit labor cost momentum — cut forecast error on core services inflation by roughly 40% in backtests.

The Phillips curve is the empirical and theoretical relationship between inflation (or wage growth) and unemployment (or labor slack). Named after economist A.W. Phillips's 1958 UK wage study, it became the backbone of postwar macro forecasting: tight labor markets push prices up; slack pulls them down. Central banks still lean on Phillips-style frameworks when setting monetary policy, even as flat-curve episodes and supply shocks force constant refinement. This guide covers the original curve, expectations-augmented models, NAIRU, wage versus price specifications, why the curve flattens, links to stagflation and disinflation, the Harbor Manufacturing refactor, a technique decision table, pitfalls, and a production checklist.

What the Phillips curve describes

At its core, the Phillips curve posits an inverse relationship between the unemployment rate and the rate of wage or price inflation. When unemployment is low, workers bargain from strength, firms pass higher labor costs into prices, and inflation rises. When unemployment is high, wage pressure eases and inflation falls.

Phillips originally plotted UK nominal wage growth against unemployment from 1861–1957 and found a stable negative slope. Paul Samuelson and Robert Solow extended the idea to price inflation in the United States during the 1960s, feeding the policy view that policymakers could “buy” lower unemployment with modestly higher inflation.

Modern usage distinguishes:

Price Phillips curve — links CPI/PCE inflation to unemployment or output gap.
Wage Phillips curve — links average hourly earnings or employment cost index growth to labor slack; often leads price inflation in services-heavy economies.
Expectations-augmented Phillips curve — adds expected inflation so that only unexpected slack moves inflation away from anchor.

From simple tradeoff to NAIRU

The 1970s shattered the naive tradeoff. Oil shocks and embedded expectations produced simultaneous high inflation and high unemployment — the stagflation regime. Milton Friedman and Edmund Phelps argued there is a natural rate of unemployment (often called NAIRU: non-accelerating inflation rate of unemployment) below which inflation accelerates even if unemployment is stable.

The expectations-augmented Phillips curve can be written conceptually as:

Inflation ≈ Expected inflation + (β) × (Unemployment gap) + supply shocks

Where the unemployment gap is actual unemployment minus NAIRU (or output gap as an alternative slack measure). If expectations are well anchored at 2%, only persistent slack or tightness deviates inflation from target. Attempts to hold unemployment below NAIRU indefinitely raise expected inflation, shifting the whole curve up — the lesson of the Great Inflation.

Measuring slack beyond U-3

Participation rate — hidden slack when discouraged workers re-enter.
U-6 underemployment — part-time for economic reasons adds pressure not visible in headline unemployment.
Job openings / quits (JOLTS) — high quits signal worker bargaining power even before wages move.
Output gap — GDP vs potential from CBO or OECD; used in DSGE and Fed staff models.

Why the Phillips curve flattens

Since the 1990s, many economies saw a flatter Phillips curve: large swings in unemployment produced smaller inflation responses. Explanations include:

Globalization and import competition — domestic wage pressure less likely to pass through to goods prices.
Well-anchored inflation expectations — central bank credibility caps second-round effects.
Labor market institutions — weaker unions, gig work, and automatic stabilizers changed wage-setting.
Measurement issues — core vs headline, shelter imputation, and hedonics blur the signal.
Supply shocks dominating — energy, chips, and pandemic logistics moved inflation without labor tightness (or with it).

Flat does not mean dead. Post-pandemic services inflation showed the wage channel reasserting when unemployment hit 3.5% and job openings exceeded unemployed workers two-to-one. Models that declared the Phillips curve obsolete in 2019 underweighted it in 2022–2023.

Policy and markets: how the curve is used

Central banks embed Phillips-style blocks in forecasting models. The Fed's reaction function implicitly trades off employment vs inflation relative to dual mandate goals. When unemployment is above NAIRU and inflation is below target, disinflationary easing is easier. When both inflation and unemployment are elevated, policymakers face the stagflation dilemma — no clean Phillips tradeoff.

Investors use Phillips logic for:

Wage print sensitivity — average hourly earnings and ECI moves reprice rate paths.
NFP surprise analysis — jobs plus wage component beats/misses drive front-end yields.
Sector rotation — tight labor + sticky services CPI favors pricing-power equities; slack favors duration.
Real wage tracking — nominal wage growth minus CPI shapes consumption and political pressure for easing.

Harbor Manufacturing wage sleeve refactor

After the M2-only miss, Harbor Manufacturing rebuilt its inflation-risk overlay:

Dual block: retain M2–velocity trigger for goods and commodity inflation; add wage Phillips block for core services.
Slack panel: U-3 minus CBO NAIRU estimate, prime-age participation gap, and JOLTS quits rate z-score.
Wage momentum: three-month annualized AHE for production workers plus ECI private wages year-over-year.
Pass-through lag: distribute wage impulse into core PCE services over 4–6 quarters via estimated beta.
Supply overlay: when oil or import prices spike > 2 standard deviations, widen confidence bands — do not attribute all inflation to slack.
Exit rule: trim services inflation hedges when unemployment gap turns positive for two quarters and wage momentum falls below 3% annualized.

Out-of-sample from 2010–2024, the dual block beat M2-only on core services MAE by ~40% and matched simple VAR models with fewer parameters. The team still publishes a “Phillips confidence” flag: high when quits and wage momentum align, low during supply-shock quarters.

Technique decision table: Phillips vs alternatives

Approach	Best when	Watch out for
Wage Phillips + slack panel	Services-heavy inflation, tight labor regimes	Flat curve episodes; NAIRU uncertainty
M2 / velocity monetarist	Fiscal/monetary surge, goods inflation	Misses wage-led services when V is stable
Output gap (GDP vs potential)	Holistic demand pressure, DSGE-style models	Potential GDP revisions move the gap
Breakeven / survey expectations	Forward-looking anchor, market pricing	Risk premia; not causal slack measure
Import prices / FX pass-through	Goods disinflation, dollar strength episodes	Weak on domestic shelter and healthcare
Single-indicator (CPI spot)	Tactical trading around prints	Backward-looking; no turning-point lead

Common pitfalls

Treating NAIRU as fixed. CBO and Fed estimates drift with demographics and policy; use ranges.
Using headline unemployment alone. Participation and U-6 hide slack or tightness.
Ignoring expectations. High anchored expectations mean unemployment must rise more to cool inflation.
Attributing supply shocks to slack. Oil spikes inflate CPI without labor tightness — widen error bands.
Confusing level vs change. Stable low unemployment can still accelerate wages if already below NAIRU.
One-country export. Euro area Phillips slopes differ from U.S. services-heavy structure.
Declaring the curve dead after one flat decade. Post-2021 showed the wage channel still bites.

Production checklist

Pick specification: wage Phillips vs price Phillips vs both.
Source unemployment (U-3), U-6, participation, and JOLTS quits monthly.
Estimate or import NAIRU / output gap (CBO, Fed, OECD).
Compute unemployment gap and three-month wage momentum (AHE, ECI).
Regress core services PCE on gap, wage momentum, and lagged inflation.
Add supply-shock dummy for energy / import price spikes.
Cross-check against M2–velocity block for goods inflation.
Stress-test: unemployment +1pp vs −1pp from baseline.
Monitor real wage growth (nominal wages minus CPI).
Flag Phillips confidence: high when quits and wages align with gap.
Re-estimate after major revisions (benchmark GDP, NAIRU updates).
Document regime notes for flat-curve vs steep-curve periods.

Key takeaways

The Phillips curve links slack to inflation. Low unemployment tends to raise wages and prices; high slack does the opposite.
Expectations matter. NAIRU is the floor below which inflation accelerates if expectations are not anchored.
Wage channel leads services. In modern economies, labor costs drive sticky core inflation.
The curve can flatten. Globalization and credibility reduce slope but do not eliminate the relationship.
Combine with money and supply blocks. Phillips alone missed 1970s oil; M2 alone missed 2022 services.