The Beveridge curve

When workers are looking and jobs are open, how well does the labor market bring them together — and did the pandemic permanently damage that matching process?

In search-and-matching models the labor market is a two-sided search problem: unemployed workers and vacant jobs meet through a matching function, and both can be plentiful at once if matching is slow. The Beveridge curve is the resulting locus: vacancies high when unemployment is low (tight markets) and vice versa, a downward-sloping curve traced out by demand over the cycle. Its position encodes efficiency — an outward shift means more vacancies are needed to employ the same share of the workforce, from mismatch in skills, geography, or search intensity. Theory also predicts loops: after a shock, vacancies recover before unemployment falls, so the economy circles counterclockwise back toward the curve.

Job openings from JOLTS (monthly, from December 2000), nonfarm payrolls, and the unemployment rate, Jan 2001 through May 2026. The vacancy rate is constructed as openings ÷ (payrolls + openings) × 100 — the JOLTS job-openings-rate definition. The unemployment series skips October 2025 (missed federal release); that month is dropped from the pairing rather than interpolated.

Each month since January 2001 is plotted as a point (unemployment rate on the horizontal axis, vacancy rate on the vertical), colored by period: the 2001–09 era spanning two recessions, the long 2010s expansion, the 2020–21 pandemic shock and reopening, and 2022–present. No curve is fitted — the relationship is visibly nonlinear and the object of interest is the cloud’s position, not a slope.

The pandemic produced both of the curve’s canonical motions in rapid succession. First a violent move along it (unemployment to 14.8% in April 2020 while vacancies fell only modestly), then an outward shift: through the reopening, the same unemployment rate coexisted with roughly 2.5% more vacancy than the 2010s norm — matching had genuinely deteriorated while workers and jobs re-sorted. Since 2022 the economy has descended the vertical segment of the curve: vacancies fell from 7.5% toward 4.6% while unemployment rose only about 0.6% — the historically unusual “soft-landing” path, cooling mostly through fewer openings rather than more layoffs.

The outward-shift comparison depends on which months are matched: it uses months with unemployment within ±0.3pp of 6%, and widening or narrowing that band moves the contrast by a few tenths. Whether the curve has fully returned to its pre-pandemic position is genuinely unresolved — the latest points sit near the top of the 2010s cloud, close enough that normal JOLTS revisions could move the verdict either way.

• Openings ≠ search intensity. A posted vacancy is free to leave up; recruiting effort per opening varies with the cycle, and low-cost online posting has made an opening in 2026 a different object from one in 2001. Part of any apparent shift is measurement, not matching.
• JOLTS revisions and coverage. The survey is revised substantially, has a modest sample for monthly readings, and its response rate has fallen sharply since 2020 — recent months are the least settled part of the picture.
• Position, not mechanism. The curve cannot by itself say why it shifted — skills mismatch, geographic mismatch, enhanced benefits, and reallocation all produce similar outward movements.

The calculation lives in apps/web/src/app/research/beveridge-curve/page.tsx and uses lib/econ; every number on this page is recomputed from the committed snapshots on each build. Data series and vintages are listed on the Data page.