Research
Job Openings Stopped Meaning What They Used To
For most of this century, hiring was better than the vacancy count implied. It is now consistently worse. The relationship inverted gradually, and not for the reason the obvious story suggests.
Published 2026-08-22 · 294 months analysed · data retrieved 2026-08-22
The finding
What we found.
Between 2001–2005 and 2021–now, the relationship between how many job openings exist and how good hiring conditions actually are has completely inverted — from hiring running 35.5 percentile points better than the vacancy supply implied, to 29.4 points worse.
In June 2026 the gap stands at 46 percentile points, the 98th percentile of 294 comparable months since 2000.
The obvious explanation — that job openings have stopped being a useful indicator — does not survive testing. Openings still track hiring about as closely as they ever did. What has changed is the level: the same abundance of vacancies that once accompanied a fast-moving labour market now accompanies a middling one.
The chart
Twenty-five years of divergence.
Above the line, hiring and search duration look worse than the supply of openings implies. Below it, better.
Reading this: above the line, hiring and search duration look worse than the supply of openings implies. Below it, better. The break is March 2020 to February 2021, excluded as structurally unreliable.
The series runs from -53 in October 2001 to +51 in May 2026, and currently reads +46.
What changed
A drift, not an event.
It is tempting to date this to 2024, when the gap first crossed +40 in June 2024. The decade averages say otherwise.
- 2000s
- 109 months, -53 to -6.5
- 2010s
- 120 months, -12 to 26.5
- 2020s
- 65 months, 7.5 to 51
The sign flipped during the 2010s, roughly a decade before the readings that prompted this analysis. The gap first exceeded +20 in July 2015 and has done so in 58 months since.
The persistence is as striking as the level. The longest uninterrupted run above +20 is 35 months, from July 2023 to June 2026 — the current one. Its mirror image sits early in the series: 60 consecutive months below −20, from July 2003 to June 2008, when hiring was persistently better than the vacancy count suggested.
The test
Have job openings stopped predicting hiring?
This was our hypothesis. It is wrong, and the way it is wrong is the most useful thing in this report.
If the vacancy count were losing its informational value, the two signals would be drifting apart in their movements. We measured the correlation between the opportunity signal and the hiring-and-duration signal inside each five-year block, so that a change in level could not disguise itself as a change in relationship.
| Period | Correlation | Openings | Hiring & duration | Gap |
|---|---|---|---|---|
| 2001–2005 | 0.87 | 64.1 | 28.6 | -35.5 |
| 2006–2010 | 0.98 | 71 | 51 | -20 |
| 2011–2015 | 0.94 | 72.2 | 75 | +2.7 |
| 2016–2020 | 0.89 | 25.2 | 39.2 | +14 |
| 2021–now | 0.90 | 14.6 | 44 | +29.4 |
The correlation never falls below 0.87 and reaches 0.985. The most recent block is not the weakest. Openings and hiring still move together.
The level did all the work. The openings signal fell 49.5 percentile points across the period — vacancies became abundant by historical standards — while hiring and duration moved up only 15.4.
So the precise claim is not that openings broke. It is that the exchange rate changed: a given quantity of advertised vacancies now buys materially worse hiring outcomes than the same quantity did twenty years ago. Anyone using vacancy counts as a shorthand for labour-market health is using a ruler whose units have shifted underneath them.
How unusual
And how much of that depends on us.
Across 294 comparable months the gap has run from -53 to 51, with a median of 0.5. The current reading is in the top 2 per cent.
Under our published measure, no month before June 2024 exceeded +40 in the 294 months analysed. That is a striking sentence, and we tested whether it survives being defined differently. It does not.
| Definition | Percentile now | Months > +40 | Of those, pre-2024 |
|---|---|---|---|
| Mean of hiring and duration | 98th | 12 | 0 |
| Hiring only | 92nd | 31 | 3 |
| Duration only | 95th | 15 | 12 |
| Worse of the two | 88th | 43 | 15 |
What this means for the headline
The divergence is present under every definition — the current reading sits at the 88th percentile even under the most permissive one. The “never before in 23 years” framing is not: it belongs to our specific construction, and under a duration-only definition 12 earlier months cross the same line. We would rather this were quoted as a percentile than as a record.
A second sighting
The same pattern, across states.
The national analysis is a time series: one country, 25 years. A cross-section is an independent way to look for the same thing.
In December 2025, ranking the 51 states and the District of Columbia by competition for each opening produces an order almost unrelated to ranking them by how fast employers are hiring — Spearman 0.006. Vacancy abundance and hiring speed are close to independent across states, just as the gap between them is close to independent of the headline index over time (r = 0.037).
This is corroboration, not proof. Two views of overlapping data are not two experiments, and the state series covers ten years rather than twenty-five. But a pattern that shows up in both the time series and the cross-section is harder to dismiss as an artefact of one construction.
Explanations
What could be causing it.
We can measure the divergence. We cannot identify its cause, and these are the candidates we cannot choose between.
Postings staying open longer
A vacancy that remains advertised without being filled is counted in every month it is open, so a slower process inflates the stock of openings without any change in hiring.
Changes in how vacancies are posted
Cheaper posting, duplicate listings across platforms, and speculative advertising would all raise measured openings relative to actual hiring.
Slower or more selective hiring processes
More interview rounds, more approvals, or a higher bar for filling a role would lengthen searches while leaving the vacancy count untouched.
Mismatch between advertised roles and available workers
Skills, seniority, location, or pay expectations that do not line up would leave openings unfilled and searches long at the same time.
Composition of who is unemployed
A shift in which industries and which workers are searching would change duration without changing the supply of vacancies.
What this research cannot tell you
- Which of the hypotheses above is responsible, or in what proportion.
- Whether employers are advertising roles they do not intend to fill.
- Whether the change is permanent or a phase of the current cycle.
- Whether any particular industry, occupation or region is driving it.
- Whether job seekers face worse odds per application than they used to.
Method
How this was done.
Five monthly BLS series — JOLTS job openings and hires rate, and Current Population Survey unemployment level, median duration and long-term share — from 2000-12 to 2026-06. Each of the three dimensions of the Job Search Friction Index is expressed as a difficulty percentile against 294 comparable months.
The disconnect is the mean of the hiring and duration percentiles minus the opportunity percentile, computed from the rounded values published on every surface, so the figure can be reproduced by hand. March 2020 to February 2021 is excluded throughout.
Correlations are Pearson within each five-year block, so a shift in level cannot present itself as a change in relationship. The state cross-section uses Spearman rank correlation over the 51 places.
No new data was collected. Every figure here can be recomputed from the published files.
Cite this research
Maid Dizdarevic, “Job Openings Stopped Meaning What They Used To”, Jobsearch.ing, 2026-08-22. Analysis of 294 months of U.S. Bureau of Labor Statistics JOLTS and Current Population Survey data, 2000-12 to 2026-06. https://jobsearch.ing/research/job-openings-hiring-disconnect//