Job openings-to-employment ratio

Download the data and code

• joe_monthly.csv908 months · 49 KBThe full monthly series with every input column.
  Download CSV
• update_joe.py9 KBPython script that rebuilds the CSV from FRED. Needs pandas.
  Download Python
• update_joe.m9 KBMATLAB script that does the same and draws the chart with recession shading.
  Download MATLAB

Both scripts download the latest figures each time they run, so the series can be brought up to date after any JOLTS release. The 1951 to 2000 help-wanted data are fixed history and are written into each script. The two scripts give identical output.

Read the code

Loading the script…

How the series is built

No single survey covers 1951 to today, so the series joins two sources at January 2001. Both are put on the same footing: job openings, V, divided by the number of people employed, E.

Before you use it

• Levels before 2001. From 2001 the series matches the one plotted by Kothari and O’Doherty. Before 2001 it runs about 7% above their plotted levels. The gap is a constant ratio, so the help-wanted index was scaled differently there, but the ups and downs are the same. This version joins smoothly at January 2001.
• Revisions. BLS revises JOLTS every year, and each run of the scripts takes the latest published figures. Recent revisions cut job openings for late 2024 and 2025 by 300,000 to 460,000 a month, lowering JOE in those months by up to 0.3 percentage points.
• October 2025 is missing. The household survey was not carried out during the federal government shutdown, so there is no employment figure for that month. The gap is left in rather than filled with an estimate.
• Publication lag. JOLTS for a month is released about five weeks after the month ends. For out-of-sample forecasting use joe_lagged_pct, which holds only the reading that was public at each month end (a one-month lag to 2000 and a two-month lag from 2001). It reproduces the timing, not the first-release values.
• January breaks. Each January the household survey moves to new population estimates without revising earlier months, which can shift the employment level by several hundred thousand.
• Not the BLS openings rate. BLS publishes its own job openings rate using payroll employment plus openings. JOE uses household-survey employment, as Kothari and O’Doherty do, so the two differ slightly.

Why it matters

Kothari and O’Doherty (2023) find that JOE is one of the strongest predictors of the US equity premium found so far. A high ratio is followed by lower excess market returns, and JOE beats more than two dozen standard predictors both in sample and out of sample. They explain the result with a production-based model in which hiring is costly when the labour market is tight.

Extending the series lets that result be tested on data the paper never saw, including the post-pandemic peak in March 2022 and the slow cooling since.

How to cite

If you use the series, please cite the paper that introduced it and the source of the pre-2001 data. A reference for this page is included in case it helps.

@article{KothariODoherty2023,
  author  = {Kothari, Pratik and O'Doherty, Michael S.},
  title   = {Job postings and aggregate stock returns},
  journal = {Journal of Financial Markets},
  volume  = {64},
  pages   = {100804},
  year    = {2023},
  doi     = {10.1016/j.finmar.2023.100804}
}

@article{Barnichon2010,
  author  = {Barnichon, Regis},
  title   = {Building a composite {Help-Wanted Index}},
  journal = {Economics Letters},
  volume  = {109},
  number  = {3},
  pages   = {175--178},
  year    = {2010},
  doi     = {10.1016/j.econlet.2010.08.029}
}

@misc{PybisJOE,
  author       = {Pybis, Sam},
  title        = {Job openings-to-employment ratio, {United States}, monthly, 1951 to August 2026},
  year         = {2026},
  howpublished = {\url{https://sampybis.co.uk/updated-data/joe.html}},
  note         = {Updated 20 October 2026}
}