Labour market data sources that arrive before official statistics

Alternative labour market data sources that are timelier than official statistics come in four kinds: weekly unemployment insurance claims from the government itself, payroll counts from a payroll processor or from tax records, indexes of job postings, and the postings of individual companies. This guide sets each one beside the monthly report by what it observes, how often and with what delay, from each publisher's own methodology. Nothing on this page is investment advice.

How the official jobs report is made and revised

The US Employment Situation combines two surveys. The establishment survey counts jobs on nonfarm payrolls in the pay period that includes the 12th; the household survey measures the labour force in the calendar week containing the 12th. Preliminary figures typically appear on the third Friday after the reference period ends, and the payroll estimates are revised twice.

The payroll count comes from the payroll records of a sample of about 119,000 businesses and government agencies, and the household figures from about 60,000 households, as the technical note of the release sets out. The two latest months rest on incomplete returns, so each is revised in the next two releases. Once a year the estimates are re-anchored to full counts drawn mainly from unemployment insurance tax records, as the bureau's answers on the payroll survey explain, and over the prior ten years that benchmark revision averaged 0.2 per cent of total nonfarm employment in absolute terms.

Revisions are large enough to matter. The release for September 2026, published on Friday 2 October 2026, revised the change in US total nonfarm payroll employment for July from +21,000 to -10,000 and for August from +162,000 to +133,000: 60,000 lower for the two months combined.

Job openings come from a separate monthly survey, JOLTS, counted on the last business day of the month. A position counts when it exists with work available, could start within 30 days and is being actively recruited for from outside the establishment. The August 2026 figures came on Tuesday 29 September 2026. Every count of postings below is read against that definition.

Which alternative labour market data sources are timelier than official statistics?

Four kinds are published more often than the monthly payroll report: weekly unemployment insurance claims from the Department of Labor, out five days after each week ends; ADP's weekly estimate of private payroll employment; daily postings indexes and postings data from publishers such as Indeed and Lightcast; and the dated postings of individual companies. Each observes something different.

The table sets out each kind from its publisher's own methodology, with two official series of the United Kingdom beside them.

Revisions run two ways by design. The payroll survey, the claims and the PAYE estimates are revised as fuller returns arrive, which brings each closer to the full count; a write-once record keeps what was known on each day. A test should say which of the two it needs.

Which real-time labour market indicators come at a higher frequency than official statistics?

Three come at a higher frequency than monthly official statistics: daily postings indexes such as Indeed's, refreshed each week; the weekly unemployment insurance claims, which are themselves official; and ADP's weekly NER Pulse. Company-level postings add a daily record per employer. Frequency is not the same as delay, so record both for every series you use.

Smoothing slows a series down. Indeed's daily index is a seven-day trailing average and ADP's Pulse a four-week moving average, and the claims release carries four-week moving averages beside the weekly figures, as the St. Louis Fed's page for the release lists.

How much timelier are ADP private payroll data, weekly initial unemployment claims and job postings?

The lead depends on the week compared. Initial claims for the week ending 26 September 2026 came out five days later, on 1 October. ADP's Pulse for the four weeks ending 19 September came 17 days later. Lightcast states 36 hours from posting to publication for sites it collects daily. The BLS September report came 20 days after the 12th.

The claims dates come from the St. Louis Fed's series of initial claims and continued claims, both updated on 1 October 2026, when continued claims ran to the week ending 19 September. A weekly source leads only for weeks the monthly report has not reached: the Pulse of 6 October covered four weeks that contain 12 September, and the September report had come four days earlier.

Payroll tax, unemployment claims and job postings as timelier labour market data sources for nowcasting

Payroll tax records, unemployment claims and job postings are each published within days or weeks of the period they describe, which makes them candidates for a nowcast. The UK's Office for National Statistics publishes a flash estimate of employment and pay from HMRC's PAYE tax records two to three weeks after the month, with about 15 per cent of it imputed.

The PAYE figures cover the whole payrolled population, so they are not sample based and do not move with response rates, and they are revised as submissions arrive for the rest of the tax year, as the ONS user guide sets out. A nowcast on any of these sources is a model you specify and test against the official figure you choose.

Alternative labour market data beside the BLS jobs report: ADP payrolls, Indeed and Lightcast job postings

Beside the BLS jobs report, ADP publishes private payroll employment from its own anonymised payroll data, Indeed Hiring Lab a seasonally adjusted daily index of postings on Indeed, and Lightcast postings collected from public job boards and company websites. The first measures employment change; the other two count advertised roles, while the BLS counts job openings in a separate survey.

ADP Research produces the National Employment Report with the Stanford Digital Economy Lab, on a reference week that includes the 12th, and publishes it monthly on a Wednesday. Its NER Pulse, out on Tuesdays between monthly reports, is the weekly change in employment as a four-week moving average, seasonally adjusted, published with a two-week lag and carrying 12 weeks of history, as its release of 6 October 2026 states.

Indeed Hiring Lab publishes its Job Postings Index in a data repository: the percentage change in seasonally adjusted postings since 1 February 2020, so a reading of 101 is 1 per cent above that day. A new posting is one on Indeed for seven days or fewer. Hiring Lab adopted its current methodology in November 2024 and revised the history, and its data FAQ notes that postings are not a precise count of available jobs.

Lightcast collects postings from public job boards and company websites, visits the busiest sites most often, and removes duplicates within each source and then across sources over 60 days. It revisits sites for up to 14 days, reclassifies postings every four weeks with the history included, and notes that a posting is not necessarily a vacancy.

Which data providers publish weekly labour market indicators from job postings?

Indeed Hiring Lab publishes a daily postings index refreshed each week; Lightcast publishes postings data collected from job boards and company websites; the UK's Office for National Statistics published an experimental weekly index of online job adverts from Adzuna data until October 2024; and Fokals delivers Market Series, weekly hiring series as of each Sunday, built from company-level postings.

The ONS method indexed a point-in-time count of Adzuna's adverts to the February 2020 average and could publish six days after each snapshot; the dataset has not been updated since 24 October 2024. It states that one advert can carry several jobs and can stay up after a post is filled.

Company-level postings and weekly series from Fokals

Fokals delivers hiring data at company level, collected from what companies publish on their own careers pages. Job Postings holds every role a company publishes, dated by when it was first and last seen; Hiring Activity gives each company's open, new and closed postings for every closed day; Market Series aggregates them weekly.

Each posting carries its job function, seniority, role flags, location, work mode, advertised pay on one annual US-dollar scale and the tools named, as the hiring dataset page lists. Market Series turns the company tables into weekly series by industry, country and size band over 7, 30 and 180 days, on same-store cohorts. For those series in a thematic view, see sector hiring trends for macro and thematic research.

Every record carries the time it was observed. Daily and weekly rows are written once after the period closes and never revised, and a posting already on a board at its first observation is a baseline, never a new posting. The record is point-in-time by construction, delivered by REST API or as bulk files, with every field defined in the data dictionary. A posting records an intention to hire, which is how it differs from a payroll count.

Worked example: putting Acme Robotics' postings on the claims week

Official and alternative series rarely share a week. Claims weeks run Sunday to Saturday, the payroll reference is the pay period that includes the 12th, and the weekly Fokals datasets run Monday to Sunday in UTC. This illustrative example re-cuts one company's daily postings into Saturday-ending weeks, so that they sit beside the claims calendar.

Assumptions:

• Acme Robotics is an invented company, and every figure for it is illustrative.
• Its Hiring Activity rows give new and closed postings for each closed UTC day, and they sum to the weekly totals below.
• Its board was first read before these weeks, so every posting that opened in them counts as new.

The working:

1. New postings fell from 16 to 8 in the week ending 19 September: down 8, or 50 per cent.
2. The four-week average, the convention of ADP's Pulse, is (10 + 12 + 14 + 16) / 4 = 13.0 for the weeks ending 22 August to 12 September, and (12 + 14 + 16 + 8) / 4 = 12.5 for the weeks ending 29 August to 19 September.
3. The average moved by the newest week less the week that dropped out, over four: (8 - 10) / 4 = -0.5. That is about 4 per cent of 13.0, against 50 per cent week on week.
4. Open postings reconcile: 143 + 8 - 14 = 137. A gap would flag postings found on a first read, which add to open postings without being new.

The query makes the same cut on your own load of Hiring Activity.

with weekly as (
  select
    day + (6 - extract(dow from day))::int as week_ending,
    sum(new_postings)    as new_postings,
    sum(closed_postings) as closed_postings,
    max(case when extract(dow from day) = 6 then open_postings end) as open_at_end
  from company_hiring_daily
  where company_id = :company_id
    and day between :first_sunday and :last_saturday
  group by 1
)
select
  week_ending,
  new_postings,
  closed_postings,
  open_at_end,
  avg(new_postings) over (order by week_ending
                          rows between 3 preceding and current row) as new_avg_4wk,
  lag(open_at_end) over (order by week_ending)
    + new_postings - closed_postings                                 as open_check
from weekly
order by week_ending;

Acme's rows for the week ending Saturday 19 September 2026 are written once, after each UTC day closes. On the Department of Labor's Thursday schedule, initial claims for that week were due on 24 September; the BLS September report came on 2 October, and ADP's Pulse for the four weeks ending 19 September on 6 October. A 7-day Market Series window as of a Sunday covers the Monday to that Sunday, so it shares six days with the claims week that ended the day before.

What alternative labour market data can macro hedge funds read ahead of the jobs report?

Ahead of the monthly jobs report, a macro team can read weekly initial and continued claims, ADP's weekly Pulse, daily postings indexes and company-level postings. Each has its own week, delay and revision rule, so the useful step is to record those properties for every series before setting it beside the payroll figure.

A checklist to fill in for every series, from its publisher's methodology:

• What is observed: a claim, a person on a payroll, an advert, or an opening as JOLTS defines it.
• The population: every employer, the private employers on one processor's payroll, one site's postings, or the employers you track.
• The week: Saturday-ending claims weeks, the pay period including the 12th, Monday-to-Sunday UTC weeks.
• The delay from the end of the period to publication, measured on real release dates.
• Smoothing: trailing averages, seasonal adjustment and how its factors are set.
• Revisions: whether a published value can change, and when, or whether rows are written once.
• Breaks: what happens to the history when the method changes, as with Indeed's revision in November 2024.

Can hedge funds use alternative data for a nonfarm payrolls prediction?

A fund can set any of these series beside the payroll estimate, but this page makes no claim that any of them predicts nonfarm payrolls or any market, and nothing in it is investment advice. In the methodology pages we read, we found no measure from a statistics office or central bank of how closely any of them tracks the first estimate.

If you build such a test, decide whether its target is the first payroll print or a later one, since the September 2026 release moved July and August by 60,000 combined. Then run it only on values that were known on each day, or look-ahead bias enters the result.

Frequently asked questions

The queries and code on this page are examples to adapt. Test them in your own environment before you rely on them.

What this page says about the products it names was checked against their public documentation on 7 October 2026. Product and company names are trademarks of their owners. Fokals is not affiliated with them or endorsed by them.