ADP vs NFP vs jobless claims: which jobs number moves the dollar?

Three readings of the US labor market come out in the same week, and they often disagree. On 30 September 2026, ADP reported a gain of 90,000 private-sector jobs, almost exactly the 89,000 our calendar carried for the official payrolls two days later. The official figure came in at 29,000. In our price data, payrolls moved the dollar several times as much as an ordinary day, and the other two far less.

Why are there three jobs numbers in the same week?

Because three different organizations count three different things, on three different days.

In UTC, that is 12:15 and 12:30 until the US clocks change in November, and an hour later in winter. The day can move too: the June report came out on Thursday 2 July 2026, the day before the Independence Day holiday.

What does each one actually measure?

The ADP National Employment Report comes from Automatic Data Processing (ADP), a company that processes payrolls. It describes itself as "an independent measure of the labor market based on the anonymized weekly payroll data of more than 26 million private-sector employees in the United States". It does not count government jobs, and it is not a survey: it reads the payrolls ADP already runs.

Since late 2025 ADP also publishes a weekly version, the NER Pulse, named after the National Employment Report. It usually comes out on Tuesdays, except in the week of the monthly report, and its figure is a four-week moving average of the weekly change, with a lag of about two weeks.

Nonfarm payrolls come from the Bureau of Labor Statistics (BLS) establishment survey, which each month covers about 119,000 businesses and government agencies. Government jobs are included. Farm workers, private household workers, unpaid family workers and the unincorporated self-employed are not.

The same report carries the unemployment rate, but that figure comes from a different survey, of about 60,000 households. So the two headlines of the same Friday can point in different directions, and both can be right.

Initial jobless claims count the people filing for state unemployment benefits for the first time. The release of Thursday 8 October covered the week ending Saturday 3 October, and the continuing claims in the same release run one week older. They measure layoffs, not hiring.

Does ADP predict NFP?

Not in the four months we have. The table compares each first print with the forecast our economic calendar carried on the day; the month is the one the data refers to.

In June and July both missed their forecasts in the same direction. In August ADP missed and payrolls beat theirs by 107,000; in September ADP beat and payrolls missed by 60,000. They agreed in two months out of four, and four months are too few to call it a pattern.

The gap between the two prints ran from 41,000 to 124,000 jobs. Part of it is built in, because ADP leaves out government and the BLS does not. And both get revised: ADP's August figure was later cut from 38,000 to 36,000, and the September payrolls report revised July and August down by a combined 60,000.

Which one moves the dollar?

We measured the high-to-low range of EUR/USD and USD/JPY in the fifteen minutes after each release, on one-minute bars from 14 August to 16 October 2026. The comparison is an ordinary day: the median of the 25 weekdays in that period with none of these releases, measured over the same fifteen minutes.

Payrolls moved EUR/USD between 3.6 and 5.2 times as much as an ordinary day at that hour. Monthly ADP moved it a little more than usual, and weekly ADP no more than an ordinary day.

Claims also moved it a little more than usual, but their spread is wide: from 4.7 to 35.7 pips in EUR/USD. The two largest, 35.7 on 10 September and 28.1 on 15 October, came when US producer prices were released at the same minute, and on the 15th retail sales as well. A fifteen-minute window cannot separate releases that share the same minute.

How do you read the three together?

Use each one for the question it actually answers.

• Claims tell you about layoffs. In every weekly release from late June to mid-October, they stayed between 187,000 and 215,000, while payrolls ranged from −23,000 to +162,000. Few layoffs do not mean strong hiring.
• ADP is an early look at the private sector, not a forecast of the headline. And the weekly figure is a per-week average: for example, 20,000 a week is not a miss against a monthly forecast of 90,000.
• Payrolls are the number the market settles on. On 4 September they came in at nearly three times the forecast and the dollar still lost the week. On 2 October a 29,000 print took an October Fed hike off the table.

In all three, what moves price is the surprise against the forecast, not the level of the number. Both payrolls reports in our price sample were large surprises, which helps explain the size of the reaction.

What this post does not give you

It does not give you a way to forecast payrolls from ADP. Four months is a small sample, and in two of them ADP pointed the wrong way.

The price table rests on two payrolls releases and two monthly ADP reports, all inside one two-month window. It shows the order of magnitude, not a rule. Other releases in the same minute, such as Canada's jobs report on 4 September, are not separated out.

It does not break the gap between ADP and the BLS into its parts either: government jobs, sample differences and revisions all play a part, and neither publisher splits the gap for you.

The takeaway

Claims tell you whether layoffs are rising, ADP gives you an early private-sector read, and payrolls decide. Judge each one against its own forecast, and do not trade the Wednesday release as if it were Friday's.

ADP figures as published by ADP Research, including its NER Pulse weekly series; payrolls survey coverage and exclusions from the BLS technical note to the Employment Situation; claims timing from the Department of Labor's weekly release. Forecasts and first prints from our economic calendar; price ranges are measured on our own one-minute bars.