The NFP problem

Non-Farm Payrolls are one of the most impactful events on the economic calendar. The first Friday of the month consistently has traders glued to their screens in anticipation of the latest reading on the US labour market. The publication obviously matters to a lot of people and therefore to financial markets as a whole. With that being the case, it is not unfair to ask how reliable these figures really are.

NFP revisions

NFP figures undergo a series of revisions following their initial release. The first Friday delivers a first estimate of the number of jobs created over the course of the prior month. One month later, this figure receives a first correction. One month after that, the figure receives a second correction. On top of these monthly revisions, every summer the Bureau of Labor Statistics publishes a preliminary annual benchmark revision, before publishing a final benchmark revision around six months later. One initial print, two revisions, two benchmark adjustments.

How big of a gap is typically observed between the different figures? At times, the word “gap” simply doesn’t cover it. Chasm would be a better descriptive. NFP numbers have always undergone significant revisions following their initial release, but the last few years have been bad enough to render the original publication almost completely useless. The period between April 2023 and March 2024 initially reported 2.9 million new jobs in the United States, a figure that was later revised downwards by 589,000 during the final benchmark adjustment. A pretty severe correction at first glance, but the following period missed by a country mile. Between April 2024 and March 2025, initial estimates reported 1.8 million new jobs, which was then revised downwards by 898,000 in the final count.

This means that initial NFP reports overestimated US job creation in 2024 by a factor of two. Not only does this discredit the prevailing narrative of a recovering labour market at the time, but it also raises the question of the Federal Reserve’s ability to properly dictate monetary policy. If the Fed is making decisions based on faulty data, then how dependable can those decisions really be? In the central bank’s defence, a shortfall of a million jobs is not catastrophic compared to the 160 million people currently in employment in the United States, representing a mere 0.6% discrepancy. It is also true that the Fed relies on other sources of information than the BLS. With that said, monetary authorities are still relying, albeit partially, on incorrect data. The question remains: why are initial non-farm payroll estimates so poor?

Data collection

Figuring out the number of people with a job is not as easy as it should be. One might think that there is a national database that the government can query to get a reliable number, but this is not the case. Typically, the body responsible for compiling such information uses survey data to reach a verdict. This is as simple as asking businesses how many employees they have on payroll (establishment survey), as well as asking households how many people within are currently employed (household survey). It seems primitive but this the basis of data collection in most countries. The problem is that there is sometimes a large discrepancy between the figure provided by the establishment survey and that provided by the household survey. An additional problem is that such techniques are ill-equipped to capture the increasing prevalence of “grey area” work, such as freelancing, remote work, gig economy jobs, all of which became far more common during and after the COVID years. A final problem is the recent collapse in survey response rates, which used to be around 80-90%, but have subsequently fallen to around 50%, leaving the Bureau of Labor Statistics largely in the dark. This increases the reliance on modelling, which is evidently sub-optimal.

This is why the benchmark adjustments exist. Rather than relying on surveys and modelling, the annual revisions lean on more robust data sources, such as tax records and census data. Such “hard data” take more time to collate, hence the longer time frames at play.

Of course, none of the above addresses the credibility of the initial NFP report. If the number of jobs can be halved or doubled further down the line, then why bother paying attention to the initial release? As far as traders are concerned, the importance stems from the difference between the expected figure, which is priced in, and the actual figure, which is not, hence the tradeable event. As far as the greater macroeconomic environment is concerned, such data prints are not as useful as they should be and as such face rightful scepticism.

The unemployment problem

Non-farm payrolls deliver another crucial piece of economic data: the unemployment rate. This is where things get arguably worse.

The rate is defined as the number of unemployed people divided by the workforce. The workforce is typically defined as the population over the age of 16, minus retirees, students, military personnel, caregivers, institutionalised people and finally, discouraged workers. A discouraged worker is anyone who wants a job but has not actively searched for one in the last four weeks. Who can blame them?

Discouraged workers are not taken into consideration when calculating the unemployment rate, which can make such figures look better than they really are. In the case of the United States, the labour force participation rate stands at 62%, meaning over a third of the adult population is not considered part of the work force at all. Almost 40% of the US population have no impact whatsoever on the apparent health or unhealth of the labour market. That’s over 100 million people. Many of them are retired, in higher education or busy raising children, but millions of others are not. The BLS attempts to offer a more realistic statistic in the form of the U-6 unemployment rate, which includes broader sections of the population, but even that fails to include anyone out of work for more than twelve months. Just as with HR departments, a one-year resume gap is a death sentence.

Large swathes of the population disappear into a statistical no-man’s land when it comes to employment figures, further eroding confidence in the NFP reports that provide them. The fact of the matter is that a growing number of people have given up, not only on finding a job, but on participating in society under any form. These people are invisible. These people do not appear in economic charts, demographic tables, census surveys, electoral polling or as marketing targets. These people lie on the other side of the broken social contract, perhaps never to be seen again. A far larger problem than one suggested by basic unemployment rates.

Final thoughts

Non-farm payrolls have long been a pillar of the economic calendar, but if the underlying figures continue to be this inaccurate, then eventually people will stop paying attention. It has not happened yet, but one day the lack of credibility in such reports will have institutions shrugging their shoulders in disinterest, turning instead to revised publications and alternative data sources. Traders will follow soon after.