Employment Data and Non-Farm Payrolls

Non-Farm Payrolls reports the net change in paid employment across the United States, excluding farm work, private household staff, the self-employed and unpaid family workers. It lands at 08:30 US Eastern on the first Friday of most months, which is 13:30 UTC in winter and 12:30 UTC while the United States is on daylight saving, alongside the unemployment rate and average hourly earnings. It is watched harder than almost any other release because employment sits directly inside the Federal Reserve's mandate.

A 100,000 miss, and eight seconds

Consensus is 180,000. The print is 80,000. In the first eight seconds EUR/USD moves 45 pips higher, the two-year Treasury yield drops 9 basis points, and the spread on the pair goes from 0.8 pips to somewhere between 4 and 10 depending on your broker and how fast their liquidity providers come back.

Work out what that costs on a normal position. A 0.5 lot on EUR/USD is $5 per pip. A spread that widens by 6 pips costs $30 on entry alone if you trade into it. If you had a stop 20 pips away on the wrong side, it did not fill at 20 pips away; it filled wherever the book was when price swept through, and a 6 pip slip on that stop is another $30. Neither number is large. Both are entirely avoidable by not being in the market at the release second. The pip calculator converts these into your own account currency in a few seconds.

The reason the move is so fast is that the number is a genuine unknown. Unlike GDP, which is assembled from data already published, payrolls is a fresh survey. Forecast dispersion is wide, typically 60,000 to 100,000 between the highest and lowest estimate, and the historical average absolute miss against consensus is large enough that the consensus is better understood as a midpoint than a prediction.

Three numbers, in order of how much they matter

1. Average hourly earnings. Wage growth feeds services inflation, which is the part of the basket policy can influence. A wage surprise often outweighs a headline jobs surprise.
2. The headline payroll change, read together with revisions to the previous two months. A 200,000 print alongside a 90,000 downward revision is a 110,000 print in disguise.
3. The unemployment rate, read with the participation rate. Unemployment can fall because people found work or because they stopped looking, and those are opposite signals.

There is a fourth item most people skip and should not: the birth-death model. A portion of the headline is an estimate of jobs created and destroyed by businesses too new or too recently closed to be in the survey sample. In a normal month it is a modest adjustment. At a turning point in the cycle it is systematically wrong in the same direction for several months running, because a model calibrated on past business formation cannot detect that formation has stopped. That is one of the main sources of the benchmark revisions discussed below.