News trading on gold: what 11 NFP releases actually did

Finding the release without a calendar

NFP lands on the first Friday of the month at 08:30 US Eastern. That is fixed, which means no economic calendar is needed to locate it in a year of bars — useful, because the calendar feeds available to us only carry the current week.

The one thing that could not be assumed was the broker’s clock. So it was derived: taking the largest-range minute on each first Friday put the cluster at 12:30 in summer months and 13:30 in winter, which is exactly 08:30 Eastern under daylight saving and standard time. That identifies server time as UTC, and it checks itself — a wrong assumption would not line up with a daylight-saving boundary. The 6 March 2026 release, two days before US clocks changed, reads 13:30. The 5 June release reads 12:30.

Every release, measured

Baseline for comparison is the median across all 352,587 minutes in the sample: a range of $1.793 and a spread of 160 points.

The average is not the story. The spread of outcomes is.

Median release-minute range: $4.81, which is 2.7× an ordinary minute. That is a smaller number than the reputation suggests, and on its own it would make NFP sound manageable.

It is not the number that matters. The largest release moved $40.48 in sixty seconds — 22.6× normal — and the smallest moved $2.03, which is an unremarkable minute. Same event, same time of day, a twenty-fold difference in outcome, and no way to tell them apart beforehand.

That is the actual risk of trading a release. Not that the average move is large, but that a position sized for the median can meet the maximum.

It is also over quickly. Counting consecutive minutes above twice the normal range, the median was 2 minutes and the longest was 24. Whatever the release is going to do, it has largely done inside the first few minutes — which is exactly the window where a retail order is least likely to fill at the price on the screen.

Does the first move continue? Yes, and here is the control

This is the question underneath “trade the news”, and it cannot be answered by looking at the releases alone. In a market that trended upward across the sample, most minutes continue in their own direction. Saying “81.8% of NFP moves continued” without that context would be meaningless.

So 30,000 randomly chosen minutes were measured identically: take the direction of the minute, then check whether price is still on that side of its open some minutes later.

The gap is there at every horizon, and it does not decay the way the control does. An hour after the release, gold was still moving the way the release minute moved 81.8% of the time against a 54.4% baseline.

The sample is 11 releases. That is small enough that the 95% confidence interval on the 60-minute figure runs from roughly 59% to 100%. The lower bound still sits above the control, which is why this is reported as a finding rather than noise — but it is a suggestive finding on eleven events, not a settled one, and anyone quoting it should quote the sample size with it.

A correction to our own spread research

We publish a study saying the XAUUSD spread is flat at $0.16 across all 24 hours. That is true across all bars, and this data shows what it hides.

For the first six releases in the sample the release-minute spread was exactly the normal 160 points — no widening at all. From February 2026 onward it ran 1.5× to 2.25×, peaking at 360 points on the March release.

The honest reading is that this is a change in the account or the broker’s pricing partway through the sample, not a law of markets. A market-wide effect would not switch on in one month and stay on. So we are not going to tell you “spreads always widen at news” — on this feed, for six consecutive releases, they did not. What we can say is that once it started it did not stop, and that sizing a trade on the assumption of a $0.16 spread during a release is betting on which half of the sample you are in.

What we do with this

Our bot does not trade through scheduled high-impact releases. It blocks entries in a window around them, and the reason is execution rather than direction — the continuation edge above is real but it is not worth the tail. A position sized for a $4.81 minute meeting a $40.48 one is not a risk a stop loss reliably contains, because price can travel through the stop rather than to it.

That is the same argument as our scalping cost study: gold’s problem for short-horizon trading is rarely the entry logic and almost always the distance between where you intended to be filled and where you were. If you do trade releases, size for the maximum rather than the median, and use a server-side stop.

Method and limits

352,587 one-minute XAUUSD bars from a live MetaTrader 5 feed on a raw-spread account, 2025-07-31 to 2026-08-04. Eleven NFP releases fall in that window. April 2026 is absent because the first Friday was Good Friday and the market was shut.

Only NFP is measured. CPI and FOMC releases move gold too and are not in this dataset, so nothing here should be read as covering “news” in general. One symbol, one broker, one year, eleven events. The control is 30,000 random minutes from the same archive, measured with the identical definition, which is the part that makes the continuation figure mean anything.

The script is research-news-spikes.py and it runs on the same published archive as our other studies — see the research index. If you think the release-minute definition or the continuation test is wrong, the inputs are public.

Common questions