Does Gold Move on the Jobs Report? We Measured 181 Releases
August 7, 2026 · Market Data · 8 min read
The Claim Everybody Repeats
Search for gold and the monthly jobs report and you will find the same story told a hundred ways: a hot payroll number pushes gold down, a weak one pushes it up. The logic is tidy. Strong labor market means the Federal Reserve can stay restrictive, real yields hold up, and a non-yielding metal looks worse by comparison.
Tidy logic is not evidence. So we measured it.
What We Measured
We took gold spot prices (XAUUSD) daily from June 2011 through August 2026, and identified every US Employment Situation release date in that window — 181 of them. Release dates were derived from the Bureau of Labor Statistics scheduling rule rather than eyeballed: the report lands on the third Friday following the reference week containing the 12th of the reference month, moved earlier when that Friday is a federal holiday.
Then we asked one narrow question: does gold behave differently on release days than on any other day?
The Answer: No Direction, Slightly More Movement
| Release days | All other days | |
|---|---|---|
| Observations | 181 | 3,754 |
| Mean daily return | +0.026% | +0.026% |
| Standard deviation | 1.14% | 1.04% |
| Days finishing higher | 54% | 53% |
The average move on jobs day is identical to the average move on a random Tuesday. The difference between them is +0.000% with a t-statistic of 0.00. Removing the COVID period changes nothing meaningful.
What does change is the spread. Release-day standard deviation runs about 1.10 times a normal day. Gold moves more on jobs day — it just doesn't reliably move in a particular direction.
We Also Tested Direction Against the Number Itself
A skeptic would say: of course the average is zero, because good and bad reports cancel out. Fair. So we sorted releases by a payroll surprise measure and looked at gold's response in each bucket.
The correlation between the payroll surprise measure and gold's release-day return was +0.071 (t = 0.80, n = 129) — statistically indistinguishable from zero. The bucket-by-bucket response was not even monotonic: the middle bucket showed a larger average gold move than the strongest and weakest ones. There is no clean gradient here.
An important caveat on that test, stated plainly because it limits the conclusion: markets react to the surprise relative to expectations, not the raw number. We did not have a consensus-forecast series, so we constructed a proxy from seasonally-adjusted payroll momentum. That proxy captures realized labor-market direction, not genuine surprise. Read this as "no evidence found with the data available" rather than "proven to have no effect."
The One Result That Looked Interesting, And Why We're Not Claiming It
Restricting to 2024 onward, release-day returns averaged −0.36% against +0.13% on other days, a gap with a t-statistic of −2.22. That looks like something.
We do not think it is, for three reasons:
- The sign test fails. Only 12 of 30 release days finished higher — a one-sided p-value of 0.18. Nowhere near conventional significance.
- It is concentrated. Three prints during the February–April 2026 unwind (−4.55%, −1.92%, −2.43%) carry most of the effect.
- We looked three times. Full sample, ex-COVID, and recent — testing multiple sample splits and reporting the one that crossed a threshold is exactly how spurious findings get published.
Reporting this honestly matters more than reporting it excitedly.
Where Labor Data Did Line Up With Gold
The relationship that showed up is slower-moving. Sorting months by trailing 12-month payroll growth (excluding COVID, June 2012 through August 2026) and looking at gold's trailing 12-month return in each quartile produced a monotonic pattern: the weakest hiring quartile coincided with an average +29.9% gold return over the following twelve months, the strongest with −3.0%.
That is a real pattern in the sample. It is also badly underpowered, and we would rather say so than let a reader over-interpret it. Those are overlapping 12-month windows drawn from a 15-year price history — roughly 135 observations containing about eleven independent ones. Restricted to non-overlapping annual observations the correlation weakens to −0.46 with a t-statistic of −1.85 across 15 data points. That is suggestive. It is not established, and nothing about it describes what will happen next.
Why "No Edge" Is Still Worth Knowing
A finding that something doesn't work is not a wasted study. Three practical takeaways:
Release day is a range event, not a direction event. Historically, the distribution widened without shifting. Anyone treating the jobs report as a directional catalyst was, over this sample, reading a coin flip with a bigger coin.
Narrative fits the data after the fact. Gold rose on 54% of release days and fell on 46%. Both outcomes generate confident next-day explanations. The explanations are describing noise roughly half the time.
The macro backdrop moved slower than the headlines. The relationship that showed any strength operated on a twelve-month horizon, not a one-day one — and even that one is estimated from about eleven independent observations.
Method and Limitations
Stated openly, because a study you can't audit is a story:
- Gold spot (XAUUSD) daily closes, June 2011 to August 2026. Weekend rows were excluded so returns represent trading-day moves.
- Release dates derived from the published BLS rule, validated against known releases including holiday-shifted ones.
- The 2025 US government shutdown disrupted data collection — the October 2025 household survey was never conducted. Computed release dates for a few months in late 2025 are likely wrong.
- Payroll levels contain an annual benchmark-revision break, so comparisons that cross the revision are not clean.
- Gold price history begins in 2011 here. Nothing in this analysis speaks to the 1970s or 1980s.
- All statistics are descriptive. Overlapping windows make several of the t-statistics unreliable as formal tests, which is why they are labeled where it applies.
The Broader Point
Most market commentary is generated on a deadline, asserted rather than measured, and never checked afterward. Checking is not difficult — it took a price series, a scheduling rule, and a willingness to accept a boring answer.
Equity Rank applies the same standard to company valuation: 19 methods, every input shown, and the model's confidence stated rather than implied. When the data doesn't support a conclusion, saying so is the product.
This article is educational and informational. It describes historical data and is not investment advice, nor a description of how any asset will behave in the future. Past patterns do not establish future results. Equity Rank is not a registered investment adviser.