Can you trade the news? I tested 2,751 events against gold

Every trader has had the thought: CPI comes out in ten minutes. If the number is hot, gold drops. I just need to be positioned — or fast. It feels like the closest thing to easy money in markets.

I tested it properly. 2,751 high-impact US news events over 11.5 years — every jobs report, CPI, Fed decision, GDP, retail sales, ISM — matched against 4 million one-minute gold prices (2015–2026). Four claims, four verdicts. Short version: the dream dies, but it dies in an unusually instructive way.

Claim 1: "Big news moves gold"

Verdict: TRUE — violently, and in a predictable direction.

Here is a real one. January 11, 2024: US inflation printed 0.3% against a 0.2% forecast. This is what gold did, minute by minute:

That's the pattern everywhere. Across all 2,751 events, gold moved in the direction the surprise implied — strong economy or hot inflation pushes gold down, weak data pushes it up — 73–86% of the time. That is a real, statistically massive relationship (t = 13). The events rank like this:

Median move in the 3 minutes around the release, in basis points (1 bp = 0.01% — at $3,300 gold, 1 bp is roughly 3 pips, or $0.33). The jobs-report minute also carries wage growth and the unemployment rate — one minute, three numbers.

So far the dream holds. News moves gold, the direction is knowable from the number, and two releases per month do most of the work. Now the bad news.

Claim 2: "So I'll watch the number come out and jump in"

Verdict: NO. The move is over before you can click.

The 73% direction accuracy exists only in the first three minutes — and mostly in the first seconds. I measured what's left if you enter at the earliest realistic human moment: one minute after the release, once you've seen the number and reacted.

Share of events where trading in the surprise direction was profitable, by entry time after the release. 701 events with a clear surprise, 2015–2026.

There is no drift. Gold finishes digesting a news surprise in about three minutes, and then the surprise tells you nothing — not at 30 minutes, not at 4 hours, not the next day. I also tested simply following the direction of the initial spike without predicting anything: same result, coin flip everywhere. Academic studies found the same thing with 2000s data. It hasn't changed.

Claim 3: "Then I'll predict the number and position before the release"

Verdict: NO. Nothing knowable in advance predicts the surprise.

This is the version you asked about if you've ever thought "forecasters always lowball the jobs number." Sometimes they do — for a while. Over eleven years, no release beats or misses its forecast often enough to bet on:

Share of releases where the actual came in above consensus, 2015–2026, ties excluded. The shaded zone is roughly what pure chance produces at these sample sizes.

I also tested the other pre-positioning idea — trading the gap between the forecast and the previous month's number, which is fully public before the release. Hit rate: 46–49%. Dead. And there is no sneaky pre-release drift to ride either: gold goes essentially nowhere in the hour before a release (−0.5 bp on average).

This is the deep reason the whole idea fails: the forecast is already the market's position. Whatever is predictable about the number is priced in before you ever saw the calendar.

Claim 4: "Fine — I'll use a fast robot, or straddle it with stop orders"

Verdict: The costs are engineered to eat exactly this trade.

In the release second, the gold spread at retail brokers widens from roughly 20 pips to 100–200 pips, and market orders slip 15+ pips beyond that. Compare the sizes:

Approximate figures in basis points. Spread and slippage from broker data during NFP/CPI seconds; the retail range is 5–10+ bp and can be far worse.

Straddle strategies (buy-stop above, sell-stop below) fail the same way in practice: the whipsaw triggers both legs, and fills land far beyond your stop levels. The backtest that made it look good used normal spreads; the live market doesn't.

What actually survives

A negative result this clean is the signal — it tells you what not to do, and it leaves three things standing:

1. A risk rule, not an entry rule. "Hot CPI → gold drops 20–30 bp within three minutes, ~85% reliably" is genuinely high-conviction — as a warning. Don't hold a position against the tail into the jobs-report or CPI minute. Time your entries and exits around the calendar instead of through it.
2. A ranking. Jobs report and CPI are 4–5× everything else. If you protect a position around two events per month, it's those two.
3. A closed door. Eleven years, 2,751 events, every version of the idea — react after, predict before, follow the spike, straddle it — tested and dead. You don't need to wonder anymore, and neither do I.

How this was done

FXStreet's economic calendar (every high-impact US event 2015–2026 with forecast, actual and exact UTC timestamp) matched against 1-minute XAUUSD data cross-validated between two independent feeds. Roughly 40 hypotheses were examined — so isolated "almost significant" rows were treated as noise, which they nearly always are. The findings match the academic literature: gold completes news price-discovery in minutes (Elder, Miao & Ramchander 2012), consensus-forecast biases exist but are priced out (Campbell & Sharpe 2009), and the documented pre-announcement drift lives in index and bond futures — not gold.

One methodological trap worth sharing: my price feed's documentation claimed timestamps in fixed EST, but they were actually daylight-saving-aware. The first run showed a beautiful "delayed reaction" one hour after every summer release — which was the release itself, shifted. If your backtest finds news drift at exactly +60 minutes, check your clocks before your conclusions.