We Measured 25 FOMC Days on Gold: What Actually Happens

We had the data sitting in the pipeline, so we measured it: every FOMC day from January 2023 to July 2026 — 25 of them — against 771 control days (594 after cleaning), all ranges normalized by ATR so that a wild week and a dead week are comparable.

What we found is a day with a remarkably consistent shape and — we want to be upfront about this — no directional edge at all. If you came here hoping for “Gold rises on FOMC,” close the tab now. If you want to know when the range actually happens, how compressed the morning really is, and where the whipsaw trap sits, read on.

Method, before the numbers

• Sample: 25 FOMC days, January 2023 → July 2026. XAU/USD spot (Dukascopy), intraday data.
• Control group: 771 non-FOMC days over the same window, 594 of them “clean” after filtering.
• Normalization: every range is expressed as a multiple of ATR, then FOMC days are compared to the same clock window on control days. Raw dollar ranges would just measure whether 2024 was more volatile than 2023; normalized ranges isolate the FOMC effect itself.
• The announcement: 2:00 pm ET — 20:00 Paris time, the clock we use throughout (our timing studies are all Paris-time based; see the volume-by-hour study).
• n = 25 is small. It’s every FOMC meeting in the window — you can’t manufacture more Fed days — but it caps how fine-grained the conclusions can be. We report p-values so you can judge each claim on its strength.

Finding 1 — The morning is genuinely dead (0.67× normal)

The folklore says “nothing happens before the Fed.” For once, the folklore is right — and now it has a number.

The pre-announcement session runs at 0.78× a normal day’s equivalent window, and the morning window specifically at 0.67× — a compression that is statistically unambiguous (p=0.0001). Liquidity providers widen out, positioning goes quiet, and the market simply refuses to move ahead of the statement.

This has a direct practical consequence: a morning “breakout” on FOMC day is fighting a market that historically does one-third less than usual. Range-based setups calibrated on normal mornings are operating in an environment that isn’t there.

Finding 2 — 79% of the day happens after 20:00 Paris

Here is the number that should reorganize your Fed-day schedule: on FOMC days, roughly 79% of the day’s total range is printed after the 20:00 Paris (2pm ET) announcement. On normal days, that same evening slice accounts for about 20% of the day’s range.

Four-fifths of the day, compressed into the final hours. The post-announcement window runs at 4.35× its normal-day equivalent (p < 1e-16 — this is the single most statistically overwhelming result in our entire pipeline). Meanwhile the full FOMC day is only 1.24× a normal day (p=0.006).

Sit with that combination for a second, because it’s the whole story: FOMC days are only modestly larger in total, but violently redistributed. The Fed doesn’t so much add range as it relocates it — starving the morning at 0.67× and detonating the evening at 4.35×. Trade the morning like a normal day and you get chopped by compression; treat the whole day as “crazy volatility” and you’re wrong too. The day is quiet-then-loud, on schedule.

Finding 3 — The +30 minute trap

The first minutes after the statement look dramatic and measure smaller than they feel: the median move at +5 minutes is about 0.16 ATR (~$5.7), and by +15 minutes the median is around $7.2.

Then comes the part that fills the loss columns. Tracking the net progress of the move from the announcement forward, there is a local trough at roughly +30 minutes — the point where the initial move has, on the median path, partially given itself back. The +30 minute zone lines up with the press conference, and it behaves like a whipsaw window: initial spike, retracement deep enough to hit reactive stops, and only then the durable move of the day.

The trader’s translation: the window where the urge to act peaks — right after the statement, into the presser — is exactly the window where median net progress stalls. Chasing the +5 minute candle has meant, on the median FOMC day, watching the position go nowhere or backward for the next half hour before the real move develops.

We measure a similar spike-then-fade anatomy around 8:30 ET data releases, but FOMC’s version has its own clock — the presser makes the +30 minute trough a Fed-day signature.

The honest part: there is no directional edge

We looked for one. Does Gold systematically rally on FOMC? Fade the first move? Over-extend and snap back? Squeeze in a predictable direction?

No. We found no systematic over-extension or squeeze pattern, and no directional tendency worth reporting on 25 events. Anyone selling you “Gold does X on Fed day” is selling a story that this dataset — every meeting for 3.5 years — does not support. And with n=25, even a mild-looking directional lean would be statistically fragile; we’re not going to dress one up as an edge.

The edge that exists is structural, not directional: compression before, explosion after, on a published schedule. That is tradeable knowledge — it just isn’t a signal.

What does that mean concretely?

• The morning is for preparation, not positioning. At 0.67× normal range, the expected value of grinding the pre-Fed chop is historically poor. Marking levels beats trading them.
• The event is the evening. If your process needs the day’s range to work, ~79% of it arrives after 20:00 Paris. Being at full attention at 14:00 and gone by 19:00 is exactly backwards on Fed day.
• The +30 minute window deserves structural respect. Whatever your entry logic, the median path says initial move → retrace trough around +30 → durable move. Plans that must act in the first five minutes are volunteering for the whipsaw.
• Size for 4.35×, not 1.24×. A stop calibrated to “FOMC days are a bit bigger” meets a post-announcement window doing four times its normal range.

This structural read — where today’s expected range stands, what implied vol is pricing, what the macro backdrop is doing — is what Range & ADR, the GVZ Vol Monitor and the Gold Macro Dashboard put on the chart. None of them predicts the Fed. They tell you which regime you’re standing in, with the historical distributions attached.

Limits, stated plainly

• n = 25. That’s every FOMC day in the window, and it’s still a small sample. The compression/expansion structure clears very demanding p-values (down to <1e-16) precisely because the effect is enormous — but finer slices (by hike vs. hold, by surprise size) would shred the sample and we won’t pretend otherwise.
• These are range observations, not a backtest. Multiples of ATR measured on historical data — no entries, stops, targets, spreads, or slippage. Nothing here is a strategy’s expectancy.
• Medians hide tails. The median +5 minute move is ~$5.7; the tail FOMC day is a different animal, and one violent meeting can dwarf the median path.
• One instrument, one era. XAU/USD spot, 2023–mid-2026 — a specific hiking-and-holding cycle. A different policy era can reshape the pattern. Historical frequencies are not guarantees.

Bottom line

Across 25 measured FOMC days, Gold’s Fed day has a signature: a morning at 0.67× normal range, a day only 1.24× normal in total, but ~79% of it after 20:00 Paris at 4.35× normal pace — with a median whipsaw trough around +30 minutes before the durable move. And no directional edge, because there isn’t one to find.

The myth-testing continues elsewhere in this series: do liquidity sweeps really reverse? (no more often than a coin flip) and does Asia predict London? (no — and we’ll show you why publishing that “no” matters).

Delta Terminal ResearchStudies run on 3.5 years of XAU/USD spot data, developed hand-in-hand with a full-time Gold day trader. We publish the negative results too.