gold FOMC trading strategy

Trading Gold Around FOMC Decisions: 268 Meetings, Tested

We measured what gold actually did around 268 FOMC meetings from 1994 to 2026 using LBMA PM fixings, with the reaction measured from the first fix published AFTER the statement, not the meeting day. Gold fell on 52.9% of hikes and rose on 52.9% of cuts. FOMC moves really are 1.52x larger than a normal trading day, but nothing about the direction is predictable.

By Auriatech Research Desk · · Last updated 2026-11-12 · 3 min read
FOMCFederal Reserveevent tradinggoldmacrobacktest Contains our own backtest data
What this is. A measurement, not a strategy. Every FOMC meeting from February 1994 to October 2026 was reconstructed against the LBMA Gold Price PM auction, and the reaction to each decision was measured from the first fix published after the statement was released. The sample is 268 meetings: 51 rate hikes, 34 rate cuts, and 183 meetings where the target was unchanged or no decision was attached. Every rule for matching a policy decision to a meeting, and every number below, is specified so you can rebuild it and disagree with it.

There is a ritual that a lot of gold traders observe. The Fed announces its rate decision, the market moves, and the folklore says gold falls on hikes and rallies on cuts. It is a rule simple enough to fit on a sticky note, and it has just enough truth in it to survive every attempt to check.

This article checks it against 268 FOMC meetings, using the same benchmark price that actually settles most of the physical market, and it finds that the rule does not survive the check. What does survive is smaller and more useful: FOMC days really are a different regime of volatility, and that is something you can size for even when you cannot call the direction.

The short version

  • Gold fell on 52.9% of rate hikes and rose on 52.9% of rate cuts. Both are coin flips.
  • The average reaction to a hike was −0.013% and to a cut −0.169% — the wrong sign for the cuts. Every confidence interval crosses zero.
  • FOMC moves really are bigger: 1.52× a normal trading day, with a 95% interval of 1.21 to 1.86 that does not contain 1.0. This is the one finding here that is solid.
  • Nothing is “already priced in” in a way you can trade. Regressing the reaction on the pre-meeting drift gives R² = 0.0007. The pre-meeting drift and the reaction moved in opposite directions 55.8% of the time.
  • The biggest moves were mostly not about the decision. The largest single reaction, +7.08%, came on a meeting where the Fed changed nothing.
  • The timing detail that changes everything: the LBMA PM fix is set at 15:00 London, and the statement comes out at 14:00 US Eastern — 18:00 or 19:00 London. The fix printed on the meeting date cannot contain the decision. If your backtest uses the meeting-day price, it is measuring the wrong thing.

The data

Item Value
Price source LBMA Gold Price PM, USD per troy ounce
Meeting source Federal Reserve FOMC calendars, per-year historical pages
Decision source Federal Reserve Open Market Operations tables
Sample 268 meetings, 1994-02-04 to 2026-10-28
Scheduled / unscheduled 263 / 5
Rate hikes / cuts / no change 51 / 34 / 183
Baseline days (event era) 8,239 daily returns
Baseline annualised volatility 16.46%
Full-sample annualised volatility 19.31%
Decisions attached to a meeting 85
Decisions outside any meeting 7

Why the reaction is measured the way it is

This is the part that most event-study gets wrong, and getting it wrong flatters the result.

The LBMA Gold Price PM auction is set at 15:00 London time. The FOMC statement is released at 14:00 US Eastern time, which is 18:00 or 19:00 London depending on whether the two countries are on the same daylight saving schedule — three or four hours after the fix that carries the meeting’s date.

So the meeting-day fix is fixed before the decision exists. A backtest that says “gold moved X% on FOMC day” using that fix is measuring the run-up into the meeting, not the reaction to the announcement.

This study defines the reaction as:

  • t_pre = the last business day on or before the meeting date
  • t_post = the first business day after the meeting date
  • reaction = the return from the t_pre fix to the t_post fix

The same rule is applied to every meeting, including the ones that fell on a Sunday or on a UK bank holiday when no fix was published. There is no special-casing, because a special case would itself be a choice about which meetings to include.

Matching decisions to meetings

The Federal Reserve’s Open Market Operations table does not use a single date convention across the whole sample. Decisions from 1994 to 2007 are dated on the announcement day; decisions from 2015 onward are dated the day the new target became effective. Matching on exact date equality alone would have captured only 53 of the 92 in-sample decisions and silently reclassified the other 39 as no-change meetings.

Each decision is therefore matched to the meeting on the same day, or failing that to the meeting one or two days earlier:

Match Count
Same day 53
One day earlier 31
Two days earlier 1
Attached to a meeting 85

The two-day case is real rather than a fudge: the statement for the March 2, 2020 emergency meeting is dated March 3, and the table shows March 4. FOMC meetings are weeks apart, so a two-day lookback can never claim a meeting belonging to a different decision.

Seven decisions in the sample did not come from any meeting — the intermeeting moves, such as the 50bp cut on 2001-01-03 or the 75bp cut on 2008-01-22. Those are real policy events, but they are not FOMC announcements and they are excluded from the reaction statistics. They are listed separately on the dataset page.

Result 1: the folk rule is a coin flip

Action Meetings Mean reaction Median Std. dev. Rose / fell 95% interval
Rate hike 51 −0.013% −0.093% 1.343% fell 52.9% −0.281% to +0.288%
Rate cut 34 −0.169% +0.134% 1.648% rose 52.9% −0.557% to +0.203%
No change 183 +0.100% +0.026% 1.632% rose 50.3% −0.076% to +0.304%

Read the last two columns together, because they are the whole story.

The direction is unknowable from the decision. Gold fell on 52.9% of hikes. Gold rose on 52.9% of cuts. The rule “hikes are bearish for gold, cuts are bullish” gets you a hit rate of about 53%, and you have to accept a standard deviation of 1.3% to 1.6% on each event. On a 52.9% hit rate with that dispersion, the expected value of trading the rule is negative once the spread is paid.

The signs are not even consistent. The mean reaction to a cut is −0.169%, which is the wrong sign for the folk rule, while the median is +0.134%, which is the right sign. A mean and a median that disagree on sign means the distribution is skewed: a few large cuts produced outsized moves and dragged the mean down, while the typical cut was mildly positive.

Note also that the no-change meetings moved more on average than the hikes did (+0.100% vs −0.013%), with a comparable standard deviation. If the decision mattered, the meetings where nothing was decided should be the quiet ones.

Result 2: what does change on FOMC days is the size of the move

This is the finding that survives, and it is more useful than the one the folklore claims.

The unconditional daily return distribution over the event era has a mean absolute move of 0.712%. The mean absolute reaction across all 268 meetings is 1.083%.

The ratio is 1.52×, with a 95% interval of 1.21 to 1.86.

The interval is a circular moving-block bootstrap over the event reactions (12-month blocks, 2,000 resamples), with the baseline held fixed because it is estimated from 8,239 days and its own sampling error is negligible next to 268 event observations. The interval does not contain 1.0. FOMC days really are larger than ordinary trading days.

Ranking each meeting’s absolute move against the absolute daily return distribution:

Measure Value
Median meeting, percentile of a normal day’s move 64.7th
Meetings above the 95th percentile of a normal day 13.4% (36 of 268)
Mean absolute reaction 1.083%
Baseline mean absolute daily move 0.712%

And the effect persists after the meeting:

Window Annualised volatility before After Ratio
10 days around the meeting 13.58% 14.85% 1.20 (95% interval 1.14 to 1.27)

Volatility was 1.20× higher in the ten days after the meeting than in the ten days before it, and it was higher after 54.5% of meetings. Broken down by decision type, the post-meeting volatility ratio was 1.18 for hikes, 1.18 for no-change meetings, and 1.34 for cuts.

That last figure is worth pausing on. A cut produces the largest sustained volatility expansion, and yet the mean reaction to a cut is negative. If you traded the folk rule and sized on the cut’s volatility, you would have found the volatility and lost the direction at the same time.

(The per-action ratios above are the ratio of the two group means; the per-meeting ratios average to 1.184 for hikes, 1.184 for no-change meetings and 1.341 for cuts, so the ordering is the same either way.)

Result 3: nothing is “already priced in”

The most common explanation for why an event does not pay is that the market has already priced it in. That is a testable claim, and here it fails.

Regressing the reaction on the drift into the meeting:

Statistic Value
Slope −0.0456 (95% interval −0.357 to +0.252)
R² 0.0007
t statistic −0.434
Meetings 268

The R² is 0.0007. The pre-meeting drift explains 0.07% of the variance in the reaction. Widening the window to three days before the meeting drops it further, to R² = 0.0004.

The sign test is the more intuitive version of the same result: the pre-meeting drift and the reaction moved in opposite directions 55.8% of the time (148 of 265 meetings with a non-zero drift). If the move into the meeting were the market positioning itself for the announcement, the reaction would tend to continue the drift. It tends to do the opposite, slightly more than half the time — which is what you would expect from a coin flip.

The same conclusion appears in the pre-meeting drift itself. Across all meetings, gold rose by +0.084% on average in the session before the meeting-day fix, and by +0.357% over the ten trading days up to it. That drift is a fact about the market, but it carries no information about what happens after the statement.

Result 4: the biggest moves were mostly not about the decision

Date Reaction Decision
2009-03-18 +7.08% no change
2013-09-18 +4.96% no change
2012-01-25 +4.67% no change
2008-09-16 +4.30% no change
2006-06-29 +4.12% hike
2013-06-19 −5.85% no change
2026-03-18 −5.54% no change
2020-03-15 −4.81% cut (100bp)
2008-03-18 −4.79% cut (75bp)
2021-06-16 −4.41% no change

The largest single reaction in the sample, +7.08%, came on a meeting where the Fed changed nothing. The same is true of the second, third and fourth largest up moves. Of the five largest up moves, four were no-change meetings; of the five largest down moves, two were cuts and the rest were no-change meetings.

This is the single strongest argument against trading the decision: the events that produce the biggest moves are largely the events where the decision was least informative. What moved gold on 2009-03-18 was not the rate decision — it was everything else that came out in the same hours. A rule that says “hikes are bearish, cuts are bullish” has nothing to say about any of the five biggest moves in the sample.

Across the whole sample, 105 meetings produced a move larger than 1% and 41 produced one larger than 2%.

Result 5: the size of the decision does not explain the size of the move

Size of move Meetings Mean reaction 95% interval Mean absolute move
1–25bp 60 −0.095% −0.322% to +0.136% 0.979%
26–50bp 17 +0.236% −0.088% to +0.549% 0.818%
51–100bp 8 −0.591% −1.903% to +0.745% 2.153%
Over 100bp 0 — not computable —

A tempting story is that bigger moves get bigger reactions. The direction column certainly does not support it: the 51–100bp bucket has the most negative mean reaction, the 26–50bp bucket has the most positive, and the sign flips again in the smallest bucket.

The magnitude column is more sensible — the 51–100bp bucket has the largest mean absolute move at 2.153% — but with 8 observations and an interval from −1.903% to +0.745%, this is not a finding. It is a suggestion, and the sample cannot carry it. There were no moves larger than 100bp in the sample at all, so that row is empty rather than zero.

Result 6: the decade split

Decade Meetings Mean reaction Std. dev. 95% interval Mean pre-meeting drift
1990s 48 −0.070% 0.625% −0.251% to +0.069% +0.145%
2000s 81 +0.170% 1.578% −0.148% to +0.496% +0.124%
2010s 83 +0.025% 1.871% −0.236% to +0.263% +0.011%
2020s 56 −0.010% 1.707% −0.261% to +0.219% +0.083%

Every decade’s interval crosses zero. The sign flips from the 1990s to the 2000s to the 2010s to the 2020s, which is the same pattern the site’s other macro datasets show: a relationship that survives one decade and reverses in the next is a period effect, not a mechanism.

The standard deviation column tells a more useful story. The 1990s reaction was three times tighter than the 2010s. FOMC-day volatility itself is a regime that changes — an approach to event trading built on one decade’s volatility would be badly miscalibrated in the next.

Result 7: what happens after the meeting

Every meeting, forward windows from the post-meeting fix:

Action 1 day 3 days 5 days 10 days
Hike (51) −0.013% +0.251% +0.318% +0.448%
Cut (34) −0.169% +0.028% +0.572% −0.060%
No change (183) +0.100% −0.050% −0.031% +0.140%
All meetings (268) +0.044% +0.017% +0.112% +0.173%

Out of 16 forward windows tested (four rows — three decision types plus the all-meetings row — at four horizons), exactly one has a confidence interval that excludes zero: the 5-day window after a cut, at +0.572% with an interval from +0.128% to +1.048%.

One window out of sixteen is what you expect from noise. If the FOMC decision carried a genuine directional signal you would expect the significant ones to cluster in the direction the folklore predicts, and to survive in adjacent horizons. One isolated cell, in the opposite direction to the “cuts are bullish immediately” reading, is a coincidence waiting to be overfitted.

Result 8: the cost side of the ledger

The direction is not knowable, so the only question that matters for an event-trading approach is whether the move is large enough to pay for being wrong. Two things are worth separating here, and the first is the one people usually get wrong.

The folk rule already has negative expectancy before you pay anything. Applying the rule literally — sell gold on hikes, buy gold on cuts — the mean outcome is −0.013% on hikes and −0.169% on cuts. Both are negative. You are not starting from a small positive edge that costs eat into; you are starting from a small negative one.

And the moves are large relative to the cost of trading them. Taking an illustrative $4,050.90/oz gold price (the last LBMA PM fix in the sample, 2026-11-10) and a $0.30 round-trip spread, one trade costs about 0.0074% of notional. The mean absolute reaction of 1.083% is roughly 146 times that cost.

So the cost is not the reason the folk rule fails. You can afford to be wrong on FOMC days, because the moves are large relative to the spread. The problem is that being right 53% of the time is still losing money: the expected move does not compensate for the occasions when you are wrong by 1.6% or more, and the edge you are paying for is already negative before the spread is deducted.

Note that this calculation assumes a $0.30 spread on gold, which is generous, and that it is a round-trip figure. Real spreads widen around announcements, and the pre-announcement spread is what you actually pay — which is another argument for treating FOMC as a volatility event to be sized for, not a direction to be expressed.

What this study is not

It measures the LBMA PM fix, which is a benchmark auction price, not a tradable quote. A position transacted at a broker’s price around the announcement will not fill at the fix, and the fix-to-fix reaction is a cleaner measure of the announcement’s information content than any broker-based series would be.

It measures a single asset. Whether a decision is a directional signal for gold says nothing about whether it is a signal for silver, for the dollar, or for rates — and at monthly frequency neither metal shows a cut-month effect: gold returns a cut month ahead of a non-cut month by 0.060 points (p = 0.857) and silver by 0.044 points (p = 0.940), with both bootstrap intervals crossing zero.

The confidence intervals here are for the mean of a noisy distribution. They describe the average reaction across many meetings. They do not bound the outcome of any individual meeting: 93.3% of the 268 reactions landed within ±3% and 98.9% within ±5%, but the tail is real — 18 meetings moved more than 3% and 11 more than 4%, and a single meeting is perfectly capable of moving five or six times the sample average in either direction.

And the sample contains only 34 rate cuts. That is enough to say the “cuts are bullish” rule does not hold. It is not enough to say anything reliable about what cuts do to gold, and nothing here should be read that way.

If you trade gold around FOMC anyway

  • Treat it as volatility, not direction. The one robust finding is that FOMC days move 1.52× more than ordinary days, and that volatility stays elevated for ten days afterwards at 1.20×. If your edge is elsewhere, an FOMC meeting is a reason to check your size, not a reason to press a button.
  • Do not trade the meeting-day fix. If you backtest this, measure from the first fix published after the statement. The meeting-day fix is set hours before the announcement exists.
  • Watch for the no-change meetings. They produced the four largest up moves in the sample and a higher average reaction than the hikes. A “decision-based” filter would have removed them from your tradeable set and with them most of the tail.
  • Know the intermeeting moves. Seven decisions in this sample came outside any FOMC meeting, including the 75bp cut on 2008-01-22 and the 50bp cut on 2001-01-03. If your calendar only lists FOMC dates, it misses a class of event entirely.
  • Recalibrate when the regime changes. The reaction’s standard deviation was 0.625% in the 1990s and 1.871% in the 2010s. A stop or a position size built on one of those is wrong in the other.

How the numbers were produced

  • Price series: LBMA Gold Price PM, 14,723 daily fixes from 1968-04-01 to 2026-11-10. The event-era baseline uses the 8,239 daily returns from 1994-01-01 to 2026-11-10.
  • Meetings: the Federal Reserve’s own per-year FOMC calendars, including the 5 unscheduled meetings in the sample.
  • Decisions: the Federal Reserve’s Open Market Operations tables, current and archived.
  • Confidence intervals: circular moving-block bootstrap with 12-month blocks and 2,000 resamples for the event means and the magnitude ratio. The block length preserves the serial correlation that overlapping forward windows create; the sample is far too small for an iid bootstrap to be honest here.
  • Verification: every number on this page was recomputed by a second, independently written implementation that reads the published per-meeting CSV and the raw LBMA series, using numpy rather than the production script’s pure-python loops. 109 checks, 0 failures. Where the two implementations disagree by a bootstrap’s Monte Carlo error, the interval was widened to match the observed noise rather than the other way round.

The full per-meeting table, the upcoming meeting dates and the complete methodology are on the dataset page, and the raw data is downloadable as gold-fomc-events.csv.


Educational content only. Not investment advice. Past performance is not indicative of future results. Trading forex and CFDs carries substantial risk of loss.

Frequently asked questions

Does gold fall when the Fed raises rates?

Not reliably. Across the 51 rate hikes in the sample, gold fell on 52.9% of them and rose on the rest. The average reaction to a hike was -0.013%, with a confidence interval from -0.281% to +0.288%, and the standard deviation of the reaction was 1.343%. A 52.9% hit rate on a market with that much dispersion is a coin flip with extra steps.

Does gold rally when the Fed cuts rates?

Also not reliably. Gold rose on 52.9% of the 34 rate cuts in the sample. The average reaction to a cut was -0.169% — the wrong sign for the folk rule — with a confidence interval from -0.557% to +0.203%. The median was +0.134%, which is a good illustration of why the mean alone is a poor summary here.

What is the biggest single FOMC move for gold in this sample?

The largest reaction was +7.08% on 2009-03-18, a meeting where the Fed made no change to the target rate at all. The largest decline was -5.85% on 2013-06-19, also a no-change meeting. Of the five largest up moves, four were no-change meetings; of the five largest down moves, two were cuts and the rest were no-change meetings. The biggest moves in the sample were mostly not about the rate decision itself.

Are FOMC moves actually bigger than normal trading days?

Yes, and this is the one thing the data does support. The mean absolute reaction across all 268 meetings was 1.083% against 0.712% for an unconditional trading day, a ratio of 1.52 with a 95% interval of 1.21 to 1.86. The interval does not contain 1.0. The median meeting ranks at the 64.7th percentile of ordinary daily moves and 13.4% of meetings exceed the 95th percentile of a normal day.

Is the decision already priced in before the announcement?

No detectable relationship. Regressing the reaction on the drift into the meeting gives a slope of -0.0456 with an R-squared of 0.0007 across 268 meetings, which is essentially zero explanatory power. Widening the pre-meeting window to three days drops the R-squared further, to 0.0004. The sign of the pre-meeting drift also fails to predict the sign of the reaction: the two moved in opposite directions 55.8% of the time, close to what chance alone would give.

So should I trade gold around FOMC meetings at all?

If you do, treat it as a volatility event rather than a directional one, and size for the fact that the direction is not knowable from the decision. The reaction's standard deviation was 1.343% for hikes, 1.648% for cuts and 1.632% for no-change meetings. One practical note on timing: the LBMA PM fix is set at 15:00 London, and the FOMC statement is released at 14:00 US Eastern, which is 18:00 or 19:00 London. A fix printed on the meeting date is set three or four hours before the decision exists, so this study measures the reaction from the first fix published after the statement.

Sources

  1. Federal Reserve — FOMC meeting calendars and information (meeting dates, 1994-2027) — accessed 2026-11-12
  2. Federal Reserve — Historical Materials by Year (per-year FOMC calendars, 1994-2026) — accessed 2026-11-12
  3. Federal Reserve — Open Market Operations table, 2003-2026 (target rate changes) — accessed 2026-11-12
  4. Federal Reserve — Open Market Operations table archive (target rate changes before 2003) — accessed 2026-11-12
  5. Federal Reserve — Federal Reserve announces time changes for FOMC statements and news conferences (statements released at 2 p.m. ET) — accessed 2026-11-12
  6. LBMA — About LBMA Daily Auction Prices (gold price set at 10:30 and 15:00 London time) — accessed 2026-11-12
  7. ICE Benchmark Administration — LBMA Gold and Silver Price (auction start times, publication and non-publication days) — accessed 2026-11-12

Use the free tools

Open-source calculators — no signup required.

Use the free tools
Risk disclosure — read before acting on anything above

All content on this site is provided for educational and informational purposes only. Nothing published here constitutes investment advice, a recommendation, or an offer to buy or sell any financial instrument. Trading foreign exchange, CFDs, and precious metals involves a high level of risk and may result in the loss of your entire capital. Past performance of any strategy, robot, or expert advisor (EA) — whether simulated, backtested, or live — is not indicative of future results. You are solely responsible for your trading decisions. If you are unsure whether a strategy is suitable for you, seek independent advice from a licensed financial professional in your jurisdiction.

Educational content only. Not investment advice. Trading forex and CFDs carries substantial risk of loss.