Research · Dataset
Gold and the Fed cycle
"Buy gold when the Fed cuts" is one of the most repeated claims in gold commentary. It is also one of the least tested. This page puts 800 months of gold against the actual Federal Reserve policy record — 274 cutting months, 187 unchanged and 338 hiking months — and asks four separate questions: does a cut month differ from an ordinary month, what happens after a cut, does gold anticipate the cut, and does any of it hold up decade by decade.
Data through 2026-09 · Gold source updated November 03, 2026 · Policy rate from fred.stlouisfed.org/series/FEDFUNDS
Source and method
How a month is labelled. Month t is a cut when the monthly average effective funds rate in t is more than 1 bp below month t-1, a hike when it is more than 1 bp above, otherwise unchanged. The label uses only information available at the start of month t. The label uses only the rate change that was already published at the start of the month, so there is no look-ahead in the classification itself.
What a "month return" is. Gold here is a monthly average price from the World Bank Pink Sheet, and the funds rate is the monthly average of the daily effective rate. A month return is the change in a monthly average — not an entry and exit you could have executed, and not a number that reflects when inside the month the Fed announced. No transaction cost, financing or storage cost is modelled anywhere on this page.
Sample caveat. The effective funds rate series starts in 1954-07 and the Pink Sheet gold series in 1960-01, so the overlap begins in 1960-01. The 1960s are distorted by the Bretton Woods peg ($35/oz until 1971-08) and are flagged wherever they appear.
The headline: a cut month is not a special month
The gap between cut months and every other month is 0.060 percentage points — roughly one hundredth of one percent, which is smaller than the rounding error in most gold forecasts. A naive t-test gives p = 0.857, but monthly gold returns are strongly autocorrelated, so that test is inflated by construction. The number to trust is the block bootstrap: resampling 12-month blocks (which preserves the within-year dependence between monthly returns) 2,000 times puts the 95% interval on the difference at -0.92 to 0.83 percentage points, and it crosses zero comfortably.
The obvious follow-up is that cuts cluster, so the naive test double counts them. Taking every twelfth cut month and comparing with non-cut months spaced at least twelve months apart gives a difference of -0.034 percentage points (p = 0.980, n = 23 vs 62). The sign flips and the effect stays indistinguishable from zero.
What this does not mean
- It does not mean the Fed has no effect on gold. The correlation between gold and the level of real rates is real and negative; that is measured separately on the macro correlation page. What this page tests is a much more specific claim: that the decision to cut, identifiable in the monthly rate data, is followed by a distinguishable move in gold.
- It also does not mean you cannot trade a Fed decision. It means the trade cannot be justified by a monthly-average statistic of this kind. If you want to trade the decision, you are trading the intraday reaction, which this data cannot see.
Does a bigger cut count for more?
A one-basis-point wobble in the monthly effective rate is not a policy decision. Requiring a real move changes how many months count as a cut, so the answer should not depend on where the threshold is drawn. Here it does.
| Minimum move to count as a cut | Cut months | Gold in cut months | Gold otherwise | Difference | p | 95% bootstrap |
|---|---|---|---|---|---|---|
| 1 bp | 274 | +0.74% | +0.68% | +0.060 pp | 0.857 | -0.92 to 0.83 |
| 10 bp | 179 | +0.97% | +0.63% | +0.348 pp | 0.355 | -0.57 to 1.19 |
| 25 bp | 113 | +1.33% | +0.60% | +0.729 pp | 0.144 | -0.30 to 1.87 |
| 50 bp | 52 | +1.17% | +0.67% | +0.500 pp | 0.509 | -0.54 to 2.17 |
At a 25-basis-point threshold the difference widens to +0.73 percentage points — the largest of the four, and still not significant (p = 0.144), with a bootstrap interval of -0.30 to 1.87 that crosses zero. Tightening the definition to 50 bp shrinks the effect back to +0.500 points on only 52 months.
The pattern worth noting: the effect is largest at a mid threshold and smaller on both sides. That is the shape of a statistic that is being defined rather than measured — the answer moves when you change the question.
What happens after a cut
The claim a trader actually uses is forward-looking: gold goes up in the months following a cut. Measured on every cut month, with forward returns taken from the cut month's own price.
| Horizon after a cut month | Gold after a cut | Gold after any other month | Difference | Naive p | 95% bootstrap |
|---|---|---|---|---|---|
| 3 months | +2.78% | +1.96% | +0.82 pp | 0.2598 | -1.93 to 3.24 |
| 6 months | +3.89% | +5.08% | -1.19 pp | 0.2735 | -5.55 to 3.34 |
| 12 months | +6.54% | +11.90% | -5.36 pp | 0.0019 | -14.90 to 1.96 |
At three and six months the difference is small and the interval spans zero. At twelve months the naive test looks convincing (p = 0.0019) — but the naive test is exactly the test that fails here, because the twelve-month forward windows of consecutive cut months overlap almost completely. They are not independent observations. The block bootstrap, which resamples time in contiguous blocks and therefore keeps the overlap structure intact, gives -14.90 to 1.96 percentage points — crossing zero, and pointing downward.
This is the single most common statistical error in macro backtests of this kind, and it is why the same claim can look significant in one study and vanish in the next.
Cut cycles: the first cut of an easing campaign
Individual cut months are noisy. A "cycle" is a run of consecutive cutting months, and the tradable event is the first of them. The table below covers the 83 cycles that began in 1972 or later — before August 1971 the US was still on the gold standard and the price was a policy artefact, so those cycles are excluded. 26 of the 83 cycles lasted three months or more; the shorter ones are administrative wobbles rather than campaigns.
Here the claim finally has some support — and the support is in the median, not the mean. After the first cut of a cycle, gold's twelve-month median move is +0.80% and it finishes higher in 53.7% of cycles. The mean is +6.47% because a handful of large cycles (1973-74 and 1979-80) drag it upward, not because the typical cycle produced a large gain.
The dispersion is the point. Across the 83 post-1972 cycles the twelve-month outcome ranges from -25.6% to +116.3%, with an annualised standard deviation of 77.7%. A rule that says "gold goes up after the first cut" was right 53.7% of the time — which is close to a coin flip, and leaves the other 46.3% of cycles with a loss, including several where the Fed cut into a rising inflation spiral.
The cycles that lasted three months or more
| First cut | Last cut | Months | Rate cut | Gold over cycle | +3m | +6m | +12m |
|---|---|---|---|---|---|---|---|
| 1973-12 | 1974-02 | 3 | -98 bp | +57.9% | +57.0% | +43.9% | +72.0% |
| 1974-08 | 1975-05 | 10 | -679 bp | +16.8% | +17.4% | +16.1% | +5.2% |
| 1975-10 | 1976-02 | 5 | -105 bp | -9.0% | -8.4% | -10.5% | -18.9% |
| 1976-07 | 1977-01 | 7 | -70 bp | +4.8% | -1.7% | +11.9% | +21.2% |
| 1980-05 | 1980-07 | 3 | -195 bp | +24.4% | +22.0% | +21.6% | -6.6% |
| 1981-07 | 1981-12 | 6 | -667 bp | -11.1% | +7.1% | -6.1% | -17.1% |
| 1982-05 | 1982-08 | 4 | -433 bp | +4.0% | +9.0% | +24.3% | +31.1% |
| 1982-10 | 1983-02 | 5 | -120 bp | +12.4% | +14.0% | +2.6% | -6.6% |
| 1984-09 | 1985-01 | 5 | -295 bp | -12.9% | -5.9% | -7.7% | -5.0% |
| 1985-04 | 1985-06 | 3 | -74 bp | +1.0% | -3.1% | -0.3% | +4.0% |
| 1986-01 | 1986-05 | 5 | -129 bp | +6.2% | -2.0% | +0.3% | +17.6% |
| 1986-07 | 1986-10 | 4 | -71 bp | +23.6% | +21.8% | +17.2% | +29.6% |
| 1989-05 | 1989-08 | 4 | -82 bp | -5.0% | -1.6% | +6.2% | -0.5% |
| 1989-10 | 1990-01 | 4 | -61 bp | +13.3% | +11.7% | +1.9% | +3.8% |
| 1990-10 | 1991-05 | 8 | -233 bp | -8.2% | +0.8% | -6.0% | -5.8% |
| 1991-07 | 1992-01 | 7 | -179 bp | -3.5% | -2.5% | -3.8% | -4.1% |
| 1995-12 | 1996-02 | 3 | -38 bp | +5.2% | +2.3% | -0.5% | -4.7% |
| 1996-12 | 1997-02 | 3 | -10 bp | -8.2% | -4.6% | -7.6% | -21.7% |
| 1998-02 | 1998-04 | 3 | -6 bp | +6.6% | +0.7% | -4.4% | -3.4% |
| 1998-09 | 1999-01 | 5 | -88 bp | +1.1% | +1.0% | -1.0% | -8.3% |
| 2000-12 | 2002-01 | 14 | -467 bp | +6.0% | -3.0% | -0.4% | +1.9% |
| 2007-08 | 2008-05 | 10 | -304 bp | +33.7% | +21.2% | +38.7% | +26.2% |
| 2008-09 | 2008-12 | 4 | -165 bp | -2.7% | -1.7% | +11.3% | +20.1% |
| 2019-08 | 2019-11 | 4 | -58 bp | +4.1% | -1.9% | +6.5% | +31.3% |
| 2024-09 | 2025-01 | 5 | -80 bp | +9.7% | +3.0% | +16.0% | +42.7% |
| 2025-09 | 2026-01 | 4 | -58 bp | +41.1% | +29.6% | +28.7% | —% |
Read the worst rows rather than the best: 1981-07 (the Volcker squeeze — the Fed cut into a collapsing economy and gold fell -17.1% over the following year), and 1975-10, where the first cut of that cycle was followed by -18.9% over twelve months. A cut is not a signal about gold; what the cut was for is.
Does gold anticipate the cut?
The "Fed put is priced in" claim says the market moves before the decision. The test: what did gold do in the six months before the first cut of each cycle?
Gold's six-month move before the first cut averaged +1.39% with a median of +0.29% — and it was higher in 51.8% of cycles. There is no anticipation signal in the run-up to a cut.
The third card is the more interesting one. Gold's six-month move before a hiking month averaged +6.20% — several times the pre-cut figure. Gold's big pre-policy moves happen on the way up in rates, not on the way down. That is consistent with the rest of this site's macro work: gold's dominant driver is the real rate, and the run-up to a hiking cycle is the period in which the market reprices real yields upward. A cut, by itself, changes very little about the real rate — and if inflation is accelerating, cutting can raise the expected real rate.
The pre-1972 sample behaves differently (+1.14% over 109 cycles) only because the pegged price contributes zeroes. That is why the cycle analysis above uses the post-1972 sample throughout.
Is it stable across decades?
A full-sample average can be a weighted blend of decades that disagree. Here they disagree.
| Decade | Cut months | Gold in cut months | Gold otherwise | Difference | Cut months up |
|---|---|---|---|---|---|
| 1960s | 47 | -0.15% | +0.12% | -0.274 pp | 6.4% |
| 1970s | 46 | +1.63% | +2.78% | -1.144 pp | 41.3% |
| 1980s | 54 | +0.42% | -0.10% | +0.519 pp | 51.9% |
| 1990s | 59 | +0.08% | -0.60% | +0.681 pp | 44.1% |
| 2000s | 40 | +0.99% | +1.37% | -0.376 pp | 45.0% |
| 2010s | 17 | +1.04% | +0.15% | +0.887 pp | 64.7% |
| 2020s | 11 | +4.63% | +0.91% | +3.724 pp | 72.7% |
The sign of the cut-month premium flips: negative in the 1960s (-0.27 pp), 1970s (-1.14 pp) and 2000s (-0.38 pp), positive in the 1980s (+0.52 pp), 1990s (+0.68 pp), 2010s (+0.89 pp) and 2020s (+3.72 pp). A rule whose sign depends on the decade is a description of history, not a trading rule.
The 2020s row is the one that fuels the current version of the claim (11 cut months, +3.72 pp difference, gold up in 72.7% of them). It is also the smallest sample on the page. The 2020s cuts came with a collapse in real yields and a liquidity scramble, which is a very different regime from anything in the previous forty years.
Also worth noting: in the 1960s the cut months were up only 6.4% of the time — but that decade's gold price was administratively pegged, so it is not a measurement of anything.
The silver control
Silver comes from the same workbook, on the same dates, with the same method. If the Fed effect were specific to gold, silver should not show it. If both show it, it is a commodity-wide dollar effect rather than a gold effect.
Neither metal shows a cut-month effect: gold +0.060 points (p = 0.857), silver +0.044 points (p = 0.940), and both bootstrap intervals cross zero. In cut months gold and silver returned together with a correlation of 0.600 — which is what you'd expect if the cut itself carries no metal-specific information and both are simply following the dollar.
This is the same control structure used on the gold/silver ratio page and the seasonality page: when two series from the same source show the same thing, the finding is about the source, not about gold.
The recessions themselves
The most common form of the claim is the loosest: gold goes up in recessions. The NBER recession record in FRED gives 9 recessions since 1960 to test against. Gold's mean month inside a recession was +1.05% against +0.54% outside one — a difference of +0.52 points with p = 0.442. Directionally the claim holds; statistically it is not established.
| Recession | Months | Gold worst drawdown in it | Gold +12m after | Rate at start | Rate at end |
|---|---|---|---|---|---|
| 1960-05 → 1961-02 | 10 | 0.00% | +0.0% | 3.85% | 2.54% |
| 1970-01 → 1970-11 | 11 | -16.67% | +22.9% | 8.98% | 5.60% |
| 1973-12 → 1975-03 | 16 | -10.83% | +24.3% | 9.95% | 5.54% |
| 1980-02 → 1980-07 | 6 | -23.85% | -20.4% | 14.13% | 9.03% |
| 1981-08 → 1982-11 | 16 | -37.00% | +21.3% | 17.82% | 9.20% |
| 1990-08 → 1991-03 | 8 | -12.95% | -5.2% | 8.13% | 6.12% |
| 2001-04 → 2001-11 | 8 | -4.06% | +22.7% | 4.80% | 2.09% |
| 2008-01 → 2009-06 | 18 | -5.58% | +62.0% | 3.94% | 0.21% |
| 2020-03 → 2020-04 | 2 | -0.31% | +10.6% | 0.65% | 0.05% |
Two rows decide this. In the 1981-08 → 1982-11 recession gold fell 37.0% from its pre-recession peak while the Fed cut from 17.82% to 9.20% — the fastest rate decline in the sample, with gold ending the following year +21.3%. And in 1980-02 → 1980-07 gold fell 23.9% and was -20.4% a year later. These were inflation-driven recessions: the Fed cut because inflation was already running, and real yields kept rising.
The 2008-01 → 2009-06 row is the case the claim was built on: gold's worst drawdown in the recession was only 5.6% and it was +62.0% over the following year, with the funds rate falling from 3.94% to 0.21%. One episode, one outcome, and it is the reason the rule exists.
How to use this — and how not to
- This page tests a narrow claim — that the Fed's monthly rate decision carries information about gold that survives a proper statistical treatment — and finds that it does not. It is not a claim that macro policy is irrelevant to gold; see the macro correlation page for the real-rate and dollar relationships, which are measured on changes rather than on decision dates.
- The monthly-average price is not an executable entry or exit. A strategy that acted on a Fed announcement would be trading intraday, and this data cannot tell you whether that would have worked.
- The post-cut median of +0.80% over twelve months, positive in 53.7% of cycles, is not a forecast and not a performance claim. It is the historical shape of an event that also produced a 46.3% loss rate.
- Nothing on this page is investment advice, and no statistic here establishes a forward edge. Past policy cycles do not repeat with the same monetary regime, and the 2020s sample is too small to extrapolate.
For the mechanism behind the real-rate and dollar relationships, see Gold, Real Yields and the Dollar. For sizing a position around a policy scenario, see Gold Position Sizing and the risk calculators. For how gold behaved in each of the nine recessions in raw price terms, see Gold Price History. For the balance-sheet side of the same policy story — Fed total assets against gold, tested on levels and on changes — see Gold and the Fed balance sheet.
Frequently asked questions
Does gold rally when the Fed cuts rates?
Not reliably, and the difference is small. Across 274 cut months the mean gold return was 0.7% against 0.7% in non-cut months, a gap of +0.06 pp with a naive p of 0.8574. The cut-month distribution is also wider than the hold-month distribution: 15.1% annualised volatility in cut months against 12.1% in hold months.
What about gold returns in the months after a cut?
The forward returns are more positive than the cut-month return, but the gap versus non-cut periods is not statistically distinguishable from zero. Over 3 months, cut months average 2.8% against 2% for non-cut months, a difference of +0.82 pp with p = 0.2598 and a bootstrap interval from -1.9 to 3.2 percentage points that crosses zero. Over 6 months the cut-month mean is 3.9% against 5.1%. The level of returns is high; the cut-specific part is not.
Is gold already pricing a cut before the Fed delivers it?
Partly, and that is where the asymmetry sits. In the 6 months before a first cut gold averaged 1.1% with a positive share of 42.2%, and in the 3 months before it averaged 0.4% with a positive share of 39.5%. By contrast the 6 months before a hike averaged 6.2%. Pre-cut performance is real but it is weak evidence of a cut-specific edge, since it is close to what an unchanged rate path would produce.
What does gold do during cutting cycles as a whole?
Across the 111 identified cutting cycles, of which 56 had two or more cuts, the mean cumulative gold return over the cycle was 2% with a positive share of 44.1%. Looking forward from the cycle start, the 3-month mean is 3%, the 6-month mean 4.6% and the 12-month mean 5.5%, all with positive shares around 45 to 50%. A cutting cycle is a decent environment for gold in absolute terms and an unremarkable one relative to everything else.
Does gold outperform during recessions?
Only slightly, and the sample is small. Across the 9 US recessions in the sample, gold averaged 1.1% per month against 51.5% of non-recession months being better, and only 3 of the 9 recessions saw positive gold returns. Nine recessions is too few to support a strong claim in either direction.
How much of the sample is distorted by the Bretton Woods gold peg?
A meaningful part, which is why the post-peg figures are published alongside. The official price was fixed at $35/oz until August 1971, so pre-1972 gold "returns" are trading noise around an administered price rather than price discovery. Restricting to first cuts from 1972-01 onward, the 6 months before a cut averaged 1.4% with a positive share of 51.8%, and the 3 months before a cut averaged 0.3% with a positive share of 47%.
What exactly counts as a cut month here?
A month in which the monthly average effective federal funds rate is more than 1 basis point below the prior month's average is labelled a cut; more than 1 basis point above is a hike; anything else is unchanged. The label uses only information available at the start of the month. On this definition the sample contains 274 cut months, 338 hike months and 187 unchanged months from 1960-01 to 2026-09.