Research · Dataset
Gold and the Fed balance sheet
"QE raises gold" is one of the most repeated claims in gold commentary, and until now it had not been measured on this desk. This page puts 287 months of gold against the Federal Reserve's total assets and asks the question properly: not whether the levels move together — they do, at 0.874, but so does everything with a dollar sign — but whether balance sheet growth carries any information about gold returns, through every QE and QT episode in the sample, and whether trading it would have worked.
Data through 2026-10 · Gold source updated November 03, 2026 · Fed total assets from fred.stlouisfed.org/series/WALCL
Source and method
The Fed publishes total assets as a weekly series from December 2002, so this study starts there. That is a short sample relative to the other pages on this site and every conclusion is scoped to it.
The headline: two numbers, opposite stories
The level correlation of 0.874 is the number usually quoted, and it is real: the Fed's balance sheet went from $732,059 billion dollars at the end of 2002-12 to $6,745,635 billion at the end of 2026-10, and gold rose over the same window. R² is 0.764 and the Spearman rank correlation is 0.839, so the relationship is not an artefact of a few outliers.
It is also close to meaningless as evidence of a mechanism, because both series trend upward. When the change is measured instead of the level, the correlation is -0.144 — the opposite sign — and once the overlapping 12-month windows are accounted for with a Newey-West HAC estimator the t-statistic is -1.04 with a p-value of 0.297. That is not distinguishable from zero. The naive t-test on the same data gives p = 0.017, which would look significant; the HAC correction is what makes it honest, because each 12-month change shares eleven months with its neighbours.
The block-bootstrap 95% interval for the 12-month change correlation runs from -0.414 to 0.170. It is wide enough to contain both "no relationship" and a meaningful negative one, which is the honest description of what this sample can resolve.
The Fed balance sheet peaked before the gold price did. From that peak the balance sheet has shrunk while gold has risen, which is the largest single contradiction of the simple "QE lifts gold" story in the sample.
Every QE and QT episode in the sample
Averages hide the mechanism. The table below is the actual record of what the balance sheet and gold did in each expansion and contraction the Fed has run since 2002.
| Episode | Window | Fed total assets | Gold |
|---|---|---|---|
| GFC balance-sheet expansion | 2008-03 to 2009-03 | +131.7% | -4.5% |
| QE1 | 2009-03 to 2010-06 | +12.5% | +33.4% |
| QE2 | 2010-11 to 2014-10 | +91.3% | -10.8% |
| QT1 | 2014-10 to 2019-09 | -14.0% | +23.6% |
| COVID QE | 2020-02 to 2021-12 | +110.6% | +12.1% |
| QT2 | 2021-12 to 2025-12 | -24.2% | +140.7% |
| Since the QT2 low | 2025-12 to 2026-10 | +1.6% | -4.7% |
The three rows that break the story are the interesting ones.
- QE2 (2010-11 to 2014-10), the largest QE programme in Fed history: the balance sheet grew 91.3% and gold fell 10.8%.
- QT1 (2014-10 to 2019-09): the balance sheet shrank 14.0% and gold rose 23.6%.
- QT2 (2021-12 to 2025-12): the balance sheet shrank 24.2% and gold rose 140.7%.
QE1 (2009-03 to 2010-06) is the episode that looks like the claim, with gold up 33.4%. But the balance sheet grew only 12.5% over that window, which is less than the crisis expansion that had already happened in the year before it, and it overlaps the tail of the financial crisis rather than a fresh policy regime.
The pandemic is the one period where the relationship looks strong, and it is exactly the period where the two are hardest to separate: the Fed expanded the balance sheet 110.6% while gold rose 12.1%, but the Fed also cut rates to zero, the dollar weakened sharply, and real yields went deeply negative — three things that happened together and each of which has its own claim on gold.
Is the relationship stable over time?
It is not, and that is the second reason to distrust it.
| Period | Months | Correlation | Avg 12m balance sheet change | Avg 12m gold return |
|---|---|---|---|---|
| 2003-2009 | 73 | -0.514 | +17.0% | +16.4% |
| QE era | 72 | -0.225 | +22.3% | +6.2% |
| 2015-2019 | 60 | -0.209 | -2.0% | +1.9% |
| 2020-2021 | 24 | +0.706 | +36.2% | +12.9% |
| 2022-2026 | 58 | -0.364 | -4.0% | +18.7% |
The correlation flips sign in 1 of the 5 sub-periods. In 2022-2026 — the period with the largest balance sheet contraction and the largest gold advance in the sample — the correlation is -0.364. It is strongest and positive only in 2020-2021, at +0.706, which is the pandemic window described above.
The same instability shows up by decade: 2000s gives -0.514, 2010s gives -0.146 and 2020s gives -0.072. All three are negative. A relationship that does not hold its sign across a decade boundary is not a structural property of gold.
Is it really the dollar doing the work?
The most likely alternative explanation is that the balance sheet is not a driver at all but a proxy for something else — most obviously the dollar, and secondarily real yields. That is testable.
| Relationship, 12-month changes | Correlation |
|---|---|
| Balance sheet growth vs gold return | -0.133 |
| Balance sheet growth vs silver return | -0.176 |
| Balance sheet growth vs dollar change | +0.169 |
| Dollar change vs gold return | -0.581 |
| Real 10-year yield change vs gold return | -0.512 |
| Balance sheet vs gold, controlling for the dollar | -0.044 |
| Balance sheet vs gold, controlling for dollar and real yield | -0.120 |
Over the 238-month control panel (2007-01 to 2026-10), the dollar explains gold far better than the balance sheet does: -0.581 against -0.133. Once the dollar is controlled for, the balance sheet's partial correlation with gold falls to -0.044, which is effectively nothing. Controlling for the dollar and the real yield together gives -0.120.
The silver control points the same way. If the balance sheet were driving precious metals, silver should show it too; silver's correlation with balance sheet growth is -0.176, essentially the same weak negative as gold.
The partial correlation is the correlation between 12-month balance sheet growth and the 12-month gold return after removing the linear effect of the control series. This is a decomposition of overlapping stories, not an identification strategy: the Fed expands its balance sheet in response to the same conditions that move the dollar and real rates, so the controls are not independent of the variable being tested.
Does it work as a forecast?
A correlation is not a strategy. This test uses a strictly forward window: the signal is 12-month balance sheet growth known at the end of month t, and the target is the gold return from month t+1 to month t+12, the year after the signal window closes. The two never overlap.
The difference is -5.69 percentage points, with a block-bootstrap 95% interval from -15.33 to 5.57 percentage points. That interval crosses zero, so the difference is not statistically distinguishable from noise. It also points the wrong way for the claim: gold's average forward year was better when the balance sheet was shrinking.
The contemporaneous version of the same split, where the balance sheet change and the gold return share a window, gives -2.74 percentage points with a 95% interval of -13.85 to 7.54. Also crosses zero. For completeness, the unconditional average forward gold return across the whole forecast sample was +10.1%.
What would trading it have produced?
Each month, if 12-month Fed balance sheet growth is above the median of everything observed so far, hold gold for the next month; otherwise hold cash. The threshold uses only prior history. The threshold is recomputed from prior history only, so there is no lookahead.
| Measure | Balance sheet rule | Buy and hold gold |
|---|---|---|
| Annualised return | +2.71% | +10.34% |
| Maximum drawdown | -35.28% | -39.28% |
| Average monthly return | +0.223% | +0.823% |
| Months tested | 214 | 214 |
The rule was in the market 44.9% of the time and returned +2.71% a year against +10.34% for simply holding gold over the same 214 months. It cut the drawdown from -39.3% to -35.3%, which is the only thing it did well, and gave up most of the return for it.
The lead-lag scan offers no rescue either. Scanning balance sheet growth against the gold return from 12 months before to 12 months after, the strongest absolute correlation is 0.229 at a lag of -12 months — the edge of the scan rather than a peak inside it. A signal strongest at the boundary of the window you tested is not a lead-lag relationship.
What this does and does not say
It does not say QE is irrelevant to gold. The narrower and defensible claim is that the balance sheet on its own is not a usable signal for gold at this frequency, and that any effect it has is likely to run through the dollar and real rates rather than as a separate channel. The two are not independent by construction: the Fed expands its balance sheet in response to the same conditions that move those, so this page cannot fully separate them.
It is possible that a real balance sheet effect exists but operates with a lag longer than this sample can resolve, or that it was swamped by the dollar channel during the periods that dominate the sample. Both are plausible and neither can be tested with the data published here.
The limits are worth stating plainly. The sample is short: 287 months, because the Fed publishes total assets as a weekly series from December 2002. It covers exactly one QE cycle, one QT cycle and one pandemic. "Total assets" is a single aggregate that hides what was actually bought, so this page cannot separate asset purchases from the lending programmes that ran alongside them. Gold is a monthly average auction price, not an executable quote, and no spread, financing or storage cost is included anywhere.
Frequently asked questions
Does quantitative easing actually raise the gold price?
Not in any way you could trade. On levels the correlation between gold and the Fed's total assets is 0.874, which looks overwhelming. On 12-month changes it is -0.144 — the opposite sign, and not statistically distinguishable from zero once the overlapping windows are accounted for (HAC t-statistic -1.04, p = 0.297). The level number is a statement about two series that both trended upward over the same 287 months. The change number is the one that tests the claim, and it does not support it.
Why is the correlation on levels so much higher than on changes?
Because both series rose over the sample. The Fed's balance sheet went from 732059 billion dollars at the start of 2002-12 to 6745635 billion at the end of 2026-10, and gold rose over the same window. Two series that both trend up will correlate strongly on levels even if there is no mechanism connecting them. This is the same trap the money supply page documents, and the same reason the page reports the change statistics first. The correlation on levels is robust to how the monthly balance sheet is sampled: taking the monthly average instead of the month-end observation gives 0.874, so it is not an artefact of one observation rule.
What actually happened during QE2, the largest QE programme?
The balance sheet grew 91.3% while gold fell 10.8%. That is the single cleanest contradiction of the "QE lifts gold" story inside the sample: the Fed expanded its balance sheet from 2346001 to 4486754 billion dollars and gold went the other way. QE1 looks like the claim at first glance, with gold up 33.4%, but the balance sheet grew only 12.5% over that window, which is less than a year of the run-up and mostly the crisis response that had already happened.
What happened during the quantitative tightening era?
Gold rose anyway. The balance sheet contracted 14% during QT1 (2014-10 to 2019-09) while gold gained 23.6%. During QT2 (2021-12 to 2025-12) the balance sheet fell 24.2% and gold rose 140.7%. The clearest single stretch is from the April 2022 balance sheet peak to 2026-10: the balance sheet fell 24.5% while gold rose 111.9%, over 54 months. A driver that shrinks while the price doubles is not driving the price.
Is the relationship at least stable over time?
No, and that is the second reason to distrust it. The correlation between 12-month balance sheet growth and the 12-month gold return is -0.514 in 2003-2009, -0.225 during the QE era, -0.209 in 2015-2019, and -0.364 in 2022-2026. It flips sign in 1 of the 5 sub-periods. The one interval where it looks strong and positive is 2020-2021, at 0.706, and that window contains the pandemic: the Fed expanded the balance sheet 36.2% and gold rose 12.9%. Two extraordinary things happened at once, which is exactly the situation where a correlation is least able to tell you which one mattered.
Could the balance sheet be working through some other channel?
We tested the most likely one. The dollar explains gold far better than the balance sheet does: over the 238-month control panel, the correlation between the 12-month change in the broad dollar index and the 12-month gold return is -0.581, against -0.133 for the balance sheet. Once you control for the dollar, the balance sheet's relationship with gold falls to a partial correlation of -0.044 — essentially nothing. Controlling for the dollar and the 10-year real yield together gives -0.12. The honest reading is that whatever the balance sheet is doing to gold, it is doing it through the dollar and real rates, not as a separate channel.
Does it work as a forecast?
No. Taking 12-month balance sheet growth at the end of month t as a signal and measuring the gold return over the following year — a window that does not overlap the signal — gives a forward correlation of -0.07 (p = 0.247) over 274 observations. Splitting the sample by whether the balance sheet was expanding or contracting at the signal date gives 8.1% average forward gold return when it was expanding and 13.8% when it was contracting, a difference of -5.69 percentage points with a block-bootstrap 95% interval of -15.33 to 5.57 percentage points. That interval crosses zero, so the difference is not distinguishable from noise — and note that it points the wrong way for the claim anyway.
What did the walk-forward test say?
It lost money. The rule holds gold for the next month whenever 12-month balance sheet growth is above the median of everything observed so far, and holds cash otherwise, using only prior history. Over 214 months it was in the market 44.9% of the time and produced a 2.71% annualised return with a -35.3% maximum drawdown. Simply holding gold over the same window returned 10.34% a year with a -39.3% drawdown. The rule cut the drawdown slightly and gave up most of the return.
Does the balance sheet lead gold?
No useful lead. Scanning 12-month balance sheet growth against the 12-month gold return from 12 months before to 12 months after, the strongest absolute correlation is 0.229 at a lag of -12 months, which is the boundary of the scan rather than a peak in the middle. A signal that is strongest at the edge of the window you tested is not a lead-lag relationship. The correlation at lag zero is -0.144.
Does this mean QE is irrelevant to gold?
It means the balance sheet alone is not a usable signal for gold, which is a narrower claim. A plausible mechanism still exists: Fed purchases lower the supply of long-duration assets available to the private sector, and gold is one of the few long-duration real assets with no credit risk. But a mechanism that is not visible in the data at the frequency the data allows is not something you can trade on, and this page measures what the data shows rather than what the mechanism might be. It is also possible that the effect is real but operates with a lag longer than this sample can resolve, or that it was swamped by the dollar and real-rate channel that dominates the period.
What are the limits of this study?
Four, and they are significant. First, the sample is short: the Fed publishes total assets as a weekly series from December 2002, so this is 287 months against the 287+ months available to the longer studies on this site, and it covers exactly one QE cycle, one QT cycle and one pandemic. Second, "total assets" is a single aggregate number that hides what the Fed actually bought; the page therefore cannot separate the effect of purchases from the effect of the lending programmes that ran alongside them, and the balance sheet grew 131.7% during the 2008-2009 crisis expansion while gold fell 4.5%. Third, the balance sheet is not independent of the dollar and real rates by construction — the Fed expands it in response to the same conditions that move those — so the partial correlations are a decomposition of an overlapping story, not an identification strategy. Fourth, gold here is a monthly average auction price, not an executable quote, and no spread, financing or storage cost is included.
Sources
- World Bank Commodity Price Data (The Pink Sheet) — monthly gold and silver prices, updated November 03, 2026
- FRED WALCL — Federal Reserve total assets, weekly
- FRED WRESBAL — reserve balances with Federal Reserve Banks
- FRED RRPONTSYD — overnight reverse repo operations
- FRED DTWEXAFEGS — nominal broad US dollar index
- FRED DFII10 — 10-year TIPS constant maturity
- Reproduction:
scripts/gold_fed_balance_sheet_analysis.py, verified byscripts/verify_gold_fed_balance_sheet.py
Educational research only. Nothing on this page is investment advice, and past behaviour does not predict future returns.