Research · Dataset
Central bank gold reserves and the gold price
Almost every gold bull-market story since 2010 ends with the same sentence: central banks are buying, and that is why the price is up. It is a good story. It is also testable. This page pulls the official gold holdings reported to the IMF, puts them next to the World Bank monthly gold price, and asks the question directly: does the buying lead the price? The answer is no, and the way it fails is more interesting than the answer.
786 monthly observations, 1960-01 to 2025-06 · Reserve series: IMF International Financial Statistics (IFS), via DBnomics · Gold price updated November 03, 2026
Source and method
Official reserves are the reported holdings of monetary authorities. They are not the same thing as net purchases: a country can sell gold without it leaving the banking system, and reported holdings are revised. Monthly changes are used as a proxy for net buying and are labelled as such.
The gold price is a monthly average, not a tradeable quote. A monthly average smooths intramonth moves, so a strategy built on these numbers would not reproduce the fills it would get in the market. Reserve data are reported holdings, which the IMF revises; the series is the current vintage, not the vintage that was published at the time.
The answer, first
Gold rose 9,480% over the sample while official reserves rose 2.03%. Those are not two views of the same fact. The tonnage was essentially flat; the market value of those tonnes rose from $40 billion to $3,920 billion, almost entirely because the price of the metal went up.
The trap: the level correlation looks damning until you see it
The level correlation of -0.3683 is the number most often quoted — a negative coefficient that reads as "central banks sold gold as the price rose". It is an artefact. Reserves are a nearly flat line over 65 years and gold is a strongly trending one; correlating a flat series against a trend produces a spurious negative coefficient that is really just measuring the gold price's own path.
The change-on-change correlation is the honest one, and it is -0.0592. The Spearman version is -0.1078. Both say the same thing: month to month, reserve buying and the gold price are close to unrelated.
Does the buying lead the price?
| Lag (months) | Correlation | Observations |
|---|---|---|
| reserves lead by 12 | -0.0440 | 773 |
| reserves lead by 11 | -0.0734 | 774 |
| reserves lead by 10 | -0.0454 | 775 |
| reserves lead by 9 | -0.0188 | 776 |
| reserves lead by 8 | 0.0155 | 777 |
| reserves lead by 7 | 0.0210 | 778 |
| reserves lead by 6 | -0.0476 | 779 |
| reserves lead by 5 | -0.0483 | 780 |
| reserves lead by 4 | -0.0168 | 781 |
| reserves lead by 3 | -0.0242 | 782 |
| reserves lead by 2 | -0.0521 | 783 |
| reserves lead by 1 | -0.0208 | 784 |
| same month | -0.0592 | 785 |
| gold leads by 1 | -0.0169 | 784 |
| gold leads by 2 | 0.0200 | 783 |
| gold leads by 3 | 0.0137 | 782 |
| gold leads by 4 | 0.0199 | 781 |
| gold leads by 5 | 0.0122 | 780 |
| gold leads by 6 | 0.0163 | 779 |
| gold leads by 7 | 0.0039 | 778 |
| gold leads by 8 | -0.0017 | 777 |
| gold leads by 9 | 0.0273 | 776 |
| gold leads by 10 | 0.0307 | 775 |
| gold leads by 11 | 0.0169 | 774 |
| gold leads by 12 | 0.0059 | 773 |
Negative lag means the reserve change happens before the gold move; positive lag means gold moves first. A lag of −11 means reserve changes eleven months earlier are paired with the gold return.
The strongest reading anywhere in the ±12-month sweep is -0.0734 at lag -11, on 774 observations. There is no lead, no lag, and no edge to trade.
Era by era
| Era | Gold, per year | Gold, total | Reserves, Moz | Reserves, % |
|---|---|---|---|---|
| Bretton Woods to the 1970s inflation shock | +13.96% | +1,437.1% | +7.2 | +0.63% |
| Volcker disinflation and the 1980s bear market | -3.86% | -32.3% | -9.7 | -0.84% |
| The great 1999-2009 gold bull market | +13.42% | +295.5% | -107.5 | -9.87% |
| The 2010s decade | +2.86% | +32.3% | +131.1 | +13.28% |
| The 2020s | +15.16% | +114.8% | +49.9 | +4.46% |
Read the two right-hand columns against the two on the left. The 1999–2009 bull market was the best decade for gold at 13.42% a year and the worst for official reserves at -9.87%. The 2020s are the best era here for gold at 15.16% a year, and official reserves rose only 4.46% over 5.4 years. The one era where central banks bought hard, the 2010s, was the era gold went sideways: 2.86% a year.
Annualised figures are compound monthly growth rates over the stated window; total figures are not annualised. Reserve changes are changes in reported official holdings.
Who sold, who bought
The aggregate world series hides the interesting part, because the direction of travel flipped between the 1999–2008 and 2010–2024 windows — and the popular narrative only covers the second one.
The largest reductions, 1999 → 2008
| Central bank | Dec 1999 (Moz) | Dec 2008 (Moz) | Change (Moz) | Change (%) |
|---|---|---|---|---|
| Switzerland | 83.3 | 33.4 | -49.8 | -59.8% |
| France | 97.2 | 80.1 | -17.1 | -17.6% |
| Netherlands | 31.6 | 19.7 | -11.9 | -37.6% |
| Spain | 16.8 | 9.1 | -7.8 | -46.2% |
| Portugal | 19.5 | 12.3 | -7.2 | -36.9% |
| Austria | 13.1 | 9.0 | -4.1 | -31.3% |
| Brazil | 3.7 | 1.1 | -2.6 | -70.5% |
| Germany | 111.5 | 109.7 | -1.8 | -1.6% |
Switzerland cut 49.8 Moz (59.8%) and France 17.1 Moz (17.6%). The United States barely moved, from 261.7 Moz to 261.5 Moz. This was a period when the gold price rose 295% and the official sector was a net seller.
The largest increases, 2010 → 2024
| Central bank | Dec 2010 (Moz) | Dec 2024 (Moz) | Change (Moz) | Change (%) |
|---|---|---|---|---|
| China | 33.9 | 73.3 | +39.4 | +116.3% |
| Turkey | 3.7 | 24.5 | +20.8 | +556.3% |
| Poland | 3.3 | 14.4 | +11.1 | +335.5% |
| India | 17.9 | 28.2 | +10.2 | +57.1% |
| Mexico | 0.2 | 3.9 | +3.6 | +1,604.0% |
| Brazil | 1.1 | 4.2 | +3.1 | +285.8% |
| Singapore | 4.1 | 7.1 | +3.0 | +72.7% |
| Korea, Rep. | 0.5 | 3.4 | +2.9 | +623.5% |
China added 39.4 Moz (116%), Turkey 20.8 Moz (556%) and Poland 11.1 Moz (336%). Germany, meanwhile, went the other way over exactly the same window: 109.3 Moz to 107.8 Moz. "Central banks are buying" is true in aggregate for the 2010s and untrue as a description of what most central banks did.
The biggest holders
| Central bank | Latest (Moz) | Latest (tonnes) | Historical peak (Moz) | Peak month | Drawdown from peak (%) |
|---|---|---|---|---|---|
| United States | 261.5 | 8,133.5 | 664.3 | 1952-12 | -60.6% |
| Germany | 107.7 | 3,350.3 | 131.7 | 1969-11 | -27.9% |
| Italy | 78.8 | 2,451.9 | 85.2 | 1970-07 | -22.0% |
| France | 78.4 | 2,437 | 149.7 | 1966-09 | -47.7% |
| Russia | 75.1 | 2,335.9 | 75.1 | 2024-04 | +0.0% |
| China | 73.9 | 2,298.5 | 73.9 | 2025-06 | +0.0% |
| Switzerland | 33.4 | 1,039.9 | 88.3 | 1967-12 | -62.1% |
| India | 28.3 | 880 | 28.3 | 2025-06 | +0.0% |
| Japan | 27.2 | 846 | 27.2 | 2022-01 | +0.0% |
| Turkey | 25.4 | 788.6 | 26.9 | 2023-02 | -21.6% |
| Netherlands | 19.7 | 612.5 | 54.8 | 1978-12 | -64.1% |
| Poland | 16.6 | 515.5 | 16.6 | 2025-06 | +0.0% |
| Taiwan, China | 13.6 | 423.9 | 13.6 | 2024-10 | +0.0% |
| Portugal | 12.3 | 382.7 | 27.9 | 1974-09 | -56.0% |
| Saudi Arabia | 10.4 | 323.1 | 10.4 | 2016-01 | +0.0% |
| Spain | 9.1 | 281.6 | 23.1 | 1965-08 | -60.9% |
| Austria | 9.0 | 280 | 21.5 | 1971-07 | -64.1% |
| Belgium | 7.3 | 227.4 | 45.7 | 1971-07 | -87.5% |
| Singapore | 6.6 | 204.1 | 7.7 | 2024-04 | -15.2% |
| Brazil | 4.2 | 129.7 | 9.3 | 1959-12 | -98.9% |
| Sweden | 4.0 | 125.7 | 7.9 | 1955-12 | -48.7% |
| South Africa | 4.0 | 125.5 | 40.3 | 1969-04 | -92.5% |
| Mexico | 3.9 | 120.2 | 5.9 | 1950-12 | -98.6% |
| Korea, Rep. | 3.4 | 104.4 | 3.4 | 2019-05 | -0.0% |
| Indonesia | 2.5 | 78.6 | 8.0 | 1951-12 | -99.3% |
| Canada | 0.0 | 0 | 32.9 | 1965-12 | -100.0% |
"Latest" is each country's most recent observation in the source series, which is 2025-06 for the world aggregate and can be a month or two later for individual countries that report faster. Canada is shown at zero because its reported official holdings are now nil — it sold its position down rather than holding it.
The United States is the one holder in the study far below its own historical peak: 664.3 Moz in 1952-12, bottoming at 261.5 Moz in 1998-11, a drawdown of 60.6%. Every other large holder is at or near its own record.
How much of the world's gold is in central bank hands?
Central banks hold 36,359.1 tonnes of the 224,200 tonnes of gold that has ever been mined — 16.2% of the above-ground stock. Physical investment gold as a whole (central banks, bars, coins, ETFs and OTC) is 45.4%. The remaining 175,400 tonnes sit in reserves and resources.
The above-ground figures are published by the World Gold Council (end-Q3 2026); the central bank tonnage is computed from the IMF series in this dataset. The World Gold Council also notes that around two-thirds of all gold ever mined has been mined since 1950.
Gold as a share of official reserves
| Year | Official gold (Moz) | Gold (tonnes) | Gold price ($/oz) | Reserve market value ($bn) | Gold share of reserves (%) |
|---|---|---|---|---|---|
| 2014 | 1,037.5 | 32,271 | 1,201 | 1,246.1 | 9.48 |
| 2015 | 1,061.6 | 33,018.2 | 1,076 | 1,142.2 | 9.27 |
| 2016 | 1,080.1 | 33,593.7 | 1,157 | 1,249.6 | 10.27 |
| 2017 | 1,093.2 | 34,003.4 | 1,264 | 1,381.9 | 10.57 |
| 2018 | 1,100.6 | 34,233.3 | 1,250 | 1,375.8 | 10.53 |
| 2019 | 1,118.3 | 34,781.8 | 1,479 | 1,653.9 | 12.05 |
| 2020 | 1,135.1 | 35,305.3 | 1,858 | 2,109 | 14.02 |
| 2021 | 1,142.4 | 35,533.2 | 1,790 | 2,044.9 | 12.92 |
| 2022 | 1,141.6 | 35,508.3 | 1,798 | 2,052.6 | 13.87 |
| 2023 | 1,157.5 | 36,003.2 | 2,026 | 2,345.2 | 15.13 |
| 2024 | 1,165.4 | 36,248.1 | 2,648 | 3,086 | 19.21 |
| 2025 | 1,169.0 | 36,359.1 | 3,353 | 3,919.6 | 22.59 |
Gold is 22.59% of total official reserves in 2025-06, against a sample median of 29.60% — 7.0 percentage points below it. The share peaked at 79.48% in 1980-02 and bottomed at 9.21% in 2014-11.
This share falls during gold bull markets for a purely mechanical reason: the numerator is revalued at market while the tonnage barely moves. In 2025-06 the official gold holdings were worth $3,920 billion against $12,917 billion of non-gold reserves.
The unit test
Computed gold market value (Moz x USD/oz) plus the IMF "total reserves excluding gold" series reproduces the IMF "total reserves at market price" series to within this maximum relative deviation. It is the unit test for the share calculation. The full-sample deviation is inflated by 1970-83, when official and free-market gold prices coexisted and the IMF carried gold at official parity; the modern-era figure is the one that applies to recent data.
Maximum relative deviation: 17.1% over the full sample, 5.84% from 1985-01 onward.
The 1980 peak, again
The sharpest version of the "central banks were buying all the way up" claim comes from the 1980 price peak. The window 1979-01 to 1985-12 is the one to look at: gold went from $227 to $322 per ounce while official reserves fell 17.6 Moz (1.51%). The wider 1970s era is the same story: gold up 1,437.1%, reserves up 0.63%.
The Volcker disinflation and the 1980s bear market era runs the other way on both counts — gold -3.86% a year, reserves -0.84%.
What this dataset does and does not support
Supported: official gold reserves were essentially flat over 65 years (2.03%) while gold rose 9,480%; the official sector was a net seller through the 1999–2008 bull market (-91.2 Moz across 25 banks) and a net buyer over 2010–2024 (+95.3 Moz); the month-to-month correlation between reserve changes and gold returns is -0.0592, and no lag in a ±12-month sweep produces a reading above -0.0734.
Not supported: the claim that central bank buying caused or led the gold price. It does not in these data. Central bank demand is real in the 2010s and 2020s, and it is a genuine part of the physical market — but on the published IMF series it neither leads nor predicts the price, and during the strongest modern bull market the aggregate official sector was selling.
This is a description of what the published data show, not a forecast and not investment advice.
Frequently asked questions
How much gold do central banks actually hold?
Official gold reserves reported to the IMF came to 1,168.972 million fine troy ounces in 2025-06, or 36,359.1 metric tonnes. That is 16.2% of the World Gold Council's estimate of total above-ground gold stock (224,200 tonnes). The market value of those holdings at the 2025-06 gold price was $3,920 billion.
Has central bank buying driven the gold price?
The data do not support that claim on its own terms. Over the 785 monthly changes in the sample, the correlation between the change in official reserves and the monthly gold return is -0.0592, and the rank correlation is -0.1078. Both are close to zero. The strongest reading across a ±12-month lead-lag sweep is -0.0734 at lag -11, which is not distinguishable from noise. Central bank buying is real and large in the 2010s and 2020s, but on this data it does not lead or predict the gold price.
Is the negative level correlation between gold and official reserves meaningful?
No, and this is the main statistical trap on this page. The correlation between the level of official reserves and the gold price over the full sample is -0.3683, and it looks like evidence that central banks sold into the gold rally. It is not. Official reserves barely moved over 65 years — from 1,145.663 Moz to 1,168.972 Moz, a change of 2.03% — while gold went from $35 to $3353, up 9480%. A nearly flat series correlated against a strongly trending one produces a spurious negative coefficient. The change-on-change correlation of -0.0592 is the honest number.
What happened during the great 1999-2009 gold bull market?
Official reserves fell. From 1999-01 to 2009-12, gold rose 295.5%, from $287 to $1135 per ounce, while official reserves fell 107.5 million ounces, or 9.9%. That is the clearest single refutation of "central banks drove the bull market": during the strongest sustained gold bull market in modern history, the aggregate official sector was a net seller.
Which countries sold during the 1999-2008 bull market?
Of the 25 central banks in this study with data for both December 1999 and December 2008, 15 reduced their holdings and the aggregate change was -91.2 Moz. The largest reductions were Switzerland (-49.8 Moz, -59.8%), France (-17.1 Moz, -17.6%) and the Netherlands. The United States barely moved, from 261.7 Moz to 261.5 Moz.
Who has been buying since 2010?
The picture reverses. Between December 2010 and December 2024, 15 of 25 central banks in the study increased their holdings, for a net change of 95.3 Moz. The largest increases were China (39.4 Moz, 116.3%), Turkey (20.8 Moz, 556.3%) and Poland (11.1 Moz, 335.5%). Note what the same data show for Germany over that window: holdings fell from 109.3 Moz to 107.8 Moz.
Are central banks buying as much as the headlines say?
Not in aggregate terms, and the difference is worth spelling out. Official reserves in the study's 26-country set rose from 91.2 Moz net sold during 1999-2008 to 95.3 Moz net bought during 2010-2024. Over the whole 65-year sample, official reserves rose just 2.03% while gold rose 9480%. The scale of "central bank demand" is a function of which window and which countries you pick.
What share of official reserves is gold now?
Gold was 22.59% of total official reserves in 2025-06, against a median of 29.6% over the 786-month sample — a gap of -7 percentage points. The share peaked at 79.48% in 1980-02 and bottomed at 9.21% in 2014-11. The share falls in a gold bull market purely because the price rises while the tonnage barely moves: in 2025-06 the gold holdings were worth $3,919.6 billion against $12,917.3 billion of non-gold reserves.
How reliable is that gold share calculation?
It is unit-tested. The computed gold market value (million ounces times the gold price) plus the IMF's "total reserves excluding gold" series is compared against the IMF's own "total reserves at market price" series: the maximum relative deviation is 17.1% over the full sample and 5.84% from 1985-01 onward, which is the figure that applies to modern data. The larger full-sample deviation is a 1970-83 artefact, when official and free-market gold prices coexisted and the IMF carried gold at official parity.
Did central banks sell during the 1980 price peak?
They did. In the window 1979-01 to 1985-12, gold went from $227 to $322 and official reserves fell 17.6 Moz, or 1.51%. The wider 1970s era shows the same pattern: gold rose 1437.1% from 1960-01 to 1980-12 while reserves changed by 0.63%. The official sector did not buy its way into either episode.
Which eras were the strongest for gold, and what did reserves do?
Across the pre-specified eras: Bretton Woods to the 1970s inflation shock — gold 13.96% a year, reserves 0.63%; Volcker disinflation and the 1980s bear market — gold -3.86% a year, reserves -0.84%; The great 1999-2009 gold bull market — gold 13.42% a year, reserves -9.87%; The 2010s decade — gold 2.86% a year, reserves 13.28%; The 2020s — gold 15.16% a year, reserves 4.46%. The two eras with the strongest central bank buying are not the two eras with the best gold returns.
How large are the biggest holders?
By reported official holdings in 2025-06: United States 261.5 Moz (8,133.5 tonnes), Germany 107.7 Moz, China 73.9 Moz (2,298.5 tonnes), Switzerland 33.4 Moz. The US is the only one in the study far below its own historical peak: holdings peaked at 664.3 Moz in 1952-12 and reached a trough of 261.5 Moz in 1998-11, a drawdown of 60.6% from the peak.
Does gold held by central banks count as part of the market?
Yes, which is why the stock comparison matters. The World Gold Council puts total above-ground gold stock at 224,200 tonnes, of which physical investment gold (central banks, bars, coins, ETFs and OTC) is 101,700 tonnes, or 45.4%. Central bank holdings alone are 16.2%. Below ground sit a further 175,400 tonnes of reserves and resources, 78.2% of the above-ground stock.
What does this dataset not tell us?
Several important things. Official reserves are reported holdings of monetary authorities, which are not the same as net purchases: a country can sell gold without it leaving the banking system, and reported holdings are revised. Reported holdings also say nothing about who owns the metal behind them — a central bank holding is not a central bank's own balance sheet decision about gold. The series runs to 2025-06, and the gold price series is a monthly average, so it will not reproduce a tradeable entry or exit point.
Is this a forecast or investment advice?
No. Everything here is a description of what the published IMF and World Bank data show over the period covered, recomputed so that every number can be re-derived from the published CSV files. The correlation of -0.0592 between reserve changes and gold returns says nothing about what gold does next, and nothing on this page is investment advice.
Sources
- IMF International Financial Statistics (IFS), via DBnomics — indicator
RAFAGOLDV_OZT (official gold reserves, millions of fine troy ounces), world aggregate and 35 individual central banks, retrieved through the DBnomics mirror of the IMF's International Financial Statistics. - World Bank Commodity Price Data (The Pink Sheet), Monthly Prices sheet, gold column — the same price source used by the rest of this desk. Updated November 03, 2026.
- World Gold Council, How Much Gold Has Been Mined? — above-ground stock, end-Q3 2026. Accessed 2026-11-15.
-
Analysis script:
scripts/gold_central_bank_analysis.py. Independent recomputation with a separate numpy implementation:scripts/verify_gold_central_bank.py. - Related pages on this desk: Gold vs. real yields and the dollar, Gold and the money supply, Gold and FOMC decisions, Gold and the yield curve inversion, and How much gold should you hold.
Prices are nominal US dollars. Nothing on this page is investment advice or a forecast.