Research · Dataset
Gold long-horizon holding periods
There is a sentence that appears in almost every gold advertisement: gold has never lost money over a ten-year period. This page tests it the only way it can be tested properly — by starting in every single month of the price history and holding for five, ten and twenty years, then counting how many of those paths ended below where they began. The claim does not survive. What it actually takes for the claim to look true is more interesting than the claim.
800 monthly observations, 1960-01 to 2026-09 · 14,723 daily benchmark fixings from 1968-04-01 to 2026-11-10 · Gold price updated November 03, 2026
Source and method
Holding-period returns are computed on prices, not on a chained return series. Monthly gold is a monthly AVERAGE price, so these are not executable quotes; the daily benchmark fix series is reported alongside to show the conclusion does not depend on the sampling frequency. No dividend, carry, storage cost, tax or trading cost is included. For spot gold the first four are close to zero or negative, which flatters these figures slightly for an actual holder. The real series pairs a monthly average price with a monthly CPI index, so monthly real returns are an approximation. No statistic on this page is hand-entered.
Two things this page does not do. It does not include storage, insurance, spread or tax costs — for spot gold these are small, but they are not zero, so a physical holder's real return is slightly worse than the figures here. And it does not claim gold is or is not good for your portfolio. It measures the shape of the path.
The answer, first
The claim is false as stated. Starting in any month of the sample and holding for ten years, 132 of 680 paths lost money — 19.4% of all possible start dates. The worst was 1980-07 to 1990-07, which returned -43.55%.
Twenty years is kinder but not kind enough: 52 of 560 twenty-year windows were negative (9.3%), with a worst case of -59.35%.
And in real terms — which is what actually matters if you intend to spend the money — the picture is worse, not better: 261 ten-year windows lost purchasing power, 38.4% of all start dates. Inflation takes its cut from the nominal return, and a decade of high inflation is enough to turn a flat nominal decade into a losing real one.
Where the losing windows are
Almost every negative window starts in one place, and it is worth being precise about where. Of the 132 negative ten-year windows, 89 start before 1990 and 43 start from 1990 onward. The negative windows cluster on the 1980 price spike, which is the only episode in the sample large enough to matter over a decade.
| Start decade | Windows | Losing | Worst | Best | Median |
|---|---|---|---|---|---|
| 1960s | 120 | 0 (0%) | +0.0% | +1,200.0% | +278.6% |
| 1970s | 120 | 2 (2%) | -10.1% | +1,828.6% | +194.9% |
| 1980s | 120 | 87 (73%) | -43.5% | +26.1% | -17.1% |
| 1990s | 120 | 41 (34%) | -31.0% | +301.1% | +9.8% |
| 2000s | 120 | 0 (0%) | +30.3% | +546.7% | +187.1% |
| 2010s | 80 | 2 (3%) | -4.8% | +357.2% | +45.7% |
Ten-year windows grouped by the decade in which the holding period started. A 1970s start decade means the holder bought in the 1970s and sold in the 1980s.
The modern counter-examples
The negatives are not exclusively a 1980s story. The most recent negative ten-year window starts in 2012-10 and ends in 2022-10 at -4.75% — the 2011–2012 spike unwinding, a completely different episode from the 1980 one.
| Start | End | 10-year return |
|---|---|---|
| 2012-09 | 2022-09 | -3.67% |
| 2012-10 | 2022-10 | -4.75% |
How long the 1980 peak took to get the money back
Gold peaked at $675.00 per ounce in 1980-01. The nominal price took 26.3 years to regain that level, reaching it in 2006-05. Adjusting for inflation, the wait was 45.1 years — 2025-02 — and the 1980 peak expressed in 2026-09 prices is worth $ 2,910 per ounce, which gold only passed in 2025-03.
This is the mechanism behind the losing windows. A holder who bought in the 1980 spike and held was not merely down for a decade — they were down for a generation, in real terms, and the outcome of their holding period depended almost entirely on a single price they happened to catch. The claim that gold cannot lose money over long horizons quietly assumes the holder did not buy the spike.
Does the answer depend on how you sample the price?
Monthly averages smooth intramonth moves, which is a real limitation for anything that trades. So the same test was run on the LBMA Gold Price PM benchmark fix — 14,723 daily fixings from 1968-04-01 to 2026-11-10 — where the only thing that changes is the sampling frequency.
| Series | Windows | Losing | Share losing | Worst window |
|---|---|---|---|---|
| Monthly average, 10 years | 680 | 132 | 19.4% | -43.55% (1980-07 → 1990-07) |
| Daily fixings, 10 years | 12,203 | 2,735 | 22.4% | -50.46% (1980-01-21 → 1990-02-06) |
| Monthly average, 20 years | 560 | 52 | 9.3% | -59.35% (1980-09 → 2000-09) |
| Daily fixings, 20 years | 9,683 | 1,111 | 11.5% | -64.44% (1980-01-21 → 2000-02-22) |
The conclusion is the same at both frequencies. Monthly averages find 19.4% of ten-year windows losing; daily fixings find 22.4%. Daily sampling produces slightly more negative outcomes and a deeper worst case (-50.46% against -43.55%), because it catches the intramonth lows that a monthly average steps over. The monthly series is not manufacturing the negative windows — if anything it hides some of them.
When the claim does hold
There is a version of the claim that survives, and it is worth stating plainly because it is what the advertisements are really describing. If you restrict the start dates, the negatives disappear.
| Start dates restricted to | 10-year windows | Losing | Worst outcome |
|---|---|---|---|
| Everything | 548 | 132 | -43.55% (1980-07 → 1990-07) |
| From 1990 | 320 | 43 | -30.99% (1991-01 → 2001-01) |
| From 2000 | 200 | 2 | -4.75% (2012-10 → 2022-10) |
Over twenty years the same story is sharper: from 2000 onward there are 0 negative twenty-year windows out of 80, and the worst outcome is +391.3%.
So the claim holds — for the sample that excludes the 1980 spike. That is not a property of gold. It is a property of the start dates that were selected, and it is exactly the kind of selection that a chart on a sales page performs without telling you. Start at 1980 instead and the same test produces large losses, because that is where the spike sits.
The decade table, which is the honest version
| Decade | Per year | Total | Best year | Worst year | Losing years |
|---|---|---|---|---|---|
| 1960s | +0.00% | +0.0% | +17.1% | -14.6% | 1 of 9 |
| 1970s | +29.24% | +1,200.0% | +118.8% | -24.5% | 2 of 10 |
| 1980s | -1.06% | -10.1% | +24.3% | -23.8% | 5 of 10 |
| 1990s | -3.62% | -30.8% | +14.3% | -21.7% | 7 of 10 |
| 2000s | +14.90% | +301.1% | +39.1% | -4.2% | 1 of 10 |
| 2010s | +2.68% | +30.3% | +22.6% | -27.5% | 4 of 10 |
| 2020s | +19.51% | +191.3% | +62.7% | -3.7% | 1 of 6 |
Gold compounded at -1.06% a year in the 1980s and -3.62% in the 1990s. It compounded at +14.90% in the 2000s, +2.68% in the 2010s and +19.51% so far in the 2020s. Two consecutive decades of negative returns in the middle of the twentieth century is what the claim is quietly asking you to forget.
Note also the worst single year in the 2010s: -27.5%, in a decade that ended higher. Long-horizon returns are averages over paths that repeatedly went underwater on the way. Anyone who held through them experienced a very different product from the compound rate.
What this dataset does and does not support
- It supports the statement that a negative ten-year gold return is possible and has occurred — 132 times in 680 windows, with a worst case of -43.55%.
- It supports the statement that a holder who bought the 1980 spike waited 45.1 years to recover in real terms.
- It does not support the opposite claim either. Gold's long-horizon record is not a guarantee of anything, and a page that used this dataset to argue you will lose money would be cherry-picking the start dates just as hard as the advertisements do.
- It does not settle what any individual should hold. That depends on the portfolio, the horizon and the currency — see our allocation study and the inflation-adjusted dataset for the parts this page cannot cover.
- It counts how many windows lost money, but it does not describe how those losses arrive. They come in long runs rather than scattered years, a negative decade is followed by another negative decade 93.94% of the time, and the wait back to breakeven from a bad entry can run decades. The companion guide Gold Long-Horizon Holding Periods works through the run length, the conditional probability, the full return distribution and the recovery waiting time, computed from the same published CSVs.
Frequently asked questions
Has gold ever had a negative ten-year return?
Yes. Starting in any month of the sample and holding for ten years, 132 of 680 windows lost money — 19.4% of them. The worst ten-year window ran from 1980-07 to 1990-07 and returned -43.55%. On the LBMA daily benchmark fix, the same test over 12,203 ten-year windows finds 2,735 negative outcomes (22.4%), with a worst case of -50.46%.
What about twenty years?
Over twenty years the claim comes closer to holding, but it still fails. 52 of 560 twenty-year windows were negative (9.3%), and the worst returned -59.35%. On daily fixings the twenty-year test finds 1,111 negative windows out of 9,683 (11.5%), worst case -64.44%.
Why do all the losing decades cluster in the 1980s?
Because the 1980 price spike was the only episode large enough to matter over a decade, and it did not start recovering until well after the crash. Of the 132 negative ten-year windows, 89 start before 1990 and 43 start from 1990 onward. The most recent negative ten-year window begins in 2012-10 and ends in 2022-10 at -4.75%, which is the 2011–2012 spike unwinding rather than the 1980 one.
How long did it take to get the money back after the 1980 peak?
The peak was January 1980 at $675 per ounce. Nominal price only regained that level in 2006-05, which is 26.3 years later. Adjusted for inflation the wait was 45.1 years, to 2025-02. In today's money the 1980 peak is worth $2,910 per ounce, and gold only passed that in 2025-03.
Is the claim true if you use inflation-adjusted returns?
It is worse, not better. Deflating gold by CPI turns 132 negative ten-year windows into 261, which is 38.4% of all ten-year windows. Inflation takes a cut out of the nominal return, and a decade of high inflation is enough to turn a flat nominal decade into a losing real one.
When does the claim actually hold?
Only when you choose the start date. Restricting ten-year windows to start dates from 2000 onward leaves 2 of 200 windows; from 1990 onward, 43 of 320 windows. That is not a property of gold — it is a property of the sample you chose. Start from 1980 and the same test produces large losses, because that is where the 1980 spike sits.
Does the answer change if you use daily prices instead of monthly averages?
No. On the LBMA Gold Price PM benchmark fix, 14,723 daily fixings from 1968-04-01 to 2026-11-10, the ten-year test finds 2,735 negative windows out of 12,203 (22.4%) against 19.4% on monthly averages. The monthly series smooths intramonth moves; it does not create the negative outcomes.
Does this mean gold is a bad investment?
No, and this page does not claim it. The decade table shows gold compounded at 2.68% a year in the 2010s and 19.51% so far in the 2020s, after -3.62% a year in the 1990s and -1.06% in the 1980s. The finding is about the shape of the path and the length of the drawdown, not about whether gold belongs in a portfolio. A 45.1-year wait to recover in real terms is a cost that has to be priced, not a reason to avoid the asset.
What are the limitations of this dataset?
Gold prices here are monthly averages from the World Bank Pink Sheet, not executable quotes, and no storage, insurance, spread or tax cost is included. For spot gold those costs are small but they are not zero, so the real returns of a physical holder would be slightly worse than the figures here. The 1960–1971 price was administratively pegged, which removes genuine price discovery from the early sample. The CPI-deflated series pairs a monthly average price with a monthly index, so monthly real returns are an approximation.
Sources
- World Bank Commodity Price Data (The Pink Sheet), Monthly Prices sheet, column 70 (Gold, $/troy oz), monthly average. Workbook dated November 03, 2026.
- Federal Reserve Bank of St. Louis, FRED series CPIAUCSL — Consumer Price Index for All Urban Consumers, index 1982-1984=100, monthly.
- LBMA Gold Price PM benchmark fix, daily. Used as an independent sampling check on the monthly series.
-
Method and full recomputation:
scripts/gold_long_horizon_analysis.py. Independent verification:scripts/verify_gold_long_horizon.py.
Nothing on this page is a forecast, a recommendation, or a statement about future returns. It is a description of what the price history actually contains. Past returns do not establish future results, and no holding period protects against loss.