Key takeaways
- Gold beat US inflation in 53.1% of rolling one-year windows since January 1975, 52.2% of five-year windows and 53.4% of ten-year windows. At twenty years it was 62.4%.
- Realised inflation explains 4.8% of the variation in gold's real return over one year, 4.4% over five and 5.2% over ten. The World Gold Council's own figure since 1971 is 16%.
- The worst twenty-year window began in January 1980 and lost 7.87% a year in real terms. Gold ended that window with about a fifth of the purchasing power it started with.
- Moving the start date from 1975 to 1988 lifts gold's twenty-year record against US inflation from 62.4% to 100%, with no change to the method.
- In sterling, gold beat UK CPI in 71.7% of ten-year windows since January 1988, measured on a sample that excludes the 1980 peak entirely.
Does gold beat inflation? The record at one, five, ten and twenty years
Gold beat US consumer price inflation in 53.1% of rolling one-year windows between January 1975 and July 2026. Over five years it was 52.2%. Over ten years, 53.4%. Over twenty years, 62.4%.
Three of those four numbers sit within four points of a coin toss. That's the answer, and everything below is about why it's harder to interpret than it looks.
The test divides the London Bullion Market Association's afternoon gold fixing, averaged by month, by the US consumer price index. When that ratio rises across a window, gold gained purchasing power. When it falls, gold lost purchasing power, whatever the dollar price did. Over the full 51.4 years gold returned 6.30% a year in dollars against inflation of 3.68%, which is 2.53% a year in real terms.
A positive long-run real return is a different claim from an inflation hedge. A hedge is supposed to move with the thing it hedges. Exposing that difference is what the four holding periods do.
Why the test starts in January 1975, and what it was checked against
The US price of gold was set by the government for most of the twentieth century, so a series that runs back through those decades measures policy rather than a market. Claude Erb and Campbell Harvey drew the same boundary in The Golden Dilemma, their 2013 National Bureau of Economic Research working paper. Their sample "starts in 1975 because that is when U.S. citizens were once again able to own and trade gold".
Two independent price series were used here, and they agree closely. The LBMA fixing and the World Bank's monthly Pink Sheet gold price differ by less than half a percentage point at every date checked, including $176.27 against $176.00 in January 1975 and $4,073.92 against $4,073.00 in July 2026. That matters because the whole result rests on the price series being right.
The method also reproduces a published figure. Erb and Harvey report that the ratio of the gold price to the CPI index "reached a low value of 1.46 in March of 2001 and a high value of 8.73 in January 1980", using a different price source. The same calculation on LBMA data gives 8.68 in January 1980 and 1.47 in April 2001. Close enough to trust the machinery.
The gold inflation hedge is close to a coin toss below twenty years
Here is the record, window by window. Each row counts every overlapping monthly start date for which a full holding period existed inside the sample.
| Holding period | Windows | Beat US CPI | Median real return | Worst | Best |
|---|---|---|---|---|---|
| 1 year | 606 | 53.1% | 1.55% a year | -42.65% | 160.86% |
| 5 years | 558 | 52.2% | 1.28% a year | -19.86% | 24.89% |
| 10 years | 498 | 53.4% | 1.01% a year | -9.79% | 17.58% |
| 20 years | 378 | 62.4% | 3.18% a year | -7.87% | 9.02% |
The chart plots the third column. What it shows is a flat line that lifts only at the far right, and the lift arrives at a horizon most people don't hold anything for.
The spread is the part worth sitting with. A ten-year holding period is long by any household standard, and it still contained outcomes from -9.79% a year to +17.58% a year in real terms. The worst ten-year window started in July 1980. The best started in August 2001. Neither of those dates was chosen for its inflation rate.
Inflation explains almost none of gold's real return
If gold tracked inflation, then knowing what inflation did over a window would tell you something about what gold's real return did. Regressing one on the other across the same windows gives an R-squared of 4.8% at one year, 4.4% at five years and 5.2% at ten years. R-squared is just the share of the variation in one number that moves with the other, so those readings say that roughly 95% of gold's real return came from somewhere other than inflation.
This is not a contrarian reading. The World Gold Council publishes the same finding, in research that elsewhere argues gold is "systematically" underweighted in private portfolios. In Beyond CPI, published on 21 April 2021, it writes that "for all the talk of gold as an inflation hedge, its relationship to changes in the US CPI is surprisingly poor" and that "since 1971, only 16% of the variation in gold prices can be explained by changes in CPI inflation".
Their 16% and the 4.4% to 5.2% here are measuring slightly different things, on different samples and different frequencies. They point the same way. Two sources with opposite incentives agree that the link is weak.
Where the relationship does get strong, it runs backwards
At twenty years the R-squared jumps to 40.6%, which looks at first like the gold inflation hedge finally showing up. It isn't. The slope is negative. Across twenty-year windows since 1975, gold's real return was lower when realised inflation was higher.
Sorting those windows by realised inflation makes it concrete. In the quartile with the lowest inflation, between 2.03% and 2.38% a year, gold beat inflation in 100% of windows with a median real return of 5.17% a year. In the quartile with the highest inflation, between 3.15% and 5.44% a year, gold beat inflation in 14.9% of windows, with a median real return of -2.79% a year.
Read literally, that says gold did worst against inflation exactly when inflation was worst. It's the opposite of the claim.
The high-inflation result rests on ten start years, and it should not be pushed hard
Read carefully, though, it says something narrower, and this is where the evidence thins out. Those overlapping twenty-year windows are not independent observations. The 94 windows in the highest-inflation quartile come from just 10 distinct start years, all of them between 1975 and 1986. They are largely the same stretch of history counted many times over.
What that quartile really contains is one episode: the run into January 1980 and the long unwinding afterwards. Gold rose 160.86% in real terms in the year to January 1980, then fell 42.65% in the year to July 1981. The twenty years from January 1980 lost 7.87% a year in real terms, which left an ounce of gold buying about a fifth of what it bought at the start.
So the honest version is not that high inflation is bad for gold. It's that the one high-inflation period in the modern record also happened to contain the most expensive entry point in the modern record, and the sample cannot separate the two. That distinction is the difference between a finding and a story. The gold drawdown that followed 1980 is long enough to swallow most investing lifetimes.
Move the start date and the verdict flips
Rerun the twenty-year test with a later start date and nothing about gold changes, but the answer does. Starting in 1975, gold beat inflation in 62.4% of twenty-year windows. Starting in 1988, it beat inflation in 100% of them, with a median real return of 5.36% a year.
Every start year from 1977 onwards raises the figure, without exception. The whole difference is whether the sample includes the windows that begin at or just after the 1980 peak. Nothing about the relationship between gold and prices is doing the work. The start date is doing all of it.
That's the strongest thing this exercise establishes. A claim whose truth depends on where you begin measuring, rather than on what inflation actually did, isn't describing a hedge. It's describing a price history. Ranked against other inflation hedge assets, gold's distinguishing feature is that its payoff is not contractually tied to a price index at all.
The sterling record is kinder, for the same reason
UK readers are exposed to gold in pounds, against UK prices. The LBMA publishes a sterling fixing, and the Office for National Statistics publishes the CPI index back to January 1988, which sets the sample.
On that sample gold beat UK CPI in 57.9% of one-year windows, 64.0% of five-year windows, 71.7% of ten-year windows and 100% of twenty-year windows. Over the full period gold returned 6.55% a year in sterling against UK inflation of 2.85%, so 3.59% a year in real terms.
Every one of those numbers is better than the dollar equivalent, and none of it is a sterling effect. Running the US test over the matched period from January 1988 gives 100% at twenty years too. The UK series simply starts after the peak. The apparent difference between the two countries is the same start-date effect wearing a different currency.
The best case for gold doesn't rest on inflation either
The strongest counter-argument isn't that these numbers are wrong. It's that CPI was never the right yardstick. The World Gold Council makes exactly this case in its Gold Long-Term Expected Return work, published on 17 October 2024. It notes that conventional models "land on an expected long-run real return ranging between 0% and 1%", and argues instead that gold's long-run return "has been well above inflation for over 50 years", more closely mirroring global gross domestic product.
That's a serious argument and the realised numbers support the premise: 2.53% a year in real terms over 51.4 years is well above zero. But notice what it concedes. The gold industry's own best case for gold says the driver is economic growth, not the price level. If gold is priced off global output rather than off CPI, then "gold protects against inflation" is not the mechanism its own advocates are defending.
The Council's Beyond CPI paper puts a harder number on the same point. Measured from the previous gold price peak in the third quarter of 1980, it reports 2.6% to 3.8% a year for gold against 2.9% a year for CPI. On its own accounting, from the last comparable starting point, gold roughly matched inflation across four decades and did not clearly beat it.
Claim against evidence
| The claim | What the record shows | Verdict |
|---|---|---|
| Gold protects against inflation over a year | Beat CPI in 53.1% of windows; inflation explains 4.8% of the real return | Not supported |
| Gold protects against inflation over five to ten years | Beat CPI in 52.2% and 53.4% of windows; R-squared 4.4% and 5.2% | Not supported |
| Gold protects against inflation over twenty years | Beat CPI in 62.4% of windows, but 100% if the sample starts in 1988 | Depends entirely on the start date |
| Gold does best when inflation is worst | At twenty years it beat CPI in 14.9% of the highest-inflation windows | Contradicted, on 10 start years |
| Gold has delivered a positive long-run real return | 2.53% a year in real terms from January 1975 to July 2026 | Supported |
Where the starting point sits today
Since the argument turns out to be about entry price rather than inflation, the current entry price is the relevant number. The ratio of the gold price to the US CPI index averaged 4.16 over the sample. It was 3.38 in January 1975, 8.68 at the January 1980 peak and 1.47 at the April 2001 low.
In July 2026 it stood at 12.20, and it reached 15.36 in February 2026. The real price of gold is higher now than at any point in the modern record, including 1980. Erb and Harvey found that when this ratio was above average, subsequent real gold returns tended to be below average, and they were careful to call that a "known unknown" rather than a rule.
None of that forecasts anything. It does mean that anyone reading the twenty-year row of the table is reading an average of start dates, most of which were considerably cheaper than today's. Loss recovery arithmetic applies to purchasing power as much as to nominal prices.
What would change this conclusion
Three things would, and each is testable rather than rhetorical.
The first is a second high-inflation episode with a different starting valuation. The whole ambiguity here comes from one confounded period. An inflationary stretch that begins with the real price of gold near its 4.16 average, rather than near a peak, would separate the two effects for the first time since 1975. That test hasn't been run by history yet.
The second is the yardstick. The case that gold tracks global output rather than CPI is checkable. If gold's real return lines up with world GDP growth across the same windows where it fails to line up with inflation, the World Gold Council's framing survives and the inflation framing does not.
The third is the sample itself. Fifty-one years is one monetary regime, one reserve currency and one inflation history. Erb and Harvey's own conclusion is that gold "may be an effective hedge if the investment horizon is measured in centuries", and the modern record is not long enough to test that. A backtest is a description of what happened, not a distribution of what could have. On this evidence the gold inflation hedge is not something the last five decades of data support at any horizon a person actually holds for, and the strongest defence of gold has already moved on to a different argument.