Does gold protect you from inflation?

Published 10 September 2026

This is the sentence you hear when prices start climbing, and it has a real core. An ounce of gold really has held its buying power over stretches longer than any of us get. The question a buyer faces is narrower. Over the five, ten or twenty years you would own it, does it keep up? We split the sentence into the three things it rests on. We set the bar for each one before measuring anything. Then we counted every holding period since the gold price became a market price.

The answer, in one sentence

Gold really has held its buying power over stretches longer than a human life — that part holds, on two independent datasets. Over the five, ten or twenty years you would actually own it, it kept up with inflation in roughly six windows out of ten, which is a real tilt in your favour but not the reliability the word "protects" suggests.

Since 1900 the real dollar price of an ounce has risen 5.2-fold, about 1.3% a year.⁵ On our own series it ran 1,007.8% ahead of US inflation between the end of 1928 and the end of 2025.³ Before counting anything, we set this promise's bar at three windows in four. Gold cleared inflation in 58.0% of five-year windows, 57.8% of ten-year windows and 62.9% of twenty-year windows. A typical twenty-year window turned €10,000 into €15,931 in today's money.³

The misses are what the word has to carry. The worst twenty-year window left €2,308 of €10,000. Sixteen start years in a row, 1979 through 1994, ended a decade behind inflation. The third promise — that gold shows up precisely when inflation does — clears the bar we set on both counts, comfortably on the average and barely on the frequency, and we are barely sure of it either way. Six of the eleven high-inflation years since 1972 were positive, against 55.6% of all years in that period.

So the honest reading is one promise kept and two kept on average or in part: a long-horizon store of value with wide swings, rather than a shield you can time. Taken as a whole we are not very sure of that answer — the long-run coin is the firm one, and the two a buyer actually cares about are the two we are least certain of. The bar for each coin is written out under How we checked, so you can judge our judgement rather than take it.

What the explainer already settled

A week ago we published an explainer on this subject, and it deliberately stopped short of a verdict. It laid out what inflation does to money and what gold is as a thing you own. It also showed why two well-known findings that look contradictory are both true. Gold and inflation do lean the same way over long stretches; gold is still unreliable over the horizons people invest over. That piece answers "how does this work". This one answers "so does it protect me". If the words real, nominal and hedge are not yet comfortable, read that one first — we will not repeat it here.

One measurement note before the numbers, because it runs through all of them. The series we count on are American: dollars, and US consumer prices. We write the translations in euros because that is what most readers count in, so read "€10,000" as a unit of buying power rather than a currency conversion. For a euro buyer the dollar-euro rate sits in between, and we have not calculated that.

Where the idea comes from: coins really were gold

The sentence has a long memory behind it. For most of history gold was not an investment you could hold an opinion about — it was the money, and a coin was worth what the metal in it was worth. That is why the idea feels obvious, and it is also why the older part of the record measures something different from the newer part.

Figure 3 puts the dollar price of an ounce next to the US price level, both starting at 100 in 1928. For most of the first forty-three years the gold line hardly moves on its own — not because the market judged gold to be worth the same, but because the dollar price was set by law. It was raised by decree in 1933 and again in 1934, and for the thirty-three years that followed the year-end price never left the range $31.69 to $35.27, with a median of $34.84. Only at the very end of the period does it come loose, as the fixed system breaks down. The data source says so itself, in a footnote to the very column we used: before 1971, gold prices were fixed and mostly stable.³

That has a consequence worth stating plainly. Between the end of 1928 and the end of 1971 the price of an ounce went from $20.66 to $43.62 — a little more than a doubling, and the two decrees alone took it from $20.69 to $34.69 — while US consumer prices went up 2.4 times. So across those forty-three years an ounce lost about 12% of its buying power. During the era when gold most obviously was money, it did not protect anyone from inflation; the law would not let it. Only after 1971, when the link was cut, does the line become a price that buyers and sellers set. And then it does something the price level never does: it swings, hard, in both directions, for decades at a time.

Over a century, the promise holds

Start with the part that survives every test we could think of. Between the end of 1928 and the end of 2025, an ounce ran 1,007.8% ahead of US inflation. That is about eleven times the buying power it started with, or 2.5% a year on top of prices. Measure it from 1971 instead, when the price became a market price, and it is 1,161.1% — 4.8% a year.³

Big cumulative numbers are easy to produce by picking a flattering start or end, so we tried to break this one. Ending in 2024 instead of 2025 — cutting out an exceptionally strong final year — still leaves 584.7%. The two obvious bad moments to have started both survive: end-1980, at the top of the previous gold wave, leaves 95.9% over the forty-five years since, or 1.5% a year — the thinnest yearly rate of any start year we tried. The weakest cumulative one is end-2012, and it still leaves 84.7%. Every start year from 1928 to 2015 and every end year we tried came out positive.

It also holds on a completely separate dataset. The UBS Global Investment Returns Yearbook uses a different price vendor and starts twenty-eight years earlier. It reports that the real dollar gold price has risen 5.2-fold since 1900, an annualised 1.3% a year. Over the fifty-four years since Bretton Woods it reports 4.7% a year — against our own 4.8% for the same idea. They also give 5.8% for a British investor and 4.3% for a Swiss one.⁵ Over that window the objection that we are really just measuring a weak dollar does not survive. This is the first coin, and it is filled.

One honest complication belongs here rather than in the small print. Erb and Harvey, whose work carries much of the rest of this piece, write that gold "has held its value over the last 2,500 years" — and they mean something specific by it. If the price merely moves with inflation, then the real return over the very long run is zero, and a real price far above its own long-run level should eventually come back down.⁶ Our measured windows run clearly above zero: 2.5% a year since 1928, 4.8% since 1971. Those two are not the same claim, and one of the ways they can both be true is if we are measuring up to a moment when gold is historically dear. Hold that thought; it comes back below.

Over the years you actually hold it

Now the coin that decides what happens to your own money. We took every possible holding period since 1971: every five-year stretch, every ten, every twenty. Of each one we asked a single question. Did an ounce at least keep pace with inflation over exactly those years?

The bar was set before we counted, and it was this: a thing you can plan on should fail at most one time in four. Gold cleared inflation in 58.0% of the fifty five-year windows, 57.8% of the forty-five ten-year windows and 62.9% of the thirty-five twenty-year windows. Roughly six in ten, where we had asked for three in four. Figure 4 shows the count, with the same test run on US shares alongside for scale.

The middle outcome was a gain, and a decent one: the typical twenty-year window turned €10,000 of buying power into €15,931. The problem is the spread around that middle, and above all how long the bad stretches lasted. Figure 5 shows every twenty-year window by the year it started. Someone who bought at the end of 1980 had €2,308 of buying power left in 2000 — not a crash they could wait out, but the outcome after two full decades. Thirteen of the thirty-five twenty-year windows ended behind inflation, and all thirteen began between 1974 and 1987.

On a ten-year horizon the run is starker still: sixteen start years in a row, 1979 through 1994, ended their decade behind inflation. Anyone who bought at any point in those sixteen years and held for ten got less buying power back than they put in.

That is the honest shape of the second coin. It is not that gold fails — the middle window gains, and a majority of windows keep up. It is that the failures come in one long wave rather than scattered around, so "roughly six in ten" understates how it feels to live through the wrong one. Half a coin: a real effect, not a reliable one.

Those navy dots are worth a sentence of their own, because they are the price of the calm. Over the same thirty-five twenty-year windows, US shares with dividends reinvested beat inflation every single time, with a median of +288.2%; gold came out ahead of them in eight of the thirty-five. That is not what you should have done — shares and an ounce of metal are not the same kind of thing to own, and the eight windows gold won include the most recent one, 2005 to 2025. It is simply what being protected cost, measured over the same years.

Does it show up exactly when inflation does?

The third promise is the one people mean most literally: when inflation runs hot, gold is the thing that carries you. This is where the evidence is genuinely thin in both directions, so it is worth being slow.

The one study in our set that measures inflation surges separately defines a regime carefully: headline inflation both accelerating and above 5%, with episodes shorter than six months excluded. It measures eight of them across ninety-five years. In those regimes gold returned 13% a year after inflation, against −1% in ordinary times.¹ That is a large difference, and it is why the idea has real support.

But look at what carries it. Gold has no figure at all for four of the eight regimes. All four are before 1971, when the price was still fixed by law; the authors do not say why the cells are empty. Of the four remaining, the biggest is the 1972–74 oil embargo, at 166%. It is printed in grey italics — which in that table means a spot price the authors excluded from their own totals. So the 13% rests on three episodes, ninety-five months in all, one of which was negative. The authors report a hit rate of 67% — which, given that only three of gold's cells count, means two of three. Their statistical test of whether inflationary and ordinary times differ comes out at 1.6, below the level normally treated as clear-cut.

And on the very next page, in a footnote, the same authors point back to earlier work by two of the researchers we also use here — work that calls gold "too volatile to be a reliable inflation hedge" and says its record since 1975 was "largely driven by a single year, 1979". They report that judgement without disputing it.¹

Our own count on the American series says something very similar. Figure 6 shows the eleven years since 1972 in which US inflation topped 5%. The average was +14.3% after inflation, which clears the bar we set. But the median was only +2.4%, six of the eleven were positive, and one year — 1979, an almost exact doubling — supplies most of that average. Six out of eleven is 54.5%; in all years since 1972 gold beat inflation 55.6% of the time. On frequency, a hot year was no different from any other year.

And the most recent test went the other way. 2021 cost 10.1% of buying power and 2022 another 5.5%; 2023's +9.6% did not quite make up for them. Over the three years together, −6.9%. An independent measurement points the same way: of the 28 years in which inflation exceeded 3%, gold returns were negative in 13 of them.⁵ That source does not say which market or which years its count covers, or whether those returns are before or after inflation, so treat it as a second opinion rather than a second measurement of the same thing.

One comparison from the Neville study is worth having, because it is measured on the same idea and it is the strongest thing that can be said for gold here. Across the eight regimes it covers, a 30-year US government bond lost 8% a year after inflation and came out ahead in two of the eight. Inflation-linked bonds gained 2% a year and came out ahead in three of the five regimes for which they have a figure. Gold's 13% is the largest of the three and rests on the thinnest base — three regimes against five and eight.¹ So against a long fixed-rate bond the bar is low, and gold clears it; against an inflation-linked bond the comparison is a good deal less one-sided than the headline numbers make it look.

So the third coin is half filled, and the reason matters. It clears the bar we wrote down: the average is positive and a majority of episodes are positive. That is why we did not leave it empty — raising a bar after seeing the result is exactly what we criticise elsewhere. What it does not clear is the second hurdle. Three regimes, eleven years, one dominant observation and a recent episode that went the other way: we are barely sure of the finding at all.

The price you pay at the door

Here is the thing that surprised us most, and it reframes the whole question. The inflation itself explained very little about what gold then did. Over ten-year stretches, realised inflation had "close to no impact" on gold's return — while the real price you paid at the start had a great deal.⁴ High real prices have historically been followed by low or negative real returns over the next ten years.

We checked the shape of that on our own data in Figure 7, with two deliberate constraints. Each starting year is measured against gold's own average real price up to that point, so the test uses only what a buyer could have known at the time; and the count starts in 1976 rather than 1971, so that every year has at least five years of history to average over. Buying below that running average returned +74.5% on average over the next ten years; buying above it, +9.9%. The odds of merely keeping up, though, were the same either way: 55%. The entry price moved the size of the outcome a great deal, and the odds hardly at all. The direction survives dropping the second constraint, but the two figures do not: counting from 1971, the dear group averages +30.4% rather than +9.9%, and the odds split 55% against 60% instead of matching.

The researchers who built this framework have skin in the argument, and they are careful about it. In 2012 they judged the real price high, expected weak returns, and the real price then fell more than 12% over the next five years — one call, made in advance, that came good.⁶ In 2024 they wrote that at a real price of 7.3 the expected ten-year return was negative "irrespective of the sample" used.⁴ That line is fitted to the same history it is describing, which is worth remembering when it points at a price outside that history.

Their 2025 paper, writing in December 2025, says plainly that prices were then at or near all-time highs, and that no historical ten-year return matches them. Their gloomy reading, they add, "may be pessimistic" if demand for gold has genuinely shifted.⁶ They have already measured part of that shift: since gold funds arrived, the same relationship sits about five percentage points higher, so at $2,500 an ounce their line gives −7% a year for the era before those funds and −2% for the era after. (That $2,500 is in 2025 money, while the 7.3 above is a ratio expressed in 1982 dollars — the two figures come from different papers and are not directly comparable.) We have no forecast. Theirs is explicit and conditional — low or negative on the past record, unless demand has structurally changed, and they say they cannot tell which.

You can read that gauge yourself, which is the useful part. The real price is simply the gold price divided by the consumer price index, so it is a ratio rather than an amount: because the index is based on 1982, a reading of 7 works out at $700 in 1982 dollars — the researchers put the same point as "$730 in 1982 dollars" for their reading of 7.3.⁴ They put it at 7.3 in March 2024, and gold has risen a good deal since. Two things follow, and they pull in opposite directions. It is the highest that gauge has been on our series, which on their framework is when to expect least from the next decade. And if you already own gold, that same high reading is why your holding has done well — the warning and the good news are the same fact.⁴

How you hold it, and what that costs

If the answer is "yes, under conditions", the next question is how you would own the stuff. That changes what is left of the protection. There are three common routes, and they differ less on cost than on what you are actually holding.

One line is missing from that table on purpose: tax. It differs by country and, more awkwardly, by route — the same metal can be taxed one way inside a fund and another way in a drawer. We have not calculated it, so read the table as a comparison before tax, not as the whole cost.

Figure 8 puts those charges in euros, and the honest headline is that they are small next to the swings. Twenty years of a 0.36% storage charge takes €10,000 to €9,304; at 0.10% it takes it to €9,802. A single round trip through a dealer at an 8% spread costs €800 in one go. Real money, worth minding — and not the reason the promise struggles.

Rerun the whole count with a 0.36% yearly charge subtracted and the ten-year result moves from 26 windows out of 45 to 24, and the twenty-year result from 22 out of 35 to 21; the five-year count does not move at all. The route costs you a few percent. The timing cost the worst buyer 77%.

One thing on that table deserves more than a cell. Physical gold's distinguishing feature is that it is nobody's promise — there is no issuer who can fail to pay you. Worth knowing that this is not the same as being beyond reach: for more than forty years, from 1933 to 1975, American citizens were not allowed to hold it at all.⁶ A tracker is convenient and cheap, and it brings that counterparty back in. The product document of one large European gold ETC says so in its own words. It is a bearer note. It carries no deposit protection, total loss of the capital is possible, and the claims on the gold belong to the issuer rather than to you. It is rated 5 on a risk scale of 7.⁷

None of that makes it a bad way to hold gold. It does mean the two routes are not the same thing wearing different price tags.

What the gold industry itself says

It would be unfair to check this claim without letting its strongest advocate make the case. So we read the World Gold Council's own research paper on exactly this question, published in April 2021 — before the inflation episode of 2021 to 2023 that we use above.² The Council is the industry body for gold miners and is funded by them, which is why nothing in this piece rests on a figure of theirs. What is striking is how much of our answer they reached themselves.

They called gold's relationship to changes in US consumer prices "surprisingly poor" and a "weak linear relation". They described gold, on inflation specifically, as "somewhat of a blunt tool". Their own statistical test found the gold price "not significantly cointegrated with US CPI" over the full window from 1971 to 2020. In plain terms: they found no evidence of a stable long-run link between the two — which is not the same as proving there is none.

Their scorecard of inflation hedges is more favourable, and it belongs here too. Inflation-linked bonds and property funds came out "the most consistent hedges against inflation" with the highest average returns, with gold third or second depending on the definition — but, they add, "averaging across all metrics, gold ranks consistently well either as first or joint second". They call that scorecard "indicative" themselves, because of what they describe as a small sample and a period with little high inflation. It runs from January 1998 to December 2020, which means it contains no 1970s data at all.

Their case then moved somewhere else: gold tracks the money supply rather than the price index, and so protects against currency debasement and asset-price inflation rather than the shopping basket. That is a defensible argument and it may well be right. It is also a different claim from the one on this page, and a much harder one for a buyer to check. Worth knowing when you next hear the short version — and worth knowing that the Council has kept developing its case since, so what you read on their site today will not be the paper we tested.

What this means for you

Start with what holds. If you own gold, you own something that has kept its buying power over horizons longer than any of us get: 5.2-fold in real terms since 1900, about 1.3% a year.⁵ None of our sources disputes that it held its value over such spans, though they do not all mean the same thing by it. It is also the rare asset that is nobody's promise. That is a genuine thing to own.

Now the part to hold more loosely. Inflation moves with about 2% volatility a year, gold with about 15%, and that gap is why a single year of high inflation tells you so little about gold. The middle outcome over twenty years was a gain; the worst one left €2,308 of €10,000, and the waits were long. If your horizon is under ten years, the swings simply dominate the signal.

Then what the route costs. Costs shave; they do not break. Twenty years of a 0.36% storage charge takes €10,000 to €9,304, and one trip in and out through a physical dealer at 8% costs more than that in a single day.

A tracker is cheap to hold and hands you back a counterparty; physical has no counterparty and is expensive to trade and awkward to sell. Neither is a trap. They are different trade-offs, and the size of your holding decides which one bites.

And it is worth knowing what the protection cost against the obvious alternative. Over the same twenty-year windows, US shares with dividends reinvested beat inflation every single time, with a median of +288.2%; gold came out ahead of them in 8 of those 35 windows. That is not what you should have done. It is the price of the calm, and whether it is worth paying is a question about you, not about the data.

If someone tells you gold protects you from inflation, one question does most of the work: over which years, and after which costs? Not one of our eight sources reports a gold return over an inflation period after costs — every net figure on this page is our own arithmetic. A second question is nearly as useful: which series, and what is inside it? Over January 1975 to August 2025, one set of researchers prints four published gold series side by side. Two ordinary price measures give almost exactly the same answer, 6.28% and 6.29% a year before inflation. A commodity index that only rolls futures contracts gives 1.30%; the same index counted as a total return gives 5.86%, and they spell out that this 4.56-point gap is interest earned on the cash behind the contracts, not gold.⁶ Four numbers, all correctly labelled "gold", five percentage points apart end to end.

If you want to look into it further

Three habits travel beyond this subject. First, ask which window a claim is measured over, then ask what the neighbouring windows say. The difference between "gold quadrupled" and "gold lost three-quarters of its buying power" is a start date, not a fact about gold. Second, when someone shows you an average, ask for the middle value and the count. An average of +14.3% built from six good years and five bad ones, with one year doing most of the lifting, is a different animal from a steady +14.3%.

Third, find out what the thing costs to buy, to keep and to sell before you compare its returns to anything. Published returns are almost always gross, and yours will not be.

Sources

1. Neville, Henry, Teun Draaisma, Ben Funnell, Campbell R. Harvey and Otto Van Hemert, "The Best Strategies for Inflationary Times", The Journal of Portfolio Management 47(8), August 2021. Author-hosted PDF at people.duke.edu (SSRN 3813202 is the same paper behind a wall). Our source for gold's +13% a year after inflation during inflationary regimes against −1% outside them, the hit rate of 67%, the t-statistic of 1.6, the 30-year Treasury's −8%, the definition of an inflationary regime, and the authors' own footnote calling gold "too volatile to be a reliable inflation hedge". Four of the five authors worked at Man Group, which sells inflation strategies; the paper says so.
2. World Gold Council, "Beyond CPI: gold as a strategic inflation hedge", Investment Update, 21 April 2021. gold.org (we read the publisher's own HTML version; the PDF download is blocked). The industry body for gold miners, and therefore the place where the claim is stated most carefully. We quote what they argue and the reservations they state about their own analysis. No figure on this page rests on this source.
3. Damodaran, Aswath (NYU Stern), "Historical Returns on Stocks, Bonds and Bills: 1928–2025", histretSP.xls, updated 1 January 2026. pages.stern.nyu.edu. The year-end gold price, the US CPI series and the S&P 500 total return in one file, which is what every count and every window on this page is built from. The rolling-window arithmetic is ours, not his.
4. Erb, Claude B. and Campbell R. Harvey, "Is There Still a Golden Dilemma?", SSRN working paper 4807895, version of 7 May 2024. ssrn.com/abstract=4807895. Our source for the finding that realised ten-year inflation had "close to no impact" on ten-year gold returns while the real price paid did, and for the negative expectation at a real price of 7.3. Not peer-reviewed. At least one author discloses a financial relationship of potential relevance.
5. Dimson, Elroy, Paul Marsh and Mike Staunton, UBS Global Investment Returns Yearbook 2026 — public summary edition (data through end-2025; PDF created 24 February 2026), section 6, "Gold as a hedge?". Published by UBS at ubs.com/…/giry2026-summary-public.pdf. Our copy is the PDF created 24 February 2026, taken from a web archive capture of that same publisher address; at the time of writing the address itself serves a later revision of the same summary, created 7 September 2026, in which every passage we cite below appears unchanged. Our source for "the relationship between gold and inflation is weak", the 28 years with inflation above 3% of which 13 saw negative gold returns, the 5.2-fold real rise since 1900 at 1.3% a year, and the post-Bretton-Woods figures of 4.7% (US), 5.8% (UK) and 4.3% (Switzerland). We read only the free summary; the full Yearbook is a paid publication and we have not read it. UBS sponsors it.
6. Erb, Claude B. and Campbell R. Harvey, "Understanding Gold", SSRN working paper, version of 10 December 2025. ssrn.com/abstract=5525138. Our source for the volatility mismatch (inflation about 2% a year against gold's 15%), the dealer spread of 5–8% on physical gold, the fund fees of 10 to 40 basis points, the 2012 call, the "may be pessimistic" caveat, and the four published gold series and what separates them, and the 1933–1975 ban on private gold holding in the United States. Not peer-reviewed; same authors and same disclosure as source 4.
7. Deutsche Börse Commodities GmbH, Basisinformationsblatt (PRIIPs key information document) for Xetra-Gold, ISIN DE000A0S9GB0, drawn up 10 September 2026; and the same issuer's "Gold ETCs: the most important cost models". Our source for the legal form of a gold ETC, the risk rating, the storage charge of 0.36% a year, and the 0.1–0.3% management fees and 0.3–0.4% spreads it reports elsewhere in the market. The issuer describes a market in which its own product does well, so we use it for its own published terms and not for its claims about the alternatives. The key information document itself is not published at a stable address; we fetched it from the issuer on 10 September 2026 and it is archived in our working file.
8. BullionVault (Galmarley Ltd), "BullionVault Tariff", read 10 September 2026. bullionvault.com. Our source for the gold-backed account route: 0.5% commission each way below $75,000, 0.12% a year custody including insurance with a $4 monthly floor, the 0.3% currency-switching fee on euro orders, and the physical-delivery charges: about 1% plus 2–5% for insurance and transport on its standard 100 g bar route, 2.5% for whole 400 oz bars, and 7.5% for other amounts. A seller's own price list, used only for its own prices.

The three coins, one by one. Figure 2 splits the claim into three things that would all have to be true. Each has an explicit bar, written down before we measured, together with what we measured against it and how sure we are — so you can judge our judgement rather than take it. How the three states work, because it decides two of the three coins here: a coin is filled only if it clears its bar and we are at least fairly sure of the finding. A coin that misses its bar is dotted when there is nothing in between — the promise is contradicted, or simply unsupported. It is half filled when the evidence carries the promise on average or in part but not reliably, or when a cleared bar comes with certainty we do not have. That three-way rule is our standing one and it is not specific to gold. 1. Over the very long run an ounce keeps its value — filled. Bar: the real change is at least zero over both the full series (end-1928 to end-2025) and the free-market era (end-1971 to end-2025), and the picture does not depend on one flattering start year. How sure: fairly sure. Measured: +1,007.8% and +1,161.1%. Every start year from 1928 to 2015 and every end year from 2019 to 2025 that we tried is positive; ending in 2024 rather than 2025 still gives +584.7%; starting at end-1980 gives +95.9% over forty-five years, and end-2012, the weakest start year in that range, gives +84.7%. A separate dataset with a different price vendor reaches 4.7% a year post-Bretton-Woods against our 4.8%, in three currencies. Two things travel with this coin. Part of the 1928 figure is a legally fixed price rather than a market one, which is why we report the 1971 window alongside it — and across that fixed-price era an ounce actually lost about 12% of its buying power. And our main research sources hold that the very-long-run real return is zero, which is not the same statement as ours; the section above sets the two side by side rather than stacking them. 2. It keeps it over the years you actually hold it — half filled. Bar: at least 75% of rolling five-, ten- and twenty-year windows since 1971 keep pace with inflation, before and after costs. How sure: not very sure. Measured: 58.0%, 57.8% and 62.9% before costs; 58.0%, 53.3% and 60.0% after a 0.36% yearly charge. We had written the bar as "after the cheapest route's annual charge", which is 0.10%; we tested at 0.36% instead, because that is the charge an issuer states outright. It is the harsher test: at 0.10% the ten-year count is 26 of 45 rather than 24. The bar is missed on every horizon either way, so the coin is not filled — and it is not dotted either, because there is a real middle: the median window gained on all three horizons (+14.2%, +17.7%, +59.3%), so a typical holder kept up and then some. An effect that is genuine on average and unreliable in practice is what the half state is for. Two things sit against the count. The windows overlap, so 35 twenty-year windows are not 35 independent observations; and the 75% bar is our choice, not a property of the claim — though US shares do clear it on all three horizons, so "nothing clears this bar" is not available to us. On the shortest horizon that margin is thin: 76.0% is 38 windows out of 50, and one window fewer would put shares under the bar too. 3. It protects just when inflation runs hot — half filled. Bar: average real return above zero during high-inflation episodes and positive in a majority of them. How sure: barely sure. Measured: the bar is cleared on both counts, on both measures — +13% a year with a 67% hit rate in the published regimes, and +14.3% average with 6 of 11 positive in our own year count. We did not empty the coin, because raising a bar after seeing the result is the thing we criticise in others. What the coin does not clear is the certainty hurdle: the published 13% rests on three of eight regimes and 95 months, one of them negative, with a t-statistic of 1.6; our own average is carried by 1979; the share of positive years, 54.5%, is indistinguishable from the 55.6% of all years since 1972; and the most recent episode, 2021–2023, cost 6.9% of buying power over three years.

How we checked. Verified 10 September 2026. This check runs on eight sources, all of which we opened and read ourselves; two of them were fetched by hand beforehand, one from the author's own university page and one from the publisher's website, because the automated route ran into a block on the way. Every figure we took from a published document was pulled out twice, independently and blind, and then reconciled cell by cell against the source itself — 296 cells compared, 21 differences settled, and none of those differences was a number inside a table. The price series behind our own counts is different in kind: it is a machine extraction from one archived spreadsheet, checked cell by cell against the original rather than read twice by hand. Where a table was rotated or a chart was an image, we read it by position at 400 dpi rather than trusting a text conversion, which is how we caught that a plain-text extraction of the key table puts one column in a different order and would have given gold a statistic of 3.1 instead of 1.6. How we chose what to read. We worked from named anchors rather than running a systematic search of the literature, so this is not a review of everything published on gold and inflation. Where a source's own figure comes from somewhere else, we say so: the 5–8% dealer spread is a trade-press survey of one retailer's bars, cited by the researchers rather than measured by them, and a competing product's issuer puts very different numbers on the same market without giving a source — which is why we carry no figure for the spread on small bars. What is ours rather than a source's. Every rolling-window count, every euro translation and every after-cost figure on this page is our own arithmetic on one archived series; no source in this dossier reports a gold return over an inflation period after costs. Figure 7 is our reading of a pattern Erb and Harvey report, not their figure, and the two-out-of-three behind the 67% hit rate is our reconstruction of their table rather than something they print. Two measurement notes travel with all of it: the gold price is a year-end price while the inflation series is an annual average, a half-year offset that matters little over twenty years and more over single years such as 1979; and the price level is chained by us from annual inflation rates, because the source gives no index level. Where the sources disagree. Erb and Harvey's own text says gold delivered 12% during inflation surges while the table they print on the facing page says 13% — 12% is the silver row. We use 13%, which is what the original study prints in two separate exhibits. They also reproduce that table without the note explaining that the largest gold figure in it is excluded from the totals, and add a colour scheme the original does not have. On costs, the two documents of one ETC issuer are irreconcilable: its key information document reports €7 of total costs on €10,000 after one year and €3 after five, lists no ongoing-cost line at all, and its own cost table does not add up — €4 is unaccounted for — while the same issuer's website states a storage charge of 0.36% a year, which on €10,000 is €36. We report both and use the higher one in our example, because that is the charge the issuer describes as actually levied. And that same issuer supplies most of our tracker column, which is worth knowing: it is one company's terms, not a survey of the market. The World Gold Council's headline calls gold a proven long-term hedge while its own tests find no evidence of a stable long-run link to consumer prices; and the Council disputes the zero-real-return premise that Erb and Harvey build on, so those two sources are not evidence that stacks. What we left out. Gold as a crisis hedge and gold as portfolio diversification are different claims and are not tested here. We did not calculate tax, which differs by country and by route, and we did not convert anything into euros at an exchange rate. Every count here assumes one purchase on the start date; buying a little at a time is a different question and we have not measured it. The horizon counts and the high-inflation years are American throughout — dollars and US consumer prices — while only the long-run coin is checked in more than one currency. Who has an interest. The World Gold Council is funded by gold miners; dealers earn the spread; issuers earn the annual charge; platforms earn commission and storage. And the mirror: we sell nothing, and a brand called "Proof over promises" has its own pull towards a claim that falls over. That is why the bars were written down first. Not investment advice — a check of a widely repeated claim. Think we got something wrong? Tell us at info@proofofreturns.com — we correct in public.