Gold in a Retirement Portfolio: What Data Shows
Quick answer: In this model, adding a slice of gold to a retirement mix can lift the score, lower it, or barely move it. Which one happens depends almost entirely on what gold is assumed to earn after inflation. Under the default assumptions (1971 to 2025 real returns, a 60/40 stock and bond plan, the default example inputs), the plan with no gold produced 87% successful paths. With a ten percent gold share earning what the dataset says gold earned, that became 93%. With the same share earning nothing after inflation, it became 85%. Both results belong together, and neither is a forecast.
Education, not financial advice. This page explains a method and shows what a model produced from stated assumptions. It does not tell you what to buy, sell, save or spend. For your own situation, ask a qualified professional.
What gold did in past rough decades
Start with what the file shows, because the argument for gold usually starts there. Our Monte Carlo simulation retirement calculator draws its numbers from an annual dataset that carries stocks, ten-year Treasury bonds and gold, along with the inflation series used to turn each nominal return into a real one, meaning a return measured after inflation, in today's purchasing power. The site's own default uses the 55 years from 1971 to 2025. A second set runs from 1928 to 2025, which is 98 years.
Over the default period, the average real return on gold in the file was 7.4% a year as a simple average and 5.0% a year compounded, with a yearly spread (standard deviation) of 24%. Stocks in the same years came to 8.4% and 7.0%, with a spread of 17%. Bonds came to 2.5% and 2.0%, with a spread of 10%. Gold's yearly moves were weakly opposite to stocks' moves, with a correlation of -0.20 on a scale where zero means no linear relationship. In the longer set the same measures read 4.1% and 2.5% for gold, with a spread of 19% and a stock correlation of -0.07. Those are descriptions of one sample, not promises about the next one.
Now the decade people point to. The worst ten-year stretch for stocks in this file, in real terms, ran from 1999 to 2008. Over those ten years stocks' cumulative real return was -32%. In the same ten years gold's cumulative real return was +136%, and bonds' was +48%. Anyone holding some gold in that decade would have seen it rise while the stock portion fell. That is a fact about those ten years in this dataset. It does not say gold will rise the next time stocks fall, and the very next section shows why that would be a stretch.
Gold also fell in many years. Since 1971, its real return was negative in 24 of the 55 years, which is 44% of them. For stocks the share was 25%. The last year in the file, 2025, shows a real gold return of 62%, and a single year that large counts toward every average above. That is one more reason to treat the averages as a description of a short sample and not as a steady rate.
What gold did not do
Gold did not hold its inflation-adjusted value in a straight line. Build a real gold index from the same file, starting in 1971, and it reaches its high in 1980. It then falls 77% to a low in 2001. It does not climb back to its 1980 real level until 2011. That is 31 years in which gold's value after inflation sat below where it stood in 1980. For a saver drawing money out through those years, a holding with no real gain would have been a quiet drag rather than a source of growth.
Put the two facts side by side, because they are the same asset. In one decade, from 1999 to 2008, gold rose a great deal while stocks fell. Across the long stretch from 1980 to 2011, gold's real value went nowhere. A plan that leans on the first fact without the second is leaning on half of the record.
There is a second thing the record does not show. It says nothing about why gold moved in either period. The dataset lists returns, not causes, and this page adds no explanation from memory. If a reason for a past move is offered to you somewhere, ask where it comes from.
Two limits of the gold column
The first limit is the period before 1971. The dataset's own footnote says that before 1971 gold prices were fixed under the gold standard and were mostly stable. That matters for the longer set. Its gold numbers blend a stretch when the price was held in place with a stretch when it moved freely, so the longer set describes a mixture of two different arrangements. It is a reason the site's default starts in 1971, a choice we made and state openly.
The second limit is the price series itself. The gold sheet lists one price per ounce per year. For years before 1970 it uses an average price for the year, and from 1970 onward it uses LBMA prices. A yearly price smooths away swings inside the year. Swings that happen inside a year cannot appear in a yearly series, and we cannot measure that gap from this file. Read the gold spread above as what a yearly series can show, not as the full range of what gold prices did.
Both limits point the same way: the gold row of the model is built from less information than the stock and bond rows, and it should be read with that in mind. The page on how this simulation works lists these and the other known limits in one place.
How the gold slider and the with-gold card work
In the retirement Monte Carlo calculator, the mix has three linked shares, stocks, bonds and gold, and they always add up to a whole. The gold share starts at zero. Nothing about gold enters your result until you move it. When you do, the tool reruns the same random draws with your chosen share, and the with-gold card prints both success rates and the signed difference between them.
The card states the difference whichever way it falls. If gold lowers the score under the assumptions you picked, it says so with a minus sign. If it raises the score, it says so with a plus sign. If nothing changes, it says no change. It does not color one outcome as good and the other as bad, and it adds the same sentence either way: whether gold helps depends on the assumptions above it.
Under the default plan and the 1971 to 2025 assumptions, with a ten percent gold share earning the dataset's gold return, the difference was +6.0 points. With the same share earning nothing after inflation, the difference was -2.5 points. Under the longer 1928 to 2025 assumptions the plan without gold produced 83%, and the ten percent share gave +1.8 points with the dataset's gold return and -4.3 points with a flat one.
What this number leaves out. It comes from a model with the assumptions shown, in annual steps and after-inflation returns. It does not model taxes, health or long-term care costs, inflation that differs from the model, or how you would act in a bad year. See what a Monte Carlo simulation cannot see.
A sweep from no gold to a large share
One reading of the slider is not enough, so we ran the default plan at several gold shares. The table below shows every result with the dataset's gold return, and again with gold earning nothing after inflation. The plan, the draws and the stock and bond assumptions are the same in every row, and the gold share is taken out of stocks and bonds in proportion. The inputs are the default example inputs (age 45, retiring at 65, planning to 95; $300,000 saved, $10,000 saved a year, $65,000 a year of spending in today's dollars, $24,000 a year of Social Security from 67, 60% stocks and 40% bonds), under the default 1971 to 2025 real-return assumptions unless the row says otherwise.
| Gold share | 1971–2025, dataset gold return | 1971–2025, gold earns nothing after inflation | 1928–2025, dataset gold return | 1928–2025, gold earns nothing after inflation |
|---|---|---|---|---|
| 0% gold | 87% | 87% | 83% | 83% |
| 5% gold | 91% | 87% | 84% | 81% |
| 10% gold | 93% | 85% | 85% | 78% |
| 20% gold | 96% | 77% | 85% | 71% |
| 30% gold | 96% | 62% | 83% | 58% |
Read the 1971 to 2025 column first. With no gold, the plan produced 87%. At five percent gold, the dataset-return case gave 91%, at ten percent 93%, and at twenty and thirty percent 96% and 96%. Now the matching flat-gold case, where gold earns nothing after inflation: at ten percent it gave 85%, at twenty percent 77%, and at thirty percent 62%. The first set of numbers climbs and the second set falls. The same plan, the same draws and the same gold share produced both, and the only thing that changed was what gold was assumed to earn.
The longer 1928 to 2025 set tells the same two-sided story with smaller moves at first. With no gold it gave 83%. With the dataset's return the sweep read 84% at five percent gold, 85% at ten, 85% at twenty and 83% at thirty. With flat gold it read 78% at ten percent, 71% at twenty and 58% at thirty. Here the dataset-return case barely moves and then slips back by the largest share, while the flat case drops steadily.
What this number leaves out. It comes from a model with the assumptions shown, in annual steps and after-inflation returns. It does not model taxes, health or long-term care costs, inflation that differs from the model, or how you would act in a bad year. See what a Monte Carlo simulation cannot see.
Why the answer depends on the assumption
The pattern follows from the inputs, and you can see it in the numbers above without any hidden mechanism. In the 1971 to 2025 set, gold's average real return of 7.4% sits far above bonds' 2.5% and a little below stocks' 8.4%. Its yearly moves are weakly opposite to stocks' moves. A gold slice carved out of stocks and bonds in proportion therefore replaces some of each, and in this model the mix ends with a similar or higher average and less swing from year to year. Under those inputs the score rises.
Take the average away and the picture reverses. If gold earns nothing after inflation, a gold slice replaces money that earned something with money that earned zero, and every extra point of gold lowers the average of the whole mix. The score falls, and it falls faster as the slice grows. In the model, then, the effect of gold is mostly a statement about gold's assumed return, with the weak link to stocks adding something on top.
So the honest summary is not "gold helps" and not "gold hurts". It is that a gold slice carries the whole weight of one assumption, and the record offers evidence for both. In the past, gold's real value rose sharply in a decade when stocks fell. It also stayed below its 1980 level for 31 years. The stated averages describe the whole sample, and they cannot tell you how long a flat stretch like that could last.
The timing of those stretches is the point that connects gold to the rest of this site. A saver drawing money out while a holding sits below its earlier value is selling into weakness, which is the mechanism described on the page about sequence of returns risk. A rise in a good decade and a long flat patch in a bad one have different effects on a plan depending on when they land. The page on how long your money will last in retirement shows how much the spending horizon alone changes a score, which is a useful comparison for any single holding.
Gold, buckets and other ways to split a portfolio
Gold is one of several ways to divide a portfolio. The bucket strategy divides money by when it will be spent, using cash, bonds and stocks, and its page tests that split in the same model. Gold, by contrast, divides money by what it is, which is a different question. Neither split removes the need for the model to be told what each part is assumed to earn.
If you decide to look into holding physical gold inside an IRA, some practical questions come up that a simulation does not touch: which coins and bars qualify, where they are kept, and how a company handles buying them back. The page on the Augusta Precious Metals gold IRA records what one company says about those questions on its own pages, with the date we read them, and it states plainly who that company's stated minimum fits.
What this can't tell you
This page can't tell you what gold will do. It cannot tell you whether the next bad decade for stocks will look like 1999 to 2008 or whether a long flat stretch like the one after 1980 will return. The dataset has 55 yearly observations in the default set and 98 in the longer one, which is a small sample for a question about decades. Overlapping decades inside it are not independent evidence.
It also cannot see your situation: what your other assets are, how much of your spending is covered by pensions or Social Security, what taxes apply to the account you use, or whether you would stay with a holding through a flat stretch. The model treats a year as a step and does not capture dealer spreads, storage costs or the price at which you would sell, all of which sit outside the annual return numbers. For the wider list of what a simulation leaves out, read what a Monte Carlo simulation cannot see.
Finally, a result on this page is the output of a model from stated assumptions. A higher number in a table is a statement about those assumptions and not a sign that a holding is wise, and a lower number does not make it unwise.
Questions about gold in the model
Does adding gold always raise the success rate?
No. In the default setup a ten percent gold share raised the rate when gold earned what the dataset says it did, and lowered it when gold was set to earn nothing after inflation. The result depends on the return you assume for gold, so the calculator shows the difference with its sign either way.
Why does the default gold share start at zero?
Because nothing about gold should shape your result unless you choose it. With the slider at zero, the score reflects only the stock and bond assumptions. When you move it, the with-gold card appears and states the difference from the no-gold case without calling it good or bad.
Is the gold column built from daily prices?
No. The file lists one gold price per ounce for each year, an average for the year before 1970 and LBMA prices from 1970 onward. That smooths swings within a year, so the model may understate how bumpy gold can be.
Does a high score with gold mean gold works?
No. A higher score means that, under those assumptions, more of the simulated paths finished with money left. It rests on the return assumed for gold, and the same page shows a flat-return case where the score fell. Neither number is a forecast.