Gold in a retirement portfolio

No asset splits a room like gold. To one camp it is a barbarous relic, with no cash flow and nothing productive behind it. To the other it is the only honest money and the ultimate insurance. Each camp is right about its own half of the picture, and that is exactly what makes gold so hard to size.

For someone living off a portfolio, the question is not the one an accumulator asks. A withdrawal plan has enemies of its own: persistent inflation, crises of confidence, regimes where stocks and bonds fall together (Market regimes (growth × inflation, sticky bears) and why they matter). Gold, empirically, is one of the very few assets that works in exactly those holes in the standard defense (Defensive assets: the map and the roles). This article looks at it without romance and without contempt. What the long data really say about its return and its behavior, sample traps of the post-1971 record included. What it does to a withdrawal plan with numbers on it, through ERN's work on gold as a sequence hedge. The three distinct jobs people give it, and the right size for each. How to hold it in practice: funds, coins, and the tax bill. And the classic mistakes, the worst of them being over-sizing after the fact, at the end of a golden decade.

What the long data really say§

A secular return near zero in real terms, and that is normal. Over the longest measurable stretches (Dimson-Marsh-Staunton, Roy Jastram's series in The Golden Constant), gold keeps its purchasing power to within about 0 to 1% real a year. An ounce buys roughly the same basket of real goods it bought a century ago. No more, because nothing is produced. No less, because nothing dilutes it fast: the stock grows only about 1.5% a year through mining. That is exactly the profile of a store of value, and precisely not that of a portfolio engine (Forward-looking expected returns (Morningstar, Vanguard, the investment banks)). In the language of finance theory, gold has no risk premium, and that is not a defect (Where returns come from: risk premia).

The behavior: dormant, then explosive, regime by regime. The path is not a slow 0.5% a year. It alternates very long dead stretches with violent episodes. The key ones. 1971-1980: up 1,200% in nominal terms, as the administered price of $35 was cut loose into a stagflation. That is the founding event of gold's reputation, and part of it cannot happen twice. 1980-2000: −70% in real terms, twenty years of purgatory through the great disinflation. 2000-2011: five times over, on post-bubble mistrust and then a financial crisis. 2011-2015: −40%. 2022-2025: a draw in dollars in the year bonds had their worst vintage on record, then new nominal highs. In that regime of inflation and rising rates, gold cost nothing at the exact moment the standard defender was failing, and for anyone whose spending currency was not the dollar, the dollar's rise added to it. Its volatility runs 15 to 20%, an equity index's, for an expected return of zero. Judge it on its own and you will always reject it. It can only be judged inside a portfolio.

Gold in dollars, deflated by US CPI, from December 1969 to December 2025. Each bar is a ten-year holding window, one every five years; the last one covers only six.

The correlations, where its real case is made. Over the long run its correlation to stocks is about 0, and to bonds about 0 as well. More to the point, it does not climb toward 1 in a crisis, unlike almost everything else (Fat tails, crises, and the Student-t). Through the worst months for global equities, gold is on average flat or up: +5% in 2008, the year everything else lost 40%, and +25% in 2020. That is requirement number 2 on the defensive spec (Defensive assets: the map and the roles). A decorrelation that survives stress is a rare property, and worth its weight in carry.

🔬 Gold and the withdrawal plan: what the simulations measure

ERN gave Part 34 of its series to our exact question, "Using Gold as a Hedge against Sequence Risk". On the data since 1971, adding 10 to 20% gold to a stock and bond portfolio raises the safe withdrawal rate of the worst vintages. That holds above all for the 1960s and 1970s start years, the stagflation nightmare (Bengen, the Trinity study, and the birth of the safe withdrawal rate). Gold there replaces part of the bond sleeve to good effect, since inflation simply destroyed bonds for those cohorts.

On the disinflationary vintages (1982 and after), it costs a little. Netted out, it is an option on rare regimes, bought at a fair price.

Portfolio Charts finds the same thing in withdrawal-rate terms: mixes holding 15 to 25% gold sit on the SWR and drawdown frontier of almost every country tested. A decumulation simulator reproduces that result, as long as you look in the right place. The A/B of gold against no gold is nearly invisible in the central model. It shows up in the models that contain the long inflations, the replay of the stagflation vintages first among them (Historical windows, bootstrap, parametric: the three families of models).

Then comes the sample warning, always the same one. Half the measured benefit comes from a single decade, the 1970s, and part of that decade cannot repeat. Cut the enthusiasm in half on common sense alone, and enough of it is left.

The three roles, and the size of each§

Confusing the roles is what produces the bad sizes. Keep them apart.

Role 1: the anti-regime block of the portfolio (5 to 10%). This is the central role in this book. The job is to cover the "persistent inflation" and "crisis of confidence" rows of the table of defenses (Defensive assets: the map and the roles), the two rows a typical retail portfolio leaves wide open. The useful size the simulations measure is 5 to 10%. Below 5% the effect is cosmetic. Above 10% you are no longer hedging, you are betting. The Golden Butterfly and Browne go to 20 to 25%, the historical upper limit, with the epistemic dependence already criticized (All-weather portfolios: Browne, All Weather, Golden Butterfly, Dragon). The modern way to split the inflation row is half gold, half linkers, the inflation-linked bonds (Inflation-linked bonds: the only contract written in real terms). Two different mechanisms, confidence and contract, against the same enemy.

Role 2: the political tail reserve (2 to 5%, outside the portfolio). This is physical gold you hold yourself, in coins, as insurance against the scenarios that live outside the model: a major monetary crisis, a break in the financial system (Hyperinflation and extreme scenarios, Simulator traps: independence, US bias, survivorship, "what is not in the model does not exist"). It is a matter of temperament more than of optimization. It cannot be simulated, it gets capped at a few percent, and it carries costs of its own: storage, insurance, the premium on coins.

Role 3: the tactical bet (0%). Buying gold "because it is going up" has no place in a withdrawal plan, like any other bet on timing (Valuations, the CAPE, and what they say about your withdrawal rate). The gold in the plan gets sized once, rebalanced at the bands, and never listened to.

Owning gold in the US: funds, coins, and the tax bill§

Physically backed gold ETFs are the standard route, GLD and IAU being the best known. A share is a claim on allocated bullion sitting in a vault, the funds trade all day, and the spread is small. Compare expense ratios before picking one: the range across these funds is wide enough to matter over thirty years (Building it with US-listed ETFs: the retiree's shopping list). The currency question that occupies European investors does not arise here, since gold quotes in dollars and you spend dollars.

The tax treatment is where the surprise lives. The IRS treats gold as a collectible, physical metal and physically backed fund alike. These funds are grantor trusts, so you are treated as owning your slice of the bullion directly, and a gain on your shares is a gain on collectibles. Long-term collectible gains carry a maximum federal rate of 28%, against 20% at the top for ordinary long-term gains (IRS Topic 409). It is a cap and not a flat rate: if your ordinary long-term rate is lower, you keep it. Hold for a year or less and the gain is short-term, taxed as ordinary income like anything else. The 3.8% net investment income tax rides on top above the usual income thresholds. Location matters more here than for almost any other holding, so decide where the gold sits before you buy it (US taxes in the withdrawal phase).

One piece of plumbing is easy to miss. The trust sells a little gold every month to pay its own expenses, and your share of those sales is a small taxable event that quietly lowers your cost basis. The sponsor publishes a yearly schedule for it. The amounts are tiny; the bookkeeping is not optional.

Inside a tax-advantaged account, the collectibles rate never appears. Nothing is taxed year by year, what comes out of a traditional IRA is ordinary income whatever earned it, and a qualified Roth withdrawal is not taxed at all. A gold ETF buys like any other security in an ordinary brokerage IRA. Holding the metal itself is a different animal: the code carves out an exception for certain coins and for bullion of a defined fineness, but only where a qualified trustee has physical possession (26 U.S.C. section 408(m)(3)). In practice that means a self-directed IRA, a custodian and a depository, all of which charge. Metal in a safe at home is not that exception, and an IRA that buys it is treated as having made a distribution.

Physical gold in your own hands, coins (American Gold Eagle, Krugerrand, Maple Leaf) or small bars bought from a dealer, is the vehicle for role 2. Two prices matter and neither is spot. You pay a premium over spot going in, and you sell into the dealer's bid coming out, typically a few percent each way and wider on small coins than on larger bars. A round trip is expensive, which is another reason this sleeve is bought once and left alone. Then come storage, a home safe or a bank box, and insurance, since a standard homeowner's policy usually caps coverage on bullion at a token amount. Keep every purchase invoice: your basis is what you can document.

The paperwork, briefly. A broker reports your fund sale on Form 1099-B, basis included. A coin dealer may or may not send a 1099-B, since only certain metals and quantities trigger one, and either way the gain is yours to report. Any dealer taking more than $10,000 in cash, in one transaction or in related ones, files Form 8300. None of that is a trap for an honest holder. It is only a reason to keep records. Rates and thresholds move: the ones above were checked against irs.gov and the tax code at the time of writing, so verify the current figures before acting on them.

What to leave alone. Gold mining stocks first: they are a volatile lever on gold, with company risk on top and an equity beta in a crash, so they miss the defensive spec. Unbacked certificates and notes next: they carry issuer risk, and issuer risk is precisely what the gold was bought against. And after those, the whole family of exotic paper gold.

The classic mistakes§

Sizing by the rear-view mirror. Mistake number one, and it runs both ways. Holding 0% after twenty years of purgatory, like the 1999 investor who "knows" gold is dead, right before it goes up five times. Or holding 25% after a golden decade, like the 2011 investor extrapolating right before the 40% fall, or the 2025 investor doing the same thing again. The size comes from the job, which is covering regimes, and never from the recent shape of the chart: 5 to 10%, written into the plan, and that is the end of it.

Judging gold on its own over three years. "My gold fund is down 12% while stocks were up 30%": yes. That is the carry on the insurance, budgeted in advance. Gold is judged at the portfolio level, over the crossing of hostile regimes, meaning the years spent inside one and not a single crash. Its natural judges are the stagflation vintages, 1966 and 1973, and the long-inflation models. Not the annual statement.

Gold as an answer to anxiety rather than to a regime. Raising the dose at every spike of fear, geopolitical or banking, is role 3 dressed up as role 2, and it always ends in paying full price for the fear. A written plan protects you from yourself, here more than anywhere.

Forgetting that gold does not pay withdrawals. In the mechanics of decumulation, gold is a reserve and a rebalancing asset. You do not live on gold, you live on flows and on liquid sleeves (The cash buffer: size, cost, and what it is really for). Gold gets sold at the bands, or in the regimes where it happens to be the one holding sitting on a gain (Drawing on a buffer and refilling it: the rules that work).

🧮 Ten percent gold, tested against the models

The plan: $1.5M, $51,000 a year, Vanguard corridor, 45 years. Variant A: 70% stocks, 30% intermediate bonds. Variant B: 70/20/10 with gold (a physically backed fund, taxable account). Now the verdicts. Central case, 4.1% failure for A against 3.9% for B, next to nothing, as expected. Sequence stress, from 7.2 down to 6.1%. Long-inflation models, from 10.8 down to 8.9%. That is where gold earns its place: in the futures that contain a stagflation. In the vintage replay, 1966 goes from "exhausted in year 27" to "ridden out, battered", and 2000 barely moves. Median wealth at 45 years: down 5%. The typical decision: B, with half the sleeve converted into linkers if the taxable account already carries a heavy tax bill, same row of the defense through a different mechanism. And one line in the plan: "gold 10%, bands plus or minus 5 points, not up for discussion in a drawdown".

The same A/B, model by model. Each bar joins the two variants of one plan on the ruin axis, and its length is the whole subject. In the central case there is next to nothing to see, which is the honest answer both to "gold is useless" and to "gold saves everything". The bar grows as the model turns hostile, out to the two points won on the long-inflation futures. The 1966 vintage is read apart, at the bottom, because a dated replay gives a verdict and not a probability.

The essentials§

Gold is a store of value (about 0 to 1% real over the centuries, equity-like volatility, long dead stretches and violent episodes), not an investment. It gives up expected return to cover the inflation and confidence regimes: the holes in the standard defense.
Its rare property is a decorrelation that survives a crisis (2008, +5%; 2022, a draw in dollars while bonds came apart): the test that disqualifies most "defensives".
The simulations (ERN Part 34, Portfolio Charts, any simulator that replays the stagflation vintages) credit 10 to 20% gold with a real improvement in the worst vintages, stagflation above all, at a modest central cost. Mark that down for how much it leans on the 1970s.
Three roles, three sizes: the anti-regime block at 5 to 10% (half gold, half linkers), the political tail reserve at 2 to 5% in metal (keep the invoices), the tactical bet at 0%.
In practice: a physically backed fund, taxed as a collectible with a long-term rate capped above the stock rate in a taxable account and that cap gone inside a tax-advantaged one; coins for role 2, with their premiums and their storage. The portfolio return on gold comes from mechanical rebalancing. It is written into the plan, and it does not get a vote.

Going further§

Roy Jastram, The Golden Constant (1977, updated): the purchasing power of gold across four centuries.
Early Retirement Now, Part 34 ("Using Gold as a Hedge against Sequence Risk"): the retiree's exact question (ERN's Safe Withdrawal Rate series: a reader's guide).
Erb & Harvey, "The Golden Dilemma" (2013): the skeptical academic reading, an indispensable counterweight.
portfoliocharts.com: gold in every mix, country by country.
In this book: Defensive assets: the map and the roles (the table of roles), Inflation-linked bonds: the only contract written in real terms (the other half of the inflation row), All-weather portfolios: Browne, All Weather, Golden Butterfly, Dragon (the historical doses), Hyperinflation and extreme scenarios (role 2 in full).