Investing in Gold

On January 30, 2026, two days after the gold price printed its all-time high, the front-month futures contract on our tape fell 11.4% in a single session, from $5,318.40 to $4,713.90. It was the worst day in the 500 sessions we hold and a larger one-day fall than the 9% of April 15, 2013, which had itself been the largest since 1983, and it came with no bank failure, no new war and no change in interest rates. The trade had become crowded, and the crowd left through one door.

Over the next five and a half months the price fell to $3,985.60. In that window the world’s central banks bought 289 tonnes, the most they have ever bought in a second quarter, and Poland alone bought 51 tonnes into the decline. Then, in August, the Western funds that had sold March’s bottom returned with $18 billion in a month, the second-largest inflow the World Gold Council has recorded, and ETF holdings made a new record of 4,189 tonnes.

That sequence is the gold market in 2026: two buyers on two clocks, a price that swings between them, and a retail investor usually on the wrong side of the second clock. The hub’s flagship, Investing in Precious Metals, surveys silver and the platinum-group metals alongside gold; this guide stays on gold, from who bought the run to what each way of owning it costs and how each is taxed, with twenty ounces put through three wrappers for a decade.

Our tape holds daily closes of the front-month gold futures contract, the nearest-dated one, from Yahoo Finance (GC=F); the window we chart is the last 500 sessions, September 12, 2024 to September 8, 2026, and it is the fixed reference for the rest of the guide. Futures closes differ from the London spot benchmark by tens of dollars and from intraday records by more, so every figure from our tape is labelled as ours.

The milestones, as first daily closes: above $3,000 on March 17, 2025; $3,500 on September 2; $4,000 on October 8; $4,500 on December 26; $5,000 on January 26, 2026. The highest close was $5,318.40 on January 29. Press reports put the spot record at $5,589.38 on January 28 (CBS News; other feeds show up to $5,608 and some date the intraday peak January 29), and the once-daily LBMA afternoon benchmark peaked near $5,405. Calendar 2025 on our tape was a gain of 64.5%, from $2,629.20 to $4,325.60, in a year the World Gold Council counted 53 all-time highs and an average price of $3,431, up 44% on 2024.

Then the top. From the January 29 close the contract fell 25.1% to $3,985.60 on July 16; on a spot basis, press reports put the fall near 30%. It was a staircase: the 11.4% day on January 30, a 5.9% day on March 19 as record ETF outflows landed, a June slide as the Federal Reserve turned hawkish that produced the first close below $4,000 on our tape on June 24, and a spot-market break below $4,000 on July 13, the day Governor Christopher Waller said rates might need to rise. Since then: a post-low high of $4,640.80 on August 24, and $4,443.90 on September 8, 16.4% below the January close and 11.5% above the July low.

Two numbers describe the character of the move better than the price. The annualised standard deviation of daily returns on our tape is 24.3% across the window and 31.3% in 2026 to date, the volatility of an equity index in a bad year; gold’s reputation as the calm asset does not survive its own tape. And the one-year change is +22.1% from $3,638.10 on September 8, 2025, so a reader who bought a year ago is comfortably ahead and one who bought in the last week of January is down a sixth, both reading the same commentary.

Our Precious Metals sub-index, an equal composite of the four metals rebased to 100 on September 2, 2025, stood at 139.05 on September 8, up 39.8% on its published one-year change. Gold fell least of the four from the January highs; silver, down 42% from its own high on our tape, has its own guide, Investing in Silver.

Where the metal came from and where it went explains both the run and the correction. The World Gold Council’s Gold Demand Trends is the market’s balance sheet; the Council is an industry body funded by miners, so read its commentary accordingly, but its tonnage series is the one everyone, including the central banks, uses.

In 2025 total demand including over-the-counter buying was 5,002 tonnes, the first year above 5,000, worth a record $555 billion, up 45%. Investment demand of 2,175 tonnes was the driver: ETFs added 801 tonnes, the second-strongest year in tonnes and a record $89 billion in dollars, and the fourth quarter alone saw 420 tonnes of bar and coin buying, a twelve-year high. Central banks took 863 tonnes, below the 1,000-tonne years of 2022 to 2024 but nearly double their 2010–2021 average of 473. Jewellery was weak in tonnes and a record $172 billion in value: households in India and China bought lighter pieces at higher prices, which is what a price-elastic buyer does.

Supply is what makes gold behave like money rather than copper. Mine production set a record of 3,672 tonnes and grew about 1% against a dollar price up about two-thirds over the year, because a mine takes a decade to build. Recycling rose only 3% to 1,404 tonnes: households that melted jewellery at about $1,570 in 2011 held at $3,400 in 2025. That reluctance to sell is one of the most bullish facts in the data and the easiest to reverse, because the above-ground stock is roughly 216,000 tonnes, about 45% of it jewellery, and mine output adds only 1.7% a year to it. The price is set by who wants to hold the inventory, not by who is digging.

The first half of 2026 shows the same machine in a softer gear. Second-quarter demand was flat on the year at 1,269 tonnes; the half came to 2,522 tonnes worth $380 billion. Investment fell to 262 tonnes because ETFs shed 45 tonnes, while bar and coin buying held steady at 307 tonnes; the official sector bought its record 289 tonnes. The average London afternoon price for the quarter was $4,506.29, 8% below the first quarter’s record, and the 16% fall was the worst quarter since 2013. Retail and fund demand rises and falls with the price, official demand rises when the price falls, and the second is now large enough to catch the first.

The marginal buyer of the decade is the official sector, and that is the single most important input to any view on the price. Central banks were net sellers every year from 1989 to 2009. They turned buyer in 2010, and from 2022 they bought at a pace with no precedent in the floating-rate era: roughly 1,082 tonnes in 2022 as first reported (since revised as high as 1,136), 1,037 in 2023, about 1,045 in 2024, and 863 in 2025. The Council revises these as late reporting arrives and the revisions can be large: the first quarter of 2026 was first reported at 244 tonnes and cut to 57 in July. Treat every quarterly official-sector figure as provisional for a year.

The National Bank of Poland is the buyer to study, because it publishes its reasoning. It was the largest official buyer in 2024 and again in 2025, when it added 102 tonnes; it added 51 tonnes in the second quarter of 2026, 82 tonnes in the first half, taking reserves to 632.4 tonnes at the end of June, and another 8 tonnes in July, 90 for the year, against a stated target of 700 and a board-approved plan for 150 more.

The bank’s stated reason is that gold is the one reserve asset that cannot be frozen or defaulted on, a direct reference to February 2022, when roughly half of Russia’s foreign reserves were immobilised within days of the invasion of Ukraine. Every central bank outside the Western alliance drew the same conclusion: a Treasury is a claim on a counterparty, and a bar in your own vault is not.

The People’s Bank of China has reported an increase in every month since late 2024, twenty-two in a row to August 2026: 33 tonnes in the second quarter on the Council’s count, its largest quarterly addition since the end of 2023, then 19.9 tonnes in July and 20.2 tonnes in August, taking declared holdings to 2,386.6 tonnes. China paused its reporting for six months of 2024 while, by most estimates, buying heavily, and the market treats its published number as a floor.

The holders are a different list: the United States about 8,133 tonnes, Germany about 3,350, Italy and France about 2,450 each, Russia about 2,330, China a reported 2,390, Switzerland about 1,040. The old holders are not buying and the new holders are; monetary gold is moving from the West to everyone else at market prices.

The European Central Bank’s June 2026 review of the euro’s international role put gold at 27% of global official reserves at the end of 2025, ahead of US Treasuries at 22% and the euro at 15%; gold passed the euro in 2024 and Treasuries in 2025. The Council’s 2026 survey of 76 reserve managers found a record 45% expecting their own institution to add gold within twelve months and 74% expecting the dollar’s share of reserves to fall over five years; a survey is not a purchase, but this one has been matched by purchases for four years running.

A thousand tonnes a year is about 27% of mine output taken by buyers who do not sell on a 16% dip, which shrinks the tradable float for everyone else and puts a bid under the market when the fast money leaves. The risk is equally simple: central banks have changed their minds before, and the Bank of England’s sale announcement of May 7, 1999 came within three months of the bottom of a twenty-year bear market. Section 8 has that history.

The fast variable is Western fund money, and the shape of the spike and the correction belongs to it. The Council publishes gold ETF flows monthly, and 2026 is the cleanest example in the series of money arriving late and leaving early.

January’s global inflow of $19 billion was the largest month in the data, $7 billion of it from North America and nearly $10 billion from Asia, a record for that region, with China the second-largest source of inflows in the world behind the United States. Holdings peaked at 4,176 tonnes on February 27.

March saw $12 billion leave, the largest monthly outflow on record. May and June were net outflows too, June’s $8.9 billion the heaviest since March; assets fell to $604 billion in May, North American funds finished the half with $7.7 billion of net outflows, their weakest first half since 2013, and the half-year net was just $8 billion. July turned modestly positive at $3 billion. Then August: $18 billion in, the second-largest month on record, led by European funds ($7.9 billion, with records in the UK and France) and a North American return after the spring’s outflows, taking assets to $615 billion and holdings to a new record of 4,189 tonnes.

Read that against the price. The record inflow arrived in the month of the record price; the record outflow in the month of the first big leg down; the second-largest inflow in August, after the price had already recovered 11% from the July low. The ETF investor, in aggregate and in public, bought the top, sold near the bottom and bought back on the way up; the central banks did the reverse in every quarter. Neither is stupid; a reserve manager fills a strategic allocation over a decade and a fund investor responds to a chart over a quarter. The flows explain the shape of the price; the official bid explains the level.

Most commentary still runs the model that worked from 2006 to 2021, so the five forces that move gold need weighing as they stand now. Gold pays nothing, so its price is a vote on what else is on offer. Real interest rates were the textbook driver: a higher yield on the 10-year TIPS (Treasury inflation-protected securities) raises the cost of holding a metal that yields zero.

That relationship held for fifteen years and broke in 2022 and 2023, when the 10-year TIPS yield went from below −1% to about 2.4%, its highest since 2008, and gold held near $1,800 and then made new highs, because the marginal buyer had changed; a reserve manager in Warsaw is not comparing gold to TIPS but to the political risk of holding Treasuries at all. Real yields still steer the Western flow, which is why a hawkish Fed produced the June and July 2026 declines, but they no longer set the floor.

The official sector, covered in Section 3, sets the floor now, and is why it rose from about $1,630 in the autumn of 2022 to above $3,900 in 2026 without a Western retail mania until the very end. The dollar and the reserve question work more slowly: gold is priced in dollars, and three-quarters of reserve managers expect to hold fewer of them in five years. Investment flows, covered in Section 4, explain the shape of the price rather than its level; record ETF inflows accompanied the tops of 2011, 2020 and 2026.

Jewellery and the East are the price-elastic residual: demand thins as the price rises, which is one reason tops form, and returns when it falls, which is one reason floors hold. The Shanghai Gold Exchange premium over London is the cleanest daily read on the largest physical market. A normal premium is $5–15 an ounce; it ran at $20–80 in the 2024 demand surge; in the week to July 24, 2026 it was $3–6, the sound of Chinese households on strike at $4,500, and ICBC and other large Chinese banks stopped offering leveraged precious-metals trading to retail clients after that week. By September 8, 2026 the premium was about $3, or 0.06%, on MetalMetric’s tracker, whose six-month range ran from a $40 discount to a $52 premium.

In January 2026 all five forces pointed the same way at once, which is what a blow-off top, a final vertical rise on record inflows, looks like. By July three had turned and the two that had not, the official bid and the reserve question, held $4,000. Expect corrections to be deep and floors to be high for as long as the official bid lasts.

The number the industry quotes for a long-term holder is true and misleading at once. Gold is up about 127-fold since the United States closed the gold window on August 15, 1971 at $35 an ounce: 9.2% a year, about 5.1% after inflation, flattered by a start price fixed by law for 37 years and an end three months after a record. Here is the same calculation from every start date that matters, to our tape’s $4,443.90 on September 8, 2026, deflated by the Bureau of Labor Statistics’ consumer price index (the July 2026 CPI-U print of 333.918 against 40.8 in August 1971 and 77.8 in January 1980).

Read the two ends. Someone who bought at the January 21, 1980 peak of $850 has, forty-six years later and after the largest bull market in a generation, compounded at 3.6% a year in dollars and 0.4% a year in real terms; that $850 is about $3,648 in today’s money, and gold only passed it, in real terms, in 2025. Someone who bought in August 1999 at $252, the summer the Bank of England began its auctions, has made 11.2% nominal and 8.4% real, better than the S&P 500’s real total return from the same unlucky start, which on our arithmetic is in the region of 5%. The 2011 buyer at $1,921 has made 3.1% real over fifteen years; the 2020 buyer at $2,067, 8.8% real in six.

Against equities over the long windows, gold loses by a wide margin. Mercer Advisors’ comparison has the S&P 500 returning 11.7% nominal and 8.2% real a year from 1980 through 2023 against gold’s 3.2% nominal, roughly zero after inflation.

Two things to hold onto. Gold’s long-run real return is positive but modest and almost entirely concentrated in three episodes, 1971 to 1980, 2001 to 2011 and 2019 to 2026; outside those windows it was flat to down for a combined thirty years. And none of these figures subtract the cost of owning gold: the premium going in, the spread coming out, storage or an ETF fee every year, and 28% of the gain at the end. Section 15 shows what that does to a decade in dollars.

The strongest academic argument against buying gold at today’s price is Claude Erb and Campbell Harvey’s The Golden Dilemma, published in the Financial Analysts Journal in 2013 and updated by the same authors in May 2024 as Is There Still a Golden Dilemma?; the argument that their framework has been overtaken comes from the authors themselves. Most popular reasons to own gold fail their tests: it is a poor inflation hedge over any horizon shorter than a century, having fallen in real terms through two decades of positive inflation after 1980.

What gold has is a real price, the nominal price divided by the consumer price index, that wanders far from its long-run average and eventually returns. When the real price has been high, subsequent ten-year real returns have been poor; when low, good. That is the signature of a mean-reverting series, and it is the most robust finding in the gold literature.

The real price is now above its 1980 record. Against the $3,648 that the January 1980 peak represents in July 2026 dollars, our tape’s $4,443.90 is 22% higher, and the 2024 update found that the ETF, retail and official-sector money of the last two decades had, on the authors’ measure, roughly doubled the real price relative to the era before that money existed. In their framework that is the single most bearish fact available: every previous visit to this neighbourhood, in 1980 and 2011, was followed by a decade in which gold lost to inflation.

The authors name the other side themselves, which is why the paper is called a dilemma: if emerging-market central banks raised their gold holdings toward the share the old Western holders keep, the demand would run to tens of thousands of tonnes and the real price could go far higher before it reverted. In 2013 that was a hypothetical; by 2026 it is the observed behaviour of the official sector at about a thousand tonnes a year. Mean reversion says the price is too high. The buyer under the market says the mean has moved. Both are true, on different horizons.

Gold’s annualised volatility since 1975 has run roughly 15–20%, which is equity-like, and its drawdowns are equity-sized and far longer; both its cycle and its crisis behaviour are sold with more confidence than the record supports. The five regimes of the floating-rate era, each of which turned on an official-sector decision, are the map.

1971–1980. Nixon closes the gold window on August 15, 1971 at $35; Americans may own gold again from December 31, 1974, for the first time since Executive Order 6102 of 1933; two oil shocks and double-digit inflation take the price to $850 on January 21, 1980.

1980–1999. Paul Volcker’s 20% federal funds rate ends it. Gold loses 70% nominal and about 86% real over twenty years, central banks sell throughout, and the Bank of England auctions 395 tonnes between 1999 and 2002 at an average near $275; the September 26, 1999 Washington Agreement capping official sales produces a 25% rally in a fortnight. The 1980 nominal high is not regained until January 2008 and the real high not until 2025.

2001–2011. From $256 in April 2001 to $1,921 on September 6, 2011, on the dot-com bust, the launch of GLD in 2004, the 2008 crisis, quantitative easing and the first wave of emerging-market official buying, which turned the official sector from seller to buyer in 2010.

2011–2015. Down 45% to about $1,050 by December 2015 as Western ETFs shed well over a third of their holdings from the December 2012 peak of about 2,630 tonnes.

2019–2026. The Fed pivot, the pandemic and $2,067 in August 2020, then the Russian reserve freeze of February 2022, the official-sector surge, $5,000 in January 2026 and the 25% correction. The two previous bull markets ended when the Fed turned decisively hawkish and when the marginal official buyer went quiet while ETF money left. Both signals are worth watching now, and only the Fed has flickered.

The crisis record

2008. Gold peaked at $1,011 in March, fell about 30% to around $712 by late October as the liquidity squeeze forced funds to sell whatever had a bid, then finished the year up about 5% while the S&P 500 lost 37%; March 2020 repeated the pattern in a week. Gold hedged the crisis, not the panic: in the weeks when correlations go to one, gold goes with them.

2022. The direct test of the inflation-hedge claim: US inflation above 9%, the Fed up 425 basis points, and gold roughly flat while the S&P fell 18% and the US bond aggregate 13%. Gold did not rally into inflation; it preserved capital in the one year a 60/40 portfolio had nowhere to hide, which is the only claim the record supports.

2026. A 25% fall with no equity crash to blame: a crisis hedge that loses a quarter in a non-crisis is also, at times, a crowded trade. Over long windows gold’s correlation with global equities is low and slightly positive on the Council’s calculations, negative in sell-offs and one in liquidations. Gold diversifies. It does not insure.

How you own gold matters more than whether you own it, which is the claim in the deck. The choice of metal moves your return by a few points a year. The choice between a coin in a drawer, an ETF share, a futures contract, a vault account and a mining stock moves your annual cost from 0.1% to 4%, your tax rate from 20% to 28% to 37%, and your counterparty from nobody to a London bullion bank to a Nevada mine manager. Same underlying, eight different assets.

Sovereign coins and small bars. Full title, no counterparty, no fee unless you vault them; a premium of about 1.5% to 4% on a one-ounce coin, a spread on sale, and the 28% collectibles rate. The only wrapper with no counterparty, and the only one you can lose in a fire.

Large bars in an allocated vault, kilo bars at about 1.5% over spot in a Brink’s or Loomis vault at 0.12–0.20% a year with a bar list in your name, or an allocated vault account such as BullionVault, which sells fractions of Good Delivery bars, the 400-ounce bars the London market trades, at the professional spot price with 0.5% commission each way and 0.12% storage. Unallocated and pool accounts: a claim on a pool of metal held by a bank or dealer, often free because you are an unsecured creditor and the holder can use your metal. Section 13 has the failures.

Physical ETFs: GLD, IAU, GLDM, SGOL and their peers, grantor trusts (a trust holding one asset that passes its tax character straight through to you) that hold allocated bars in London and sell a little gold each year to pay a fee of 0.09% to 0.40%. A one-cent spread, instant liquidity, no delivery, and a 28% tax rate, because the IRS looks through the share to the metal. Futures: leverage, the deepest liquidity in the market, and the 60/40 treatment of Section 1256 contracts, at the cost of a roll every two months (selling the expiring contract and buying the next) that charges the short-term interest rate as contango, the premium of a later-dated contract over spot.

Miners: operating leverage on the gold price, a dividend and equity tax at 20%, with everything that can go wrong at a mine. Royalty and streaming companies: gold-price upside with no operating costs and equity tax, at a valuation that already prices most of that in. Tokenised gold (Paxos’s PAXG, Tether’s XAUT): a blockchain claim on a vaulted ounce, about 37 tonnes across the two as of May 2026 on Coin Metrics data, with the issuer as counterparty and unsettled tax treatment.

The chart is the cost dimension; Section 14 has the tax dimension, which splits the same list into a collectibles group at 28%, a Section 1256 group at a blended 26.8% charged every year, and an equity group at 20%. The counterparty dimension runs from nobody, through a vault operator, a bullion bank, an ETF custodian and a clearing house, to a mining company’s balance sheet. Every step toward convenience is a step toward a counterparty; every step toward the metal is a step toward a premium and a storage bill. The design problem is picking the point on that line that matches what the gold is for.

Here is what a physical buyer actually pays and gets back, with quotes dated to this week. The physical trade runs on the premium: the amount over spot a dealer charges for a specific product, covering the mint’s fabrication charge, the dealer’s margin and the scarcity of that product on that day.

The one-ounce coins

The American Gold Eagle (US Mint, 22-karat, a full ounce of gold in a heavier 1.09-ounce coin) is the most liquid coin in the United States. On FindBullionPrices’ dealer comparison retrieved on September 9, 2026, at a melt value of $4,402.12, the cheapest random-year Eagle was $4,462.32, a premium of $60.20 or 1.4% over melt, while the average of sixteen dealer offers was $4,581.02, about 4.1% over; the lowest available premium had ranged from −0.25% to 1.45% over the previous thirty days and averaged 0.86%.

A current-year 2026 Eagle was dearer: the cheapest offer on the same site was $166.41 over melt, 3.8%, inside the 3–6% band that site calls normal for a new-date coin. (The pages reprice hourly with spot, so treat the dollar figures as a snapshot and the percentages as the lesson.)

Two lessons. The date on the coin is worth nothing at resale and costs 2–3% at purchase, so buy random-year. And the gap between the cheapest dealer and the average dealer is larger than the premium itself, so ten minutes comparing three quotes is the best-paid work in physical gold.

The American Buffalo (24-karat, .9999 fine) trades at a similar or slightly higher premium and, like the Eagle, is not on the dealer-reporting list. The Canadian Maple Leaf (.9999, with a micro-engraved security mark) usually runs a little under the Eagle, the Krugerrand (22-karat, minted since 1967, the most widely held bullion coin in the world) a little under that, and the Austrian Philharmonic, Britannia and Kangaroo sit in the same band, with one to two percent between top and bottom; that ordering is from 2026 dealer guides (USAGOLD, SD Bullion, MintBuilder) rather than same-day quotes, which we did not retrieve for these coins. Fractional coins carry premiums of 5–15% and are a poor investment and a good gift.

Bars

Premiums fall with size. A one-ounce minted bar from PAMP, Valcambi or Argor-Heraeus in a sealed assay card runs roughly 2–4% over spot; a 100-gram bar 1–2%; a kilo bar (32.15 ounces, about $143,000 at our tape’s price) had a melt value of $141,388 on FindBullionPrices on September 9, 2026, with the cheapest tracked offer about 1.5% over melt and the average offer about 3% over, and buyback bids on refiner kilo bars near 98.5% of spot. The trade-off is divisibility and verification: a kilo bar sells whole, to a buyer who can test it, and the refiner’s brand plus intact assay packaging is what makes the test cheap.

The spread, and the dealers

Premiums are not symmetric. In the panic-buying weeks of March and April 2020 Eagle premiums went above $100, 6–7% at the time, and fractional coins 15–25%, on FindBullionPrices’ record; in the 2026 correction the lowest random-year premium touched zero and briefly went below it (a thirty-day low of −0.25%), with dealers offering at $60–100 over and bidding near spot, the spread widening as inventory backed up. The round trip on a one-ounce coin through a national dealer is about 2–4% in a normal market and 6–8% in a stressed one.

JM Bullion, APMEX and SD Bullion are the large online houses, APMEX typically 0.5–1.5% dearer on the same product, each publishing a buyback price; a local coin shop often bids closer to spot on Eagles and pays cash, which raises the reporting question in Section 14. Whatever the source, the invoice with your name on it is your cost basis at sale and your provenance at resale.

SPDR Gold Shares (GLD) launched on November 18, 2004 and made gold a mainstream portfolio asset; it held about $175 billion at the January 2026 peak on a Motley Fool comparison of January 24 and roughly $140 billion by June, still more gold than most central banks hold. iShares Gold Trust (IAU) reported net assets of about $62 billion on June 11, 2026. What follows is how the trust works, which one to own, and the two things most holders find out too late.

The mechanism

Each is a grantor trust. The sponsor (State Street for GLD and GLDM, BlackRock for IAU and IAUM, abrdn for SGOL) appoints a bullion-bank custodian in London that holds allocated Good Delivery bars on a published bar list, and authorised broker-dealers create and redeem shares against metal, which keeps the share price tied to gold. The trust earns nothing, so it pays its fee by selling gold: a GLD share represented a tenth of an ounce in 2004 and about 0.092 of an ounce now, the decline compounding at the expense ratio. Every basis point is an ounce leaving.

The ladder

GLD charges 0.40% and is the most liquid, with the options and lending markets institutions use. IAU charges 0.25%. State Street’s own GLDM charges 0.10% and exists because IAU was taking the retail flow; BlackRock’s IAUM charges 0.09% for the same reason; abrdn’s SGOL and GraniteShares’ BAR charge 0.17% and hold bars in Swiss and London vaults. For a buy-and-hold investor the lowest fee wins, because the funds are otherwise near-identical claims on the same kind of bars; for a trader, GLD’s spread and options depth are worth the extra thirty basis points.

The two surprises

The first is tax. Because the trust is a grantor trust, the IRS looks through the share to the metal, and a long-term gain on GLD or GLDM is taxed at the 28% collectibles rate, not the 20% equity rate; SPDR’s own tax FAQ says so. And because the trust sells a little gold each month to pay its fee, every shareholder has a small reportable gain or loss on that sold gold every year, which the sponsor publishes in a tax schedule and most holders ignore at some small risk.

The second is custody: a share in a New York-listed trust holding bars in a London bank vault, with sub-custodians the trust does not fully control, and a prospectus that says so at length. If the reason you own gold is that you distrust banks and governments, it is the wrong wrapper. In every other scenario it is the cheapest and most liquid one, and for most readers the first position belongs there, inside a Roth or traditional IRA where the 28% problem disappears; the IRS ruled in 2007 (PLR 200732026) that an IRA may hold the physical trusts without triggering the collectibles prohibition.

None of these three wrappers is a claim on a specific bar, and each is a different asset from the metal. The COMEX gold contract is 100 troy ounces, deliverable in approved New York-area depositories; the micro contract (MGC) is 10 ounces and is the one a private investor can size. Since January 13, 2026 CME has set gold margins as a percentage of contract value rather than a fixed dollar amount, 5% for a standard account and 5.5% for higher-risk positions (silver 9%), so the collateral required rises with the price.

Most contracts settle by offset, but delivery is real: tariff fears pulled 25.4 million ounces into COMEX warehouses between December 2024 and March 2025, to a record 43.3 million; stocks were 27.4 million on September 3, 2026, 15.1 million of them registered for delivery, and registered stocks falling while open interest (the number of contracts outstanding) rises is when the futures market moves.

What futures cost and how they are taxed

A long-only holder rolls into a contract priced above spot by roughly the short-term interest rate less the gold lease rate. On July 30, 2026 that contango was $161.70 an ounce from the August 2026 to the June 2027 contract, about 4.7% a year, with annualised carry of 4.0–5.1% along the curve against a three-month bill yield of 3.86% (metalcharts.org curve reading). If your collateral sits in Treasury bills, the interest offsets most of the contango and the position tracks spot minus frictions; if it does not, the roll is a fee larger than any ETF’s.

The tax is Section 1256: 60% of any gain long-term and 40% short-term regardless of holding period, a blended top federal rate of about 26.8%, lower than the collectibles rate even for a day trade. The catch is mark-to-market: every open position is deemed sold on December 31 and the tax is due that year. For a trader that is a gift; for a ten-year holder it is a compounding cost, and Section 15 shows it makes the futures the worst of three wrappers for a buy-and-hold.

Miners

A gold miner is a call option on the gold price with a strike at its cost of production, and 2026 is the year the option went deep in the money. In the second quarter, on an average price of about $4,512, Agnico Eagle produced at an all-in sustaining cost (the industry’s measure of the full cash cost of an ounce, sustaining capital included) of $1,459 an ounce, Newmont at $1,621 and Barrick at $1,866, margins of 59% to 68% of revenue at businesses that a decade ago struggled to earn their cost of capital at $1,200.

The leverage cuts both ways: the VanEck Gold Miners ETF (GDX, 0.51% fee) fell 21% in that quarter while gold fell 16%, and was still up about 50% over the trailing year. Miners move 1.5–2 times as far as gold in both directions and then give some back through cost inflation, dilution and acquisitions; the majors’ 2026 cost guidance was up 10–25% on 2025 (Agnico Eagle $1,475 against $1,339; Newmont $1,680 against $1,358; Barrick $1,760–1,950 against $1,581). Over 2011–2024 the miners badly lagged the metal; over 2024–2026 they crushed it. Both will happen again; Investing in Gold Miners and Royalty Companies, on this hub, works the 2006–2026 record and the tax difference through.

Royalty and streaming

Franco-Nevada, Wheaton Precious Metals, Royal Gold and Osisko finance mines for a share of revenue or the right to buy ounces at a fixed low price, carrying gold-price upside with no operating costs, margins in the 80–90% range and dividends that rose through the 2013–2018 bear market, at valuations often two to three times the producers’ multiples. Both are equities: taxed at 20% with qualified dividends, with a balance sheet you can read and a management you can fire, but also a cost curve, a union, a permitting regime and a government. They belong in the equity budget, not the gold sleeve.

Physical gold is lost three ways: a bad bar, a bad custodian, and a claim that was never metal. The terminology first, because the failures hide inside it. Allocated means specific bars, listed by serial number, held in your name, off the custodian’s balance sheet. Segregated means physically separated in your own box. Unallocated means you are an unsecured creditor of whoever holds the pool.

The menu, with published fees

At home, a scheduled-valuables insurance rider is typically quoted at 1–2% of value a year, the dearest storage on this list. A bank safe-deposit box costs $50–300 a year, is uninsured, and sits inside the institution you may be hedging; it is nonetheless the right home for a mid-sized coin holding.

The professional vaults publish schedules, charted in Section 9: Brink’s at $1.00–1.20 per $1,000 of value a year commingled and $2.00 segregated, insured; BullionVault at 0.12% a year with a $48 minimum in Zurich, London, New York, Toronto or Singapore, plus 0.5% dealing commission each way falling to 0.1% above $75,000; Delaware Depository at about 0.50% commingled and 1.50% segregated on IRA custodians’ schedules; the Perth Mint, owned by the Government of Western Australia, at 1.00% allocated and nothing unallocated; Swiss private vaults at 0.5–1.2% on 2026 surveys.

Counterfeits

Tungsten has a density of 19.25 grams per cubic centimetre against gold’s 19.30, so a plated tungsten slug passes the weight and specific-gravity tests that used to be enough, and cored fakes have turned up in one-ounce to kilo formats.

The dealer’s testing stack: XRF reads the alloy at the surface and penetrates only about 0.05 millimetres, so a plated bar passes it; the Sigma Metalytics verifier pushes an electromagnetic field several millimetres in and measures resistivity, and because tungsten conducts less than half as well as gold a cored bar fails in a second; ultrasound measures the speed of sound through the bar, about 3,240 metres a second in gold against 5,170 in tungsten, and is the standard for large bars. For a coin, weight and dimensions against the mint’s specification plus a Sigma pass is sufficient. For a bar, the cheapest protection is provenance: a refiner bar in its sealed card from a dealer whose name is on the invoice.

The failures

Northwest Territorial Mint filed for bankruptcy in 2016 owing tens of millions of dollars to customers who had paid for metal that was never bought; Bullion Direct in Texas collapsed in 2015 with a vault holding a small fraction of what its customers believed was there. In both cases the customers thought they owned allocated metal. Unallocated bank accounts and dealer pool accounts are the same structure in respectable clothing: cheap because the holder can lend and under-reserve against your metal, and legally a deposit that queues with the other creditors when the holder fails.

Four questions settle it: is there a bar list with serial numbers, is the metal held by a third-party vault operator rather than the seller, is it insured in your name, and can you take delivery. A no to any of them means you own a promise. Executive Order 6102 of 1933, which required Americans to sell their gold to the Federal Reserve at $20.67 before the government revalued it to $35, is the oldest counterparty story of all and the reason a certain kind of buyer keeps a portion abroad; Section 14 is what that costs in paperwork.

Tax is the largest single cost in every column of the worked example that follows, so here is the US treatment of every wrapper precisely. The Internal Revenue Code defines gold bullion and coins as collectibles, and Section 1(h) taxes gain on a collectible held more than a year (the “28-percent rate gain” of Section 1(h)(4)) at a maximum of 28%, against 20% for stocks. The 3.8% net investment income tax applies on top above $200,000 of income for a single filer and $250,000 for a joint return, so the effective federal top rate is 31.8% before state tax. Gains on holdings of a year or less are ordinary income at up to 37%. Losses are capital losses and offset gains in the normal way.

The chart is the map. The grantor-trust ETFs follow the metal at 28%, plus the small annual expense-sale gain the sponsor reports (Section 11); futures are Section 1256 contracts at a blended 26.8%, marked to market every December (Section 12); miners and royalty companies are equities at 20% with qualified dividends. The odd result is that the wrapper Washington taxes most gently is the leveraged one, and the one it taxes hardest is the coin in your hand.

IRAs

Section 408(m) lets an IRA hold American Eagles by name (since 1986) and, since 1998, gold of at least .995 fineness, provided the metal is held by the IRA trustee at an approved depository. A self-directed gold IRA therefore involves a custodian, a dealer and a depository, with fees that typically total $200–400 a year on top of the dealer’s premium, and a great deal of aggressive marketing.

The Tax Court’s 2021 McNulty decision (157 T.C. No. 10) confirmed that a home-storage IRA, with the coins in a safe at home through an LLC, is a taxable distribution of the coins. The simpler route is the one the IRS opened in a 2007 private letter ruling (PLR 200732026): GLDM or IAUM inside a Roth is gold at a tenth of a percent with no 28% problem at all, and in the worked example it is the best column by more than $15,000.

Sales tax, reporting and the border

Forty-five states exempt bullion and coins from sales tax, most with a minimum purchase or fineness condition, and the map moved twice this summer in the buyer’s favour: Maryland, which had narrowed its exemption in 2025, restored it statewide for transactions of $1,000 or more from July 1, 2026 (SB 309, signed May 26, 2026), and Virginia’s exemption, due to sunset on June 30, 2026, was extended to June 30, 2028 in the 2026 budget.

The five that still tax it, plus the District of Columbia, are Hawaii, Maine, New Mexico, Vermont and Washington, which repealed a forty-year exemption from January 1, 2026 at 6.5% plus local rates. A buyer in a taxing state pays 4–10% going in and recovers none of it; for those readers the ETF or a vault in another state is the only sensible route.

Dealers file Form 1099-B when a customer sells them items on the IRS reportable list in reportable quantities: gold bars of a kilo or more, 25 or more Krugerrands, Maple Leafs or Mexican Onzas, and larger silver and platinum lots. American Eagles and Buffaloes are not on the list in any quantity, which is a large part of why Eagles are the US coin of choice; it is dealer reporting, not a tax exemption. A dealer receiving more than $10,000 in cash files Form 8300. Buying is not reported at all.

At the border: metal in a foreign storage account, which is what most vault programs are, generally goes on the FBAR, the Treasury’s foreign-account report, once your aggregate foreign accounts exceed $10,000; a foreign safe-deposit box with no account relationship does not. If the point of a Swiss vault is privacy from the US government, the account form of it is precisely what the US government requires you to declare.

The same position, twenty troy ounces of gold, about $88,900 at our tape’s $4,443.90, goes through three wrappers for a decade with every cost and every tax shown in dollars. Assume gold compounds at 5% a year in nominal terms to $7,238.64 in September 2036, a rate between the 3.6% a buyer at the 1980 peak has earned and the 9.2% earned since 1971, and well below the recent run; top federal rates of 28% on collectibles, 26.8% blended on Section 1256 gains and 37% on ordinary income; no state tax, no NIIT; and a sale at spot. Then the same three with gold flat, because a wrapper’s cost shows most clearly when the metal does nothing.

Twenty American Gold Eagles in a safe-deposit box

Buy random-year coins at $100 over spot, between this week’s cheapest offer (1.4% over, about $60) and the average dealer’s (4.1%, about $180): 20 × $4,543.90 = $90,878. Rent a box at $150 a year: $1,500 over the decade, uninsured. Sell to a dealer at spot, a fair assumption for Eagles in a normal market: 20 × $7,238.64 = $144,773. Gain $144,773 − $90,878 = $53,895; tax at 28% = $15,091. Net after tax and storage: $128,182, or 3.50% a year on the cash laid out.

Twenty ounces of GLDM in a taxable brokerage account

Buy $88,878 of shares at spot, no premium, no commission. The 0.10% fee is paid in gold, so after ten years the shares represent 20 × 0.999 to the tenth power = 19.80 ounces, worth 19.80 × $7,238.64 = $143,332. Gain $54,454; tax at 28% = $15,247. Net: $128,085, or 3.72% a year on the smaller outlay. The same shares in GLD at 0.40% would keep 19.21 ounces and net $125,027; the $3,058 between the two funds is the case for reading the expense ratio. The same shares in a Roth IRA would net the full $143,332.

Two micro futures contracts, fully collateralised

Hold two 10-ounce MGC contracts and keep the full $88,878 in Treasury bills at 3.5%, rolling every two months. Assume the curve charges the same 3.5% as contango, so bill interest and roll cost cancel and the position tracks spot less about $50 a year of commissions and spreads. The difference is timing: each year’s futures gain is taxed at 26.8% that December and each year’s bill interest at 37%, with the after-tax cash reinvested. Over the decade the futures tax comes to $4,360 and the interest tax to $13,388, paid as you go. Net in 2036: $123,583, or 3.35% a year, the worst of the three despite the lowest headline rate, because the collateral income is taxed as ordinary income every year and nothing is deferred.

That assumes the roll and the bill yield cancel. On the curve as it stood on July 30, 2026, with bills at 3.86% and contango at 4.7%, the same position ends at about $115,300, and at about $78,500 if gold is flat; a roll that costs more than the collateral earns is the futures holder’s version of an expense ratio, and a large one.

When gold does nothing

Flat at $4,443.90 for ten years, the Eagles come back at $88,878 less the $1,500 box: $87,378, down 3.9% on the cash out. GLDM keeps 19.80 ounces worth $87,993, down 1.0%. The futures end at $84,966, down 4.4%, because the bill interest is taxed at 37% every year while the equal-and-opposite contango loss is worth only 26.8% as a deduction; the asymmetry alone costs about $320 a year on a position that is a wash before tax. The flat case is the one to size by: coins cost about 0.4% a year to hold, collateralised futures about 0.45%, the low-fee ETF a quarter of either.

Three lessons. First, the wrappers converge over a decade: coins and the low-fee ETF finish within $100 of each other in the growth case, and the premium that looks painful on day one is noise by year five. Second, tax is the largest cost in every taxable column, $14,000 and up, and the same 28% for coins and ETF; the Roth column beats the best taxable column by $15,247, more than every fee, premium and storage charge in the table combined. The structure decision is worth more than the wrapper decision. Third, the wrapper with the best rate has the worst outcome for a holder, because the rate is charged every year. A ten-year holder should not be in futures; a ten-week trader should not be in coins.

The sequence below is the one that loses the least money; most bad outcomes in this asset come from skipping a step. The readings that would change our view follow it.

1. Decide what the position is for. A diversifier belongs in a low-fee ETF in a tax-advantaged account; crisis insurance belongs in sovereign coins you can hold; a trade belongs in a futures or brokerage account with a stop. Buying the wrapper for one purpose and judging it by another is the commonest error in gold.
2. Start in the cheapest liquid wrapper, GLDM or IAUM at a tenth of a percent, inside an IRA if you have room, sized at half your target weight. You now own gold at institutional cost with no premium to recover, and can learn how the price behaves with real money.
3. Add the physical layer slowly. Random-year Eagles, Buffaloes or Maples from the cheapest of three dealer quotes on the day, bought monthly rather than in a lump. Test the first coin from any new source with a Sigma or at a local shop.
4. Sort the storage before the fifth coin. A safe-deposit box for coins; an allocated vault with a bar list and third-party insurance for bars; a written inventory with serial numbers and invoices kept somewhere other than the metal.
5. Size it and leave it. Five percent of a diversified portfolio as a standing position, the level the Council’s 2026 strategic-asset work most often illustrates within its 4–15% range; up to ten for someone whose other assets are all financial; above ten is a view, not an allocation. Rebalance annually and after any 20% move in either direction, which meant selling in January 2026 and buying in July.
6. Read two numbers a month: the Shanghai premium and the Council’s central-bank data. Together they tell you whether the two buyers that set the floor are present.

What we are watching, with thresholds

Central-bank purchases: a full-year 2026 figure below 600 tonnes in the Council’s January 2027 report would mean the marginal buyer has gone quiet; any quarter above 350 tonnes would confirm it has not, and the third-quarter figure, due in late October or early November, is the first test of whether the record second-quarter buying held through the summer recovery.

ETF flows: August’s $18 billion put holdings above the February record, so the fast money is back; two consecutive months of $5 billion or more of outflows would put the July low back in play.

Real yields: the 10-year TIPS yield was 2.43% on September 4, 2026 (WatchGold’s tracker; the nominal 10-year less its breakeven gave about the same reading on September 9), just under the 2.5% that is our line. It touched 2.4–2.5% only briefly, in October 2023; a quarter sustained above it is where we would expect Western fund money to leave again.

The Shanghai premium: about $3 on September 8, 2026; above $20 an ounce for a month means the Eastern physical buyer has returned; a sustained discount means the price has run ahead of the world’s largest physical market.

Our own tape: the July 16, 2026 close of $3,985.60 is the line; a close below it puts the January 2026 high in the same category as January 1980 and September 2011, and a close above $5,318.40 without official buying behind it is the signal in Section 7’s rule.