Gold Rose 16.8% While Real Yields Rose 3bp (August 2026): The Six Weeks That Broke the Discount-Rate Model

From the London 3pm auction low of $3,993.55 on 16 July 2026 to $4,663.70 on 24 August, gold gained 16.8%. Over exactly those six weeks the 10-year inflation-protected Treasury yield — the opportunity cost that is supposed to govern the price of an asset paying no coupon — rose three basis points, from 2.35% to 2.38%. In the six months before that, the same relationship worked so cleanly it looked like a law: real yields up 46 basis points, gold down 26.1%. The model did not get less true. The buyer changed.

This is a piece about a broken explanation rather than about a price. Gold ended the week lower — spot fell 2.75% to $4,474.45 on Friday 28 August after the Fed chairman's Jackson Hole keynote — and that fall is part of the evidence, not a contradiction of it. The question worth answering is why a 13.5% monthly gain happened in a month when the number that is meant to drive gold barely moved, and what that tells you about who owns the metal now.

The two series, side by side

Everything below rests on two datasets anyone can pull: the LBMA's published auction prices, which are the settlement benchmark the physical market actually clears against, and the US Treasury's daily real yield curve, which is the market-implied real cost of money. Both are official. Neither is a forecast.

Read the first two rows together and the problem is obvious. In the first, a 46 basis point rise in the real cost of holding a zero-coupon asset coincided with a 26% fall in its price — textbook, and the mechanism this site set out when gold was falling in its worst quarter since 2013. In the second, the real yield finished higher than it started and gold gained almost 17%. A three basis point move has no explanatory power over a 16.8% one. Either the channel switched off, or something larger was operating on the other side of it.

The 24 August auction at $4,663.70 was the highest since 14 May, when the same auction fixed at $4,683.05 — a three-month high on the benchmark, consistent with the more-than-three-month spot high of $4,696.18 recorded on 25 August. Gold is still 15.5% below its 29 January auction record of $5,405, and up 5.0% for the year to 27 August against a 10-year real yield 40 basis points higher than where it started 2026.

The 19 August announcement did what the discount rate didn't

The fastest stretch of the rally has a date attached to it. On 19 August 2026 the US Treasury said it would increase the size of its liquidity-support buyback operations in longer-dated nominal coupon securities "by at least double" — from $2 billion per operation to at least $4 billion in the 10-to-20-year and 20-to-30-year sectors, running from 9 September to 4 November 2026. Five days later CNBC reported, citing two senior Treasury officials, that the Treasury General Account — which closed at $966.8bn on 24 August per the Daily Treasury Statement — is considered available to help fund those purchases, an approach Treasury Secretary Scott Bessent described as a "Treasury Twist". This site traced that sequence through the bond market in the 30-year round trip.

Watch what gold did across it. The 19 August auction fixed at $4,460.70. By 21 August it was $4,582.10, and on 24 August $4,663.70 — a 4.6% gain in three sessions. Over the same three sessions the 10-year real yield went from 2.35% to 2.38% and the 30-year nominal from 5.19% to 5.23%. Both rose. The opportunity cost of holding gold went up while gold went up 4.6%.

That is not the rate channel. It is the channel that runs through the credibility of the issuer. A sovereign whose debt manager is enlarging purchases of its own long bonds, and whose cash buffer is being discussed as a funding source for them, is a sovereign whose long-dated paper carries a different kind of risk from the one a real yield measures. Reserve managers hold government bonds as reserves. When the risk attaching to those bonds is about the issuer rather than about inflation, the natural substitute is the one reserve asset no government issues.

It is not a dollar story either

The obvious alternative explanation is the pricing currency. Gold is quoted in dollars, so a weaker dollar raises the dollar price of unchanged metal, and the dollar did weaken through most of August. On the ECB's daily reference rates the euro went from $1.1485 on 31 July to a peak of $1.1699 on 21 August and $1.1645 on 27 August — a 1.37% dollar decline over the month.

Convert the auction price into euros at those same rates and the rally survives almost intact: 3,505.96 euros on 31 July, 3,923.53 euros on 27 August, a gain of 11.9%. Of the 13.5% dollar-denominated move, roughly 1.6 percentage points was the currency and the remaining twelve was the metal. A European buyer who never held a dollar captured nearly the whole thing. That is the cleanest available test, and it says the dollar was a passenger rather than the driver.

Who the marginal buyer became

The demand data explains what the rate data cannot. The World Gold Council's Gold Demand Trends for the second quarter of 2026 records 289 tonnes of central bank net purchases against a 45-tonne outflow from exchange-traded funds, with total demand of 1,268.9 tonnes at an average auction price of $4,506.29. The Council attributed the quarter's price weakness to "upward adjustments to both inflation and interest rate expectations", particularly in North America — which is to say the real-yield channel was doing its job on the fund flows, and the official sector bought through it anyway.

The regional detail in the ETF flow data is sharper still. Across the first half of 2026, Asian funds took in a record US$12bn and European funds US$3.2bn, while North American funds shed US$7.7bn — their weakest first half since 2013. Holdings finished June at 4,047 tonnes with US$526bn of assets. July then flipped: US$3bn of inflows, holdings up 23 tonnes to 4,068 tonnes, assets up 1% to US$530bn, led by Europe.

Sequence that properly. Through the 26% drawdown, official reserve managers accumulated and Western funds liquidated. Then, from late July through August, the fund bid returned on top of official demand that had never gone away — into a market whose available float had been quietly reduced by buyers who do not sell into rate moves. A central bank adding gold is executing a reserve-allocation decision with a horizon measured in decades and no funding cost to compare it against. It has no discount rate. That is why the discount-rate model lost its grip: not because the arithmetic is wrong, but because the arithmetic describes a holder who was no longer setting the price.

Friday was the model working, briefly

None of the above means rates stopped mattering. On Friday 28 August at 10:00 a.m. Eastern, Kevin Warsh delivered his first Jackson Hole keynote as Fed chairman, described the inflation numbers as "more concerning", and closed the economic section with a standard: unless underlying inflation is moving to target "clearly and at sufficient speed", "we have work to do". Gold lost roughly $70 an ounce within ten minutes. Spot closed down 2.75% at $4,474.45 and December futures down 3% at $4,524.10. CME FedWatch pricing for a September increase moved from 36% to 56%, with December at 80%, and the dollar rose to a more-than-one-week high, per CNBC's account. The 10-year real yield closed eight basis points higher at 2.42%; the 2-year rose fourteen to 4.34%. The full repricing is set out on the Jackson Hole page.

So on a one-day horizon the opportunity-cost channel is alive and fast. Eight basis points of real yield produced 2.75% of gold. Three basis points over six weeks produced nothing, alongside 16.8% of gold. Both statements are true, and together they define what the model is for: it prices the response to news, not the level the news moves from. The speed of that response is a separate question again, and one this site examined in the dealer-hedging amplifier — a $70 move in ten minutes is what an options-heavy book does to a given surprise, in either direction.

Where this reaches a currency

Two channels, and it is worth being honest about their relative size. The first is the pricing currency: gold is quoted in dollars, so every gold quote contains a dollar quote, and the same Fed pricing that moves the real yield moves the dollar's rate factor directly. That channel is strong and it runs in both directions, which is why one speech delivered a firmer dollar and weaker gold on the same afternoon.

The second is the exporter channel, and it is weaker than gold's prominence suggests. Australia is a major gold producer, so a higher gold price is a genuine terms-of-trade gain for the Australian dollar — but gold is a much smaller share of the Australian export basket than iron ore, so the transmission is real and modest rather than decisive. The general mechanism is set out in the commodity-currency explainer. Forcing a 16.8% gold move into a currency call would be the wrong lesson to take from any of this.

What would change the picture

Three observable things, none of them a price level.

The first is North American ETF flows. If they turn decisively and persistently positive, the marginal owner becomes a rate-sensitive allocator with a funding cost to compare against, and the opportunity-cost channel regains its grip on the level as well as the day. Counter-intuitively that would mean more 28 August-style sessions, not fewer — a market owned by rate-sensitive money is a market that moves on rates.

The second is the official-sector pace. Two hundred and eighty-nine tonnes in a quarter is the price-insensitive floor underneath everything above. The Council's next quarterly report is the check on whether it is still there, and a material slowdown thins that floor without changing a single yield.

The third is the buyback programme itself. Treasury's expanded operations run from 9 September to 4 November. If they proceed without drawing on the cash account, the sovereign-credibility premium that built between 19 and 24 August has less to feed on; if the account is tapped, the question a reserve manager is asking gets louder. Either way it is a published schedule, not a guess.

And a standing caution on all of it. Every window in the table above was chosen because its endpoints are meaningful — a record, a low, a month boundary, an announcement date — but any two-series comparison can be made to say what you want by moving the endpoints. The claim here is narrow and falsifiable: across the six weeks in which gold gained 16.8%, the 10-year real yield rose. If you are new to how a rate or a commodity price reaches a currency at all, the about page explains what this site measures and why.

Educational macro context only — not investment advice.