The dollar, US yields and oil explain only about a sixth of gold's autumn slide

By Strota Newsroom · 2026-10-26 · How Strota reports

On 24 August 2026 the front-month gold contract settled at 4,697.8 dollars an ounce; by 26 October it was at 4,028.6, a drop of 14.24%. The familiar story blames a firmer dollar, higher US bond yields and an oil spike stoking fears of rate hikes. When we measured how much of the fall those three inputs can actually carry, based on how gold has responded to them in the past, the answer was: not much. Over the nine completed weeks to 23 October, gold lost 14.37 points, and the three textbook drivers account for just 2.43 of them using post-2022 relationships. The part left over is the biggest for any nine-week window we can find since 2022.

That matters because the standard narrative tends to be repeated as if it settles the question. It does not. Something else drove most of this decline, and the inputs people usually reach for were not it.

The test is simple. We gathered daily closes from December 2017 through 26 October 2026 for four series (front-month COMEX gold, Brent crude, the Dollar Index and America's 10-year Treasury yield) and turned them into Friday-to-Friday weekly changes. A linear regression then estimated how gold's weekly return has typically moved with the dollar's percentage change, the yield change in basis points and Brent's percentage change. Because central-bank buying is widely thought to have loosened gold's tie to rates after 2022, the model was fitted twice: once on 2018-2021 (209 weeks) and once on 2022 through October 2026 (252 weeks).

The fitted sensitivities look as expected for the dollar and yields. In the recent period, each 1% rise in the Dollar Index has gone with gold about 0.9019% lower, and each basis point of 10-year yield with a small decline too. Oil is the odd one: in neither period does Brent show a reliable week-to-week link to gold. The relationships have also faded. Taken together, the three inputs explained 39.6% of weekly gold moves in 2018-2021 but only 16.1% since 2022.

Now apply those weights to what actually happened. Across the nine weeks, the 10-year yield rose 25.0 bp, the Dollar Index gained 3.05% and Brent climbed 23.66%. Under the 2022-2026 fit, the dollar and yields together explain 3.36 points of gold's 14.37-point fall, while the oil surge, if anything, pushed the other way by 0.92. Net, the model accounts for 2.43 points and leaves a residual of 11.94. Even the older, tighter 2018-2021 relationships only stretch to 5.29 points, leaving 9.08 unexplained.

The timing undercuts the yields story further. Over the closing three weeks, the 10-year yield actually eased, from 5.28% at the start of October to 4.99% by 23 October, yet gold kept sliding. The week to 23 October, down 4.14%, was gold's worst since early June, when it dropped 4.96%.

To judge whether the leftover is merely noise, we computed the same nine-week residual, including the 2022 fit's average weekly drift, for every nine-week stretch since 2022. With drift included, the current one came to -16.32 points, the lowest of all 243 windows. Think of it like a doctor who can account for a fever's mild part by the known infection, but whose thermometer reads far beyond anything that infection has produced before: the reading says another cause is at work, without naming it.

Indian holders got no shelter from the currency. Measured in rupees, the fall from 24 August to 26 October was 14.67%, because the rupee barely shifted, going from 95.70 to 95.23 per dollar. For context, gold's 2026 high close was 5,318.4 on 29 January; by 26 October it sat 24.25% below that peak and 7.20% under its 2025 year-end close of 4,341.1.

We also checked a curiosity. On 21 October, Marathon Fusion said it had converted mercury into stable gold, and some coverage wondered whether lab-made metal could weigh on prices. Gold rose 0.18% that day, then fell 2.16% and 1.29% on the next two sessions. Against a daily standard deviation of 1.73% over the prior 250 sessions, those are ordinary moves, and they came in a week when Brent jumped. According to press reports of the company's own modelling, output would be about two tonnes per gigawatt-year of fusion power, set against roughly 3,672 tonnes mined in 2025, and no commercial fusion plant exists yet. No outside party had confirmed the claim, and the company has not released any figures on yield or cost. The price data cannot tell the announcement apart from noise.

So what does explain the slide? This study cannot say. The residual points to forces the three inputs miss, such as the unwinding of buying from the earlier rally: profit-taking, fund and ETF flows, or a change in central-bank pace. None of those were measured here.

The limitations are real. Front-month futures closes carry small roll effects, and rupee figures use international prices converted at USD/INR, not MCX or retail prices with import duty and GST. Weekly data miss intraday reactions, and the model is linear and looks backward. The residual identifies no cause. And the Marathon Fusion supply figures come from reports of the company's claims, not from data we hold.

More money stories

• Does Muhurat trading bring luck? We checked every Diwali session since 2012
• Tanishq vs Malabar gold plans: what your ₹5,000 a month really earns
• How to improve your odds of getting shares in a 100x-oversubscribed IPO
• Fed decision days barely move the Nifty. Gold is where the reaction shows up
• Nifty does not settle at max pain: across 120 expiries, the prior close was the better guide
• India VIX overstated the Nifty's coming volatility on 79.7% of sessions since 2014