Gold Fell 8.5% While ETF Holdings Rose. What Was the Market Pricing?

Gold fell 8.5% in September 2026 while gold ETFs added 67 tonnes, taking their holdings to a record 4,256 tonnes. We separate fund demand from futures positioning, real yields and price momentum, set out what Jeffrey Christian, Brent Johnson and Chris Vermeulen said in early October, and explain why none of it is an instruction to change a portfolio.

Published October 16, 2026 · Updated October 17, 2026 · 16 min read

Key Takeaways

• Gold ended September 2026 at US$4,176 an ounce, 8.5% lower on the month, while the gold ETFs the World Gold Council tracks added 67 tonnes, to a record 4,256 tonnes, on US$10 billion of inflows. Their assets under management still fell 7%, to US$574 billion: more ounces, each worth less.
• Futures traders went the other way. The Council reports net managed-money positions in COMEX gold down US$12 billion, or 84 tonnes, in September, and the CFTC’s weekly figures show the net long falling for six weeks in a row, to 110,987 contracts on 6 October. The group was still net long by more than seven to one.
• The US 10-year real yield rose 49 basis points in September, to 2.93%, and was 2.84% on 14 October: an easing of nine basis points, not a reversal.
• CPM Group expects gold and silver to rise over the next several months while allowing for more short-term weakness; Brent Johnson put gold near US$3,700 if the dollar index returned to 110; Chris Vermeulen’s next downside target was US$3,600, with a close below about US$3,900 as his trigger. Each is an attributed view, not a finding of the data.
• Our published Tactical Score rose from 35/100 on 2 October to 43/100 on 14 October, still Cautious, with seven of the eight points from its momentum block. Fund holdings are not one of its signals, and a demand report is not an instruction to change a portfolio.

Gold finished September 2026 at US$4,176 an ounce, 8.5% lower over the month on the World Gold Council’s figures. In the same month the gold-backed funds in the Council’s dataset added 67 tonnes of metal, taking their holdings to a record 4,256 tonnes. Both figures were published on 7 October and describe September as a whole. The month’s other large buyer, China’s central bank, is the subject of our article on its September purchase; this one looks at the funds, at the traders who were reducing their positions while the funds were adding, and at what the first half of October added to the picture.

+67 tonnes Added to gold ETF holdings worldwide in September 2026, taking them to a record 4,256 tonnes. Net inflows were US$10 billion, and every region recorded inflows. Over the same month the gold price fell 8.5%, to US$4,176 an ounce, and the funds’ assets under management fell 7%, to US$574 billion. World Gold Council, 7 October 2026.

That raises a reasonable question: if investors were buying, why did the price fall? It becomes easier to answer once “investors” stops meaning one group with one view. Evidence of buying in one part of a market is not evidence that conditions improved everywhere, and a larger quantity of metal held by the funds does not require each ounce to finish the month at a higher price.

For a portfolio investor the useful distinction is between three things: demand for an asset, the conditions that affect its price, and the purpose of owning it. They are related questions, not interchangeable ones, and September is a clear example of all three pointing in different directions.

How can gold ETF holdings rise while the gold price falls?

Fund holdings can rise while the price falls because holdings count ounces and the price sets what each ounce is worth: more ounces at a lower price can add up to less money. The World Gold Council’s September report separates four measures that are easy to confuse.

Source: World Gold Council, “UK crowned in a record quarter”, 7 October 2026. Rounded figures as reported. The four rows describe the same funds from different angles and are not four separate additions to demand.

Holdings measure metal, flows measure money and assets under management measure value. The Council’s dataset covers more than 100 physically backed gold ETFs and similar products, including closed-end funds and mutual funds, so it is a collection of structures with different terms rather than one kind of fund. The same three measures moved together in August, when they all rose; the difference between them is the subject of our review of August’s record fund flows. In September they moved apart.

Nor is every trade in a fund’s shares a purchase of gold. SPDR Gold Shares (GLD) explains in its investor FAQ that the trust’s trustee does not deal with the public: it handles creation and redemption orders with authorised participants, in blocks of 100,000 shares called baskets, and gold moves into or out of the trust with those baskets. An investor buying shares on the exchange is therefore not dealing with the trust; gold enters the trust when an authorised participant creates a new basket. The FAQ adds that creation and redemption have not taken place every trading day. GLD illustrates the mechanism; it does not describe every product in the Council’s dataset.

The conclusion September supports is therefore narrower than “everyone is buying”. One investment channel accumulated metal. The figures do not identify each investor’s motive, or show how long the positions will stay in place.

If the funds were buying, who was selling?

The published data do not identify every seller, but they show one group cutting its exposure: managed-money traders in COMEX gold futures, whose net positions fell by US$12 billion, or 84 tonnes, during September, according to the World Gold Council’s monthly market commentary. The Council’s attribution model counts a rise in yields, with the US 10-year Treasury yield up 53 basis points to 5.3%, and a 2% rise in the dollar index as major contributors to the fall in price, and says the futures liquidation likely contributed as well. Those are the Council’s analytical conclusions, not a reconstruction of every transaction. The 84 tonnes is the Council’s own measure, to the end of the month; on the CFTC’s futures-only figures used below, the managed-money net long fell by 24,429 contracts between Tuesday 25 August and Tuesday 29 September, about 76 tonnes. The dates and methods differ, and both point the same way.

Futures matter even to an investor who only ever buys bars and coins. A COMEX gold contract is a contract on 100 troy ounces, and when a large group of traders reduces its exposure to the price, the price can move without a single bar leaving an ETF vault. But the futures figure should not be subtracted from the ETF figure to produce “net gold demand”. One is a change in a defined group’s contractual exposure; the other is a change in metal held by tracked funds. They are not two lines of the same physical account.

There is a second distinction. In the CFTC’s report the long positions of all traders add up to the short positions (396,109 contracts each way on 6 October, counting spreads), so when managed money becomes less net long, the change describes that group’s position, not a market with sellers and no buyers. Our Who Holds Gold Futures chart separates the categories so each can be read on its own.

The positioning report of 9 October

The CFTC’s futures-only report for Tuesday 6 October, released on Friday 9 October, shows managed-money traders holding 127,101 long contracts in COMEX gold and 16,114 short contracts: a net long position of 110,987 contracts. Against 29 September their longs fell by 4,610 and their shorts rose by 4,721, so the net position fell by 9,331 contracts, about 933,000 ounces or 29 tonnes at 100 ounces a contract. It was the sixth weekly decline in a row.

The group remained net long by a wide margin, which is different from saying it had turned short. The date matters as much as the direction. Commitments of Traders reports normally use Tuesday’s positions and are released on the Friday, so this one says nothing about trading after 6 October, and it should not be read as an explanation of later price moves. The next report, for positions on 13 October, had not been released when this article was prepared. It is a later observation, not a number to add to the September ETF total.

Have real yields come down since September?

Only a little. The US 10-year real yield rose from 2.44% to 2.93% during September and stood at 2.84% on 14 October, nine basis points below its end-September level after a 49-basis-point rise. A basis point is a hundredth of a percentage point.

Source: US Treasury, Daily Treasury Par Real Yield Curve Rates, 2026, 10-year column. These are dated readings from one series, not intraday highs or a complete daily chart; the changes are our arithmetic.

This is the Treasury’s own real yield curve, published as a series of its own, not a figure we derived by subtracting an inflation rate from a bond yield. The World Gold Council’s September commentary pairs its model’s finding with the ordinary 10-year yield, which rose 53 basis points over the month. Bullion pays no interest, so a higher yield on the alternative raises what it costs to hold gold instead, though it is not a rule that forces gold lower. The argument, and the years when it held, are set out in the counter-arguments section of our Learning Academy.

What matters here is not the last decimal place but the difference between a modest easing and a reversal of the earlier move. By 14 October the real yield had given back less than a fifth of its September rise. Calling both of those “falling yields” would hide more than it explains.

Two Voices on demand during a falling price

Jeffrey Christian: demand can persist through price weakness

In a presentation published with a written summary on Kitco on 9 October, Jeffrey Christian of CPM Group discussed continued gold ETF buying during the decline, and CPM said it expects gold and silver prices to rise over the next several months despite continued short-term downward pressure. That is a forward-looking interpretation, not proof that a low has formed. Its relevance here is the distinction it draws between a weak price and an absence of investment demand.

Two details in CPM’s summary do not match the data used in this article, and we have not reconciled them. It puts September’s fund buying at about 2 million ounces, roughly 62 tonnes, against the Council’s 67 tonnes, which may reflect rounding or different coverage. And it refers to an increase in investors’ net long positions in COMEX gold futures without naming the category or the period, while the CFTC’s managed-money figures fell in every week from 25 August to 6 October. We use the Council’s and the CFTC’s figures for the data, and CPM’s text for its outlook.

Brent Johnson: a constructive view that includes a downside scenario

In MacroVoices episode 553, recorded on 7 October and published on 8 October, Brent Johnson was asked what gold would do if the US dollar index went back to 110, and said it would probably go lower, perhaps to around US$3,700. Earlier in the interview he allowed for a near-term pullback in the dollar while thinking it could go much higher over a longer period. In the same answer he said he had bought gold for a client the day before, that gold was at a support area, and that he expected a fairly big move in the following weeks, probably higher; if it fell instead, he said he would probably buy at the low. The US$3,700 figure was a scenario with a condition attached, not his unconditional forecast.

CPM and Johnson both allow for further weakness while holding a constructive view. Their value here lies in the assumptions and time horizons they state, not in counting optimistic commentators as evidence that gold must rise.

Is buying demand the same as price momentum?

No: demand describes who is adding to their holdings, while momentum describes the direction and persistence of the price, and in early October 2026 the two pointed different ways. The clearest statement of the momentum side came from outside our Voices. In an interview with Craig Hemke for Sprott Money, recorded on 2 October and published on 6 October, the technical analyst Chris Vermeulen described gold as trading below its 150-day moving average, with its shorter averages below that one, and called the recovery a counter-trend rally. This was his assessment at the recording date, not Sprott Money’s house view.

He called about US$3,600 an ounce the next downside target, with US$3,100 possible beyond it, and said the long-term trends were more likely to continue than to reverse. He named a trigger as well: if gold broke and closed below roughly US$3,900 for a day or two, he thought it would slide to US$3,600. He also described the opposite case: if the pullback moved sideways and then turned higher, it could start a large rally. These were chart projections with conditions attached, not outcomes the ETF data establish.

His firm’s written update of 7 October again described the metals’ bounce as counter-trend and said it remained focused on equities, especially the Nasdaq and technology-related leadership. That does not show that ETF investors sold gold to buy shares. Relative price performance is not a record of where any particular investor’s money went.

Sources: CPM Group, commentary published on Kitco, 9 October 2026; Sprott Money interview transcript, 6 October 2026; The Technical Traders, written update, 7 October 2026. These are attributed views, not GoldSilverPortfolio forecasts.

For our purposes the useful point concerns the decision being made. Christian’s argument about demand and Vermeulen’s reading of momentum address different questions. A supportive investment case does not automatically satisfy a trend-following rule for when to enter, and a weak chart does not erase reported fund buying. Vermeulen’s published approach is to own what is trending up and move to cash when trends break. That is a trading discipline to judge on its own terms, not a rule to apply automatically to an investor’s whole metals allocation. Read the method and its conditions, not only the price target.

What does September mean for the Macro Conditions Score?

Nothing directly: fund holdings are not among the 16 signals the dashboard scores, so the ETF report could not move it, and nobody adjusts the score by hand for a headline. What the score did after September can be read from its own published inputs. Our review of the week to 2 October recorded a Tactical Score of 35/100 for that Friday, in the Cautious band, and a Long-Term Structural Score of 64/100 in its 28 September update, in the Bullish band.

Seven of the eight points came from the momentum block, which blends changes over two to four weeks in the real yield, the two-year Treasury yield, inflation expectations, the dollar and the gold price; the other came from the confirmation basket. The core levels, which include the real yield itself, did not move. That is the distinction drawn earlier in a different form: conditions eased at the margin while the level of the competing yield still weighed against gold. The Long-Term Structural Score read 61/100, still Bullish, in its 5 and 12 October updates.

The two horizons can point different ways. A reader can regard a strategic holding as worthwhile while accepting that conditions for a new short-term position are less comfortable, and a tactical opportunity can appear without changing the size of the long-term allocation. The Signals page carries the current readings and their dates; the background for reading these measures together is in the market context section of our Learning Academy.

Should a fund-buying report change a portfolio decision?

Not on its own: a holdings report records activity in one part of the market, a technical assessment interprets price behaviour, and a portfolio plan sets out what a holding is meant to do, and none of them substitutes for the others. GoldSilverPortfolio favours a deliberately sized precious-metals holding as part of a diversified portfolio. The percentage, the mix of metals, the form of ownership and the strategy remain the investor’s decisions.

A tactical warning is a reason to examine a position’s risks and assumptions, not an automatic instruction to abandon a strategic holding, and strong buying data are no more an instruction to increase it. Calling a holding “strategic” does not justify one larger than the investor can tolerate. Our position is not that metals rise in every market; it is a preference for building a portfolio that allows for more than one economic outcome. Our two-horizon portfolio article develops the difference between strategic ownership and tactical decisions.

The questions that decide the role are personal: how much volatility the investor can live with, what income and cash reserves are needed, which risks already dominate the portfolio, and why the metal is held at all. The allocation section of our Learning Academy works through them. Several approaches can be considered without pretending that one report picks a winner: holding an allocation range and reviewing it at set intervals, building gradually through scheduled purchases, or separating a strategic core from a smaller tactical position. Each imposes its own discipline. Rebalancing needs a rule for when to act, scheduled buying needs a budget and patience, and tactical changes need an exit and re-entry framework and the acceptance that either decision may be wrong.

The form of ownership is a separate decision. Physical bullion can carry storage, insurance and resale costs, and other vehicles bring their own dependencies and risks; the physical-versus-paper section of our Learning Academy sets them side by side. The Financial Industry Regulatory Authority (FINRA) notes that precious metals are often promoted as safe havens, but that this does not make them immune from price declines. Nor does gold-specific evidence settle the case for silver, platinum or mining shares: a mining company has costs, management and financing risks of its own, a difference covered in our article on how gold ETFs are taxed.

What would make the next update useful?

The next review should test several observations rather than look for another reassuring headline:

• Do fund holdings keep rising when the October figures are reported?
• Does managed-money positioning stop falling in the reports after 6 October?
• Do real yields retreat by more than the nine basis points recorded by 14 October?
• Does the dollar index climb towards 110, the premise of Johnson’s US$3,700 figure, or pull back as he allowed it might in the near term?
• Does the price confirm or contradict Vermeulen’s reading, in either direction?

Two of those questions had early answers a week later. By 16 October the 10-year real yield had fallen to 2.74%, 19 basis points below its end-of-September level, and managed money’s net long position in COMEX gold had fallen for a seventh week in a row, to 107,089 contracts on 13 October; gold and silver still ended that week lower. See our review of the week to 16 October.

The answers need not agree, which is why the measurements are worth keeping separate. September does not show that demand no longer matters. It shows why a demand headline cannot stand in for an investment framework. Our part is to set out the evidence, the alternatives and the trade-offs, so that readers can decide what fits their own circumstances, including how much exposure they are prepared to keep when the price does not reward it straight away.

Sources and Methodology

Market data run to 14 October 2026, and every figure keeps its own reference date: monthly fund and price data are for September, futures positions are weekly (Tuesdays) and yields are daily. The gold price and its 8.5% fall are the World Gold Council’s month-end figures, not a COMEX settlement return or a retail bullion return after spreads, fees or currency conversion. Fund holdings, flows and assets under management are the Council’s, as reported; the split of September’s change in assets under management, the net futures positions and their changes, the tonne conversions and the basis-point changes in yields are our own arithmetic. Market figures are reported data, not an independent audit. The Macro Conditions Score readings are the dashboard’s published values, read from its record on 14 October 2026; nothing was recalculated for this article, and the ETF report played no part in them.

Christian’s views are taken from CPM Group’s written summary of its presentation, not from an independent transcription of the video. Johnson’s are from the public MacroVoices transcript, not from the separate paid report discussed in the interview. Vermeulen’s are from Sprott Money’s interview transcript and his firm’s written update; we did not reproduce his charts or test his strategy. Forecasts and scenarios are attributed opinions, not established outcomes. The futures figures are the CFTC’s disaggregated futures-only series for COMEX gold, not the futures-and-options series or the legacy non-commercial categories.

Conflicts of interest, stated plainly. The World Gold Council is a membership organisation funded by gold-mining companies, and its purpose includes developing demand for gold; its explanation of why the price fell is the analysis of an interested party and is attributed as such. CPM Group sells commercial research and advisory services, and its commentary was published on Kitco, which also sells bullion. Brent Johnson manages investments for clients. Sprott Money sells bullion, and The Technical Traders sells paid trading services. GoldSilverPortfolio has no affiliate or commission arrangement with CPM Group, Kitco, MacroVoices, Sprott Money or The Technical Traders, and their work is included for discussion, not as an endorsement of any product or strategy. GoldSilverPortfolio may earn a commission through two links elsewhere on this site: an affiliate link to SilverGoldBull, a bullion dealer, and a referral link to Wealthsimple, a brokerage. It also sells TaxStack, a paid record-keeping tool. No fund, dealer, trading signal or commentator’s forecast is presented here as a recommendation to buy or sell.

Primary sources

• World Gold Council, Johan Palmberg — “Gold Market Commentary: Go with the flow”, 7 October 2026: gold at US$4,176 an ounce at the end of September, 8.5% lower on the month; its Gold Return Attribution Model finding a rise in yields (the US 10-year Treasury yield up 53 basis points to 5.3%) and a stronger dollar (the DXY index up 2%) major contributors; net managed-money positions down US$12 billion (84 tonnes); the futures liquidation as a likely contributor to the decline
• World Gold Council — “UK crowned in a record quarter” (gold ETF flows to 30 September 2026), 7 October 2026: inflows of US$10 billion in September in every region; holdings up 67 tonnes to a record 4,256 tonnes; assets under management down 7% to US$574 billion; the definition of gold ETFs as regulated securities holding gold in physical form, including closed-end funds and mutual funds
• World Gold Council — Gold ETF Flows for August 2026, 9 September 2026: assets under management of US$615.3 billion at the end of August, as recorded in our review of August’s fund flows
• World Gold Council — Gold ETF holdings and flows dataset, coverage note: more than 100 physically backed gold ETFs and similar products
• Kitco News — “Gold ETF demand defies rising bond yields as investors pour $31 billion into record third quarter”, 12 October 2026: 67.3 tonnes worth nearly US$10 billion added in September (a second route to the Council’s figures)
• SPDR Gold Shares — GLD FAQ, January 2026, pages 4–6: the trustee creates and redeems shares only with authorised participants, in baskets of 100,000 shares; creation and redemption have not taken place every trading day
• US Commodity Futures Trading Commission — Commitments of Traders, disaggregated futures-only, COMEX gold (code 088691), weekly positions of 25 August to 6 October 2026: managed money 127,101 long and 16,114 short on 6 October, changes of −4,610 and +4,721 from 29 September; cross-checked against the CFTC’s 2026 history file
• US Commodity Futures Trading Commission — Commitments of Traders, about the reports: positions as of Tuesday, generally released on Friday at 3:30 p.m. Eastern Time
• US Treasury — Daily Treasury Par Real Yield Curve Rates, 2026: 10-year real par yields of 2.44% (31 August), 2.93% (30 September), 2.92% (2 October), 2.91% (9 October) and 2.84% (14 October)
• CPM Group — “Gold and silver ALERT: What could finally move prices?”, published on Kitco, 9 October 2026
• MacroVoices — Episode 553, “Brent Johnson: Disparate Housewives”, 8 October 2026, and its public interview transcript (recorded 7 October), pages 4–5
• Sprott Money, Craig Hemke — “Gold Could Fall to $3,600”, interview with Chris Vermeulen recorded 2 October and published 6 October 2026; transcript passages at 10:04–11:35, 13:11–14:43 and 16:05–17:24
• The Technical Traders — “Are Stocks Topping While Metals Begin to Pull Back?”, 7 October 2026, and the firm’s description of its approach
• FINRA — “4 Tips to Know Before Buying Physical Precious Metals”, 24 March 2026
• GoldSilverPortfolio — Macro Conditions Score dashboard (Engine V2), published tactical readings for 2, 9 and 14 October 2026 and Long-Term Structural Score updates of 28 September, 5 October and 12 October 2026, as recorded on 14 October 2026