Gold ETF Inflows Hit a Record 4,256t While Gold Fell 8.5%

September: gold ETFs took in US$10bn to a record 4,256 tonnes while COMEX money managers cut 84 tonnes — the divergence that framed gold's $4,150 base into the 28 October FOMC.

Levels in this article
  • $4,100
  • $4,073
  • $4,162
  • $4,150
  • $4,176
  • $4,161
  • $4,172
  • $4,181

Wednesday's CPI at 3.7% did what a hot inflation print is supposed to do to gold: the metal fell through $4,100 to a five-day low of $4,073, the 10-year yield spiked to 5.33% and the dollar printed an 18-month high. Thursday's PPI was hotter still — +1.3% m/m against +0.6% expected, the fastest since March 2022 — and gold dipped $12, round-tripped to $4,162, and closed New York in the $4,150-4,165 band as the 10-year reversed 14bp to about 5.08%.

Two hot prints in two days, and the metal is higher than before the first one: the marginal buyer is not trading the rate path, it is trading credit. An AI-linked equity and credit selloff — CoreWeave down more than 15% with its April debt near 81 cents on the dollar — pushed money into Treasuries and dragged the long end down hard. For one session, the bond market mattered more to gold than the inflation data that has punished it all month.

September was two markets wearing one price

Gold finished September at $4,176/oz, down 8.5% m/m — the 93rd percentile of monthly falls since 2004, according to the World Gold Council's 7 October commentary, Go with the flow. In the same month:

  • Global gold ETFs took in US$10bn, or 67 tonnes, lifting holdings to a record 4,256 tonnes and capping a record quarter of US$31bn.
  • COMEX managed-money net positions fell US$12bn, or 84 tonnes, and total spreading positions — offsetting long and short books across calendar months — dropped US$22bn, or 156 tonnes.
  • The dollar index rose 2% and the 10-year yield added 53 basis points to 5.3%, while the 10-year real yield went from 2.44% to 2.93%.

The WGC notes that a fall that large alongside ETF demand that strong is unprecedented in its data. Read it plainly: the September seller was the futures book, not the Western ETF holder. Real yields near 3% did the damage, but the accumulation bid never left — the useful part of the dataset for anyone trading real yields rather than headlines.

The European bid is stranger than the American one

North America led September's inflows; the surprise is the UK. UK-listed gold ETFs added 54 tonnes in Q3, with inflows in 12 of the 13 weeks to 25 September — the most consistent run since 2022. The WGC's model, built on the relationship with Western flows outside the UK, predicted 18 tonnes. That 36-tonne excess is unexplained by the old relationship, and the WGC points at the gilt term premium: investors buying insurance against the fiscal path rather than positioning for a cut.

Official-sector demand agrees. OMFIF's Global Public Investor 2026, drawn from 90 institutions with over US$10tn, found a net 30% of central banks planning to add gold, 51% citing geopolitical risk (up 11 points on 2024) and 61% expecting gold between five and six thousand dollars an ounce by June 2027. The WGC's own survey found a record 45% expecting their reserves to rise.

The counterweight: the IMF's July note Gold in Central Bank Reserves argues that gold's reserve share — 10% in January 2019 to over 22% — rose mostly through valuation, not accumulation, and belongs in a reserve book's investment tranche. Buying is lumpy too: 289 tonnes in Q2 made the first half the weakest since 2022, even as Poland reached 632 tonnes and the PBoC 2,346 tonnes. The structural bid is real, and it is not a straight line.

The selling has mostly already happened

How much futures selling is behind us matters as much as the ETF buying left. The CFTC report covering 6 October shows managed money net long 110,987 COMEX gold contracts, down 9,331 week on week — the short-side third of its range. The WGC's 13 October monitor put that net long at 357 tonnes and flagged a counter-current: ETF inflows have begun to decelerate, even as Chinese investors added longs after Golden Week.

The setup into the FOMC is therefore a market where the fast money has de-risked, the slow money is still accumulating, and price is decided by the bond market, not the Fed.

Where the levels sit

  • Resistance: $4,161-4,166 (Thursday's high zone and the 1H 200-EMA), $4,172 (the range high CPI broke), $4,181-4,188, then the 21-day at $4,216 and the 100-day at $4,248.
  • Support: $4,127, $4,107-4,112, $4,084-4,088, and Wednesday's $4,073 low. Underneath is the round number at $4,000, with July's lows at $3,996 and $3,959 below it.

Closing the week under both averages is why this is a repair inside a damaged structure, not a new uptrend: the averages did not fall, the price did. The session levels page tracks the liquidity map.

Scenario mapping

  • Base case — range, then the meeting. A light US calendar leaves gold trading levels and Fed pricing, with $4,050-4,200 the working box until the FOMC decides at 18:00 UTC on 28 October.
  • Bullish trigger: a 4-hour close back above $4,188 with the 10-year under 5.1% puts the $4,216 / $4,248 shelf in play and marks Wednesday's flush as repaired.
  • Bearish trigger: yields above 5.25% with the dollar re-bidding exposes $4,073 and then the $4,000 pivot — where TD Securities expects ETF and central bank flows to absorb supply.
  • Event risk: China's Q3 GDP on 19 October, the PBoC on 20 October, UK CPI on 21 October, flash PMIs on 23 October. Friday's Warsh IMF fireside chat is the last scheduled Fed airtime before the 17 October blackout, which is why Friday's plan gates around it.

What to take from it

  • Real yields, not CPI headlines, set the tone. September's 49bp jump in the 10-year real yield did more damage than any inflation surprise; when it stops rising, the headwind stops.
  • Follow the flow, not the price. ETF holdings rose through the worst month in a year, and the leveraged sellers now sit near the bottom third of their historical range.
  • $4,000 is the line that matters. Watch spreads and long-end yields into the 17 October blackout; a decisive break of the round number would test the accumulation thesis.

The 15 October CPI reaction and the PPI reaction trace the two-day move; the news board carries the rest of the macro coverage, and this flow data is filed under /topics/etfs.

Sources

Risk note: this is macro analysis, not investment advice, and no level here is a recommendation to buy or sell. Spot feeds disagree by $10-20 intraday, so treat every zone as a band and size so a stop costs a fixed fraction of the account.

Trading these levels? XM — MT5, gold, demo first paid link