Madhav Sethi
Gold · 15 min read

The Quiet Gold Divergence of 2025–2026: How the World's Largest Reserve Buyers Went Dark

Central banks officially reported ~16t of Q1 2026 gold buying — but really bought ~244t. Inside the off-screen bid repricing gold and the dollar.

By Madhav Sethi

The short version

  • There are now two gold markets in the data: the one central banks tell the IMF about, and the one that actually trades. In Q1 2026, central banks officially reported just ~16 tonnes of net buying — and reported ~115–129 tonnes of sales, led by Türkiye — yet the World Gold Council (WGC), using London OTC and Swiss refinery flows, estimated true official-sector buying at ~244 tonnes (roughly 15x the reported figure), up from ~208 tonnes in Q4 2025. The “cooling” was a reporting illusion, not a buyers’ strike.
  • China is the most likely largest hidden buyer. J.P. Morgan’s Gregory Shearer notes “Chinese net imports of gold have inflected higher, coming in at 317 tons in the first quarter of 2026, up by nearly three times compared to the previous quarter,” even as the People’s Bank of China admitted to only single-digit tonnes per month (“from around a one-ton-per-month pace for the six months through February to five tons in March and eight tons in April”). The share of global official buying actually reported to the IMF has collapsed from ~90% four years ago to roughly one-third — the IMF’s data is voluntary, so the blind spot is structural, not accidental.
  • The divergence signals a structural regime change. Gold has decoupled from real interest rates; the ECB’s June 2026 report says gold (27% of reserves) has overtaken US Treasuries (22%) as the world’s largest reserve asset by market value; and forecasters (Goldman Sachs $5,400, J.P. Morgan $6,300 by end-2026) attribute the price floor to a price-insensitive “conviction buyer” operating largely off-screen.

What stands out

  1. The headline “central banks are selling” narrative is misleading. The gap between reported flows (near-zero or negative in early 2026) and estimated flows (~244t in Q1 2026) is the core of the “quiet divergence.” Reported sales were driven by distressed, idiosyncratic sellers (Türkiye, Russia); the structural buyers kept buying but increasingly did not report.

  2. Detection is mechanism-based, not speculative. Analysts infer hidden buying three ways: (a) the residual between total estimated demand and visible demand (WGC/Metals Focus); (b) physical-flow tracking via Swiss refinery exports and UK/London large-bar exports (the bar size central banks favour); and (c) for China, “gap” accounting (net imports + domestic mine output − commercial/retail demand). Each method carries real uncertainty.

  3. China’s true holdings are genuinely unknown and contested. Official PBoC reserves are ~2,313t (~9% of reserves). Credible analysts estimate true holdings near ~5,000t; some go far higher. Société Générale estimated up to ~250t of Chinese official buying in 2025 vs. the ~24t officially reported — a roughly 10x gap.

  4. Motive is strategic, anchored to 2022. The freezing of ~$300bn of Russian central-bank reserves demonstrated that dollar assets held offshore can be frozen. Gold has no counterparty and cannot be frozen — and buying quietly avoids both moving the market and signalling intent.

  5. Price formation has changed. Gold’s 25-year inverse relationship with real yields broke down in 2024–2025: gold tripled even as real yields stayed positive. A large, price-insensitive official bid is now the marginal price-setter, supported by record ETF inflows and a structural fiscal-risk premium.

The fuller picture

1. What is happening: defining the divergence

For two decades, central-bank gold data was boring and roughly complete. That is no longer true. By early 2026 a wide and persistent gap had opened between two numbers:

  • Reported flows. Central banks report gold holdings to the IMF (through its International Financial Statistics, or IFS, and its COFER currency database) on a voluntary basis. On this basis, Q1 2026 looked like a stall: net reported purchases of only ~16 tonnes, with ~115–129 tonnes of reported sales. Türkiye alone sold or swapped roughly 60–70 tonnes (mostly in March), and Russia was a net seller. Taken at face value, the multi-year buying spree appeared to be ending.

  • Estimated true flows. The WGC, using its data partner Metals Focus, estimates actual official-sector demand by adding “unreported buying” to the reported figure. Drawing on London OTC market data and Swiss refinery flows, the WGC’s Gold Demand Trends Q1 2026 put true central-bank demand at ~244 tonnes — higher than Q4 2025’s ~208 tonnes, above the five-year quarterly average, and roughly 15 times the ~16t official figure. The buying did not stop; the reporting did.

The single most telling datapoint comes from J.P. Morgan’s head of base and precious metals, Gregory Shearer: Chinese net gold imports inflected sharply higher to ~317 tonnes in Q1 2026, “up by nearly three times compared to the previous quarter” — even as the PBoC reported adding only “around a one-ton-per-month pace for the six months through February to five tons in March and eight tons in April.” The metal is clearly flowing into China; very little of it appears promptly on the official balance sheet.

For context on scale: per the WGC’s Gold Demand Trends Full Year 2025, “Q4 buying helped lift full-year demand to a sizeable 863t,” with unreported buying representing “57% of the annual total” — down from a revised ~1,086 tonnes in 2024 and the record ~1,136t in 2022. So the pace genuinely eased as prices tripled, but conviction did not. The PBoC added “just 3t in Q4, the lowest quarterly reported increase since Q1’24,” for a full-year reported 27t and holdings of 2,306t (~9%). Net reported sales in Q1 2026 were also distorted by Türkiye using ~80 tonnes of gold in swaps for FX/liquidity during the Iran-conflict energy shock — a liquidity operation, not a strategic exit.

Gold’s price path frames all of this: gold broke $5,000 for the first time on 26 January 2026 and hit an all-time high of ~$5,590–5,597 on 28–29 January, then corrected hard, trading around $4,300–4,340 by mid-June 2026 (still up roughly 30% year-on-year). The “divergence” matters precisely because the off-screen buyer is widely credited with putting a floor under that volatile price.

2. How can we tell? The detection methodology

There is no satellite for gold (unlike oil). Detection rests on inference from physical flows and accounting residuals.

(a) The implied/residual method (WGC + Metals Focus). Gold that is mined or recycled must exit through identifiable demand channels — jewellery, technology, bar-and-coin, ETFs, and reported official purchases. Whatever is left over is the residual. The WGC’s verbatim definition: “Unreported buying is the residual between estimated demand and reported purchases.” Its broader “OTC and other” line “captures demand in the OTC market (for which data is not readily available), changes to inventories on commodity exchanges, any unobserved changes in fabrication inventories and any statistical residual. It is the difference between total supply and gold demand.” Limitation: a residual is a remainder, not a measurement — it absorbs all statistical error and can conflate sovereign buying with private OTC hoarding and sovereign-wealth-fund activity.

(b) Swiss refinery and London/UK flow tracking. Switzerland is the world’s largest bullion refining and transit hub; its monthly customs data shows where physical gold is going in near-real-time. Swiss exports to China rebounded to ~58 tonnes in February 2026 (from ~26t in January) and rose a further ~18% in March. Because the large (400-oz London Good Delivery) bars favoured by central banks are mainly traded in London, UK gold exports to China are used as a direct proxy for official buying. Limitation: trade flows capture all demand (jewellery, private investment, commercial banks), not just central banks, so attribution requires judgement.

(c) London OTC market. The over-the-counter (off-exchange, directly between two parties) London market is where the largest, least-visible wholesale transactions clear. The WGC uses OTC-derived data to triangulate demand that never appears on an exchange tape.

(d) China-specific “gap” accounting. Beijing’s gold buying is managed through the State Administration of Foreign Exchange (SAFE, part of the PBoC), and purchases can also be made by the sovereign wealth fund (CIC) and the military, none of which must disclose on a timely basis. China is also the world’s largest gold miner (~10% of global output), so it can add to reserves domestically without any import showing up. Analysts therefore compute: net imports + domestic mine production − (commercial-bank holdings + retail/consumer demand) = implied official buying. Plenum Research (a Beijing consultancy) used this to estimate a “gap” of ~1,351 tonnes in 2023 and ~1,382 tonnes in 2022 — more than six times China’s publicly reported purchases in those years. Separately, Société Générale’s Michael Haigh uses UK-to-China large-bar export data to estimate SAFE imported “up to 250 tonnes annually, potentially over a third of global demand.” Shanghai Gold Exchange (SGE) withdrawal data (134t in March 2026) and Shanghai Futures Exchange volumes are tracked as demand barometers. Limitation: every input is itself an estimate, and the method cannot distinguish PBoC from other state entities. As Haigh notes, gold’s opacity makes the market “unique and tricky” because “the tonnage going in and out of central banks is so impactful. Without having clarity on that, it is a bit more of an issue.”

(e) Why the IMF blind spot exists. Reporting to the IMF is voluntary; there is “no mandatory rule on reporting purchases to the IMF,” per the WGC. This is the structural root of the divergence. The share of official buying actually reported has collapsed — by WGC estimates based on Metals Focus data (reported by the Financial Times in November 2025), “only about one-third of official buying was publicly reported, down from around 90 per cent four years ago.” Central banks may stay quiet to avoid front-running their own purchases (the classic cautionary tale is the UK’s pre-announced 1999 gold sale, which depressed prices to ~$275/oz) or for political reasons — including not wanting to telegraph a dollar exit. As MKS Pamp’s Nicky Shiels put it, “in most emerging markets, it is in central banks’ interest to not fully disclose purchases.”

3. Why now? The drivers

  • Sanctions risk / weaponization of reserves. The 2022 freezing of ~$300bn of Russian central-bank reserves is the universally cited inflection point. Gold held in a domestic vault has no counterparty and cannot be frozen by a foreign government. Goldman Sachs frames this as a “structural shift in reserve management behavior” it does not expect to reverse near-term.
  • Dollar diversification. Emerging-market central banks remain structurally underweight gold versus developed peers (China ~9% of reserves vs. ~70% for the US, Germany, France, Italy), giving years of runway. Domestic purchase programmes (used by roughly half of buyers, per WGC survey work) let a country build a gold stockpile in its own currency without ever touching FX markets — a deliberate, structural mode of accumulation rather than an opportunistic trade.
  • Geopolitical conflict premium. The 2026 US–Israel–Iran conflict, the ongoing Russia–Ukraine war, and broader fragmentation have entrenched a risk premium that Goldman now treats as “sticky” — persistent rather than self-reversing.
  • Fiscal and debasement fears. G7 long-bond yields hit multi-decade highs in 2026 (US 30-year above 5%, UK 30-year gilt highest since the 1990s, Japan’s 30-year at record highs), reflecting doubts about sovereign-debt sustainability — making a no-counterparty asset attractive.
  • Strategic stealth. Buying quietly avoids moving a relatively small market and avoids signalling a dollar exit that could be politically costly.
  • De-dollarization politics / BRICS. BRICS+ has piloted a gold-anchored settlement “Unit” (a 40%-gold, 60%-currency basket) and is building non-dollar payment rails — though serious analysts judge de-dollarization will remain partial and gradual.

4. What does it mean? Interpretation

  • A neutral reserve asset is being rehabilitated. The ECB’s “International role of the euro” report (published 2 June 2026) found that “gold accounted for 27% of global central bank reserve assets at the end of 2025… up from 20% a year earlier,” while “US Treasuries fell to 22% from 25%,” with the euro steady at 15% — gold overtaking Treasuries for the first time since 1996. ECB President Christine Lagarde wrote: “Geopolitical tensions continue to drive strong central bank demand for gold.” Crucially, the ECB stressed this was driven mainly by price (gold rose ~60% in 2025): at end-2023 prices, Treasuries would still lead (~26%) with gold at ~16%. So gold’s “overtaking” is real but valuation-inflated.
  • Conviction vs. opportunistic buyers (Goldman’s framework). Goldman Sachs splits the market into conviction buyers (central banks, ETFs, speculators) who buy on a macro/hedging view regardless of price, and opportunistic buyers (EM households) who step in on dips. Its rule of thumb: every 100 tonnes of net conviction-buyer demand corresponds to ~1.7–2% price appreciation. Tellingly, Goldman discovered in 2026 that its own model had under-counted central-bank demand by more than 70% for ~8 months — confirming the off-screen buyer was even bigger than measured.
  • Price-discovery distortion. When a large, price-insensitive buyer operates off-screen, public flow data understates true demand and the market’s “fair value” signal is muddied — exactly what produced the early-2026 “central banks are selling” scare even as buying rose.
  • The real-yield decoupling. Gold’s ~25-year inverse correlation with real yields (R² ~84% in 2005–2021, per RBC) collapsed to single digits in 2022–2025; gold tripled despite positive real yields. The marginal price-setter shifted from rate-sensitive Western ETF investors to price-insensitive central banks plus a fiscal-risk hedge. (PIMCO and J.P. Morgan’s private bank argue the relationship is dormant, not dead, and may reassert if the Fed cuts.)

5. Future outlook (1–5 years)

  • Forecasts (note these are projections, not facts): J.P. Morgan raised its year-end 2026 target to $6,300/oz on 2 February 2026 (projecting central-bank/investor demand averaging ~585 tonnes per quarter), with ~$6,000 in Q4 2026. Goldman Sachs analysts Daan Struyven and Lina Thomas raised the end-2026 target from $4,900 to $5,400 in January 2026 and held it through the March sell-off, with an upside of $5,700–6,100 under accelerated de-dollarization. UBS raised its target to ~$6,200 (upside ~$7,200); Deutsche Bank reiterated $6,000. The wide dispersion reflects genuine disagreement about whether the central-bank behavioural shift is permanent.
  • Will hidden buying continue? The WGC’s 2026 Central Bank Gold Reserves Survey (conducted 5 February–19 May 2026 with YouGov; 76 respondents, a record) found 89% expect global central-bank gold reserves to rise over the next 12 months and a record 45% expect their own institution to add gold (up from 43% in 2025). As the WGC’s Shaokai Fan put it: “Central banks are still very positive on gold. In fact, more positive than ever.” Goldman assumes the accumulation trend runs ~3 more years.
  • Scenarios:
    • Base: Official buying continues at ~700–900t/yr (UBS ~800–850t; WGC ~850t; Goldman ~720t at ~60t/month); price grinds higher with high volatility.
    • Bull: Western institutions (pensions/insurers, currently <1% allocated) and EM private investors broaden the conviction bid; de-dollarization accelerates; price overshoots forecasts.
    • Bear: A durable geopolitical de-escalation plus a hawkish Fed lifts real yields, triggers ETF outflows, and — if more EM central banks follow Türkiye/Russia into forced selling — removes the demand floor.
  • What would reverse it? Credible, durable resolution of the Iran conflict; a serious US fiscal-consolidation plan; sustained higher real yields; or broad EM currency crises forcing reserve liquidation.

6. Currency, macro and geopolitical reshaping

  • Dollar reserve status. Per the IMF’s COFER release (27 March 2026), “the share of US dollar holdings decreased to 56.77 percent in 2025Q4, from 56.93 percent in 2025Q3” (euro 20.25%, renminbi 1.95%; total reserves $13.14 trillion) — down from ~72% in 2001. But the IMF and Fed both stress that much of the recent decline is valuation (FX moves), not active dollar-dumping; the dollar remains dominant (~89% of FX transactions per the 2025 BIS survey). Gold’s rise has come largely at the expense of Treasuries, not the euro (whose reserve share held at ~15%).
  • Treasury demand and sovereign debt. With G7 long yields at multi-decade highs and term premia (the extra yield investors demand for holding long-dated bonds) rising, the marginal reserve dollar is increasingly going into gold rather than long Treasuries — a feedback loop that pressures government funding costs.
  • Multipolarity. The system is becoming “less static” rather than de-dollarizing outright — a gradual repositioning driven, in OMFIF’s framing, by “a structural loss of trust,” with gold as the neutral fallback because no rival currency offers comparable depth.
  • EM central-bank behaviour. The buyer base is broadening (Poland targeting 700t and the largest 2026 buyer; Uzbekistan at ~87% gold; new entrants like Bank Negara Malaysia and the Bank of Korea), entrenching the structural bid.
  • Monetary-policy independence and inflation credibility. Holding a no-counterparty asset insulates reserves from foreign policy, but a price-driven gold reserve is volatile and yields nothing — the ECB itself flags gold as an “awkward” reserve asset.

What to watch

The benchmarks worth tracking, and what each would signal:

  1. The reported-vs-estimated gap matters more than the headline. When the story is “central banks are selling,” the tell is the WGC/Metals Focus estimate set against the IMF-reported figure. A widening gap = stealth buying continuing; a narrowing gap with falling estimates = a genuine slowdown. Threshold: if WGC estimated quarterly demand falls below ~150t for two consecutive quarters, the structural bid is genuinely weakening.
  2. China import data is the leading indicator. Chinese net imports and UK→China large-bar exports lead the official PBoC disclosures by months. A sustained drop in Chinese net imports below ~100t/quarter would be an early warning that the largest hidden buyer is pausing.
  3. The real-yield relationship is the regime signal. If gold begins falling when real yields rise (i.e., the old inverse correlation reasserts), it signals the conviction buyer has stepped back and rate-sensitive money is again marginal — a bearish structural shift.
  4. “China holds 5,000+ tonnes” is a credible estimate, not a fact. Worth distinguishing hard data (PBoC: ~2,313t reported), well-grounded estimates (SocGen ~250t/yr buying; Plenum’s gap method), and speculation (figures of 8,000–30,000t).
  5. The reversal triggers: durable Iran-conflict resolution, a US fiscal plan, a hawkish Fed lifting real yields, and any new EM central bank joining Türkiye/Russia in forced selling. Any two occurring together would materially raise the odds of the bear scenario.

Caveats

  • The residual is a remainder, not a measurement. “Unreported buying” absorbs all statistical error plus private OTC and sovereign-wealth-fund activity; it should not be read as a precise central-bank purchase figure. Both the ECB and Brookings-affiliated economists (Arslanalp, Eichengreen, and Simpson-Bell, 2025) note part of the IMF-vs-WGC gap is methodological (SWF inclusion), not pure secrecy.
  • China’s true holdings are genuinely uncertain. Estimates range from the official ~2,313t to ~5,000t (credible) to 8,000t+ (speculative). The ambiguity is partly deliberate state policy.
  • The ECB “gold overtakes Treasuries” milestone is valuation-driven. At constant 2023 prices, Treasuries still lead. This is a price effect as much as a reallocation.
  • Single-source dependence. Several China-specific figures (the ~one-third reporting share; Plenum’s 1,351t; SocGen’s 250t) trace substantially to one Financial Times investigation (November 2025) and its syndications. They are consistent across reproductions but not independently multi-sourced.
  • Forecasts are not data. All price targets ($5,400–$6,300+) are projections contingent on Fed policy and geopolitics, both uncertain. Sources with commercial interests in gold (bullion dealers, miners) tend to emphasise the bull case; the figures here lean on the WGC, IMF, ECB, OMFIF, Goldman Sachs, J.P. Morgan, UBS, and the FT/Reuters/Bloomberg over promotional ones.
  • Data lags and revisions. IMF IFS data are ~two months in arrears, and the WGC explicitly warns the current volatile environment increases the chance of future revisions.