When Foreign Official Treasury Holdings Fall, Three Separate Data Series Have to Agree First

A headline reporting that foreign central banks are "dumping" U.S. Treasuries almost always rests on one monthly number pulled from one table. The number is usually real. What it measures, how late it arrives, whether it describes reserve managers at all, and whether anything was actually sold are four separate questions — and they are answered by three different published series that frequently disagree with each other.

As of mid-August 2026, all three point in the same direction for the first time in a while. Official holdings are down on the Treasury's own survey, down on the Federal Reserve's weekly custody line, and down over a multi-year window. That convergence is worth taking seriously. It is also the exact circumstance in which the reasons matter far more than the direction, because at least four unrelated mechanisms produce an identical-looking decline.

Where the Numbers Actually Stand

Begin with what is published rather than what is asserted.

The Treasury International Capital system released June 2026 holdings on August 17, 2026. Total foreign holdings of U.S. Treasury securities stood at $9,299.0 billion. Of that, $3,778.1 billion was attributed to foreign official institutions — central banks, finance ministries, and sovereign entities that report as official. The residual, roughly $5,520.9 billion, sat with private foreign holders: asset managers, insurers, banks, and funds domiciled abroad.

A year earlier the same table read $9,093.6 billion total and $3,892.5 billion official.

Series (TIC)	June 2025	June 2026	Change
Total foreign holdings	$9,093.6bn	$9,299.0bn	+$205.4bn
Of which foreign official	$3,892.5bn	$3,778.1bn	−$114.4bn
Implied foreign private	$5,201.1bn	$5,520.9bn	+$319.8bn
Official share of total	42.8%	40.6%	−2.2pp

Two things happened simultaneously. Official holdings fell by $114.4 billion. Total foreign holdings rose by $205.4 billion. Private foreign holdings therefore rose by roughly $319.8 billion — enough to cover the official decline close to three times over. The official share of foreign-held Treasuries slipped from 42.8% to 40.6%.

Country detail sharpens the picture. China's reported holdings fell from $731.4 billion to $633.4 billion over the twelve months, a decline of $98.0 billion, or 13.4%. Japan's fell from $1,154.8 billion to $1,116.7 billion, a decline of $38.1 billion, or 3.3%. The United Kingdom — a custodial jurisdiction whose line item reflects where securities are held far more than who owns them — reported $939.9 billion in June 2026, with Belgium at $482.5 billion and Canada at $459.6 billion.

China's $98 billion decline is the figure that generates coverage. It is also less than half the increase in total foreign holdings across the same window. Both statements come from the same table, and only one of them typically survives into a headline.

Three Series, Three Different Questions

The reason careful analysts reach opposite conclusions from the same event is that "foreign central banks are selling" can be tested against three published series, each measuring something the others do not.

1. TIC holdings — comprehensive, slow, custodially blurred

The TIC survey is the broadest of the three. It covers essentially all foreign-held U.S. securities and splits official from private. It is also the slowest. June 2026 data reached the public on August 17, 2026 — roughly 48 days after the reference month closed, and nearly 80 days after it began. Any TIC-based claim about "current" official behavior is describing a quarter that has already ended.

The survey also carries an attribution problem Treasury itself flags: the data are collected from U.S.-based custodians and broker-dealers, so a security held through an overseas custody account may not be attributed to its actual owner. This is why the United Kingdom, Belgium, and Luxembourg carry balances that bear no relationship to their domestic reserve needs. A reserve manager that moves a portfolio from a New York custodian to a European one produces a decline in one country line and an increase in another, with no security changing hands.

2. Federal Reserve custody holdings — fast, narrow, valuation-sensitive

The Fed's weekly H.4.1 release carries a memorandum line for marketable U.S. Treasury securities held in custody for foreign official and international accounts. It publishes on Thursdays for the Wednesday prior — a one-day lag against TIC's 48 days.

That line stood at roughly $2.60 trillion on August 12, 2026. It was near $2,631 billion a week earlier and near $2,657 billion the week before that: a decline of about $59.8 billion across three weeks. Measured against the same week a year earlier, the series was down approximately $258 billion. Against its March 10, 2021 peak of $3,135.5 billion, it was down roughly $538.6 billion, or about 17.2%. It stood at $2,756.8 billion in the first week of January 2026, making the year-to-date decline about $160.0 billion.

This series is genuinely timely, and the three-week move is the kind of number that produces alarm. But it is also the narrowest of the three. Compare it against the TIC official figure: $3,778.1 billion of official holdings versus roughly $2.60 trillion in Fed custody implies that on the order of $1,181 billion of official Treasury holdings sit with custodians other than the New York Fed. Custody share is a choice, not a constant. A reserve manager consolidating with a commercial custodian generates a custody decline that is indistinguishable, in the weekly print, from an outright sale.

3. Currency composition — the only series that answers the reserve-status question

Neither of the first two says anything about the dollar's reserve status. Both are denominated in dollars and measure a single instrument class. A central bank that sells Treasuries and buys U.S. agency debt, U.S. equities, or dollar deposits has not reduced its dollar exposure at all — it has changed its Treasury exposure.

Currency composition belongs to the IMF's COFER database, which reports quarterly with roughly a quarter's lag and separates allocated from unallocated reserves. Two structural features of that dataset matter for interpretation. Participation is voluntary and a portion of global reserves remains unallocated, so the currency shares describe a reporting sample rather than the universe. And the shares are computed at market value in dollar terms, meaning a stronger dollar mechanically lifts the reported dollar share while a weaker dollar lowers it, with no reserve manager doing anything at all. A currency-share move that has not been decomposed into valuation and transaction effects is not evidence of a decision.

How an Official Sale Actually Reaches the Market

Assume for a moment that a genuine official reduction is underway. The path from that decision to a market price runs through four gates, and it is weaker at each one than the usual framing implies.

Gate 1 — sale or roll-off. The quietest way to reduce a Treasury portfolio is to let bills mature and decline to reinvest. No secondary-market transaction occurs, no dealer is hit, no price is printed. The holdings line falls anyway. Given the bill-heavy composition typical of reserve portfolios, a substantial share of any monthly decline can be pure roll-off, and the published tables do not separate the two.

Gate 2 — the buyer on the other side. The June 2026 data answer this directly. Over twelve months, private foreign holders added roughly $319.8 billion while official holders shed $114.4 billion. Foreign private demand did not merely offset official supply; it exceeded it by a wide margin. A sale only moves a yield if the marginal buyer demands a concession, and the composition data show a marginal buyer arriving without one.

Gate 3 — dealer balance sheet. Official sales are typically executed through primary dealers, who warehouse the position before redistributing it. The binding constraint is dealer inventory capacity, not the identity of the seller. In a period of ample capacity, a large official sale clears with minimal concession. In a stressed period, a much smaller one does not. This is why the same nominal size produces different outcomes at different moments, and why size alone is a poor predictor of market impact.

Gate 4 — the auction. Secondary-market sales and primary-market demand are linked but distinct. The observable evidence of foreign official retreat at auction is the indirect bidder share and the tail relative to the when-issued yield. Both are published within minutes of each auction, which makes them a far better real-time indicator than a survey published 48 days after the fact.

Historical Analogues Worth Separating

Two prior episodes illustrate how different the same headline can be underneath.

The first is the 2015–2016 Chinese reserve drawdown. Reserves fell sharply and reported Treasury holdings fell with them. The motive was exchange-rate defense: reserves were being converted into domestic currency purchases to slow depreciation pressure. Read as a statement about the dollar, that episode was backwards — the reserve manager was selling dollar assets precisely because dollar demand was overwhelming, and the sales were a symptom of dollar strength rather than a vote against it. A framework that treats every official sale as a confidence signal reads that episode exactly wrong.

The second is the post-2021 drift visible in the Fed custody series itself. The line peaked at $3,135.5 billion on March 10, 2021 and has fallen roughly 17.2% since, to about $2.60 trillion. That is a five-year, multi-hundred-billion decline — and it coincides with a period in which total foreign holdings of Treasuries rose. Both facts come from official series. The peak-to-present custody decline is real; so is the growth in aggregate foreign ownership. Any interpretation that can only accommodate one of them is incomplete.

The distinction that matters across both cases is between a flow decision made for cyclical reasons — currency defense, liquidity need, duration management — and a stock decision to hold a structurally different currency mix. The first shows up in months and frequently reverses. The second shows up in currency-composition data over years and does not. Nothing in a monthly holdings table distinguishes them; only the surrounding conditions do.

What This Analysis Cannot Settle

Several things in the preceding argument are genuinely unresolved, and treating them as settled would repeat the error this piece is trying to describe.

• The custody decline may partly be custodian migration, and the split is not published. The roughly $1,181 billion gap between TIC official holdings and Fed custody holdings establishes that alternatives to Fed custody are used at scale. It does not establish how much of the recent decline is migration versus sale. No public series decomposes this. Anyone claiming a precise split is estimating, not measuring.
• Valuation effects are not stripped out of either series. Both TIC holdings and the custody line are reported at market value. In a period of rising long-end yields, a portfolio that has not been touched declines in reported value. The published figures do not separate price from quantity, so some portion of any decline is arithmetic rather than behavioral.
• The private-buyer offset may not be independent. Private foreign holdings rising by $319.8 billion is reassuring only if that demand is unlevered and insensitive to hedging costs. If a material portion reflects hedged carry positions that depend on the cross-currency basis, the offset is conditional on funding conditions that can change within weeks. Holdings data cannot distinguish the two.
• Three weeks of convergence is not a trend. The custody move of about $59.8 billion since late July is large, but the series is volatile and has produced comparable swings that reversed. A short window supports a hypothesis; it does not confirm one.
• The reverse case is coherent and deserves equal weight. A serious argument runs the other way: that the multi-year custody decline, the fall in the official share of foreign holdings from 42.8% to 40.6%, and the shift toward private and potentially hedged holders together describe a Treasury investor base that is less price-insensitive than it used to be. Official reserve managers historically buy for policy reasons rather than return, which makes them stable holders through stress. Private holders reprice. Under that reading, the substitution that looks benign in the level data is a change in the quality of demand, and the market would discover it only in a stress episode. That case cannot be dismissed from the data above.

What to Watch Next Week

• The Thursday H.4.1 custody print. The relevant test is whether the three-week decline of roughly $59.8 billion extends or stabilizes. A single week's reversal would suggest the move was operational; a fourth consecutive decline of similar magnitude would shift the balance of probability toward sustained reduction.
• Indirect bidder share and auction tails. Published at each Treasury auction. If official retreat were reaching the primary market, indirect participation would soften and tails would widen against the when-issued yield. If indirect shares hold near recent averages while custody falls, the two facts together point at custodian mechanics rather than at demand.
• Cross-currency basis in yen and euro. A widening basis raises the hedged cost of holding Treasuries for exactly the foreign private buyers who absorbed $319.8 billion over the past year. This is the single most plausible channel through which the private offset weakens, and it is observable daily.
• The long end relative to the front end. Official reserve portfolios skew short. A reduction concentrated in bills should leave the long end largely unaffected. Long-end steepening alongside continued custody declines would be the more meaningful combination, and the one that would justify revisiting the benign reading.

Concrete Framework — What Comes First

Applied in order, each step either kills a claim or promotes it. Most claims die at step two.

1. Date the data before reading the direction. Identify the reference period and the publication date. A TIC-based statement published in August is describing June. If the claim is about conditions now and the source is TIC, it is 48 days stale before it is read.
2. Check the official-versus-total split, not the country line. A single country line can fall on custodian migration alone. The diagnostic pair is the foreign official total against the grand total. In the year to June 2026, official fell $114.4 billion while the grand total rose $205.4 billion — a composition change, not a withdrawal.
3. Ask what replaced it. Treasuries sold for dollar deposits, agency debt, or U.S. equities are not a reduction in dollar exposure. Only a currency-composition series answers this, and only with roughly a quarter's delay.
4. Separate flow motive from stock decision. Test whether the seller's currency was under depreciation pressure in the same window. If it was, treat the sale as exchange-rate defense until shown otherwise — the 2015–2016 precedent suggests this reading is more often correct than the confidence-signal reading.
5. Require a multi-year window before using the word structural. One month is noise. One quarter is a hypothesis. The custody series' 17.2% decline from its March 2021 peak qualifies as a trend worth naming; three weeks does not, however large the number attached to it.
6. Cross-check against a price published without a lag. If official behavior mattered to the market that week, indirect bid shares, auction tails, or the term premium would carry it. When the holdings narrative and the price evidence disagree, the price evidence is the one that did not arrive 48 days late.

The honest summary of the August 2026 data is narrower than either the alarmed or the dismissive reading. Official holders are reducing, on two independent series, across both a one-year and a five-year window. Private foreign holders have more than absorbed it. Whether that substitution is neutral or a quiet degradation in the resilience of Treasury demand is not answerable from holdings data alone — and saying so is more useful than picking a side the evidence does not yet support.

This article is analysis of publicly reported macroeconomic data and is not investment or financial advice.