COFER's 1.47 Point Dollar Slide: 0.12 Points Survive at Constant Exchange Rates

Official reserve managers hold about $13.10 trillion of foreign exchange, and 57.13 percent of it sits in US dollars. That second number gets read as a verdict. It is not a verdict. It is a ratio, and a ratio moves when either half of it moves, whether or not a single reserve manager decides anything.

The International Monetary Fund publishes that ratio quarterly through COFER, the Currency Composition of Official Foreign Exchange Reserves. Across the last five quarters, the dollar share as first published ran 57.79, then 56.32, then 56.92, then 56.77, then 57.13. Read as a sequence of decisions, that is a violent story: a collapse, a partial recovery, a stall, a snap back. Read as arithmetic, most of it never happened.

The IMF itself has said so, in plain language, in its own release notes. For the second quarter of 2025, when the published dollar share fell 1.47 percentage points, the Fund wrote that the share "would have decreased by only 0.12 percentage points (p.p.) if the exchange rates had not moved." The headline move was roughly twelve times the move that survived holding exchange rates fixed.

The Published Share Has Four Moving Parts and One of Them Is a Decision

COFER is built on a mechanical convention that is stated openly and then routinely forgotten. Reserve holdings are "reported to the IMF in U.S. dollars, converted using end-of-period market exchange rates." Every non-dollar asset in the dataset is therefore carried at its dollar value on the last business day of the quarter.

That convention creates two ways for the ratio to move without any reserve manager transacting, a third that is an actual portfolio decision, and a fourth that is not even a market event.

The first is exchange rate valuation. If the dollar appreciates during a quarter, the dollar value of every euro bund and every Japanese government bond in the world's reserve pool falls. The numerator is unchanged, the denominator shrinks, and the dollar share rises. Nobody bought a Treasury.

The second is non-exchange-rate valuation. Reserve assets are securities, and securities reprice. The IMF flagged exactly this for the most recent quarter, noting that for the yen and the Swiss franc, exchange rate effects "were less central" and pointing to "outsized movements in benchmark bond yields" instead. A bond selloff in one reserve currency lowers that currency's share the same way a depreciation would.

The third is active management, which is the only channel that represents a choice. It is also the channel that is never reported separately.

The fourth is definitional, and it is the one that caught most readers out.

Second Quarter 2025 Is the Cleanest Test Case on Record

The quarter that produced the loudest de-dollarization commentary is also the quarter for which the IMF supplied the counterfactual. The published dollar share fell from 57.79 percent to 56.32 percent. The Fund's own constant-exchange-rate calculation put the decline at 0.12 percentage points. The remaining 1.35 points was the dollar falling against the currencies it was being measured against.

The same release applied the logic symmetrically, which is the part worth noticing. The euro's published share rose 1.13 percentage points that quarter, but the Fund noted that "if the exchange rates had not moved, the euro share would have decreased by 0.04 p.p." The euro did not gain reserve ground. The euro got more expensive in dollars, and COFER prices everything in dollars.

The renminbi went the other way: a flat published share concealed an underlying gain of 0.03 percentage points at constant exchange rates. Three currencies, three published moves, three constant-rate moves with different signs. The published column and the decision column were not measuring the same thing.

The Denominator Was Redefined in 2025Q3 and Backdated Twenty-Five Years

Until late 2025, COFER split the world's reserves into "allocated" holdings, where the reporting country identified the currency, and "unallocated" holdings, which were the residual left after the reported currency detail. The famous dollar share was a share of allocated reserves only. It excluded, by construction, every reserve manager who declined to break out currencies.

That changed. The IMF states that "Starting in 2025Q3, with revisions back to 2000Q1, the IMF eliminated the 'unallocated' portion of the COFER dataset to provide a complete currency composition—expressed in both dollars and shares—that accounts for 100 percent of the world's foreign exchange reserves." The unallocated pool is now imputed across currencies using, per the Fund's technical note, "stratified mean imputation and carry-forward imputation."

The size of that redefinition is measurable, because one quarter was published under both conventions. The second quarter of 2025 printed at 56.32 percent on the allocated basis in the October 2025 release. The same quarter appears at 57.08 percent in the December 2025 release, on the new basis. The denominator change alone moved the number 0.76 percentage points, which is larger than any genuine quarter-over-quarter move in the series since.

Anyone comparing a remembered figure from 2023 against today's print is comparing two different statistics unless they re-pulled the whole series. This is the same failure mode as choosing a denominator for strategic stockpiles, where the same barrel count reads as 65 days of cover or 106 depending on which draw rate sits underneath it. The reserve number did not change. The divisor did.

Vintage Revisions Are Larger Than the Quarterly Signal

There is a second, quieter problem. COFER quarters get revised, and the revisions have recently been bigger than the quarter-over-quarter changes they sit next to.

The fourth quarter of 2025 first printed at 56.77 percent in the March 2026 release. By the July 2026 release it read 56.42 percent. That is a 0.35 percentage point revision to a single quarter, and it arrived alongside a first print for the next quarter of 57.13 percent. The 0.71 point "jump" into the first quarter of 2026 is therefore measured off a base that had itself just moved a third of a point in the opposite direction. Against the originally published 56.77, the jump was 0.36 points, not 0.71.

Total reserves moved too. 2025Q3 printed at $13.0 trillion in December and $13.03 trillion in March; 2025Q4 printed at $13.14 trillion in March and $13.15 trillion in July. Small in percentage terms, but that is the denominator.

The pattern of a headline level that shifts between vintages while the underlying path stays intact should be familiar to anyone who has tracked a monthly bank credit aggregate that grew 9.7 percent while the subsequent weekly prints went the other way. A single number pulled from a single vintage is not a trend.

What the First Quarter of 2026 Contains

The most recent print is worth working through, because the IMF gave a partial attribution and the raw exchange rate data is public.

Published: the dollar share rose to 57.13 percent from 56.42 percent, a gain of 0.71 percentage points, with the Fund attributing "exchange-rate-driven valuation effects accounting for around half of the increase in its share."

Now the exchange rates. Using the Federal Reserve's H.10 quarter-end quotations, between 31 December 2025 and 31 March 2026 the euro fell from $1.1736 to $1.1518, a 1.86 percent decline in dollar terms. The yen went from 156.80 to 159.08 per dollar, a 1.43 percent decline in dollar value. Sterling fell 1.96 percent, the Canadian dollar 1.73 percent, the Swiss franc 1.13 percent. The Fed's broad dollar index rose from 120.1166 to 121.0350.

The arithmetic of the ratio follows directly. Hold dollar holdings fixed at 56.42 percent of the pool and shrink everything else by 1.43 percent in dollar terms, and the dollar share lands at 56.77 percent. That is a gain of 0.35 percentage points from pure repricing, which is almost exactly the half the IMF attributes to exchange rates. To generate the full 0.71 point move on valuation alone, the non-dollar basket would have had to lose 2.85 percent, which is well beyond what the quarter's currency moves deliver.

The cross-section confirms it. Two reserve currencies appreciated against the dollar over the quarter: the Australian dollar, up 2.79 percent, and the renminbi, up 1.38 percent. Both gained share, by 0.09 and 0.04 percentage points respectively. The two biggest depreciators among the large holdings, the euro and the yen, lost the most share, 0.35 and 0.40 points. The ranking of share changes broadly tracks the ranking of currency moves, with one conspicuous exception: sterling fell the furthest of any reserve currency, 1.96 percent, and lost 0.01 points of share.

Where it does not track is informative. The yen lost more share than its 1.43 percent depreciation alone implies, and the Swiss franc gained share despite depreciating 1.13 percent. Those are precisely the two currencies the IMF singled out as driven by "outsized movements in benchmark bond yields" rather than exchange rates. The residual is doing what the Fund said it would do. Sterling is a third misfit the Fund did not comment on, and a 0.01 point share change against a 1.96 percent depreciation has no published explanation.

In levels, dollar holdings went from roughly $7.42 trillion to $7.48 trillion, up about $65 billion, while non-dollar holdings fell about $115 billion and the total shrank $50 billion. The dollar share rose partly because the dollar pile grew and mostly because the rest of the pile was marked down.

Where This Doesn't Apply

Three limits deserve stating, because the argument above can be overextended.

Valuation effects do not net to zero over long horizons in the way they do within a quarter. If the dollar trends in one direction for years, the cumulative valuation drag on the published share is real and persistent, and reserve managers who rebalance to currency targets will transact to offset it. Over a decade, the distinction between valuation and flow blurs because the flows are responses to the valuations.

Constant-exchange-rate figures are not neutral either. They require choosing a base date, and the choice matters. The IMF's quarterly counterfactual uses the prior quarter-end, which is a defensible convention and not the only one.

The imputation of previously unallocated reserves is an estimate, not a measurement. The Fund's own note concedes that "the allocation of previously unallocated reserves leads to modest adjustments in currency shares." Reporting is voluntary, and the non-reporters are unlikely to hold a currency mix drawn from the stratified mean of those who do report. The new series is more complete and less directly observed at the same time.

Finally, COFER does not measure dollar dominance in any operational sense. It measures the currency of denomination of one pool of official assets. Invoicing currency, trade settlement, cross-border bank claims and derivatives markets are separate datasets with separate definitions, and a reserve share is not a proxy for any of them.

What to Watch Next Week

The next COFER release covering the second quarter of 2026 is the obvious marker, and the specific things to read in it are narrow.

Check the restatement of 2025Q4 and 2026Q1 before reading the new quarter. If 2026Q1's 57.13 percent moves by more than about a quarter of a point, the first print of 2026Q2 should be treated as provisional regardless of its direction.

Check whether the Fund again supplies a constant-exchange-rate counterfactual. It did so explicitly for 2025Q2, gave a qualitative attribution for 2025Q3 and 2025Q4, and gave a proportional attribution for 2026Q1. The precision of that note varies by release, and a quarter without one should be read with correspondingly less confidence.

Run the H.10 quarter-end comparison independently. The broad dollar index closed the first quarter at 121.0350 and stood at 118.0732 on 4 September 2026, a 2.45 percent dollar depreciation spread across two quarters rather than confined to either. Pull the 30 June 2026 quarter-end quotation from H.10 and split that move between the second and third quarters before reading the new share. On the fixed-numerator test, a full 2.45 percent depreciation applied to the non-dollar basket would drag the published share down about 0.59 points on repricing alone, and half of it would cost about 0.30 points. A published decline inside that band carries no allocation information.

Watch the gap between share changes and currency changes in the cross-section. Currencies that move in share against the direction of their exchange rate are where the non-valuation story lives.

Concrete Framework

A repeatable reading order for any COFER print, or for any published share statistic built on a currency-converted denominator.

1. Identify the vintage. Note the release date of the figure being quoted and whether it was produced under the allocated-reserves convention or the 100-percent-coverage convention adopted in 2025Q3. A pre-2025Q3 figure quoted from memory is not comparable to a current print.
2. Re-pull the prior quarter from the same release. Do not difference a new print against a remembered old one. The 2025Q4 example shows a 0.35 point revision that would have doubled an apparent quarterly move.
3. Compute the quarter-end currency moves first. Pull H.10 or an equivalent primary exchange rate source for the two quarter-end dates before reading the share table. Establish what the ratio should have done on repricing alone.
4. Apply the fixed-numerator test. Take the prior share, shrink the complement by the weighted currency move, and see how much of the published change that accounts for. Anything left over is the candidate signal.
5. Read the Fund's attribution language literally. "Around half," "muted," "modest" and "nearly all" are the Fund's own decomposition, and they are more reliable than any outside reconstruction because the Fund sees the confidential country detail.
6. Check the cross-section for inconsistency. Rank currencies by exchange rate move and by share move. Currencies out of order are where active management or non-FX repricing is hiding.
7. Separate the ratio from the level. Dollar holdings rose about $65 billion in 2026Q1 while the dollar share rose 0.71 points. Those are different facts with different implications, and the same distinction between a gross line and a net one governs whether customs refunds running ahead of collections reads as a shrinking tariff take or an expanding one.
8. Set a materiality floor. Given observed revisions of up to 0.35 points and valuation effects that ran 0.35 points in 2026Q1 and 1.35 points in 2025Q2, a single-quarter share move below roughly half a point carries no allocation information. Require either a multi-quarter direction or an explicit constant-rate figure before calling it a trend.

The dollar's reserve share is a genuinely useful series. It is simply not a scoreboard, and every quarter that gets read as one produces a conclusion the underlying data does not support.

Sources

• International Monetary Fund, IMF Data Brief: Currency Composition of Official Foreign Exchange Reserves, 1 July 2026 (2026Q1). data.imf.org
• International Monetary Fund, IMF Data Brief: Currency Composition of Official Foreign Exchange Reserves, 27 March 2026 (2025Q4). data.imf.org
• International Monetary Fund, IMF Data Brief: Currency Composition of Official Foreign Exchange Reserves, 19 December 2025 (2025Q3). data.imf.org
• International Monetary Fund, IMF Data Brief: Currency Composition of Official Foreign Exchange Reserves, 1 October 2025 (2025Q2). data.imf.org
• International Monetary Fund, COFER Frequently Asked Questions, dataset IMF.STA:COFER. data.imf.org
• International Monetary Fund, Improving the Analytical Usefulness of the IMF's COFER Data, Technical Notes and Manuals 2025/014, 26 November 2025. imf.org
• Board of Governors of the Federal Reserve System, Foreign Exchange Rates H.10, releases of 5 January 2026, 6 April 2026 and 8 September 2026. federalreserve.gov