How to Read Central Bank FX Reserves (COFER and the gold share)

A fall in the dollar’s share of reserves can make a powerful headline. It does not, by itself, show that central banks sold dollars. Exchange rates can change the value of the other currencies in the denominator, and a methodological revision can change the series being compared. COFER is useful precisely when those distinctions survive the headline.

Open the IMF’s COFER dataset alongside this guide. Keep the observation quarter and release date separate. The objective is to understand what changed in reserve portfolios and what the public data leave unresolved.

Define the reserve measure first

COFER stands for Currency Composition of Official Foreign Exchange Reserves. It groups official foreign exchange holdings by currency and reports the aggregates in US dollars. Monetary gold, special drawing rights and the reserve position in the IMF are outside this currency breakdown. A gold share of total reserves consequently uses a different denominator. The dataset description establishes this scope.

“Dollar reserves” also covers more than US Treasury securities. Currency denomination and instrument type answer different questions. A comparison with Treasury holdings therefore needs a separately defined securities dataset. It cannot be reconstructed by renaming COFER’s dollar column.

The 2025 methodology change matters

Beginning with 2025 Q3, the IMF eliminated the published “unallocated” portion by estimating the missing currency composition. It revised the history back to 2000 Q1 and began publishing the share of reserves that had been imputed. These are the methodological changes documented on the COFER dataset page.

The resulting currency shares cover the full foreign exchange reserve aggregate. That does not mean every currency holding was directly reported. The IMF’s November 2025 technical note explains the estimation methods and confidentiality constraints.

For a historical comparison, download the whole series from the same release vintage. Do not join an old chart of allocated reserves to a new estimate covering all FX reserves without explaining the break. An old publication can still be cited accurately as the estimate available then; it should not silently replace the revised historical series.

A worked example of the exchange-rate effect

The IMF’s October 2, 2025 data brief reported that the dollar share of allocated reserves fell from 57.79% to 56.32% between the first and second quarters of 2025. That is a decline of 1.47 percentage points. Holding exchange rates unchanged, the IMF estimated a decline of only 0.12 percentage points.

The mechanism is straightforward. If a portfolio holds the same quantity of euros but the euro rises against the dollar, those euros occupy a larger share of a portfolio measured in dollars. The dollar share can shrink while the dollar holdings are unchanged. The IMF technical note discusses this valuation problem.

Removing exchange-rate effects narrows the interpretation. It still does not expose every transaction or a policymaker’s motive. Securities valuations, income, changes in coverage and revisions may matter; aggregate data do not supply a trade-by-trade record. Describe an exchange-rate-adjusted change as such, rather than calling the entire residual “dollar selling.”

Keep gold on a separate page of the calculation

A central bank’s gold holdings have both a quantity and a market value. A higher value can reflect a higher gold price, more metal, or both. Use reported quantities to examine accumulation, and value to assess the weight of gold within a clearly defined reserve portfolio. Neither measure substitutes for the other.

Because gold sits outside COFER, a rise in its share of total reserves cannot be read directly from COFER’s dollar share. Claims that gold “overtook” an asset also need the same valuation date and a specified asset perimeter. We examine those questions separately in central banks, gold and de-dollarisation.

What the dataset cannot tell you

COFER publishes aggregate currency composition while protecting individual reporters’ information. A country need not disclose its currency breakdown publicly for its confidential data to contribute to an aggregate. Do not infer non-participation merely because a national breakdown is unavailable. The IMF methodology note describes these disclosure limits.

A global currency-share movement also cannot reveal which particular country made a transaction. And a quarterly reserve stock does not describe trade invoicing, cross-border bank lending or payment-message traffic. Those are separate dimensions of international currency use, with separate evidence. Our analysis of de-dollarisation claims provides the wider reading path.

Build a comparison that can be checked

Record the quarter, release vintage, currency, unit and denominator before calculating the change. Check the imputed share and any revision notice. Show exchange-rate-adjusted results only when their method and inputs are available. Keep gold quantities, gold values and FX reserve shares in distinct series. If the evidence cannot separate valuation from transactions, leave that separation unresolved.

This discipline makes a modest conclusion useful: it identifies which part of a widely repeated story the data actually support. A reserve-share chart can document a change in composition. Establishing a strategic decision requires additional evidence from the institutions concerned.

Sources and revision

Primary sources are the COFER dataset and scope notes, the November 2025 methodology note, and the October 2, 2025 historical data brief. The October 8, 2026 revision corrects the obsolete allocated-only description and replaces a broker-sourced valuation claim with the IMF’s dated example. The chart deliberately retains the original publication vintage and is not spliced into the revised COFER history.