The plumbing of yen-buying intervention

Yen-buying intervention is not a monetary operation, and the Bank of Japan does not pay for it. It is the Ministry of Finance spending down a levered, dollar-denominated bond portfolio — and every route it can take to raise those dollars leaves a different fingerprint on the Federal Reserve's balance sheet. This note maps the participants, the settlement chain and the four funding routes, then sets out how to tell from public data which one was used.

The timing is not academic. On 31 July 2026 the MOF bought yen in coordination with the US Treasury — the first US participation in a yen-supporting operation since 1998 — and Minister Katayama's statement of 3 August added that Japan "plans to utilize the Federal Reserve's Foreign and International Monetary Authorities (FIMA) Repo Facility in the future." That single sentence proposes to change the plumbing, and with it the observable signature of Japanese intervention.

The cast. The decision, the execution, the cash and the custody sit in four different institutions on two continents — which is the whole reason the routes below differ.

Ownership: the Bank of Japan is the desk, not the principal

The reserves belong to the State and are managed by the Minister of Finance through the Foreign Exchange Fund Special Account — 外国為替資金特別会計, the FEFSA. Three statutes fix this:

The Bank says so itself: it "executes foreign exchange interventions in accordance with the directions of the Minister of Finance by using the Foreign Exchange Fund Special Account… the account of the Japanese Government." The operational split is MOF's FX Markets Division deciding, the BOJ's FX Division executing, and the BOJ's International Operations Division settling.

The BOJ does hold foreign currency assets of its own — ¥11.98trn (roughly $76bn) on 10 August 2026 — but under a separate Policy Board mandate for international cooperation and emergency FX liquidity to Japanese banks, restricted to central-bank deposits and foreign government paper of five years or less. That is not the war chest, and it is not available to the MOF.

Why it matters. Every argument that begins "the BOJ could just use the swap line" fails at this step. Swap dollars land on the Bank's balance sheet, and the Bank has no authority to intervene on its own account.

The balance sheet: a levered carry trade

At 31 March 2025 the FEFSA held ¥191.4trn of assets against ¥97.6trn of Financing Bills outstanding — roughly 51% debt-funded, structurally short yen and long dollars, financed in yen T-bills. Cumulative FX valuation gain through FY2024: about ¥50.3trn.

This framing matters for what intervention is. Yen-selling intervention levers the account up: issue more bills, buy more dollars. Yen-buying intervention deleverages it: sell the dollar asset, retire the yen bill. The account ends smaller on both sides.

What Japan actually holds

Japan's official reserve assets, US$ millions:

Securities are 72.0% of the total and 85.1% of the foreign currency reserves. Deposits are 12.6% — and 99.7% of those sit with foreign central banks and the BIS, leaving $432m with commercial banks, which is to say nothing. Gold's rise from $49bn to $110bn is entirely price: the volume has been fixed at 27.20mn fine troy ounces throughout.

The deposit line has barely moved across four intervention episodes: $135bn at end-2021, $162bn now. Japan does not build a cash pile before intervening. Intervention at scale is therefore always, eventually, a securities operation.

Where those dollars physically sit

The MOF template gives a single combined line for "foreign central banks and the BIS". It does not separate the New York Fed from the BIS, and the Federal Reserve publishes no country breakdown of its foreign accounts, by stated policy. So the split is inference.

What is documented is the menu. The New York Fed has provided dollar banking services to foreign official institutions since 1917 and now serves over 200 account holders across more than 550 deposit and custody accounts. Three instruments matter:

• Deposit accounts, used for time-critical official payments. Almost certainly non-interest-bearing: Regulation D pays the IORB rate only to "eligible institutions", and §204.2(y) defines that term exhaustively as depository institutions, trust companies, §25 and §25A corporations, and branches or agencies of foreign banks. Foreign central banks and governments are not on the list. Corroborated by size — foreign official deposits on the H.4.1 are only $9.5bn. Flagged as inference: there is no current Fed sentence saying so explicitly.
• The foreign repo pool, a reverse repo that sweeps end-of-day balances, in existence since the mid-1970s. Rate "generally equivalent to" the ON RRP rate, so 3.50% today. Usually overnight, though the Continuing Directive permits terms of up to 65 business days. No cap; SOMA collateral is reserved for it ahead of the domestic ON RRP. Outstanding $317.7bn on 5 August 2026 — which is 99.5% of the Fed's entire reverse-repo book, the domestic ON RRP having drained to $1.65bn.
• Custody, $2,910bn of securities for all foreign official and international accounts, of which $2,631bn marketable Treasuries, at face value (Wednesday 5 August levels). No bill/note/bond split is published.

The BIS is the alternative, and a structurally different one: its statutes do not permit current accounts in the name of governments, so it serves central banks and international institutions but not finance ministries directly. Against that it offers a term curve out to five years, SDR-denominated instruments, gold services and active management — at BIS rather than US sovereign credit.

Bills versus coupons

MOF publishes the maturity split of its securities book once a year, in the FEFSA disclosure each November. At 31 March 2025:

The ≤1yr bucket jumped 7.2 percentage points in a single year. The book was deliberately shortened. Adding demand deposits (≈$92bn), time deposits (≈$45bn) and securities maturing within a year (≈$271bn) gives roughly $409bn, about 36% of FX assets, mobilisable without a distressed sale — before touching the securities-lending book. Goldman Sachs independently put cash and equivalents at around $200bn in August 2026, a narrower definition.

What is not disclosed: the currency composition, withheld by explicit standing policy "because it could affect the FX market"; the split of the 26% non-government bucket between agencies, supranationals and ABS; and any duration figure, ever. The only official word on issuer is a 2011 Cabinet answer to the Diet stating that the Fund holds foreign-currency securities 「米国債を中心に」 — centred on US Treasuries — and that country-of-issue detail is withheld.

The chain, step by step

Take a $75bn operation at USD/JPY 156 — the scale of the April–May 2026 episode, ¥11.73trn.

Decision. The Minister decides. Nothing moves on any balance sheet. The test is the G7 one — excess volatility and disorderly movements — restated in the September 2025 US–Japan Joint Statement.

Execution. The BOJ desk sells $75bn and buys ¥11.73trn against a dealer panel. It is spot FX, so nothing settles for two business days. What changes immediately is dealer risk: they are short yen and long dollars in size, against a counterparty they know will not stop. Most of the price impact happens here, in the hedging, not in the later cash flow.

Raising the dollars. The MOF now has two days to produce $75bn. This is the fork in the road, and the subject of the next section.

Settlement, dollar leg. On T+2 the dollars move out of Japan's account at the New York Fed and into the US commercial banking system, via CLS or bilaterally over Fedwire. If the source was cash, this is a pure liability swap on the Fed's balance sheet — foreign official claims down, bank reserves up, total size unchanged.

Settlement, yen leg. The same afternoon, dealers deliver ¥11.73trn into the government's account at the Bank of Japan. Yen bank reserves fall by that amount; the government deposit rises. A real, if temporary, drain of yen liquidity. Note who has done nothing: the Bank of Japan. Its balance sheet has changed composition, not size, and no policy decision was involved.

Sterilisation, which happens by itself. The MOF now holds yen it does not need, so it stops rolling maturing Financing Bills. On redemption the government deposit falls and bank reserves rise back. The drain reverses; the stock of yen Treasury Discount Bills shrinks by ¥11.73trn. Nobody at the BOJ has to act — the sterilisation is a by-product of the funding account deleveraging. The gap between the two legs is where yen repo and front-end rates can twitch for a few days.

Net result. The FEFSA is smaller and less levered. The US private sector holds $75bn more Treasury bills and correspondingly less cash. Japanese banks hold fewer yen T-bills. The Fed's balance sheet is the same size, with liabilities shifted from a foreign official claim to domestic bank reserves. And the exchange rate has moved — by an amount that has far more to do with what dealers now believe about the reaction function than with the $75bn itself.

The four funding routes

The swap line is a fifth option only in appearance. There is no contractual bar — the 2014 USD–Yen swap agreement contains no use-of-proceeds restriction, only a recital about market functioning, and even the OIS+25bp pricing is operational policy rather than contract. The block is institutional and threefold: the FOMC applies a liquidity purpose test to swaps and a separate disorderly-markets test to intervention; New York Fed Staff Report 983 states the modern lines were "solely intended to provide U.S. dollar liquidity on a temporary basis and not to fund foreign exchange intervention"; and decisively, swap dollars arrive at the BOJ, which cannot intervene on its own account.

The history points the same way but not as a deliberate severing: the FOMC eliminated the standing swap facilities on 17 November 1998 "because of their long disuse" and in anticipation of the euro, with participants agreeing that the lines "were no longer needed for their traditional use, exchange-market intervention." Intervention demand had evaporated; the channel was not closed to stop it.

Behaviourally, the lines are not being used. The BOJ's drawings in 2026 have all been between $1m and $6m — $2m on 7 July, $2m on 21 July — recurring operational-readiness tests rather than funding. That is suggestive, not proof: the direct evidence is legal, not behavioural. What is striking is that the entire policy argument in August 2026 concerns the capped facility while the uncapped one sits idle.

The Fed channels, compared

On FIMA the Fed's own staff have written the answer down. Staff Report 983 states that the proceeds are "generally expected to result in transfer of dollars to local institutions with funding needs and/or fund FX spot interventions", with footnote 17 constraining the direction: "FX intervention funded in this way would only be in the direction of currency sales to strengthen the domestic currency vis-à-vis the U.S. dollar." That is precisely yen-buying.

The facility has been drawn in size exactly once: $60,000mn — at the cap — in the week to 22 March 2023, running off to zero by 26 April. Press attribution to the Swiss National Bank is widespread but the Fed does not disclose counterparties.

On the "no different than a swap line" framing

Secretary Bessent said on CNBC on 4 August 2026 that FIMA repo is "really no different than a swap line — the country posts collateral and we lend them the money to intervene." He is more right than the obvious rebuttal allows, and it is worth being precise about where.

On price and tenor at seven days, he is exactly right: seven-day FIMA repo is priced at one-week OIS plus 25 basis points, which is the dollar swap line's pricing formula, over the swap line's tenor. The difference is only at the overnight tenor, where FIMA is struck at the SRF rate.

Where the analogy fails is the counterparty and the collateral, which is the distinction he himself named. Swap dollars land at the Bank of Japan and can only be on-lent to Japanese banks; FIMA dollars land in the account of the institution that pledged the Treasuries, which for Japan is the MOF's — and the MOF is the principal. That, plus the FOMC's separate purpose test, is why the capped facility is usable for intervention and the uncapped one is not. The cap is a live question precisely because the substitute is unavailable.

And the US side is small

If Washington joins in, it fires from the Exchange Stabilization Fund, decided by the Treasury Secretary and executed by the New York Fed as fiscal agent — not by the Fed as a policy matter. The ESF's FX book was $18.8bn at 30 June 2026: $12.95bn of euro-denominated assets and $5.79bn of yen-denominated assets. Warehousing with the Fed is capped at $5bn.

The Fed's own SOMA holds a deliberately matched foreign currency book — $19.1bn at 31 March 2026, the latest quarterly report — and the New York Fed notes that intervention currencies "have historically come equally from the SOMA portfolio and the ESF." But SOMA participation is an FOMC decision, not the Treasury's to command. So Treasury acting alone has roughly $19bn; with the FOMC alongside, roughly $38bn. Either figure is an order of magnitude below what Japan can deploy on its own.

Which is the point: joint intervention is a signal about intent, not about size.

What can be done next

The four routes, their limits, and what each resulting operation would look like in public data — together with the size of the help Washington can actually provide.

The map makes two things explicit that the tables above leave implicit. First, Japan's constraint is not capacity but legibility: roughly $409bn is mobilisable without a distressed sale, yet three of the four routes put Treasuries into the market and announce themselves — in the front end, in duration, or in TIC six weeks later. Only FIMA repo raises dollars without selling a bond, and it is the one route with a hard cap. That is the entire content of the August 2026 argument about the $60bn limit.

Second, the US contribution is bounded by the euro book, not by willingness. Buying yen means selling something, and the ESF's $12.95bn of euro-denominated assets is the something; the $5.79bn of yen it already holds is yen owned, not yen it can buy. Treasury acting alone has about $19bn, and roughly $38bn with the FOMC alongside — against a Japanese side an order of magnitude larger.

What the episodes actually show

Official MOF amounts, against the balance-sheet traces:

There was no intervention in 2023, 2025 or Q1 2026. Note that the pool windows are not consistent — they are the intervals over which a move is visible, chosen after the fact — and the pool is never disaggregated by country. Treat that column as suggestive only. The Apr–May 2026 row is the illustration: the pool fell $22.9bn over the month, yet Japan's own reserve deposits were flat, so most or all of that decline belongs to somebody else.

May 2026 is the cleanest observation in the dataset, though not because the series agree precisely — they don't. Two of the three are TIC-sourced, and the TIC-implied figure is some $15bn short of the MOF-implied one:

The decisive line is the last two. MOF's own securities fell by almost exactly the size of the intervention while its deposits did not move at all. The bill evidence then locates which securities: Japan's short-term holdings fell $59.8bn, against a $61.0bn decline in the global foreign-official bill aggregate.

That last comparison needs a caveat the headline version usually omits. Japan's TIC line is all Japanese holders, official and private; the foreign-official aggregate is all countries but official only. The two are not the same universe, so "Japan was 98% of it" is a coincidence of magnitudes, not an identity. What it does establish is that no other reserve manager was selling bills in size that month, which makes the attribution to Japan far more comfortable than it would otherwise be.

The reading — inference, but as tight as this data ever gets — is that roughly 80% of the operation was funded out of Treasury bills, the balance out of coupons, and the cash pile was not touched. This is the single most important corrective to the received narrative. "Japan drains the foreign repo pool" fitted 2022. It does not fit 2026.

Note also what didn't happen in the coupon book. In April and May 2024, long-term net transactions were +$246m and +$171m — essentially flat. Japan does not dump duration; it sells bills. That is why the bill line is the informative one: Treasury estimates short-term net transactions with zero valuation change, and private Japanese institutions hold overwhelmingly long-duration Treasuries rather than bills, so movements in the short line are far more likely to be the reserve manager.

Signatures: telling the routes apart

The practical payoff. Each route leaves a different fingerprint across four public releases.

* Falls, but for a mechanical reason rather than a sale, and the two are distinguishable. Staff Report 983 describes FIMA collateral moving from the counterparty's custody account "to a SOMA custody account", so the custody memo should decline by the drawn amount — while the Repurchase agreements: Foreign official line rises by the same amount and the Fed's balance sheet grows. A genuine sale shows the custody fall with no offsetting asset. The H.4.1's own footnote addresses only the opposite case (securities pledged to foreign official holders against reverse repos), so this remains an inference from the staff description rather than from the release.

** A FIMA drawing lands first as a foreign official deposit at the Fed, so a brief blip is possible before the dollars are spent.

Order of arrival, for anyone building a monitor: price, instantly → H.4.1, Thursday → MOF reserves template, around the 7th → MOF intervention total, around month-end → TIC, roughly six weeks → FEFSA annual disclosure, each November.

What to watch next. If the 31 July operation and whatever follows are funded through FIMA repo as Minister Katayama signalled, the signature changes completely: no TIC bill drawdown, no custody decline, and instead a line that has printed $0 almost continuously since 2020 starts carrying a balance. That single H.4.1 row — Repurchase agreements: Foreign official — becomes the highest-frequency public read on Japanese intervention funding that has ever existed.

Known limits

1. TIC has no official/private split by country, in any publication. Japan's TIC line is not the FEFSA.
2. The Fed publishes no country breakdown of the foreign repo pool, by stated policy. Every attribution here is inference.
3. MOF publishes no currency composition and no split of the 26% non-government securities bucket, by explicit standing policy.
4. There is no official split of Japan's reserves between the FEFSA and the BOJ. The ~6% BOJ share implied above is arithmetic, at book value, with a timing mismatch.
5. The 2022 episode cannot be decomposed into flow versus valuation — those TIC columns begin only in February 2023.
6. The H.4.1 gives no bill/note/bond split of custody holdings.
7. The ~$85bn size of the 31 July 2026 operation is a Goldman Sachs estimate. MOF publishes the 30 July–27 August window around 31 August 2026.
8. Custodial bias: Japanese-owned Treasuries held through Euroclear, Clearstream or Caribbean chains do not appear in the Japan TIC line at all. Belgium, Cayman, Luxembourg and Ireland together carried $1,736bn in May 2026.
9. Attribution of the July–August 2026 foreign repo pool drawdown to Japan fails on the evidence: the largest weekly fall pre-dates the operation, MOF's end-July deposits were flat, and the decline is smaller than the reported intervention size.
10. Every H.4.1 figure here is from the release of 6 August 2026 (data to Wednesday 5 August). The FIMA repo line is the one to re-check weekly: it has printed $0 almost continuously since 2020, and Japan has publicly signalled an intention to draw on it.

On the FIMA governance question, one point often reported as unresolved is not. Press coverage states that raising the $60bn cap would need a majority of the FOMC. The Continuing Directive, as amended effective 27 January 2026, says otherwise and says it plainly: "The Subcommittee may approve changes in the rate, the maturity of the transactions, eligible Foreign Accounts counterparties… and the counterparty limit; and the Subcommittee shall keep the Committee informed of any such changes." On the governing document, the Foreign Currency Subcommittee can do it alone. That materially lowers the institutional bar to what Secretary Bessent asked for.

Sources

All accessed 13 August 2026.