The backstop Japan did not use
A repo facility exists so that Japan can raise dollars without selling US Treasuries. Japan let $123bn of them run off anyway, and borrowed almost nothing. Both of those numbers are published.
The yen carry trade is one of the few pieces of market plumbing that periodically becomes everyone's problem. Borrow in a currency with near-zero rates, convert, buy something yielding more, and pocket the difference. It works quietly for years and then unwinds violently over a few days, and the violence is never really about Japan. It is about the leverage that was funded there and the assets it was funded into.
What makes the current configuration worth writing about is that the trade's engine has been losing power for a year, and the currency has not behaved the way the textbook says it should.
The engine, compressing
A year ago the gap was 388 basis points. It is now 263. The Federal Reserve cut through the autumn and has held since; the Bank of Japan has moved the other way, doubling its policy rate from 0.50% to 1.00% in a series of small steps.
That is a 125 basis point compression in the thing that pays for the trade. Every carry position funded in yen is now earning meaningfully less for the same risk, and the direction of travel on both sides points to further compression rather than to a reversal.
The currency, which did not get the message
The textbook response to a narrowing rate differential is a stronger funding currency: as the trade becomes less attractive, positions unwind, borrowed yen gets bought back, and the yen appreciates.
It has done the opposite. The yen has gone from about 147 to the dollar to about 162 over the same period, roughly 10% weaker while the reason to be short it got 125 basis points less compelling.
That divergence is the most informative thing on the page, because it means the position is not being held for the carry. Something structural is holding it: Japanese institutional demand for foreign assets, hedging ratios that have not been rebuilt, an expectation that the compression stalls, or simply the inertia of a very large and very old position. Whatever the cause, a trade held for reasons other than its yield is a trade whose unwind trigger is not the yield either, which makes the usual monitoring variable close to useless.
What a Japanese saver actually faces
Currency moves are easier to read against something that is nobody's liability.
In dollars, gold rose about 21% over the year and has given back a good deal of a sharp move. In yen it rose about 33% and has held more of it. The gap between those two lines is the currency, and it compounds: a Japanese holder of domestic assets has lost purchasing power against a hard asset at a materially faster rate than a dollar holder has.
That is the pressure the Bank of Japan is actually responding to, and it explains why its tightening has continued even as the Federal Reserve has stopped. It is not a growth story or an inflation-target story in any conventional sense. It is a currency story with a domestic political dimension, and those tend to run further than rate-differential models expect.
The other half: the buffer is gone
The reason any of this matters beyond Japan is that a carry unwind is a dollar funding event. Positions get closed, dollars get bid, and the stress shows up in the overnight funding market rather than in the currency pair.
Which is where the second fact belongs.
For several years the overnight reverse repo facility held hundreds of billions of dollars of money-market cash that had nowhere better to go. That balance functioned as a buffer: when funding markets tightened, the cash left the facility and went into the market, and rates barely moved.
It is now effectively empty: a billion dollars against a peak measured in the hundreds of billions. Bank reserves have fallen from roughly $3.30 trillion a year ago to $2.96 trillion. And secured overnight financing has been printing at or fractionally above the effective federal funds rate rather than below it, which is what a system with abundant reserves does not do.
None of that is a crisis. It is the description of a system that has moved from abundant reserves to something closer to ample, where a shock is absorbed by the price rather than by the buffer. In that regime a large forced-buying event in dollars, a carry unwind, a quarter-end or a Treasury settlement, has considerably more room to move funding rates than it did two years ago.
A carry unwind is not a currency event that spills into funding. It is a funding event that happens to be denominated in a currency pair.
FIMA: the backstop that exists for exactly this
The instrument built for this problem is the Foreign and International Monetary Authorities repo facility, which the Fed set up in March 2020 and made permanent in July 2021 alongside its domestic standing repo facility.
The problem it solves was demonstrated live. In March 2020 foreign central banks needed dollars, and the way a foreign official raises dollars is by selling US Treasuries. They did, in size, into a market that could not absorb it, and the world's deepest bond market briefly stopped functioning. The dollar squeeze and the Treasury dysfunction were the same event.
FIMA breaks that link. A foreign central bank or international monetary authority with an account at the New York Fed can repo its Treasuries overnight for dollars instead of selling them. The securities stay where they are, the dollars arrive, and the Treasury market never sees a seller. It is deliberately priced at a penalty to the market, so it is a ceiling rather than a subsidy.
If you wanted to design an instrument for a country that holds an enormous stock of Treasuries and suddenly needs dollars, you would design this one. Japan holds the largest such stock in the world.
What Japan actually did
From the February peak of $1,239bn to June's $1,117bn, Japan's holdings fell by $123bn in four months.
Part of that is price rather than sales. TIC reports holdings at market value, and the ten-year yield rose 34 basis points over those four months, which on a stock of that size and duration accounts for something in the region of $20bn. The remaining hundred billion is a seller.
The two tall bars on the left are March and April 2023, the weeks Silicon Valley Bank failed and Credit Suisse was taken over. $60bn went out of the facility in a single week and unwound over the following month. That is what FIMA looks like when a foreign official needs dollars it did not plan to need.
Everything to the right of those two bars is the period this piece is about. The tallest point in it is $3bn, in February 2026. Thirteen of the last thirty-two months are exactly zero.
So the sequence is this: a facility exists whose entire purpose is to let a foreign official raise dollars without selling Treasuries; the largest foreign holder of Treasuries reduced its holdings by $123bn; and the facility was drawn on for an amount that rounds to nothing against that.
Japan did not use the pawn shop. It sold.
The facility's purpose is to stop a seller becoming a fire sale. It cannot stop a seller who would rather sell.
Why that is the interesting result
The popular version of this story is that an arrangement was reached, publicised, and thereby prevented a Treasury fire sale. The publicity part is real and it may well have mattered. An announced backstop changes what other people expect a seller to do, which is most of what stops an orderly sale becoming a disorderly one.
But the usage data says the backstop was not the mechanism. There are three readings and they are not equally comfortable:
Selling was cheaper. FIMA is priced at a penalty, it is overnight, and it has to be rolled. For a holder that wants dollars for months rather than days, to pay for imports or to fund intervention, repeatedly rolling an expensive overnight loan is a worse deal than selling. If so, the facility is simply mispriced for the use case it is being credited with.
Or the sales were wanted. Reducing dollar assets and buying yen is what a country defending its currency does. On that reading Japan was not raising dollars under duress at all; it was rebalancing deliberately, and the facility was irrelevant rather than declined.
Or stigma. The first institution to draw meaningfully on a lender-of-last-resort facility announces something about itself. That is the standing problem with every such arrangement, it is not solved by making the facility permanent, and it is exactly the circumstance in which a country would rather sell an asset quietly than borrow loudly.
You cannot distinguish those three from the published data. What you can say is that the facility has been drawn at scale exactly once, in a week of acute bank stress, and that the episode in this piece is not the shape of demand it has ever served. A funding squeeze that lasts a week and a currency position that runs for months are different problems, and only one of them has ever been brought to the window.
Correcting the thing that gets said about it
The claim that follows this story around is that FIMA is the Federal Reserve printing money for foreign governments, and therefore inflationary for Americans. The mechanics do not support it, and the distinction is worth being precise about.
A FIMA drawing is an overnight repurchase agreement against Treasury collateral at a penalty rate. The Fed takes in securities it already custodies, advances dollars, and the trade reverses the next day. Reserves are created for one night and then extinguished. Nothing is bought, no security leaves the private market, and the Fed's balance sheet returns to where it was.
That is a different operation from quantitative easing, where the Fed purchases Treasuries outright, permanently removes them from the market, and the reserves it creates stay in the system. The two look similar on a balance sheet line and behave nothing alike.
The honest criticism of FIMA is not monetary. It is about moral hazard and about who the Federal Reserve is ultimately a central bank for. Those are real questions, and better ones than an inflation argument that the instrument's own structure refutes. And in this instance the criticism has an additional problem: at $3bn in the busiest month, whatever FIMA did or did not do over this period, it did not do it at a scale capable of affecting US prices.
The leverage question, taken seriously
One part of the popular account deserves more respect than it usually gets from people who dismiss the rest.
A large foreign holder of Treasuries does possess a form of leverage, and it is not imaginary. The threat is not that a country sells and the US cannot fund itself. The market is far too deep for that. It is that a disorderly sale raises long yields at a politically inconvenient moment, and long yields transmit to mortgages, to equity valuations, and to the fiscal accounts through refinancing cost.
What limits that leverage is that it is close to unusable. A seller of $1.1tn of Treasuries is also the largest victim of the price decline they cause, and they are left holding a depreciating pile of the currency they just sold into. It is a threat that costs the threatener more than the target, which is why it gets discussed constantly and executed never.
Which is the same structure as the carry trade itself, and as the bond-manager story: the largest holders are the least free. Size looks like power right up until you try to use it.
What the flat line is worth
The rate differential is the variable everyone monitors and, for the reasons above, close to the least informative one: the position is evidently not being held for the carry.
The plumbing indicators are better and they are published: the cross-currency basis, Japanese insurers' hedge ratios, a sharp move in the yen on no news, secured funding rates pushing above the top of the corridor. None of them require special access and none of them are fast enough to trade on, which is roughly the right description of a risk indicator.
And there is the FIMA line, flat almost all of the time and moved properly once. March 2023 is the entire body of evidence about what the facility does under stress, and as far as it goes it is reassuring: $60bn went out in a week, came back over a month, and nobody had to sell into a market that could not take it.
What it does not tell you is anything about the case in this piece, because the two demands are not alike. One observation is a thin basis for confidence in a backstop, and a thinner one for the assumption that it will be reached for next time, an assumption the last four months have already tested once, and found wanting.
← Back to all insights