- What a carry trade actually is
- The currency risk cannot be hedged away, by construction
- Carry accrues in slivers; the currency moves in hours
- Why the unwind sells things that have nothing to do with Japan
- The route into Indian equities
- Slow drift and fast unwind are different events
- Why our model watches the speed, not the level
- What the correlation panel can and cannot say about this
- Four ways this gets read wrongly
- Where this sits in the app
- Common questions
What a carry trade actually is
A carry trade borrows money in a currency where borrowing is cheap and puts the proceeds into something that pays more. The profit is the gap between the two rates, earned for as long as the position is held. Japan has offered the cheapest large-scale funding in the developed world for a very long time, which is how the yen became the borrowing leg of choice.
Three steps, and it is worth being slow about them because everything later follows from the order.
- Borrow yen. The loan is denominated in yen. Whatever happens next, the amount owed is a fixed number of yen.
- Sell the yen and buy the currency of wherever the money is going — dollars, most often, then onward.
- Buy the asset. A Treasury, a corporate bond, a Mexican government note, an equity portfolio, a basket of emerging-market risk. Anything that yields or is expected to appreciate.
Each day the position is open, it earns the difference between what the asset pays and what the yen loan costs. That difference is the carry. It has nothing to do with the asset going up; a carry trade can be profitable on a portfolio that has not moved at all, which is exactly why so much money is willing to do it. The mechanics of how a policy rate becomes a borrowing cost are the same everywhere and are covered in how a rate change reaches you; what matters here is only that two countries can have very different ones for years at a time.
Notice what has not been mentioned yet: the price of the yen. Step two converted yen into something else and step three spent it. The loan still has to be repaid in yen, and nobody has bought that yen back.
The currency risk cannot be hedged away, by construction
The obvious objection is that a professional would simply hedge the yen exposure with a forward contract and collect the interest differential cleanly. That does not work, and the reason it does not work is the most important mechanical fact in this article.
A currency forward already prices in the interest differential. If yen borrowing is cheap and dollar deposits pay more, the forward price of the yen against the dollar is set so that no free money exists in the round trip — borrow yen, convert, deposit, and lock the conversion back — because if it did, an arbitrageur would take it until it stopped existing. Buy the hedge and it costs you, near enough, precisely the gap you were trying to earn.
Near enough, not exactly. The relation is an arbitrage, not an identity, and in stressed funding markets a gap opens between the two sides of it — the cross-currency basis. That gap is real and traders work it. It has never been wide enough to turn the fully hedged version into the reason anyone does this trade.
So the carry trade is not an interest-rate trade that happens to carry currency risk. It is a payment for accepting currency risk, and it exists only for participants who leave that leg open. The unhedged yen liability is not carelessness. It is the entire product.
Which reframes the question that most explanations skip. The carry is not compensation for patience or for cleverness. It is compensation for holding a short position in the yen, and it is collected in small daily instalments for as long as that position stays quiet.
Carry accrues in slivers; the currency moves in hours
Put numbers on that asymmetry, because it is the whole reason a carry unwind is violent rather than gradual. The figures below are illustrative — round numbers chosen so the arithmetic is easy to redo, not a quote of any actual pair of rates.
Take a position earning an interest differential of 4 percentage points a year. Spread across roughly 250 trading days, that is 0.016 percentage points a day. Sixteen thousandths of a percent. That is what a carry trade earns on a normal Tuesday.
Now suppose the yen strengthens 1.2% against the dollar in a single session. The yen loan has just become 1.2% larger measured in the currency the assets are held in. Divide: 1.2 ÷ 0.016 = 75 trading days. One session has taken back about three and a half months of the position's entire reason for existing.
That ratio — seventy-five days of carry against one bad day of currency — is not an anomaly. It is the normal shape of the trade, and it is why the position cannot be managed by waiting. There is no version of "hold on and let the carry catch up" that works on a timescale a leveraged book can survive.
Leverage does not change that ratio, since both the carry and the currency loss scale with the same notional. What leverage changes is how many bad days the position is permitted. At five times leverage, a 1.2% adverse currency move is 6% of the equity behind it; at ten times, 12%. Past some level of loss the position stops being closed by its owner and starts being closed by the broker, and that level is set by the margin agreement rather than by anyone's view of the yen.
Why the unwind sells things that have nothing to do with Japan
Here is the chain, and it is the part almost never spelled out for Indian readers.
- The yen strengthens. The reason does not matter much — a policy shift in Japan, a growth scare abroad that narrows the rate gap, or simply a crowded position starting to leave.
- Every yen-funded book takes a mark-to-market loss on the funding leg, simultaneously, worldwide. Nothing needs to have happened to the assets.
- Margin is called. Levered positions are revalued daily and the shortfall must be met in cash, usually the same day.
- Cash is raised by selling. And this is the step that matters: a book under a margin call does not sell its worst asset. It sells whatever can be sold fast and in size without moving the price too much.
- Buying the yen back to repay the loan strengthens the yen further, which deepens the loss on every remaining yen-funded position and calls more margin. The move feeds itself.
Step five is why these episodes are fast and step four is why they are indiscriminate. Together they produce the named failure mode worth carrying away from this article:
A carry trade fails on the funding leg, not on the asset leg. You can be entirely right about the Indian company you own and still watch it get sold, because the person selling it is not selling it for any reason connected to the company. They need yen by Thursday.
One honest limitation. How much yen-funded leverage is outstanding at any moment is not publicly observable in real time — it is spread across banks, hedge funds, corporate treasuries and retail margin accounts in several jurisdictions. Which means nobody, including us, can tell you how much fuel is in the tank. That is a genuine gap in what this framework can know, and anyone quoting a precise figure for the size of the carry trade is estimating.
The route into Indian equities
India does not borrow yen to any significant degree, and Indian companies mostly do not either. So the transmission is not a direct debt channel. It runs three ways, and they are worth separating because they behave differently.
Shared books. The pool of money that owns Indian shares overlaps with the pool that runs yen-funded positions. When a global risk book de-risks, it sells its positions in proportion to how the book is built — so India's share of the selling is set by India's weight in someone's allocation, not by anything happening in India.
Liquidity, which cuts the wrong way. A book raising cash sells what it can sell. India is among the more liquid emerging markets, and in an unwind that is a liability rather than a virtue. Being easy to exit makes you the first exit. It is one of the few places in markets where quality of market structure produces a worse short-run outcome.
Margin, which compounds it locally. A volatility spike raises exchange margin on derivative positions, because the exchange revalues a portfolio across a grid of scenarios and charges the worst one. Higher volatility widens the grid, so the same position costs more to hold, and Indian traders reduce positions for reasons that started in Tokyo.
Foreign flow prints are the closest thing to a receipt, and they are a weak one. They arrive at the end of the session, which makes them a record of what happened rather than a warning of what is happening — a distinction the macro reading guide makes about every lagging input on the page. And a net sale figure is silent on motive: it looks the same whether the seller was closing a view on India or raising yen. The print is consistent with this chain on days when the chain is running. It does not establish it.
Two things this does not mean. It does not mean anything changed in the cash flows of the Indian business you own; a funding shock abroad is not an earnings event at home. And it does not mean the effect is permanent — forced selling and considered selling look identical on the tape and are completely different in what they imply. Distinguishing them afterwards is possible; distinguishing them live is much harder than anyone claims.
Slow drift and fast unwind are different events
The single most useful discipline here is refusing to treat "the yen is strong" and "the yen strengthened fast" as the same observation. They are different phenomena that happen to share a ticker.
| Slow drift | Fast strengthening | |
|---|---|---|
| What it is | The rate gap between Japan and elsewhere narrowing over months | Positions being closed under pressure inside days or hours |
| Effect on a carry book | The trade gets less attractive; new positions are not put on | Existing positions become losses that must be funded today |
| Who acts | Allocators, at their own pace | Risk desks and margin systems, on someone else's timetable |
| What gets sold | Whatever the new arithmetic no longer justifies | Whatever is most liquid, regardless of what it is |
| Reaches Indian shares | Gradually, through allocation decisions | Immediately, through correlated forced selling |
| Ends when | The rate gap stabilises | Leveraged positioning is cleared — a finite quantity, but an unobservable one |
Read the right-hand column again and notice that nothing in it is about Japan. It is about leverage, margin and liquidity. Japan is only where the loan happened to be written.
The self-limiting property in the last row deserves care. An unwind runs on leveraged positions liquidating, and the supply of those is finite, so the mechanism cannot continue indefinitely. That is a statement about the mechanism and nothing more — it does not say when, it does not say what the level will be when it stops, and it does not say what happens next. Anyone converting "self-limiting" into a timing call has added something that was not there.
Why our model watches the speed, not the level
The macro page treats USD/JPY twice, on purpose, and the reason is everything above.
First, as an ordinary scored tile. USD/JPY carries a weight of 0.10 and a positive sign, meaning a rising USD/JPY — a weakening yen — reads as supportive for Indian equities, because it says the carry is intact. Its contribution is the day's percentage change divided by a scale of 1.0 and clamped to the range −1 to +1, so a 1% move saturates it. The full formula and the rest of the weights belong to the macro reading guide, which owns them.
Second, as a circuit breaker. A one-day fall of 1.2% or more in USD/JPY sets the page's regime to Stress regardless of the composite score. A one-day VIX jump of 20% or more does the same. Either fires on its own; neither waits for the average to agree.
The reason for the duplication is the arithmetic from earlier. A weighted average across a whole page of inputs is built to be steady, and a carry unwind is not steady — on the numbers above, one session takes back seventy-five days of carry. Left to the composite alone, a −1.2% day in USD/JPY saturates its own tile and still moves the score by only about 7 points on a scale that runs from −100 to +100, because that tile is 0.10 of a weight total of 1.45 on a day when every input reports. Two ordinary green tiles cancel it — a softer crude price carries 0.12 and a firm session in US futures another 0.06, and both would be scored as helpful for Indian equities. The score would print something unremarkable on the day a funding shock began. The override exists because an average is the wrong instrument for detecting a discontinuity, and no amount of reweighting fixes that — it is a property of averaging, not of the weights.
Now the part that surprises most people who look at the page carefully. The level of USD/JPY is not in the model at all. Only the change is. A yen that has been weak for two years contributes exactly nothing on a quiet day, because there is no daily move to score. The page has no opinion on whether the yen is cheap. It has an opinion on whether the yen just moved, which is the only part that transmits.
The Nikkei sits in the same family, at a weight of 0.03, and earns its place as a second read rather than a first: Japanese equities falling hard while the yen surges is the classic signature of an unwind, because a domestic Japanese problem would not usually strengthen the currency. Two tiles disagreeing with each other in that particular way is more informative than either on its own.
Every number in this section is our modelling judgement, not a fact about the world. The 0.10 weight, the 1.0 scale, the −1.2% trigger, the VIX threshold and the regime cut-offs were chosen as a considered view of what matters to Indian equities. They were not estimated from a regression and they are not constants anyone measured. A different desk would set them differently and would not be wrong to. We publish them so that a reader who disagrees can see precisely what they are disagreeing with.
What the correlation panel can and cannot say about this
USD/JPY is one of the ten drivers in the correlation panel, which computes the Pearson correlation of Nifty's daily returns against each driver over a 30, 60 or 90-session window. A positive Nifty–USD/JPY correlation is consistent with the carry channel: weak yen, carry intact, risk appetite up, Indian shares up alongside.
Consistent with. Not evidence of. The reasoning discipline the macro reading guide sets out applies with particular force to this pair, for two reasons specific to it.
The mechanism is episodic and the correlation is continuous. The carry channel does most of its work in a handful of violent sessions and very little the rest of the time. A coefficient computed over 30 sessions containing one unwind is mostly measuring that week; the same coefficient over 30 quiet sessions is measuring almost nothing. Both print a number to two decimals and look equally authoritative.
Both series respond to the same third thing. A global risk repricing moves the yen and Indian equities together without either causing the other, which is the ordinary case in macro rather than the exception.
So the correct order of reasoning is: state the channel first, then check whether the window is consistent with it. A correlation you can attach a mechanism to is mild support for that mechanism. A correlation you cannot is a coincidence with a decimal point.
Four ways this gets read wrongly
- Treating the yen's level as the signal. "The yen is strong" is not a state that transmits. The strengthening is the event; once it has happened and stopped, the funding shock is over even though the level is still there.
- Reading a carry unwind as news about India. The seller had a funding problem. Nothing was communicated about Indian earnings, and treating the sale as information about the business is reading a stranger's margin call as a research note.
- Reading tile colour as direction. Green on the macro page means helpful to Indian equities, not "the number went up" — so a falling dollar index shows green while a falling USD/JPY shows red. The pillar covers this properly; it is listed here because this pair is where it bites hardest.
- Believing the stress flag is a warning. It is not. It fires on a move that has already happened — a description of the state the market is in, not a forecast of the next one. Our own page cannot tell you a carry unwind is coming, and no honest reading of it pretends otherwise.
The trade-off underneath all four is worth stating plainly, because it is what a macro framework actually costs. Knowing this chain does not give you a way to act on it faster than the people whose margin calls create it. What it gives you is the ability to tell a funding shock from an earnings shock while it is happening, which mostly argues for doing less rather than more — the same reasoning behind holding a policy on asset allocation rather than a view, and behind what happens to people who stop investing during a drawdown. A framework that explains a fall is worth having. A framework that promises to anticipate one is selling something.
One more, quieter than the others: assuming this is the only funding channel. It is the most visible one because Japan's rate gap has been the widest for longest, but the same arithmetic applies to any cheap borrowing currency. The mechanism belongs to leverage, not to the yen. A bondholder meets a relative of the same discomfort through duration — the price falls because the rate environment moved, not because anything changed at the issuer. Different machinery, same lesson: a loss can arrive on a holding whose own facts are untouched.
Where this sits in the app
All of the above is a chain you can watch rather than a story you have to take on trust.
FNOTrader's Options Analytics app carries the macro page this article describes: USD/JPY as a scored tile alongside the dollar index, US yields, crude, credit spreads and the volatility gauges, each with its transmission mechanism written out on the tile; the composite score with its largest contributors named in words; the separate stress state with the yen and volatility overrides described here; the Nikkei as the second carry read; and the correlation panel across 30, 60 and 90-session windows. Every tile opens its own history chart from one month to five years, so a move can be read against that instrument's own range rather than against a memory of it.
The weights, scales and thresholds are the ones stated above, and they are ours. They are published rather than hidden for exactly the reason given in that section.
Common questions
What is the yen carry trade in simple terms?
Borrowing money in Japan, where borrowing is cheap, and putting it into something elsewhere that pays more. The profit is the gap between the two rates, earned daily. The catch is that the loan must be repaid in yen, and nobody bought that yen back when the money was invested.
Why does a stronger yen cause selling in markets outside Japan?
Because the loan is fixed in yen. When the yen strengthens, every yen-funded position takes a loss on the funding side at once, margin is called, and cash is raised by selling whatever is most liquid — not whatever is worst. Buying the yen back to repay strengthens it further, so the move feeds itself.
Does the yen carry trade affect Indian stocks directly?
Not through Indian borrowing, which is negligible in yen. It reaches India through shared global risk books that own Indian shares alongside yen-funded positions, through India's liquidity making it an easy place to raise cash, and through the volatility spike raising exchange margin on domestic derivative positions.
Why does speed matter more than the level of the yen?
Because a slow move in the rate gap is an allocation decision made at someone's own pace, while a fast move is a funding event handled on a margin desk's timetable. The same currency pair produces two different phenomena, and only the fast one forces indiscriminate selling.
Why does a 1.2% fall in USD/JPY trigger the stress state on the macro page?
Because a weighted average across a whole page of inputs is built to be steady and a carry unwind is not. Left to the composite alone, that day's move saturates its own tile and still shifts the score by only about 7 points out of 100, so two ordinary green tiles cancel it. The override exists so a discontinuity is not smoothed into an unremarkable score. The 1.2% threshold is our modelling judgement, not a measured constant.
Can a carry trade be hedged against currency risk?
Not while keeping the profit. A currency forward already prices in the interest differential, so the hedge costs about what the carry earns. The trade is payment for accepting the currency risk — the open yen leg is the product, not an oversight.
How large is the yen carry trade?
Nobody can say precisely in real time. The borrowing sits across banks, funds, corporate treasuries and retail margin accounts in several countries, and there is no live aggregate. Any specific figure quoted for its size is an estimate, which is worth remembering when one is presented as a fact.
Does a positive correlation between USD/JPY and Nifty prove the carry channel is working?
No. It is consistent with the channel, which is weaker and more useful. Both series can respond to the same global risk repricing without either causing the other, and a coefficient over 30 sessions is heavily influenced by whichever few violent days fall inside the window.
Continue reading
More in Macro & Markets · App: Options Analytics · Definitions: glossary · Free tools: calculators · All: every article