The FX carry trade explained, with its risks

Reference article

A carry trade borrows a currency with a low interest rate and holds one with a high rate. The rate gap is paid slowly, a little every day. The risk is that the high-rate currency falls, and when it falls it can fall fast, because many holders sell at once.

In currency markets, the word carry means the income you collect as interest for holding one currency while you owe interest in another. A carry trade borrows a currency whose interest rate is low and uses the money to hold a currency whose interest rate is high. The gap between the two rates is the carry.

This page works one example by hand, shows where the forward price fits, and then shows why carry's payoff shape fools people: many small gains, then a large, fast loss.

Every outside page below was read on 8 November 2026. This page is education only. It is not advice to buy or sell any currency.

The words first

• Exchange rate. The price of one currency in units of another. Here we write it as units of the low-rate currency per one unit of the high-rate currency.
• Funding currency. The currency you borrow. It has the low rate.
• Target currency. The currency you hold. It has the high rate.
• Spot. The exchange rate for trading now.
• Forward. An exchange rate agreed now for a trade on a later date.
• Leverage. Holding a position larger than the cash you put down. The rest is borrowed.
• Margin. The cash a broker or bank asks you to keep against a leveraged position.

A worked one-year carry trade

Take two currencies, L and H. L pays 0.5% a year. H pays 5% a year. The spot rate today is 100 L per H.

1. Borrow 1,000,000 L for one year at 0.5%.
2. Swap it at spot for 1,000,000 / 100 = 10,000 H.
3. Deposit the 10,000 H for one year at 5%. At the end you have 10,500 H.
4. At the end you owe 1,000,000 × 1.005 = 1,005,000 L.

Whether you made money depends only on the spot rate in one year. Call it $S_1$. Your result, in L, is

Spot in one year ($S_1$)	Result in L	As a share of the 1,000,000 L borrowed
105 (H rose 5%)	+97,500	+9.75%
100 (no change)	+45,000	+4.5%
95.71 (H fell 4.29%)	about 0	about 0%
95 (H fell 5%)	-7,500	-0.75%
90 (H fell 10%)	-60,000	-6%

The break-even spot solves $10{,}500 \, S_1 = 1{,}005{,}000$, so $S_1 \approx 95.71$. H can fall about 4.29% against L before the trade loses. If the rate does not move, the trade gains the rate gap, about 4.5 points. A 10% fall in H wipes that out and more. We checked every number in this table in Python.

Now add leverage. Say the broker asks you to keep only 100,000 L of your own cash as margin against the whole position. The same table, measured against your 100,000 L, reads +45% if nothing moves and -60% if H falls 10%. Leverage does not change the bet. It changes how much of your own cash one bad year takes. With a margin account, a falling position can bring a call for more margin before the year ends. You then add cash or close the position, perhaps at a bad price.

The forward price already holds the rate gap

Here is the step that turns a currency story into a quant idea. Compute the one-year forward from the two rates, the way covered interest parity says:

That is the same number as the break-even spot. It is not a coincidence. Covered interest parity says the forward must hold the rate gap, or someone can lock in a riskless profit by borrowing in one currency, lending in the other and fixing the exchange back with a forward. So a carry trade that you hedge with a forward earns nothing extra. The carry is only there if you leave the exchange rate open.

That gives a clean way to state what a carry trader believes: the spot in one year will end above the forward. In other words, the forward is a biased forecast of the future spot.

Covered interest parity is close to exact but not perfect. A BIS study, Covered interest parity lost (BIS Quarterly Review, September 2016), writes that "since the onset of the Global Financial Crisis (GFC), CIP has failed to hold", shown by a lasting gap called the cross-currency basis. The outline above still stands, but a real forward can differ from the formula by that basis.

Uncovered interest parity, and why its failure matters

Uncovered interest parity is a different claim. It says that, on average, the high-rate currency falls by about the rate gap, so the carry trade earns nothing on average. If that were true, carry would be a coin toss with costs.

Research has found that it does not hold well. Brunnermeier, Nagel and Pedersen, in Carry Trades and Currency Crashes (NBER working paper 14473, 2008), refer to it as "the uncovered interest-rate puzzle". Burnside, Eichenbaum, Kleshchelski and Rebelo, in Do Peso Problems Explain the Returns to the Carry Trade? (NBER working paper 14054, 2008), describe the carry trade's payoffs as "on average large and uncorrelated with traditional risk factors" in their sample.

"On average" means across the years they studied, not in any one year, and not in the future.

The shape of the payoff: small gains, rare crashes

The same Brunnermeier, Nagel and Pedersen paper "documents that carry traders are subject to crash risk". Exchange rate moves between high-rate and low-rate currencies are "negatively skewed". Skew measures whether the long tail of outcomes is on the loss side (negative skew) or the gain side (positive skew). The authors argue the negative skew comes from "sudden unwinding of carry trades", which tends to happen when "risk appetite and funding liquidity decrease". Funding liquidity is how easily traders can borrow to keep a position open.

A toy model shows why that shape fools people. Suppose a carry position gains 0.4% in 95 months out of 100 and loses 8% in the other 5. Then:

• The average month is $0.95 \times 0.4\% + 0.05 \times (-8\%) = -0.02\%$. It loses money on average.
• The skew is about -4.1. The volatility (standard deviation) of a month is about 1.83%.
• A 36-month record has a $0.95^{36} \approx 16\%$ chance of showing no crash at all. Such a record shows a steady 0.4% a month with almost no swings.

So a short, calm history can make a losing bet look smooth and safe. Our Python check printed a mean of -0.02%, a standard deviation of 1.831% and a skew of -4.129, and the chance of no crash month in 12, 36, 60 and 120 months as 54%, 16%, 4.6% and 0.2%.

This is also the "peso problem" in the Burnside paper's title: a rare bad event that a sample may simply not contain. Their abstract states "We argue that these payoffs reflect a peso problem." Read in full, they argue the rare event matters because it hits in bad times, not because it is a very large loss in itself.

A dated case: the yen carry unwind of August 2024

In early August 2024 many carry trades were closed at once, among them trades funded in Japanese yen.

• The trigger. The BIS Quarterly Review of September 2024, Carry off, carry on, says "the trigger was investors' outsize reaction to a somewhat disappointing US labour market release on Friday 2 August".
• The unwind. The same review says "currency carry trades unwound amid changing expectations of interest rate paths", and that "the unwinding of carry trades on 5 August coincided with a widespread sell-off across global asset classes". On that Monday it records "the yen appreciating further and the TOPIX index posting double-digit losses". TOPIX is a broad index of Japanese shares.
• The size. BIS Bulletin 90, The market turbulence and carry trade unwind of August 2024 (27 August 2024), says the size of these trades "is difficult to measure". It gives a rough middle estimate of 40 trillion yen, about 250 billion US dollars, going into the event, and says that figure is likely too low because of gaps in the data.
• What made it worse. The bulletin says the unwinding of leveraged trades "amplified the initial reaction" to the US data. It also says markets then stabilised quickly.

For a carry trader funded in yen, a rising yen is the loss case in our table: the borrowed currency gets dearer to pay back. When many traders hold the same position, one seller's loss can force the next seller's. Thuztra Learn's page on crowded positions works through that mechanism.

Carry is not only a currency idea

Koijen, Moskowitz, Pedersen and Vrugt, in Carry (NBER working paper 19325, 2013), split a security's expected return into "its 'carry' and its expected price appreciation". Carry is what you earn if prices do not move. They report that carry predicts returns across many asset classes, including global equities, global bonds, currencies, commodities, US Treasuries, credit and equity index options. They also report that carry strategies across asset classes tend to do badly in periods that coincide with global recessions.

On Thuztra Learn, the same idea appears for storable goods in commodity carry and the inverted curve, and for options in put-call parity and carry.

How a quant measures carry, step by step

This is a method for study, not a trading plan.

1. Pick a fixed list of currencies and a fixed start date. Write them down before you look at any result.
2. Each month, rank the currencies by their short-term interest rate. Use the rate known at that date, not a later revised figure.
3. Form two baskets. Hold the highest-rate group and borrow the lowest-rate group, in equal sizes. This is a long-short portfolio.
4. Charge costs. Each change of basket crosses the bid-ask spread, the gap between the buying and selling price. Include the spread and the cost of rolling forwards.
5. Report the shape, not only the average. Show the skew, the worst month, the maximum drawdown (the largest fall from a peak) and how long recovery took.
6. Name the data window in every chart. A result from 2003 to 2007 and one from 2003 to 2009 can tell opposite stories.

Why a carry backtest can mislead

A backtest runs a fixed rule on past data to see how it would have done. Carry is a hard case for backtests, for three reasons.

• The peso problem. As the toy model showed, a sample with no crash makes a negatively skewed bet look calm. A short window may hold no crisis at all.
• Currencies that stop existing. Some currencies are replaced, pegged or closed to foreign money. If your data keeps only currencies that still trade today, it drops some of the worst outcomes. That is survivorship bias.
• Rates and prices you could not have traded. A quoted interest rate is not always a rate a foreign investor could earn. Capital controls can block the trade at the moment it matters most.

Read what a backtest is, in plain words for the basics. Past results do not tell you the future.

What Thuztra Learn teaches about carry, and what it does not

Thuztra Learn is our product. Its quant trader path has a module called FX and commodities. It covers FX quoting, forwards and the two interest parities, the FX delta conventions and the smile, and commodity carry. Its risk path covers funding liquidity and crowded positions. It does not give trading signals, run a currency portfolio for you, or connect to a broker. Membership is a paid subscription: see what Thuztra Learn is and what it costs.

Check the forward yourself

Redo the worked example with two other rates, say 1% and 4%, and a spot of 1.25. Compute the forward and the break-even spot. Check that they are equal.

You know it worked if both numbers match to four decimal places. Two things go wrong most often. The exchange rate is quoted the other way up: read FX quoting, forwards and the two interest parities first. Or the rates are mixed up between the two currencies: the funding currency's rate goes on top of the fraction when the spot is quoted in funding-currency units.

Fine print: the BIS, NBER and the named authors are cited as sources only. They have no link to Thuztra. Currencies L and H are examples, not real currencies. Education only, not advice.

In short

The forward price already contains the rate gap, so a carry trade is a bet that the exchange rate will not fall as far as the forward says. Research finds the bet has often paid on average and has negative skew: long calm runs, then sharp crashes. Judge any carry result by its worst periods, its costs and its data window, not by its average.

Terms used here

• Interest
• Exchange
• Volatility
• Maximum drawdown
• Survivorship bias
• Backtest
• Liquidity
• Portfolio
• Bid-ask spread

Related

Topics

• carry trade
• FX
• interest rate differential
• covered interest parity
• uncovered interest parity
• crash risk
• negative skew

Also in the library

• Quant puzzles
• Interview questions
• Quant brainteasers
• Glossary
• Reference articles