The Carry Trade Explained: Interest Differentials, the Evidence and Crash Risk

The carry trade is a currency strategy that borrows (or sells) a low-interest-rate currency and invests in (or buys) a high-interest-rate currency, aiming to earn the interest-rate gap between them. It has made money on average for decades, but its returns arrive slowly and its losses arrive suddenly, so the long-run profit is best understood as payment for bearing crash risk rather than as free interest.

In short: textbook theory (uncovered interest parity) says the high-yield currency should depreciate by enough to cancel the interest gap. In the data it does not, so a diversified carry portfolio earned roughly 4% to 6% a year with a Sharpe ratio of about 0.4 to 0.9 depending on the sample and construction. The catch is negative skewness: in Brunnermeier, Nagel and Pedersen’s words, the trader saying is that exchange rates “go up by the stairs and down by the elevator.” Our own carry portfolio on Federal Reserve data lost 16.3% in October 2008 alone and 40% peak to trough from 2007 to 2009, taking until 2013 to recover. A retail trader holding the same idea with leverage through a broker’s overnight swap faces that crash risk multiplied, minus a financing markup.

Carry is often packaged inside “swap-positive” forex signals and grid or hold-forever expert advisors. This piece explains the mechanics, the peer-reviewed evidence and the crash arithmetic.

The misconception: “you just collect the interest”

A carry pitch usually looks like this: AUD/JPY pays a positive swap every night, so hold it and get paid while you wait. Two things are wrong with that picture.

First, the interest is not free. You are exposed to the exchange rate the whole time, and the high-yield currency is typically the one that falls hardest when global markets panic.

Second, theory says the carry should be zero on average. The fact that it is not is one of the best-known puzzles in international finance, and the leading explanations all say the same thing in different words: carry traders are being paid to take a risk that shows up rarely and badly.

The mechanics

Interest parity, covered and uncovered

Two parity conditions define the problem.

Covered interest parity (CIP) is an arbitrage relationship. The forward exchange rate must reflect the interest-rate differential, or a riskless profit is available. If US dollar rates are 4% and yen rates 0.5%, the one-year forward price of yen in dollars sits about 3.5% above spot. Buying a high-yield currency forward is therefore economically the same as borrowing the low-yield currency and depositing the proceeds in the high-yield one. Lustig, Roussanov and Verdelhan note research concluding that CIP holds at daily and lower frequencies, with exceptions in stress, notably the fall of 2008.
Uncovered interest parity (UIP) is a hypothesis, not an arbitrage. It says the expected change in the spot rate equals the interest differential, so the high-yield currency should depreciate on average by exactly the carry. If UIP held, the carry trade would earn zero on average.

The forward premium puzzle

Empirically, UIP fails. As Brunnermeier, Nagel and Pedersen put it, “empirically the reverse holds, namely, the investment currency appreciates a little on average, albeit with a low predictive R2 (see, e.g., Fama 1984).” This failure, the forward premium puzzle, is the entire source of carry profits. The return on a carry position over a period is approximately:

carry return = (foreign interest rate - domestic interest rate) - depreciation of the foreign currency

If the high-yield currency does not depreciate on average, you keep the interest gap.

How it is implemented

Institutions mostly use currency forwards or FX swaps. Lustig, Roussanov and Verdelhan build their portfolios from spot and one-month forward rates, including bid-ask spreads.
Futures carry the same economics; Brunnermeier, Nagel and Pedersen use CFTC data on speculators’ net futures positions as a proxy for carry trade activity.
Retail traders usually hold a spot or CFD position overnight and receive or pay a daily rollover (swap), which is the interest differential adjusted by the dealer’s financing markup. The markup works against you on both sides: it reduces the swap you receive on the high-yield side and increases what you pay on the other. Rollover mechanics are covered in transaction costs in retail FX.

The evidence

Brunnermeier, Nagel and Pedersen (2008): crash risk

Markus Brunnermeier, Stefan Nagel and Lasse Heje Pedersen, “Carry Trades and Currency Crashes,” NBER Macroeconomics Annual 2008, volume 23 (published 2009), pp. 313-347. They studied eight major currencies against the US dollar (AUD, CAD, JPY, NZD, NOK, CHF, GBP and the euro, spliced with the German mark before 1999) using 3-month interbank rates from 1986 to 2006.

Key findings:

High interest rates predict negative skewness. Across currencies, average skewness lines up against the average interest differential with an R squared of 81%. The yen, with the most negative differential to the dollar, had the most positive skewness; the Australian and New Zealand dollars, the classic investment currencies, had strongly negative skewness.
Diversification does not remove the crash risk. Their equally weighted long-short carry portfolios had annualized Sharpe ratios of about 0.59 to 0.78, but skewness of -0.54 to -0.98, and the negative skew did not shrink as more currencies were added. For comparison, the US stock market’s Sharpe ratio over the same period was about 0.41 to 0.44.
Speculators are long carry. Net futures positions of non-commercial traders were positively related to the interest differential.
Crashes coincide with funding stress. In weeks when the VIX (equity implied volatility) rose, carry trades tended to lose and speculators cut positions. A higher TED spread had similar effects. They link this to “liquidity spirals”: losses force leveraged traders to unwind, which deepens the losses.
The October 1998 example. They highlight the collapse of the dollar against the yen on October 7 and 8, 1998, which happened without a major fundamental news announcement. In the Fed’s H.10 data the dollar went from 131.15 yen on October 6 to 117.00 on October 9, a fall of almost 11% in three trading days.

Lustig, Roussanov and Verdelhan (2011): a priced global risk factor

Hanno Lustig, Nikolai Roussanov and Adrien Verdelhan, “Common Risk Factors in Currency Markets,” Review of Financial Studies 24, no. 11 (2011), pp. 3731-3777. They sorted up to 35 currencies, developed and emerging, into six portfolios by forward discount (that is, by interest rate) every month from November 1983 to December 2009.

The lowest-yield portfolio had an average forward discount of -297 basis points a year but appreciated only 64 basis points, a log excess return of -233 basis points. The highest-yield portfolio had a forward discount of 901 basis points but depreciated only 282, an excess return of 620 basis points.
After bid-ask spreads, the high-minus-low strategy earned 454 basis points a year with a Sharpe ratio of 0.50. On developed currencies only, the Sharpe ratio was 0.32.
Two factors explain most of the variation in currency portfolio returns: a “dollar” factor (the average return on all foreign currencies) and a carry factor, HML_FX (high minus low interest-rate currencies).
High-yield currencies load more on that carry factor, and the factor is related to changes in global equity market volatility: “High (low) interest rate currencies tend to depreciate (appreciate) when global equity volatility is high.” Their conclusion is direct: “By investing in high interest rate currencies and borrowing in low interest rate currencies, US investors load up on global risk.”

Burnside, Eichenbaum and Rebelo: diversification, peso problems and a dissent

Craig Burnside, Martin Eichenbaum and Sergio Rebelo, “Carry Trade and Momentum in Currency Markets,” Annual Review of Financial Economics (2011), studied 20 major currencies from 1976 to 2010.

An equally weighted carry portfolio earned 4.6% a year with a standard deviation of 5.1% and a Sharpe ratio of 0.89, against 6.5%, 15.7% and 0.41 for the US stock market.
Individual carry trades had the same 4.6% average payoff but 11.3% volatility, a Sharpe ratio of 0.42. Diversifying across currencies “cuts volatility by more than 50 percent.”
Their skewness estimates for the carry portfolio were negative but “not statistically significant,” a more cautious reading than Brunnermeier and coauthors.
They found the payoffs are not explained by conventional risk factors. In “Do Peso Problems Explain the Returns to the Carry Trade?” (Review of Financial Studies 24, no. 3, 2011, pp. 853-891), Burnside, Eichenbaum, Isaac Kleshchelski and Rebelo used currency options to build a hedged carry trade that cannot suffer large losses. Its average payoff was smaller than the unhedged version, and they argue the missing piece is a rare “peso event” in which losses are moderate but they happen in states of the world where money is extremely valuable.
They dated the 2008 disaster for carry to August to November 2008, when the strategy lost about 10%, its worst four months in their sample. Currency momentum gained about 24% over the same months.

The papers disagree about how to describe the risk (crash skewness, a global volatility factor, or a peso event) but agree on the practical point: the average carry return is real, and it is earned by accepting losses at the worst possible times.

Worked example 1: a carry portfolio on Federal Reserve data

We built a simple carry portfolio in Python from primary data: Federal Reserve H.10 daily exchange rates (via FRED, with the German mark from the Fed’s historical H.10 tables before 1999, converted at the official 1.95583 marks per euro) for the euro, yen, pound, Swiss franc, Canadian dollar, Australian dollar and New Zealand dollar, plus the US dollar itself, and OECD monthly short-term interest rates (via FRED: 3-month interbank rates where available; for the yen, the Bank of Japan discount rate before July 1985 and the call rate after; for the Swiss franc, the call rate until March 2024). At each month-end the portfolio goes long the two highest-yielding of the eight currencies and short the two lowest. To avoid look-ahead, both the ranking and the interest earned over the following month use the latest full month’s average rate that would already have been known, which is the month before the formation month. The monthly return is the simple percentage change of each currency against the dollar over the next month plus one-twelfth of its interest differential, averaged across the long pair minus the same for the short pair. It ignores bid-ask spreads and uses interbank rates as a stand-in for forward points, so it is a gross, unlevered return.

Period	Average return	Volatility	Sharpe ratio	Skewness
Feb 1976 to Oct 2026	4.48% a year	10.3%	0.43	-0.88
1986-2006 (BNP’s sample)	6.85% a year	10.9%	0.63	-0.96
2007 to Oct 2026	2.37% a year	10.0%	0.24	-1.02

Average returns are the mean monthly return times 12; volatility is the monthly standard deviation times the square root of 12; skewness is of monthly returns. (Using log exchange-rate changes instead, a common academic convention, gives 4.46% and a skewness of -0.95 for the full period, and shows October 2008 as -17.3 log points.)

The 1986-2006 figures are close to Brunnermeier, Nagel and Pedersen’s two-long, two-short portfolio (Sharpe ratio 0.64 on quarterly data, skewness -0.75). Some 60% of months were positive. The worst months show where carry lives and dies:

Month	Return	What happened
October 2008	-16.3%	Long AUD and NZD, short JPY and CHF during the global deleveraging
July 1986	-14.1%	Long AUD and NZD as the Australian dollar slid, short a rising yen and Swiss franc
October 1987	-12.5%	The global equity crash
January 2015	-10.7%	Short Swiss franc when the Swiss National Bank abandoned its euro floor

The 2008 episode was the worst. From August to October 2008 the portfolio lost 24.1% (31.9% through December). From its peak in mid-2007 to its trough in January 2009 it fell 40.3%, and it did not regain the 2007 peak until February 2013, more than five years later. The January 2015 loss is the subject of the Swiss franc shock of 2015.

Notice also the decline in average returns: 8.3% a year from 1996 to 2005 (Sharpe ratio 0.97), but 2.9% from 2006 to 2015 and 2.1% from 2016 to 2025, with Sharpe ratios of 0.22 and 0.36. The raw material shrank: the average gap between the two highest and two lowest short rates in our eight-currency set was 10.3 percentage points in 1976-1985, 8.2 in 1986-1995 and 5.6 in 1996-2005, but only 2.8 from 2009 to 2021, when policy rates across the developed world sat near zero. It widened again to 3.9 from 2022 to 2026.

Worked example 2: a leveraged AUD/JPY carry trade into 2008

AUD/JPY was the archetypal retail carry pair of the mid-2000s. Using H.10 rates (yen per Australian dollar computed from the two dollar rates) and OECD interest rates:

From January 2005 to December 2007, the average gap between Australian and Japanese short rates was 5.86% a year.
AUD/JPY rose from 80.10 on January 3, 2005 to a peak of 107.36 on July 19, 2007.
From 100.24 on August 1, 2008 it fell to 56.65 on October 27, 2008, a drop of 43.5% in under three months. Peak to trough, the fall was 47.2%.

Now simulate a $10,000 account that buys a fixed AUD/JPY position on January 3, 2005, earns the daily interest differential, and holds it. “Leverage” is the starting notional divided by equity.

Leverage	Equity at end of July 2008	Of which carry	Lowest equity, Oct 2008	Outcome
1x	$14,511	$2,268	$8,945	Three and a half years of gains gone
3x	$23,532	$6,803	$6,834	Lost more than all gains
5x	$32,553	$11,339	$4,724	Down 85% from July in three months
10x	$55,107	$22,678	Below zero	Account wiped out on October 27, 2008

With an illustrative 1% a year financing markup (the spread a dealer might take between the interbank differential and the swap it pays), the 5x account’s July 2008 equity falls to $30,698 and its October low to $2,744, because the markup also eats the carry that was cushioning the position. These are daily noon fixings in New York; intraday prices went lower, and a real broker would have issued margin calls and closed positions before equity reached zero. The leverage rules that govern those margin calls are explained in leverage and margin in forex.

The crash arithmetic

Strip it down to two numbers. At a 5.86% annual carry, a single 43.5% crash costs 7.4 years of interest. Ignoring all other exchange-rate moves, the trade breaks even if a crash of that size happens with a probability of about 13.5% a year, roughly once every seven to eight years. Our eight-currency portfolio, which is diversified and unlevered, had drawdowns of more than 20% in 1978, 1986-1987, 1993-1995 and 2008-2009, plus shorter shocks in 1998 (an 11% loss over August to October), January 2015 and mid-2024; a single cross such as AUD/JPY swings much further. The exact frequency matters less than the shape of the bet: a crash rate well within historical experience is enough to consume the interest, which is what a risk-premium interpretation predicts. The carry looks like free money until you price the crash.

Leverage multiplies both sides. At 5x, the carry is 29% a year and the crash is 218% of equity. A strategy that pays a steady positive swap every night and then loses more than the account in a quarter will look superb in any track record that ends before the crash; this is the same survivorship problem described in drawdown and risk of ruin.

August 2024: the unwind in real time

The most recent large carry crash is documented in BIS Bulletin No 90, “The market turbulence and carry trade unwind of August 2024” (Aquilina, Lombardi, Schrimpf and Sushko, August 2024).

The yen’s depreciation reversed in the first half of July 2024, with signs of carry unwinding on July 24. The stress peaked on August 5, when Japan’s TOPIX index “lost 12% in a single day.”
The BIS estimated the overall size of yen carry trades going into the event at a “rough middle ballpark of ¥40 trillion ($250 billion),” which it said was likely “biased down due to data gaps.” Speculative net short yen futures, about ¥2 trillion, were “only the tip of the iceberg.”
It found that “an unwinding of retail traders’ positions via margin brokerage platforms likely also played a role,” noting that Japanese retail margin traders had mostly been short the yen since the pandemic.
On August 5, the yen appreciated the most of any major currency, followed by the Swiss franc.

In the Fed’s H.10 data, USD/JPY fell from 161.73 on July 10, 2024 to 143.95 on August 5, down 11.0%, and AUD/JPY fell 14.4% over the same dates. Our eight-currency carry portfolio, then long the US and New Zealand dollars and short the yen and Swiss franc, lost 5.8% in July and August 2024. The BIS noted that the episode was “yet another example of volatility exacerbated by procyclical deleveraging and margin increases,” which is the Brunnermeier, Nagel and Pedersen mechanism, almost word for word, sixteen years later.

What can go wrong

Crash timing. Carry losses cluster in global risk-off episodes, when equities fall and volatility jumps. The trade fails exactly when other assets do.
Leverage and margin. Rising volatility raises margins and forces deleveraging, which deepens the crash. Retail accounts are liquidated at the bottom and miss any rebound.
Financing markups. A retail swap is the interest differential minus the dealer’s spread. On a modest differential, the markup can consume most of the carry while leaving the full crash risk.
Policy shocks. Funding currencies can jump on central bank decisions, as the Swiss franc did in January 2015 and the yen did around the Bank of Japan’s July 2024 rate rise.
Regime change. When interest rates converge, as they did from 2009 to 2021, the carry shrinks but the exchange-rate risk does not.
Emerging-market traps. The highest yields are often in currencies with capital controls, wide spreads or devaluation risk. Lustig and coauthors had to delete several emerging-market periods where covered interest parity broke down.
Mistaking a swap-positive track record for skill. A signal or expert advisor that holds high-yield crosses will show a smooth, rising equity curve in calm years. Ask what it did in October 2008, January 2015 and August 2024.

The short version

The carry trade earns the interest gap between currencies because uncovered interest parity fails: high-yield currencies do not depreciate enough, on average, to offset their yield. Diversified carry portfolios earned about 4% to 6% a year with Sharpe ratios of roughly 0.4 to 0.9 in the major studies, after costs in Lustig, Roussanov and Verdelhan’s case. The return is compensation for crash risk: negative skewness, losses when global volatility spikes, and unwinds amplified by leverage. Our Fed-data portfolio fell 40% from 2007 to 2009; a 5x AUD/JPY position lost 85% of its equity in three months of 2008; and the BIS documented the same mechanism in August 2024. This explains how the strategy works and what the evidence shows; it is not a recommendation.