Quant Memo
Advanced

Building a Carry Basket and Its Crash Risk

How a cross-sectional carry basket is actually built — rank, weight by risk, neutralise — and the arithmetic of the crash that pays for the premium, worked through the August 2024 yen unwind.

Prerequisites: Carry, Vol Targeting

Carry is the return you earn if prices simply stay where they are. A basket turns that into a portfolio: rank a universe by carry, own the top, fund it by shorting the bottom, scale to a risk target. It is one of the most reliable premia in finance and one of the most brutally distributed.

One definition across assets

Carry is always the slope of the forward curve — what you earn from the passage of time alone:

Ci=FnearFfarFnear×12mC_i = \frac{F_{\text{near}} - F_{\text{far}}}{F_{\text{near}}} \times \frac{12}{m}

In words: the percentage gap between the near and far contract, annualised over the mm months between them. In FX that gap is the interest-rate differential; in bonds, yield plus roll-down; in commodities, backwardation; in equity index futures, dividend yield minus financing. One formula, one ranking, every asset class comparable.

Worked example: a G10 FX basket

Take mid-2024 one-year rates, rank the ten currencies, and go long the top three against the bottom three, equally weighted within each side.

Long legRateShort legRate
NZD5.50%JPY0.10%
USD5.33%CHF1.25%
GBP5.25%SEK3.75%
Average5.36%Average1.70%
  1. Gross carry: 5.361.70=3.665.36 - 1.70 = 3.66 percentage points a year at one unit per side.
  2. Basket volatility: a three-against-three G10 basket typically realises about 7% annualised; the pairs are 8–10% vol individually but partially offset.
  3. Scale to target: at a 10% volatility target, leverage is 10/71.4310/7 \approx 1.43, lifting expected carry to 3.66×1.435.23.66 \times 1.43 \approx 5.2 points.
  4. Implied Sharpe: if spot rates were a pure random walk, 5.2/10=0.525.2 / 10 = 0.52 before costs.

That last line is the whole trade. Uncovered interest parity says step 4 should be zero: high-rate currencies ought to depreciate by exactly the differential. Empirically they do not, or not enough, and the gap is the premium.

Weight by risk, then neutralise

Equal notional is not equal risk. Weight each leg by 1/σi1/\sigma_i so a 15%-vol currency gets half the notional of a 7.5%-vol one, and cap any single leg near 20% of gross so one position cannot decide the year. Then check what the basket is accidentally long of: high-carry currencies are overwhelmingly commodity exporters and emerging markets, so a naive basket is a leveraged bet on global growth in an FX costume. Most desks demean the scores cross-sectionally and neutralise the residual global-equity exposure.

A carry basket is not a collection of independent bets. High-carry legs fall together and funding legs rally together, so the effective breadth of a ten-currency basket is closer to two positions than ten.

What you are actually short

Carry returns are negatively correlated with volatility shocks. Drag the slider below to about ρ=0.6\rho = -0.6 — realistic for monthly carry returns against changes in equity volatility — and look at the fit line. Carry loses in exactly the months the rest of the portfolio loses, which is why it is a risk premium and not an anomaly.

Correlation explorer
X →Y ↑
ρ = -0.60r² = 0.36relationship: moderate negative

Brunnermeier, Nagel and Pedersen showed the mechanism: high-carry currencies have negative skew and their crashes cluster with funding-liquidity shocks. Leverage builds quietly while volatility is low, a shock forces deleveraging, everyone unwinds the same position at once, and the move overshoots. Carry indices lost roughly 30% in the second half of 2008.

The arithmetic of an unwind: yen, August 2024

By July 2024 the short-yen trade was the most consensual position in global macro, funded at 0.10% against dollar rates above 5%. USD/JPY then fell from about 161.8 in mid-July to about 141.7 on 5 August — roughly 12% in three weeks. The Nikkei dropped more than 12% in a single session and equity volatility spiked above 60. The accounting on one short-yen-versus-dollar unit:

  • Monthly carry collected: 5.2/120.435.2 / 12 \approx 0.43 percent.
  • Loss in the unwind: about 12 percent.
  • Months of carry destroyed: 12/0.432812 / 0.43 \approx 28.

Nearly two and a half years of carry gone in three weeks — and that is the funding leg alone, before the long AUD and NZD legs sold off alongside it. That is what "picking up nickels in front of a steamroller" means quantitatively: a high hit rate, a fine Sharpe on monthly data, and a left tail that arrives once a decade and takes the decade back.

Do not size a carry basket off realised volatility alone. Volatility is lowest right before an unwind, precisely because the crowded position suppresses it, so a naive volatility-targeting rule levers you into the crash. Cap leverage independently of the volatility estimate.

Living with the tail

Three defences, none free. Buy options on the funding currency — effective and expensive, since out-of-the-money yen calls can eat most of the carry. Overlay trend, which flips short the high-carry currency as it falls and historically cuts carry's worst drawdowns roughly in half; this is why carry-and-trend blends are a standard product. Cut on funding stress, when cross-currency bases widen or volatility term structure inverts.

In interviews

Define carry as the forward-curve slope, note that the same formula ranks FX, bonds and commodities, and build the basket in four steps: rank, long-short, risk-weight, scale to a volatility target. Pre-empt the obvious challenge — why is this not free money — with the skew story, and have the yen arithmetic ready: 0.43% a month collected, 12% lost in three weeks, about 28 months of carry erased. Finish with the trend overlay and why it works: trend goes short the very currency carry is long.

Related concepts

Used in strategies

Practice in interviews

Further reading

  • Koijen, Moskowitz, Pedersen & Vrugt (2018), Carry
  • Brunnermeier, Nagel & Pedersen (2008), Carry Trades and Currency Crashes
  • Lustig, Roussanov & Verdelhan (2011), Common Risk Factors in Currency Markets
ShareTwitterLinkedIn