Carry, and what breaks it
M4 — Macro & rates
The last of the roadmap's three self-assessment targets, and the idea that ties M4 back to everything before it.
A carry trade earns the difference in yield between what you borrow and what you hold. It makes money as long as the price of the thing you hold doesn't move enough to wipe out that difference.
Two sentences, which is what the roadmap asked for. The rest of this lesson is why that innocuous-sounding trade is one of the most reliably profitable and periodically catastrophic structures in finance.
The classic form
BORROW yen at 0.10% INVEST in dollars at 5.25%
└──────────────────────────────┘
carry = +5.15% a year, for doing nothing
the catch: you are SHORT yen. If the yen strengthens 5.15%
against the dollar, a full year of carry is gone.
You are being paid to hold a currency position, and the payment accrues smoothly, daily, whether or not anything happens. That is the seduction: the P&L looks like a savings account right up until it doesn't.
Note how little movement it takes to break even — at 5.15% a year, roughly 0.43% a month. Currencies move that in a day.
Why theory says it shouldn't work
Uncovered interest parity (UIP) is the no-arbitrage argument: if dollars yield 5% more than yen, the dollar should be expected to depreciate about 5% against the yen, exactly cancelling the carry. Otherwise there is free money, and free money should not persist.
Empirically, UIP fails — and not marginally. High-yielding currencies have historically depreciated less than the differential, and often appreciated. Carry has earned positive average returns across decades and currency pairs. This is the forward premium puzzle, one of the most durable anomalies in international finance.
The resolution most people accept is a risk story, and you already have the shape for it from M2.3:
carry P&L, drawn over time
▲ ╱
│ ╱────────────────────╱─────────────╱
│ ╱ │
│ ╱ │
└────────────────────────────────────────────┼──▶
small steady gains, month after month │
▼
sudden violent loss
this is a SHORT VOLATILITY payoff — the M2.3 short-gamma shape,
the M2.4 variance risk premium, and M0.8's warning about
selling options, all wearing a currency costume
Carry is not free money; it is compensation for bearing crash risk. You are paid a steady premium to hold a position that occasionally loses several years of income at once. Whether that is a good trade depends entirely on whether the premium exceeds the crash cost, which is unknowable in advance and is why the trade never disappears.
August 2024, the yen
The yen-funded carry trade was the biggest example of the 2020s. Japan held rates near zero while the Fed hiked, so the differential became enormous, and the yen obligingly weakened for three years — carry traders earned the differential and a capital gain:
USD/JPY (yen per dollar; higher = weaker yen = carry winning)
2021-01-04 103.19 ╮
2022-01-03 115.27 │ three years of the trade working
2023-01-03 130.83 │ perfectly — carry AND appreciation
2024-01-02 141.89 │
2024-07-03 161.48 ╯ ← peak
2024-08-05 143.95 ← 33 days later
The yen appreciated 12.2% in five weeks; a yen-funded dollar position lost about 11% on the currency alone. At a 5.15% annual carry, that is more than two years of accumulated income erased in thirty-three days — and it happened alongside a global equity selloff, so the same investors were losing elsewhere simultaneously.
The trigger was mundane — a small Bank of Japan hike plus a weak US payrolls print. What made it violent was crowding and leverage: everyone held the same position, the unwind forced buying of yen, which strengthened the yen, which forced more unwinding. That feedback loop is M0.7's margin spiral and M3.5's forced-selling mechanism, in a currency.
Carry is everywhere
Once you see the shape, it recurs across every asset class, and recognising it is most of what this lesson is for:
FX carry borrow low-yield currency, hold high-yield
Credit carry hold corporate bonds, earn the spread over
governments — until defaults arrive
Curve carry borrow short, lend long, earn the term premium
(M4.4) — this is banking, and it is why an
inverted curve hurts banks
Vol carry sell options, collect the variance risk premium
(M2.4) — the most explicit version
Equity carry high-dividend stocks, funded
Every one has the same profile: steady income, rare large loss, and a tendency for everyone to be in it at once because the Sharpe ratio looks wonderful right until the drawdown. And because they are all short the same underlying risk factor, they tend to lose together — which is M3.5's tail dependence, arriving from a different direction.
When the regime changes
The deepest version of this is not a trade but an assumption. The largest carry-like position in the world is the belief that bonds hedge equities — the 60/40 portfolio, and every risk-parity structure built on it.
S&P 500 daily returns vs 10-year bond returns
window correlation
──────────────────────────────────────
2016–2019 (pre-Covid) −0.35
2020–2021 −0.37
2022 (inflation shock) +0.18
2023–2026 +0.05
For decades, bonds rose when equities fell — the negative correlation that makes 60/40 work. In 2022 the sign flipped, and it has not returned to its old level since.
The reason is exactly M4.5. When the dominant risk is growth, bad news hurts equities and helps bonds (the central bank will cut) — correlation negative. When the dominant risk is inflation, bad news hurts both, because the response is higher rates. Change what the economy is afraid of and you change the sign of the correlation that a whole industry's risk models assumed was structural.
Which is the closing point of M4, and it is really a point about M3:
A correlation estimated in one regime is not a property of the assets. It is a property of what the economy was worried about at the time.
Howard Marks' formulation, in the roadmap's recommended reading, is that you cannot predict cycles but you can know where you are in one — and that knowing whether growth or inflation is the dominant risk tells you more about how assets will co-move than any amount of historical covariance.
M5 turns from the market to the people trading it: what the major firms actually do, how they make money, and how they are organised and paid.
Source: Howard Marks, Mastering the Market Cycle — the roadmap’s pick, and the right register for this lesson. On the forward premium puzzle, Burnside et al., “The Returns to Currency Speculation”, is the readable empirical treatment. FX and correlation figures above from FRED series DEXJPUS, SP500 and DGS10.