Carry trade
A Socratic walk-through of the carry trade — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does money rush toward whichever country pays more interest, and why does that stop?
Two countries pay different interest rates on safe short-term money. One pays close to nothing; the other pays several percentage points more. Capital notices, and moves.
That much sounds obvious — water runs downhill. But it should not be obvious at all, because the money has to change currency to make the trip, and it must change back to come home. If everyone can see the gap, why has the exchange rate not already moved to cancel it? And once the flow does start, why does it run for years and then reverse over a few days rather than easing gently to a stop?
Reasoning it through
REASONING #Think of the interest differential as a gradient, and ask what a gradient normally does when something is free to flow along it. It gets flattened. The flow itself carries whatever equalises the two ends.
So look for the equalising force here. There is one, and it has a name: uncovered interest parity. If the high-rate currency is expected to weaken over the period by exactly the interest gap, then the extra interest is precisely offset by the loss on converting back, and there is nothing to collect. Under that condition the gradient is an illusion — it exists in interest but not in return.
Now test whether the force operates, and be careful, because there are two versions of the question. Hedge the currency by locking in the return rate in the forward market and the gain vanishes exactly. That is covered interest parity, held in place by arbitrage: a bank can borrow, convert, invest and sell forward at once, riskless. The hedged gradient really is flat.
The unhedged one is a different matter. If you simply take the currency risk, the empirical record is that high-interest currencies have, on average, not depreciated by the amount the differential implied — sometimes they appreciated. Economists have argued about this since the early 1980s under the name of the forward premium puzzle, and it is still not settled why it persists. So the honest summary is: the equalising force is not reliably observed, and nobody has a fully agreed explanation for its absence.
That is enough to explain the rush. If the gradient is not flattened by expected depreciation, then borrowing where money is cheap and lending where it is dear pays — and it pays without needing any view about the future, which is what makes it so easy to scale with leverage.
But now ask what the flow does to its own environment. Money moving in buys the high-rate currency, which pushes it up, which adds a capital gain on top of the interest and makes the trade look better still. The flow steepens the very gradient it is running down, at least for a while. Any student of feedback should be uneasy at that point, because a self-reinforcing flow does not have a gentle stopping condition.
So what stops it? Not the differential closing — that would be the gentle version. What stops it is the funding side. The trade is typically levered, and leverage is sized to volatility. When volatility jumps, or the cheap currency's central bank raises its rate, or a lender asks for more collateral, positions must be cut. Cutting means selling the high-rate asset and buying back the cheap currency — which pushes the cheap currency up, which worsens everyone else's position in the same trade, which forces more cutting. The unwind is the same feedback loop running backwards, and it runs much faster, because a margin call has a deadline and a yield does not.
That asymmetry is the trade's defining shape: returns that accumulate slowly and losses that arrive in a rush. It has been described as going up by the stairs and down in the elevator. The yen has played the cheap-money role repeatedly — in the 2008 unwind it appreciated sharply, and again in early August 2024, when a Bank of Japan rate rise coincided with weak United States data and a rapid deleveraging rippled through global markets.
The analogy
THE ANALOGY #Imagine two reservoirs at different heights, joined by a pipe with a valve.
Open the valve and water runs from the high pool to the low one. The flow is steady, the drop is real, and while it lasts the operator is paid per litre. But the pipe is not the only thing connecting them: the pumps that keep the upper pool topped up are rented, and the rent rises whenever the weather turns. When it does, everyone shuts their valve at once and drains their pipe back — and the surge of water going the wrong way is far more violent than the placid flow that preceded it.
Water never doubts the height difference, whereas here the "height" is partly a belief about future exchange rates, so the gradient can vanish because opinion changed rather than because anything physically levelled out — which is why the reversal can begin with no change at all in the interest rates that supposedly drive it.
Clarifying the model
THE MODEL #The carry trade is not a free lunch that markets have failed to notice. Its average return has been positive, but the distribution is strongly negatively skewed: many small gains, occasional very large losses. That is a recognisable payoff for bearing a risk, not evidence of an oversight, and one serious reading is that the carry return is compensation for exactly the crash risk that shows up in the unwinds.
Hedged and unhedged are different animals, and conflating them is the common error. The hedged version is arbitraged flat by construction. Everything interesting about the carry trade lives in the decision not to hedge — so anyone describing it as riskless has quietly deleted the only part that pays.
The mechanism of the puzzle is genuinely open. Competing accounts include a risk premium, slow-moving capital, the peso problem of rare disasters, and limits to arbitrage. They are not all the same claim, and this walk-through does not settle between them; what is well documented is the empirical pattern, not its cause.
The falsification test. If the unwind is driven by funding and leverage rather than by the differential closing, then reversals should coincide with volatility and margin conditions, not with the yield gap narrowing. In the sharp episodes, spot moves have led rather than followed the rate change, which is what the funding account predicts and the parity account does not.
A picture of it
THE PICTURE #How to readRead top to bottom as one round trip. The first four messages are the calm phase; the first note is the self-reinforcing part. Everything below the dashed funding shock reverses direction, and the final note is why the exit is crowded: every escape route runs through the same currency purchase.
What became clearer
WHAT CLEARED #A gradient free to be flowed along should flatten, and the striking fact about interest differentials is how often they do not — which is why the flow starts and why it runs for years. The reversal is not the gradient finally equalising; it is the plumbing failing. Separate the pull from the constraint, and the trade's rhythm of slow gains and abrupt losses stops looking like irrationality and starts looking like the shape the constraint imposes.
Where to go next
ONWARD #- Why covered interest parity itself has shown persistent deviations since 2008, when arbitrage was supposed to prevent them.
- How carry behaves in emerging-market currencies, where capital controls change the plumbing.
Key terms
TERMS #| Term | What it means |
|---|---|
| Uncovered interest parity | the proposition that a currency with a higher interest rate should be expected to depreciate by the amount of the differential. |
| Covered interest parity | the same relationship with the currency risk hedged in the forward market, enforced by arbitrage. |
| Forward premium puzzle | the long-observed failure of high-interest currencies to depreciate as uncovered parity predicts. |
| Funding currency | the low-interest currency borrowed to finance the position, historically often the yen or the Swiss franc. |
Every term the collection defines is gathered in the glossary.