Yen carry trade explained: how it pays, why it unwinds, and the 2026 numbers

A yen carry trade borrows Japanese yen at a low interest rate and buys higher-yielding currencies or assets abroad, pocketing the interest rate differential while betting that the yen will not rise. On 15 October 2026 US policy rates sat about 2.6 percentage points above Japan’s: 3.75% to 4.00% against 1.25%.

In 2026 USD/JPY closed at 163.87 on 23 July, the highest close in TickerLayer’s daily bars back to March 1988, then fell 6.3% to 153.53 by 11 September, first with Japanese intervention and then without. On 15 October it closed at 158.10.

What is a carry trade, and why borrow yen?

A currency carry trade borrows a low-rate currency to hold a high-rate one and earns the difference while the exchange rate holds. The yen is the classic funding leg, "the predominant carry trade funding currency" in a 2024 Bank for International Settlements (BIS) bulletin, and the Bank of Japan (BoJ) kept its rate below zero until March 2024. Any long yen cross is short the yen, so the Japanese yen carry trade runs through USD/JPY, AUD/JPY, CAD/JPY and MXN/JPY.

A USD/JPY carry trade example

Over those 50 days the currency move was more than 17 times the carry. From 2 January 2024 to 15 October 2026 the two were about equal, +11.4% and +11.0%. In a trading account the carry arrives as rollover, or forex swap, booked on positions held past 5:00 p.m. New York time, the daily cut-off of forex market hours.

Is there a yen carry trade today?

Yes, at about half the 2024 gap: the Fed’s target-range midpoint minus the BoJ rate was 5.475 points on 2 January 2024 and 2.625 on 15 October 2026. Since July 2024 the BoJ has climbed from 0.25% to 1.25% in four steps of 25 basis points; the latest, in September 2026, only offset a Fed hike two days earlier.

Market yields agree. The bond snapshot endpoint returned "rate": 4.77 for US:2Y on "date": "2026-10-15", against 1.913% for Japan’s 2-year bond in Ministry of Finance data. Carry lives at the short end of the yield curve, where the US gap is widest:

Carry-to-risk, the gap divided by annualized volatility, is a Sharpe ratio built from interest alone. With 2026 volatility (January to 15 October) it was 0.33 for USD/JPY (2.625 ÷ 7.99%), 0.41 for AUD/JPY and 0.57 for MXN/JPY.

Why is the yen so weak?

Partly because the yen still pays less to hold than the currencies it is sold for. The Bank of Japan’s interest rate, around 1.25% since 24 September 2026, sits 1.00 to 5.25 points below the six other policy rates charted above. It is below inflation too: the BoJ’s 18 September statement put core inflation "in the range of 1.5-2.0 percent" and said "real interest rates have remained at low levels". The dollar bought 11.4% more yen on 15 October 2026 than on 2 January 2024, even as the gap halved:

Why BoJ rate hikes did not lift the yen

Each 2026 hike still left a 2.625-point dollar gap, and USD/JPY rose on both decision days: 0.07% on 16 June, the hike to 1.0%, and 0.59% on 18 September, the hike to 1.25%. July 2024 was different: the BIS says the hike to 0.25% on 31 July was "perceived as a hawkish rate hike", and the pair fell 1.84% that day, 1.88% on 2 August and 1.60% on 5 August.

Yen intervention: what Japan bought and what USD/JPY did

Japan’s Ministry of Finance (MOF) buys yen by selling dollars and publishes the amounts later: ¥15.3 trillion on four days in 2024, ¥11.7 trillion on three days in April and May 2026, and ¥15.4 trillion from 30 July to 26 August. On 3 August 2026 Finance Minister Satsuki Katayama said Japan had bought yen on 31 July, US Eastern time, in coordination with the US Treasury, citing "excessive volatility and disorderly movements".

Intervention days show up as wide ranges: 3.15% to 3.54% from high to low on 29 April 2024, 1 May 2024 and 30 April 2026, against a 0.72% median since January 2024. From 1 to 11 September USD/JPY fell 4.16% with no buying reported, while USDX, the dollar index, fell only 0.55%: the move was the yen’s.

Yen carry trade unwind: 1998, 2008, 2024 and 2026 in numbers

An unwind is carry positions closing together: the yen jumps, and margin calls on leveraged positions, from hedge funds to Japanese retail accounts, force more of them shut. The BIS put the trade going into August 2024 at "a rough middle ballpark of ¥40 trillion ($250 billion)" and judged that month’s exchange-rate moves "not outsized when compared to past well-known carry crashes". Two such crashes are in the same daily bars:

The 2024 stress peaked on 5 August, when Japan’s broad stock index lost 12% in a single day and the selling spread to other assets, by the BIS’s account. In TickerLayer’s daily bars the US500 index fell 3.00% that day and BTCUSD 7.12%. The 2026 fall was the smallest in the table, yet its worst day, -2.38% on 30 July, was sharper than any USD/JPY day of the 2024 unwind.

Carry trade risk: one bad day outweighs months of carry

Held from 23 July to 11 September, a long position earned about 59 USD/JPY pips in carry and lost 1,034 on the exchange rate.

How to spot a yen carry trade unwind in daily bars

unwind_check.py pulls daily forex bars for seven yen crosses with the Python 3.9+ standard library and a TICKERLAYER_API_KEY environment variable.

"""Yen carry unwind check from daily forex bars: drawdowns, co-movement and all-down days.

Standard library only (Python 3.9+). Set TICKERLAYER_API_KEY. One request per pair.
"""
import itertools
import json
import math
import os
import statistics
import urllib.request
from datetime import date, datetime, timezone

API = "https://api.tickerlayer.com"
PAIRS = ["USDJPY", "AUDJPY", "NZDJPY", "CADJPY", "MXNJPY", "EURJPY", "GBPJPY"]


def daily_closes(symbol, start, end):
    """UTC date -> close for weekday sessions."""
    closes, offset = {}, 0
    while offset is not None:
        url = f"{API}/forex/agg/{symbol}/1/day/{start}/{end}?limit=5000&sort=asc&offset={offset}"
        req = urllib.request.Request(url, headers={"x-api-key": os.environ["TICKERLAYER_API_KEY"]})
        with urllib.request.urlopen(req, timeout=30) as res:
            body = json.load(res)
        for bar in body["results"]:
            day = datetime.fromtimestamp(bar["t"] / 1000, tz=timezone.utc).date()
            if day.weekday() < 5:  # daily forex bars include some weekend rows: drop them
                closes[day] = bar["c"]
        offset = body.get("next_offset")
    return closes


def corr(x, y):
    mx, my = statistics.mean(x), statistics.mean(y)
    num = sum((a - mx) * (b - my) for a, b in zip(x, y))
    return num / math.sqrt(sum((a - mx) ** 2 for a in x) * sum((b - my) ** 2 for b in y))


def fall(closes, a, b):
    """Highest close in [a, b] to the lowest close after it, in percent."""
    days = [d for d in sorted(closes) if a <= d <= b]
    peak = max(days, key=closes.get)
    trough = min((d for d in days if d > peak), key=closes.get)
    return (closes[trough] / closes[peak] - 1) * 100


closes = {p: daily_closes(p, "2023-12-31", "2026-10-15") for p in PAIRS}
days = sorted(set.intersection(*(set(c) for c in closes.values())))  # sessions all 7 share
rets = {p: {d: math.log(closes[p][d] / closes[p][prev]) for prev, d in zip(days, days[1:])}
        for p in PAIRS}

episodes = {"Jul-Aug 2024": (date(2024, 7, 1), date(2024, 8, 9)),
            "Jul-Sep 2026": (date(2026, 7, 1), date(2026, 9, 15))}
print("fall from high close, %".ljust(24) + "".join(f"{e:>14}" for e in episodes))
for p in PAIRS:
    print(p.ljust(24) + "".join(f"{fall(closes[p], a, b):>14.1f}" for a, b in episodes.values()))

windows = {"all sessions 2024-2026": (date(2024, 1, 2), date(2026, 10, 15)),
           "run-up, May 11-Jul 22 2026": (date(2026, 5, 11), date(2026, 7, 22)),
           "unwind, Jul 11-Aug 9 2024": (date(2024, 7, 11), date(2024, 8, 9)),
           "unwind, Jul 23-Aug 7 2026": (date(2026, 7, 23), date(2026, 8, 7)),
           "unwind, Aug 31-Sep 15 2026": (date(2026, 8, 31), date(2026, 9, 15))}
print("\naverage pairwise correlation of daily returns")
for name, (a, b) in windows.items():
    win = [d for d in days[1:] if a <= d <= b]
    avg = statistics.mean(corr([rets[x][d] for d in win], [rets[y][d] for d in win])
                          for x, y in itertools.combinations(PAIRS, 2))
    print(f"  {name:<28}{avg:.2f} ({len(win)} sessions)")

all_down = [d for d in days[1:]
            if max(rets[p][d] for p in PAIRS) < 0 and statistics.mean(rets[p][d] for p in PAIRS) < math.log(0.985)]
print("\nall 7 pairs down, average fall over 1.5%:")
print(", ".join(d.isoformat() for d in all_down))

fall from high close, %   Jul-Aug 2024  Jul-Sep 2026
USDJPY                           -10.8          -6.3
AUDJPY                           -14.1          -4.0
NZDJPY                           -13.3          -6.3
CADJPY                           -12.2          -4.9
MXNJPY                           -18.7          -4.5
EURJPY                            -9.9          -5.0
GBPJPY                           -11.8          -5.4

average pairwise correlation of daily returns
  all sessions 2024-2026      0.73 (727 sessions)
  run-up, May 11-Jul 22 2026  0.50 (53 sessions)
  unwind, Jul 11-Aug 9 2024   0.89 (22 sessions)
  unwind, Jul 23-Aug 7 2026   0.88 (12 sessions)
  unwind, Aug 31-Sep 15 2026  0.90 (12 sessions)

all 7 pairs down, average fall over 1.5%:
2024-05-01, 2024-07-11, 2024-07-31, 2024-08-02, 2024-08-05, 2024-09-27, 2025-08-01, 2026-04-30, 2026-07-30, 2026-09-03

The 2026 falls were smaller, with peaks spread from 15 July (GBP/JPY) to 1 September (MXN/JPY). The order flipped too: MXN/JPY fell most in 2024, while in 2026 the two highest yielders, MXN/JPY and AUD/JPY, fell least.

Correlation alone cannot tell an unwind from a rally: a rolling 20-session average of the correlation matrix hit 0.89 on 9 August 2024, as the crosses fell 9.0%, and on 24 October 2025, as they rose 1.8%. Falling together is common too: all seven fell on 133 of 727 sessions. Adding an average fall over 1.5% leaves the ten printed dates, among them three intervention days (1 May 2024, 11 July 2024, 30 April 2026), the BoJ hike of 31 July 2024 and 5 August 2024.