Brazilian Real Carry Trade: 20 Years, Hedged and Unhedged

The earlier articles in this series priced the hedge—how the forward is built, what it costs and which instruments carry it. This one asks whether it was worth paying—the question behind every Brazilian real carry trade. Take a dollar investor who, every quarter since September 2006, put money into Brazilian interbank deposits for three months and then brought it home. That is the simplest form of every real-denominated yield product sold to dollar holders, from tokenized receivables to real-pegged stablecoins. Run it two ways: hedged each quarter with a three-month forward, or left open to the currency. The data are B3’s curves and the Federal Reserve’s daily exchange rates, September 2006 to mid-2026.

Both columns are true at once, and the rest of this article is about why. On average, leaving the currency open paid. In the tail, it cost an investor who arrived at the wrong moment more than a decade.

What each strategy actually earns

The hedged investor earns the onshore dollar rate. As the first article in this series showed, a hedge turns the real’s interest rate into the dollar rate available inside Brazil: here it ranged from 0.4% a year (2021) to 7.2% (2008) and never produced a losing quarter. The unhedged investor earns the real’s interest rate, about 10.1% a year over the period, minus whatever the real lost against the dollar, about 4.5% a year. Compounded, that gives 5.4%. The hedged 3.1% is worth comparing with what the same dollars would have earned at home: over these windows the onshore dollar rate averaged about 1.2 points more than the US three-month interbank rate. That is the onshore premium described in the hedge-cost article, and it is the only part of the Brazilian rate a hedged investor keeps.

The price of choosing the first column over the second is the hedge cost: the annualized forward premium, which is mostly the gap between the two interest rates. Over twenty years it moved a lot:

A three-month hedge cost a median of 7.3% a year, as little as 0.6% in November 2020 when the Selic was at 2%, and as much as 13.1% in June 2015. On 2 October 2026 it was 6.66%, cheaper than in 57% of the months since 2006.

On average, carry won

If forward rates were forecasts, the two strategies would earn the same on average: the forward would have priced in exactly the depreciation that followed. They did not behave like forecasts. Over the 237 three-month windows, forwards priced in an average rise of the dollar against the real of 6.6% a year; the dollar actually rose 4.8% a year (both log averages over the windows). The real held up better than the interest gap implied, and the unhedged investor kept the difference. In 141 of the 237 windows, 59.5%, the open position beat the hedged one.

This is not a Brazilian quirk. It is the forward premium puzzle: across currencies and decades, high-interest currencies depreciate less than forwards imply, and sometimes appreciate (Fama, 1984). The standard test regresses the realized move on the forward premium; a forecast-like forward gives a slope of 1. On these Brazilian windows the point estimate is −0.10. It comes with a wide error band: twenty years of one currency cannot statistically separate carry’s edge from zero, which is why the cross-country evidence carries more weight than this single series. The strategy that harvests the gap is the carry trade: hold the high-interest currency, unhedged, and collect the difference.

In the tail, carry lost years in months

The same data, cumulated, tell the second half of the story.

The unhedged dollar more than doubled by mid-2011, then fell 41% on a quarterly basis—45% marked to market daily, from July 2011 to September 2015—and climbed back above its 2011 level only in June 2025. Over the same years the hedged line rose slowly, without a single down quarter. The worst three-month window began in August 2008: a $1,000,000 round trip lost $268,000 unhedged and earned $10,400 hedged. Measured on spot alone, the real lost 47% of its dollar value between April 2014 and September 2015, and 32% between January and May 2020.

The shape of the gains explains why the average and the tail disagree. Relative to the hedge, the open position won in six windows out of ten, by an average of 5.8 points per quarter, and lost in four, by an average of 6.7 points. The distribution is mildly skewed to the left: wins come more often, losses are larger on average, and the single worst quarter (−27%) is bigger than the best (+21%). The 16 windows in which the unhedged investor lost more than 10% in three months cluster in five episodes: 2008, 2011–12, 2014–15, 2018 and early 2020. That is the signature of carry everywhere. Carry returns are negatively skewed because the trade unwinds all at once when risk appetite and funding dry up (Brunnermeier, Nagel and Pedersen, 2008).

Twenty years in five regimes

The averages hide regimes that lasted years. Split the same quarterly chain at its turning points and the two strategies look like different asset classes:

An investor’s experience of “carry” depends almost entirely on which of these rows they arrived in. Someone who entered in late 2006 and left in mid-2011 saw a product that paid nearly 20% a year in dollars. Someone who arrived in mid-2011 spent four years losing more than a tenth of their money a year. The hedged column barely notices the regimes: its return tracks the onshore dollar rate, low when US rates were near zero and higher since 2022.

There is a third choice between the two columns. Hedging half the position each quarter would have earned 4.6% a year over the whole period, with a worst fall of 19% instead of 41%. It kept most of the carry and halved the drawdown, but it still spent 2011–2017 below its earlier peak. A partial hedge does not remove the tail; it resizes it, which is often what a product actually needs.

Pick an entry month

The explorer replays every one of the 237 three-month round trips. Each bar is the unhedged result minus the hedged one; green means carry won. It opens on the worst window; the buttons jump to the best one and to the latest. For each entry month the cards show both results on $1,000,000, the hedge cost the forward charged on the day of entry, and where the exchange rate started and ended. Comparing the last two is the forward premium puzzle in miniature: the cost is known at entry, the move is not, and the two are only loosely related.

Drag through 2009–2010 and 2016 to see carry at its best; through 2008, 2015 and early 2020 to see what “on average” leaves out.

How backtests like this one mislead

Averaging annualized quarters

The arithmetic mean of the 237 unhedged three-month returns, each annualized, is 10.1% a year. That number is an artifact: raising a quarterly move to the fourth power inflates gains more than it shrinks losses, and the average of those inflated figures says little about what an investor earned. The geometric mean across the same windows is 4.9%; the compounded quarterly chain gives 5.4%. Any “historical carry yield” quoted without saying which of these it is should be read with suspicion.

Calendar choices

Small timing choices move a twenty-year result by more than a point a year. A version that rolls 90-day deposits from each quarter-end, and so exits a day or two before the next one, earns 3.9% a year unhedged, against 5.4% for links that run exactly from quarter-end to quarter-end. The real tends to firm in the last days of a quarter, and missing those days repeatedly adds up. A backtest is only as good as its stated calendar.

Start years

By year of entry, the open position lost money on average in 2008, 2012–2015, 2018–2020 and 2024, and made large gains in 2007, 2009–2010, 2016, 2021–2023 and 2025. A five-year track record can be chosen to tell either story. Twenty years are needed to see both.

Gross of taxes and costs

Every number here is gross: before Brazil’s tax on foreign inflows (as high as 6% on fixed income in 2010–11), withholding tax, bid-ask spreads and fees. They lower both columns and do not change the shape of the comparison, but a product’s actual track record will sit below these lines.

“Hedged” is not “risk-free”

The hedged column carries Brazilian bank credit, the onshore dollar basis described in the hedge-cost article, and the margin or credit lines that a forward requires. It removes the currency; it does not remove everything.

# Inputs per month-end t (B3 TaxaSwap + FRED H.10):
#   pre90, du90 — DI "pre" rate and business days at 90 calendar days
#   doc90       — onshore dollar rate (cupom cambial) at 90 days, ACT/360 simple
#   s_t, s_t90  — BRL per USD at t and at the first date on or after t + 90 days
def round_trip(pre90, du90, doc90, s_t, s_t90):
    brl_growth = (1 + pre90) ** (du90 / 252)       # deposit in reais
    hedged = doc90 * 90 / 360                      # forward locks the onshore dollar rate
    unhedged = s_t * brl_growth / s_t90 - 1        # convert back at the future spot
    return hedged, unhedged

# Frequency statistics use all 237 overlapping monthly windows;
# cumulative figures chain non-overlapping quarter-end to quarter-end links.
# Geometric mean: exp(mean(log(1 + r_3m)) * 365 / 90) - 1, not mean(annualized r).

Who holds the tail in an RWA structure

A real-denominated yield product sold to dollar holders always contains this tail. The design question is whose balance sheet it sits on.

• The investor, unhedged. A real-pegged stablecoin or a vault that reports in dollars without hedging hands the full distribution to the holder: about 5% a year on average over twenty years, and a 45% fall that took fourteen years to recover. That is a legitimate product, but its disclosure should show the drawdown history, not only the average.
• The issuer, hedged. A structure that hedges pays the forward premium—6.7% a year at the three-month price on 2 October 2026—and passes on the dollar rate plus the asset’s credit spread. The tail moves into roll risk, margin and counterparty lines.
• A first-loss tranche. Tranching protects against defaults. It is a poor substitute for a hedge, because the shocks that crash the real tend to hit Brazilian credit at the same time.
• The borrower, in dollars. A USD-native loan moves the currency to the borrower, where it reappears as credit risk; our agri-debt adjudication prices that route.

The choice between hedge and carry is not a forecast of the real. It is a choice about who holds the left tail, and whether the people holding it know it is there. The four funding levers of an RWA deal are where that choice gets written into the structure. Funding in yuan or rand instead of dollars changes the partner currency, not the question; the final article in the series prices those hedges.

The takeaway

Over twenty years a dollar investor in Brazilian deposits earned 5.4% a year unhedged and 3.1% hedged. The open position won because forwards priced in more depreciation than happened: 6.6% a year against 4.8%, the forward premium puzzle in Brazil’s own data. It also lost 45% from 2011 to 2015 and needed until 2025 to recover, with its worst quarters bunched into five crises. Neither column is the right answer in general. The right answer for a product is the one whose holder can carry the distribution it is given—and that is a question of disclosure and structure, not of a view on the real.