Did Brazil's carry protect against the real's depreciation?

Brazilian cash carry accumulated steadily across the sample, while exchange translation moved in both directions and ended as the main offset (Fig. 1). Cumulative log points put continuously compounded interest and currency components on one additive scale. Subtracting US cash still left a positive cumulative log return, which produced the 4.25% annualized excess result. The long horizon matters because the path contained several deep reversals before that endpoint.

The average ordinary 21-business-day window earned 0.33% above US cash, and 55.2% of those windows were positive. That modest monthly edge shows compensation was not a steady shield for principal.

The pre-specified regime split classifies each 21-business-day block by conditions on its entry date. Ordinary starts earned 0.33% above US cash on average across 181 windows. The three starts with both external and domestic stress lost 5.91% on average (Fig. 2). Domestic-only starts and external-only starts were positive on average in the baseline partition.

That contrast reverses under a basic timing check. Combined stress appeared under 20 of the 21 possible block starts. Its mean return ranged from negative 5.91% to positive 10.46%. The median across those starts was positive 2.50%. Each populated combined cell held one to five observations. The baseline loss is an unstable clue, rather than an entry rule.

The thresholds remain useful descriptions of market states rather than estimates of causal effects. External stress includes a broad dollar rise of at least 2% over 20 business days. It also includes the VIX, an index of expected US stock-market volatility, at 25 or rising by at least 10 points. Domestic stress means a rise of at least 100 basis points in B3’s 504-business-day PRE rate, a roughly two-year nominal fixed-rate benchmark. The classification says when the three baseline combined-stress windows began. It does not show that those indicators caused their returns.

The worst drawdown began on July 26, 2011 and reached its trough on September 24, 2015. It stood 45.55% below the prior high. Recovery did not arrive until August 8, 2025, just over fourteen years after the peak (Fig. 3). The investor received carry throughout the slide, but the real’s depreciation dominated it.

Exchange translation during that episode removed 100.57 log percentage points, while BRL cash added 39.77 log points. Relative to the mean daily carry over the preceding 252 Brazilian business days, the currency loss equalled 9.49 years of carry. That ratio describes the economic scale of the shock: a decade of the prevailing income advantage could disappear inside one unrecovered position.

The other six drawdowns reached troughs within 90 business days, and each recovered within months rather than years. The 2011 episode accounts for the result’s concentration and duration. A full-sample average can reward patience ex post while remaining unusable for an investor whose horizon, borrowed capital, or risk limit ends inside the drawdown.

Mean excess return was positive at every tested horizon from 21 to 252 Brazilian business days (Fig. 4). The positive share rose from 55.1% at one month to 68.3% at six months. It then fell to 55.0% among twenty non-overlapping one-year windows. Longer exposure increased the time available for carry to accrue, but it did not create a monotonic protection rate.

The one-year result is based on twenty blocks. Almeida et al. (2023) provide the longer comparison: they report profitability for five-year real-dollar carry windows from 2000 to 2021, including periods of Brazilian instability. Their result and ours agree that a long holding period can reveal compensation that shorter paths obscure.

The opposing interpretation starts from the positive average itself. Carry earns a premium because the investor accepts infrequent currency crashes, so a profitable long-run record and a severe drawdown are two parts of the same compensation rather than conflicting findings. Brunnermeier, Nagel and Pedersen (2008) document negative skewness, a return distribution weighted toward rare large losses. They study a portfolio that buys high-rate currencies and funds them with low-rate currencies, and connect sudden unwinds to falling risk appetite and funding liquidity. Banerjee et al. (2026) add that carry positions can make exchange-rate responses state dependent, meaning the response changes with starting market conditions.

Brazil-specific evidence strengthens that case. Almeida et al. (2023) find profitable long-horizon real-dollar carry over 2000 to 2021, including five-year windows. The long-run 4.25% annualized excess return and the decade-scale 2011 drawdown fit a crash-risk premium more naturally than a claim that high interest rates mechanically protect the real.

The same distinction appears across fixed eras. Annualized excess return was positive in three of four blocks (Fig. 5). The exception ran from 2011 to 2015. BRL carry still contributed close to 10 log percentage points a year during that losing era.

The drawdown evidence fits crash compensation without distinguishing it from other causes of concentrated currency loss. Brazilian carry paid on average under either interpretation, while the timing and duration of losses determined whether an investor could collect that average.

The first live check is the break-even identity. Over any holding window, protection is failing when BRL depreciation exceeds local carry after subtracting the US cash return. A 21-business-day window below zero is a realized failure, while several consecutive negative windows show that the loss is lasting longer than the strategy’s primary evaluation horizon.

The second check is the pair of pre-specified stress flags. Mark broad-dollar stress after a rise of at least 2% over 20 business days. Mark volatility stress when VIX reaches 25. A rise of at least 10 points over 20 business days also qualifies. Mark domestic stress when the roughly two-year PRE rate rises at least 100 basis points over 20 business days. A simultaneous flag is a prompt to inspect the exchange loss and position horizon, not a forecast of a negative month.

The third check is drawdown duration. A 5% fall from the strategy’s previous high starts an episode under the rule used here; recovery occurs only when the index regains that high. The historical warning is the time spent underwater. A drawdown that survives beyond 90 Brazilian business days has already lasted longer than the trough time of every episode except the 2011 to 2015 decline.