Carry Trader
Carry Trader
Carry Trader
ABSTRACT
We find important differences in dollar-based and dollar-neutral
G10 carry trades. Dollar-neutral trades have positive average
returns, are highly negatively skewed, are correlated with risk
factors, and exhibit considerable downside risk. In contrast, a
diversified dollar-carry portfolio has a higher average excess
return, a higher Sharpe ratio, minimal skewness, is uncondi-
tionally uncorrelated with standard risk-factors, and exhibits
no downside risk. Distributions of drawdowns and maximum
losses from daily data indicate a role for time-varying autocor-
relation in determining negative skewness at longer horizons.
This research was supported by a grant from the Network for Study on Pensions, Aging,
and Retirement to the Columbia Business School. We thank two anonymous referees, Craig
Burnside, Mike Chernov, Victoria Dobrynskaya, Pab Jotikasthira, Campbell Harvey, Hanno
1 Introduction
This paper empirically examines returns to carry trades in the major inter-
national currency markets including their exposures to various risk factors.
A carry trade is an investment in a high interest rate currency that is funded
by borrowing in a low interest rate currency. The ‘carry’ is the ex-ante
observable positive interest differential. Returns to the carry trades are un-
certain because the exchange rate between the two currencies may change.
An individual carry trade is profitable when the high interest rate currency
depreciates relative to the low interest rate currency by less than the interest
differential.1
Carry trades are known to have high Sharpe ratios, as emphasized
by Burnside et al. (2011a). Consistent with this, our baseline carry trade
has an annualized Sharpe ratio of 0.78 over the 1976:02-2013:08 sample
period. Alternative versions, which we discuss below, have Sharpe ratios
as high as 1.02.
While such return premiums are obviously inconsistent with the theory
of uncovered interest rate parity, the academic literature offers numerous
explanations for the existence of these return premiums. Brunnermeier
et al. (2008) document that returns to standard carry trades are negatively
skewed and exhibit infrequent large losses. While their evidence is sug-
gestive that such downside risk could be priced, they do no formal asset
pricing tests. Negative skewness per se is not enough to explain the prof-
itability of carry trades as Bekaert and Panayotov (2015) develop “good”
carry trades that retain high average returns and do not have significantly
negative skewness. Lettau et al. (2014) and Dobrynskaya (2014) find that
carry trade portfolio returns are differentially exposed to the return on the
equity market when the market is significantly down. They estimate a high
price of downside risk and conclude that such downside risk explains the
Lustig, Nick Roussanov, Ivo Welch and the participants of the 2015 NBER International
Asset Pricing Summer Institute, the 2016 Vanderbilt FMRC Conference on International
Finance, and the 2016 China International Conference in Finance for helpful comments
and suggestions. We especially thank Pierre Collin-Dufresne for many substantive early
discussions that were fundamental to the development of the paper. We also thank Elessar
Chen for his research assistance, and Ken French, Jakub Jurek, Tracy Johnson, Lucio Sarno,
Maik Schmeling, and Adrien Verdelhan for providing data.
1
Koijen et al. (2015) explore the properties of ‘carry’ trades in other asset markets
by defining ‘carry’ as the expected return on an asset assuming that market conditions,
including the asset’s price, stay the same.
The Carry Trade: Risks and Drawdowns 3
profitability of the carry trade. Jurek (2014) similarly finds that currency
carry trades are strongly exposed to the returns of the Jurek and Stafford
(2015) downside risk index (DRI) portfolio, and he concludes that the
abnormal returns of currency carry trades are indistinguishable from zero
after controlling for their exposure to DRI and other equity risk factors.
Lustig et al. (2014) find that the average forward discount of currencies rel-
ative to the U.S. dollar is a particularly strong predictor of excess currency
returns. They conclude that common movements of the dollar relative to
all currencies are a primary driver of dollar-based carry trade returns.
We examine these and other potential explanations for carry trade
premiums by decomposing the standard carry trade into dollar-carry and
dollar-neutral-carry components. The carry trade, as commonly imple-
mented in academic studies, can have a large positive or negative exposure
to the U.S. dollar, depending on the level of USD interest rates relative to
the median non-USD interest rate. Our diversified, dynamic dollar strategy
captures this time-varying dollar exposure.
We show that the majority of the excess return of our basic carry trade
is attributable to the time-varying dollar component. This view supports
the analysis of Lustig et al. (2014) who develop a dollar-carry strategy
based on the average forward discount with a larger set of currencies
than ours. We find that the average excess return of the dollar-neutral
component of our basic carry trade is different from zero, but the returns
are highly negatively skewed, are correlated with standard risk factors, and
exhibit considerable downside risk. In contrast, a dynamic dollar strategy
diversified across the G10 currencies has a higher average excess return, a
higher Sharpe ratio, minimal skewness, is uncorrelated with standard risk-
factors, and exhibits no downside risk. While Lustig et al. (2014) conclude
that their multi-country affine model is consistent with the observed returns
on their dollar-carry strategy, we argue that the priced risks in such a model
would manifest themselves in statistically significant coefficients in our
asset pricing regressions, which we do not observe. We consequently do
not interpret our results through the lens of their model.
Additionally, we examine how spread-weighting and risk-rebalancing
affect the profitability and risk of the basic carry trade. Our spread weighted
portfolio conditionally invests more in the currencies with the highest inter-
est differentials relative to the dollar. We also employ a simple estimated
covariance matrix of the currency returns that allows us to reduce the
overall risk of our carry trade portfolios in recognition that traders face
4 Kent Daniel et al
limits on losses that require reductions in the sizes of trades when volatility
increases. The covariance matrix is also used in a sequential mean-variance
optimization. Spread-weighting and risk-rebalancing enhance the perfor-
mance of the carry trade. For the equal-weighted carry trade, we also
show that returns are dependent on the base currency with the dollar base
providing the largest average return.
We conclude our analysis with a study of the drawdowns to carry
trades.2 We define a drawdown to be the loss that a trader experiences
from the peak (or high-water mark) to the trough in the cumulative return
to a trading strategy. We also examine pure drawdowns, as in Sornette
(2003), which are defined to be persistent decreases in an asset price over
consecutive days. We document that carry trade drawdowns are large and
occur over substantial time intervals. We contrast these drawdowns with
the characterizations of carry trade returns by The Economist (2007) as
“picking up nickels in front of a steam roller,” and by Breedon (2001) who
noted that traders view carry trade returns as arising by “going up the stairs
and coming down the elevator.” Both of these characterizations suggest that
negative skewness in the trades is substantially due to unexpected sharp
drops. In contrast to these characterizations, our analysis of daily returns
suggests that a large fraction of the documented negative return skewness
of carry trades results from the time varying return autocorrelations of
daily carry returns: the large drawdowns in carry trade returns result from
sequences of losses rather than large single-day drops.
3
Hassan and Mano (2015) find that 70% of carry trade profitability is due to static
interest rate differentials, while Bekaert and Panayotov (2015) develop static trades that
are 50% as profitable as their more dynamic ones. Because a substantive fraction of carry
trade profitability is effectively unconditional, while rational explanations of the forward
premium anomaly typically involve dynamic, time-varying risk premiums, Hassan and
Mano (2015) argue that reconciling carry trade profitability and the forward premium
anomaly may require separate explanations.
6 Kent Daniel et al
trade. Farhi and Gabaix (2016) and Farhi et al. (2015) also argue that
the carry trade is exposed to rare crash states in which high interest rate
currencies depreciate. Burnside (2012) reviews the literature examining
the risks of carry trades.
Jordà and Taylor (2012) dismiss the profitability of the naive carry
trade based only on interest differentials as poor given its performance
in the financial crisis of 2008, but they advocate simple modifications of
the positions based on long-run exchange rate fundamentals that enhance
its profitability and protect it from downside moves indicating a market
inefficiency.
S t+1
z t+1 = [(1 + i t∗ ) − (1 + i t$ )] y t (1)
St
where
+1 if i t∗ > i t$
yt =
−1 if i t$ > i t∗
Equation (1) scales the size of returns to the carry trade either by
borrowing one dollar and investing in the foreign currency money market,
or by borrowing the appropriate amount of foreign currency to invest one
4
When it is necessary to distinguish between the dollar exchange rate versus various
currencies or the various interest rates, we superscript them with numbers.
8 Kent Daniel et al
dollar in the dollar money market. When covered interest rate parity holds,
if i t∗ > i t$ , then F t < S t ; that is the foreign currency is at a discount in
the forward market. Conversely, if i t∗ < i t$ , then F t > S t ; and the foreign
currency is at a premium in the forward market. Thus, the carry trade can
also be implemented by going long (short) in the foreign currency in the
forward market when the foreign currency is at a discount (premium) in
terms of the dollar. Let w t be the amount of foreign currency bought in
the forward market. The dollar payoff to this strategy is:
When covered interest rate parity holds, and in the absence of transac-
tion costs, the forward market strategy for implementing the carry trade in
equation (2) is exactly equivalent to the carry trade strategy in equation
(1). Unbiasedness of forward rates and uncovered interest rate parity imply
that carry trade profits should average to zero.
Uncovered interest rate parity ignores the possibility that changes in the
values of currencies are exposed to risk factors, in which case risk premiums
can arise. To incorporate risk aversion, we need to examine pricing kernels.
That is, the expected return to the carry trade will be positive if its covaria-
tion with the stochastic discount factor is negative.5
Burnside et al. (2011b), Caballero and Doyle (2012), Farhi et al. (2015),
and Jurek (2014) examine hedging the downside risks of the carry trade by
purchasing insurance in the foreign currency option markets. To examine
this analysis, let C t and Pt be the dollar prices of one-period foreign currency
call and put options with strike price K on one unit of foreign currency.
Buying one unit of foreign currency in the forward market costs F t dollars
in one period, which is an unconditional future (time t +1) cost. One can
also unconditionally buy the foreign currency forward by buying a call
option with strike price K and selling a put option with the same strike
price, in which case the future cost is K + C t (1 + i t$ ) − Pt (1 + i t$ ). To prevent
arbitrage, these unconditional future costs must be equal, which implies
F t = K + C t (1 + i t$ ) − Pt (1 + i t$ ). (7)
When S t+1 < K, [S t+1 − K + max (0, K − S t+1 )] = 0; and if S t+1 > K,
max(0, K − S t+1 ) = 0. Hence, we can write equation (8) as
H
z t+1 = [max (0, S t+1 − K) − C t (1 + i t$ )]w t ,
The ∆ of an option is the derivative of the value of the option with respect to a change
6
in the underlying spot rate. A 10∆ (25∆) call option increases in price by 0.10 (0.25)
times the small increase in the spot rate. The ∆ of a put option is negative.
The Carry Trade: Risks and Drawdowns 11
3 Data
In constructing our carry trade returns, we use data on the world’s major
currencies, the so-called G10 currencies: the Australian dollar (AUD), the
British pound (GBP), the Canadian dollar (CAD), the euro (EUR) spliced
with historical data from the Deutsche mark, the Japanese yen (JPY), the
New Zealand dollar (NZD), the Norwegian krone (NOK), the Swedish krona
(SEK), the Swiss franc (CHF), and the U.S. dollar (USD).7 All spot and
forward exchange rates are dollar denominated and are from Datastream
and IHS Global Insight. For most currencies, the beginning of the sample is
January 1976, and the end of the sample is August 2013, which provides
a total of 451 monthly observations on the carry trade. Data for the AUD
and the NZD start in October 1986. Interest rate data are eurocurrency
interest rates from Datastream.
We explicitly exclude the European currencies other than the euro (and
its precursor, the Deutsche mark), because we know that several of these
currencies, such as the Italian lira, the Portuguese escudo, and the Spanish
peseta, were relatively high interest rate currencies prior to the creation of
the euro. At that time traders engaged in the “convergence trade,” which
was a form of carry trade predicated on a bet that the euro would be
created in which case the interest rates in the high interest rate countries
would come down and those currencies would strengthen relative to the
Deutsche mark. An obvious peso problem exists in these data because there
was uncertainty about whether the euro would indeed be created. If the
euro had not succeeded, the high interest rate currencies, such as the lira,
escudo, and peseta, would have suffered large devaluations relative to the
Deutsche mark, drastically lowering the return to the convergence trade.
We also avoid emerging market currencies because nominal inter-
est rates denominated in these currencies also incorporate substantive
sovereign risk premiums. The essence of the carry trade is that the in-
vestor bears pure foreign exchange risk, not sovereign risk. Furthermore,
7
Investable carry trade indices based on G10 currencies include the iPath Optimized
Currency Carry ETN and the Powershares DB G10 Currency Harvest Fund. The Bank for
International Settlements (2014) triennial survey reports that the G10 currencies accounted
for 88% of global foreign exchange market average daily turnover in April 2013. Academic
research that focuses on the profitability of carry trade strategies in the G10 currencies
includes Burnside et al. (2011a), Christiansen et al. (2011), Farhi et al. (2015), and Jurek
(2014).
12 Kent Daniel et al
Our first strategy, designated EQ, has equal absolute value weights. The
weight on currency j is therefore:
j
EQ si g n(i t − i t$ )
w j,t = (9)
Nt
PNt EQ
the weighted sum of the returns: R EQ,t+1 = j=1 w j,t z j,t+1 where z j,t+1
is the time t + 1 payoff to investing $1 in the money market for foreign
currency j, and borrowing $1 at time t, which we implement using the
equivalent forward market transactions as discussed in Section 2.1.
The second strategy, labeled SPD, is ‘spread-weighted.’ Like the EQ
strategy, it is long (short) currencies that have a positive (negative) interest
differential relative to the dollar, but the size of the investment in a partic-
ular currency is determined by the relative magnitude of the interest-rate
differential:
j
i t − i t$
SPD
w j,t = N
Pt j
i t − i t$
j=1
The sum of absolute values of the weights in the foreign currencies is again
one, but the SPD strategy invests proportionately more in currencies that
have larger interest differentials.10
Because the return volatilities of the EQ and SPD strategies rise and fall
with changes in exchange rate volatilities, such strategies would not gener-
ally be employed by traders in FX markets who are typically constrained
by a value-at-risk requirement, defined as the maximum loss that could
be sustained with a given probability. For example, a typical value-at-risk
model constrains a trader to take positions such that the probability of
losing, say more than $1 million on any given day, is no larger than 1%.
Traders consequently must scale their investments based on some estimate
of portfolio risk. To evaluate the efficacy of such scaling we construct
“risk-rebalanced” versions of the EQ and SPD strategies, labeled EQ-RR and
SPD-RR.
Constructing these strategies requires a conditional covariance matrix
of the returns for which we use simple IGARCH models from daily data.
Let H t denote the conditional covariance matrix of returns at time t with
ij
typical element, h t , which denotes the conditional covariance between
the ith and jth currency returns realized at time t + 1. Then, the IGARCH
et al. (2011b), Clarida et al. (2009), Lettau et al. (2014), Lustig et al. (2011) and Lustig
et al. (2014).
10
Jurek (2014) spread weights by taking positions on the basis of the absolute distance
of country j’s interest rate from the average of the interest rates in countries with ranks
five and six.
The Carry Trade: Risks and Drawdowns 15
ij
model for h t is
ij j ij
h t = δ(r ti r t ) + (1 − δ)h t−1 (10)
where because of the daily horizon, we treat the product of the returns as
equivalent to the product of the innovations in the returns. We set δ = 0.06,
as suggested in J. P. Morgan (1996). To obtain the monthly covariance
matrix we multiply the daily IGARCH estimates of H t by 21.
For the EQ-RR p and SPD-RR strategies, we target a monthly standard
deviation of 5%/ 12 – corresponding to an approximate annualized stan-
dard deviation of 5% – by adjusting the dollar scale of the EQ and SPD
portfolios accordingly.
Our final strategy in this section involves sequential mean-variance
optimization and is labeled OPT. Beginning with the analysis of Meese and
Rogoff (1983), it is often argued that expected rates of currency apprecia-
tion are essentially unforecastable. Hence, we take the vector of interest
differentials, labeled µ t , to be the conditional means of the carry trade
returns, and we take positions wOP t
T
= κ t H t−1 µ t , where κ t is a scaling
factor that sets the sum of the absolute values of the weights equal to one
as in the EQ strategy. If the models of the
Æconditional moments are correct,
the conditional Sharpe ratio will equal µ0t H t−1 µ t .11
Table 1 reports the first four moments of the various carry trade strate-
gies and their (annualized) Sharpe ratios and first-order autocorrelations.
Standard errors are based on the Generalized Method of Moments of
Hansen (1982), as explained in Section B of the online appendix. Through-
out the paper, when we discuss estimated parameters, GMM standard errors,
calculated using three Newey and West (1987) lags, are in parentheses and
the associated robust t-statistics are in square brackets.
For the full sample, the carry trades for a USD-based investor have
statistically significant mean annual returns ranging from 2.10% (0.47)
for the OPT strategy, to 3.96% (0.91) for the EQ strategy, and to 6.60%
(1.31) for the SPD strategy. Jurek (2014) also finds that spread-weighting
improves the performance of the carry trades. The strategies have impres-
sive Sharpe ratios, which range from 0.78 (0.19) for the EQ strategy to
11
Ackermann et al. (2012) also use conditional mean variance modeling so their positions
are also proportional to H t−1 µ t , but they target a constant mean return of 5% per annum.
(0.05/12)
Hence, their positions satisfy wAPS t
= µ0 H −1 µ H t−1 µ t . While their conditional Sharpe ratio is
t t t
0.5
also µ0t H t−1 µ t , their scaling factor responds more aggressively to perceived changes in
the conditional Sharpe ratio than ours.
16 Kent Daniel et al
1.02 (0.19) for the SPD-RR strategy. As Brunnermeier et al. (2008) note,
each of these strategies is significantly negatively skewed, with the OPT
strategy having the most negative skewness of -0.89 (0.34). Table 1 reports
positive excess kurtosis that is statistically significant for all strategies. The
first-order autocorrelations of the strategies are low, as would be expected
in currency markets, and only for the EQ-RR strategy can we reject that the
first-order autocorrelation is zero. Of course, it is well known that this test
has very low power against interesting alternatives. The minimum monthly
returns for the strategies are all quite large, ranging from -4.01% for the
OPT strategy to -7.26% for the SPD. The maximum monthly returns range
from 3.21% for the OPT to 8.07% for the SPD. Finally, Table 1 indicates
that the carry trade strategies are profitable on between 288 months for
the EQ strategy and 303 months for the OPT strategy out of the total of
451 months.12
The preceding discussion and most academic research about the carry trade
takes the perspective of a U.S.-based investor for whom the USD is, in our
terminology, both the base currency and the measurement currency. Here,
we use measurement currency to denote the currency in which the investor
measures his or her profits. The base currency denotes the currency which
is the basis for the positions that the carry trade takes.
For example, for the EQ strategy that is most often studied in the
academic literature, the USD is most often the base currency. The EQ
strategy goes goes long (short) all currencies with an interest rate higher
(lower) than the USD interest rate. Financing of the long high-interest-rate
currency positions is done by borrowing in USD, and the capital that is
raised by shorting the low-interest-rate currencies is assumed to be invested
in USD. Were such a strategy implemented in an alternative base currency,
the returns would be different for two reasons. First, the “cutoff-rate” that
determines whether a currency is bought or sold would be the base-currency
interest rate rather than the USD interest rate. Second, the financing of the
long positions and the investing from the short positions would be done in
the alternative base currency, rather than in USD. As as extreme example,
12
The strategy returns are all positively correlated. Correlations range from .63 for EQ
and OPT to .90 for EQ and SPD. The correlations of EQ and EQ-RR and SPD and SPD-RR
are .88 and .89, respectively.
The Carry Trade: Risks and Drawdowns 17
suppose the USD interest rate is the highest, and the JPY interest rate is
the lowest in the G-10. In this case, the USD-based EQ strategy would be
long 1 unit of USD, and short 1/9th of every other currency. The JPY-based
EQ strategy, in contrast, would be long only 1/9th unit of USD, and would
be short 1 unit of JPY.
The measurement currency for the EQ carry trade that is the basis for
most academic papers is also generally the USD. Note that the measurement
currency need not be the same as the base currency.13 For example, a
European investor could implement the EQ carry trade with a USD base
currency, but measure the returns in EUR. To do so, this investor would go
long (short) all currencies with an interest rate higher (lower) than the
USD interest rate, borrow in USD to finance the long (high-interest-rate)
currency positions, and invest the short proceeds (raised by shorting the
low-interest-rate currencies) in USD. However, this investor would measure
her profits in her EUR home currency.
One might imagine that the measurement currency would have a large
effect on the performance of the carry trade; to stick to our example
from the preceding paragraph with a USD base currency and an EUR
measurement currency, one might think that when the all other currencies
appreciate relative to the EUR, that EUR-measured carry trade return would
be substantively higher than the USD-measured return. Interestingly, this
effect is small, and it is vanishingly small in continuous time for diffusion
processes. The reason is that carry-trade returns are excess returns, and
to convert the payoffs measured in USD into EUR, you convert both the
long-side and the short-side payoffs. The result of a EUR depreciation over
an interval, for example, means that the EUR-measured long-side payoff
will be higher, but the EUR-measure short-side payoff will be lower by
roughly the same amount.
There is, however, a difference in average returns to these carry trades
from changing the measurement currency that arises as a result of any
covariance between the carry-trade return and the appreciation of the
measurement currency. We provide a mathematical derivation of this in
online appendix C, and confirm empirically that this covariance almost fully
captures the differences that arise from using an alternative measurement
currency. We also compare the results presented in Table 2, which are
denominated in USD, to results denominated in the base currency, and
13
We thank the editor, Ivo Welch, for pointing out this distinction.
18 Kent Daniel et al
verify that the performance ordering that we see in this table (both average
returns and Sharpe-ratios) is independent of the currency in which the
excess-returns are denominated.
While the measurement currency doesn’t affect our results, the choice
of base currency turns out to have a large effect. Table 2 presents the
summary statistics for EQ carry trades with alternative base currencies
but where, to allow for comparisons between the columns, the excess
returns are all measured in USD. For each strategy, if the interest rate in
currency j is higher (lower) than the interest rate of the base currency,
the investor goes long (short) in the forward market of currency j as in
the USD-based EQ strategy. For the non-USD-based currencies, the mean
annualized strategy returns range from 2.42% (1.26) for the CHF to 4.70%
(1.53) for the NZD. All mean returns are statistically significant at the .06
marginal level of significance or smaller. The USD-base-currency strategy
has the second highest mean return, but the highest Sharpe ratio among
these strategies.14 Despite their lower mean returns, the non-USD-based
strategies generally have higher volatilities, and as a results their Sharpe
ratios are all smaller than the USD-based Sharpe ratio of 0.78 (0.19).
Except for the EUR, the point estimates of skewness for the alternative
base-currency carry trades are all negative, and the statistical significance
of skewness is high for the JPY, NOK, SEK, CHF, NZD, and AUD. In addition,
the excess kurtosis of each strategy is positive and statistically significant.
Only the GBP-based carry trade shows any sign of first-order autocorrelation.
Only the return volatility of the CAD-based strategy is lower than the USD-
based one, and we thus find that the maximum gains and losses on these
strategies generally exceed those of the USD-based strategy with maximum
monthly losses for the JPY, SEK, CHF, NZD, and AUD carry trades exceeding
10%. The alternative base-currency carry trades also have fewer positive
monthly returns than does the USD-base strategy.15 These results show
that carry trade profitability is not just a USD phenomenon, but that the
USD is potentially more important than other currencies in determining
the profitability. We explore this in more depth below.
14
It is unlikely that we would be able to reject equivalence of the means, given the
standard errors.
15
For the NZD and AUD, for which the sample is smaller, the percentage of positive
monthly returns is slightly smaller than for the USD.
The Carry Trade: Risks and Drawdowns 19
R t = α + β 0 Ft + "t . (11)
In most of our analysis we use market-traded risk factors that are returns
to zero-investment portfolios in which case α measures the average return
of the carry trade not explained by its unconditional exposure to the risks
included in the regression multiplied by the average returns to those risks.
Panel B of Table 3 presents results for the two foreign exchange market
risks proposed by Lustig et al. (2011) who sort 35 currencies into six port-
folios based on their interest rates relative to the dollar interest rate, with
16
The Fama-French risk factors were obtained from Kenneth French’s web site which
also describes the construction of these portfolios.
The Carry Trade: Risks and Drawdowns 21
portfolio one (six) containing the lowest (highest) interest rate currencies.
Their two risk factors are RRX ,t , the mean return on all six portfolios, and
R H M L−F X ,t , which is the return difference between portfolios 6 and 1. No-
tably, RRX ,t has a correlation of .99 with the first principal component of
the six portfolio returns, and R H M L−F X ,t has a correlation of .94 with the
second principal component.17 Given its construction, it is not surprising
that R H M L−F X ,t has significant explanatory power for our carry trade re-
turns, with t-statistics between 6.22 for the OPT portfolio and 8.85 for the
EQ-RR portfolio. The R2 ’s are also higher than with the equity risk factors,
ranging between .13 and .34. While this pure FX risk model better explains
the average returns to our strategies than do the equity risks, the α’s re-
main statistically significant and range from a low of 1.29% for the OPT
portfolio to 3.60% for the SPD-RR portfolio.18 While Lustig et al. (2011)
essentially demonstrate that the returns on their carry trade portfolios
have a reduced dimensionality, the conditioning information provided by
spread-weighting and risk-rebalancing allows those conditional trades to
demonstrate abnormal profits relative to that reduced factor space.
17
The factor return data are from Adrien Verdelhan’s web site, and the sample period is
1983:11-2013:08 for 358 observations.
18
Note however that the t-statistic of the α for the EQ portfolio falls to 1.73.
19
Sarno et al. (2012) find that reduced form affine models designed to price bond
yields, which have small bond pricing errors, are unable to capture the dynamics of the
rates of currency depreciation.
The Carry Trade: Risks and Drawdowns 23
The coefficients on both of the bond market factors are highly significant.
Positive returns on the 10-year bond that are matched by the return on the
2-year note, which would be caused by unanticipated decreases in the level
of the USD yield curve, are bad for the USD-based carry trades. Notice
also that the coefficients on the two excess bond returns are close to being
equal and opposite in sign, suggesting that unexpected positive returns on
the two-year note (i.e., decreases in the 2-year note yield) are bad for the
carry trades. Nevertheless, the R2 ’s remain between .02 and .05, as in the
equity market regressions. The statistically significant α’s, ranging from
2.15% for OPT to 6.71% for SPD, indicate that bond market risks do not
explain the carry trade returns.
20
Menkhoff et al. (2012) introduce foreign exchange volatility as a risk factor. To
develop a traded risk factor, they project FX volatility onto a set of currency returns sorted
on interest rate differentials. Because the resulting portfolio has a correlation of .80 with
R H M L−F X ,t , we find that their volatility risk factor has similar explanatory power to the pure
foreign exchange risk model described previously, and consequently, we do not report those
results here.
Table 4: Carry Trade Exposure to Equity and Volatility Risks 24
Description: This table reports regressions of five carry trade returns on risk factors as in Panels A and C of Table 3, but it also includes the
return on a variance swap as a risk factor. The sample period is 1990:02-2013:08 (283 observations). The reported α’s are annualized
percentages. Autocorrelation and heteroskedasticity consistent t-statistics from GMM are in square brackets.
Interpretation: The estimated α’s fall slightly and become statistically insignificant for the EQ and EQ-RR strategy after controlling for
exposure to both the bond-market factors and the variance swap returns. All other estimated α’s remain statistically significant.
EQ EQ-RR SPD SPD-RR OPT EQ EQ-RR SPD SPD-RR OPT
Panel A: Equity Risk Factors
α 3.11 4.15 4.51 4.54 1.38 2.87 3.76 4.22 4.20 1.32
[2.76] [2.50] [3.58] [3.34] [2.65] [2.56] [2.32] [3.41] [3.14] [2.40]
βMKT 0.09 0.16 0.06 0.07 0.03 0.08 0.15 0.05 0.06 0.02
[3.03] [3.66] [2.42] [2.66] [2.58] [2.42] [2.92] [1.81] [1.96] [1.98]
βSMB -0.04 -0.01 -0.04 -0.02 0.01 -0.04 -0.01 -0.04 -0.02 0.01
[-0.92] [-0.15] [-1.00] [-0.36] [1.01] [-0.97] [-0.22] [-1.08] [-0.45] [0.94]
βHML 0.06 0.14 0.04 0.07 0.03 0.06 0.13 0.04 0.06 0.03
[1.75] [2.77] [1.25] [1.99] [2.48] [1.60] [2.52] [1.11] [1.81] [2.37]
βVS -0.02 -0.03 -0.02 -0.03 -0.01
[-0.94] [-0.90] [-1.24] [-1.40] [-0.41]
R2 .07 .10 .04 .04 .04 .07 .11 .04 .05 .04
Panel B: Bond Risk Factors
α 2.83 3.77 4.75 4.71 1.60 2.32 2.79 4.38 4.13 1.45
[2.19] [2.08] [3.69] [3.43] [2.98] [1.72] [1.55] [3.40] [3.06] [2.61]
βMKT 0.07 0.15 0.05 0.06 0.02 0.06 0.12 0.04 0.04 0.02
[2.38] [3.05] [1.73] [2.11] [2.26] [1.85] [2.32] [1.27] [1.43] [1.59]
β10y 0.31 0.63 -0.07 0.08 -0.02 0.39 0.78 -0.01 0.16 0.01
[1.05] [1.57] [-0.26] [0.28] [-0.15] [1.26] [1.85] [-0.05] [0.58] [0.04]
β10y−2y -0.35 -0.73 0.08 -0.10 0.01 -0.44 -0.91 0.01 -0.20 -0.02
[-0.97] [-1.48] [0.24] [-0.29] [0.06] [-1.17] [-1.74] [0.03] [-0.59] [-0.13]
βVS -0.03 -0.06 -0.02 -0.04 -0.01
[-1.39] [-1.69] [-1.25] [-1.76] [-0.70]
Kent Daniel et al
R2 .04 .08 .02 .02 .02 .05 .09 .02 .03 .02
The Carry Trade: Risks and Drawdowns 25
Lustig et al. (2014) and Jurek (2014) find important differences between
dollar-based carry trades and dollar-neutral carry trades. The analysis in
Section 4 confirms that the historical efficacy of the EQ carry trade depends
on the base currency in which it is implemented. In particular, Table 2 shows
that the USD-based EQ strategy had a higher Sharpe ratio than for any other
base currency. We now extend our earlier analysis by decomposing the EQ
portfolio into dollar-neutral and dollar-carry components. We confirm that
much of the efficacy of the carry trade is attributable to the dollar-carry
component, and we further demonstrate striking differences in the risk
characteristics of the two components.
Our dollar-neutral carry trade portfolio – which we label EQ-0$ – takes
positions only in the non-USD currencies. The EQ-0$ weights are:
1 j k
+ Nt if i t > med i t
EQ−0$ j
w j,t = − 1 if i t < med i tk (12)
Nt j k
0 if i t = med i t
k
where med i t indicates the median of the non-USD interest rates at time
t, and Nt is the number of non-USD currencies in the sample at time t.
Then, assuming nine non-USD currencies, we take long positions of $(1/9)
in the four highest interest rate currencies, financed by short positions
of $(1/9) in the four lowest rate currencies. We take no position in the
median interest rate currency. Given this construction, EQ-0$ is a long-short
portfolio with no direct dollar exposure.
Tables 5 and 6 present summary statistics and analysis of the risk-
exposures of the EQ and EQ-0$ portfolios, and two additional portfolios
that capture the dollar component of the carry trade, which we describe in
Sections 5.1 and 5.2.
Table 5: Summary Statistics for the Dollar-Neutral and Pure-Dollar Carry Trades
Description: This table presents summary statistics on returns to four carry trade strategies. The first three are the equal-weighted (EQ), 26
dollar-neutral (EQ-0$) and dollar (EQ-$) strategies. EQ-$ is the difference between EQ and EQ-0$. EQ-D$ is the dynamic dollar strategy.
The sample period is 1976:02-2013:08 except for the AUD and the NZD, which start in October 1986. The reported parameters, mean,
standard deviation, skewness, excess kurtosis, and autocorrelation coefficient and their associated standard errors are simultaneous GMM
estimates, and the mean and standard deviation are annualized. The Sharpe ratio is the ratio of the annualized mean and standard
deviation, and its standard error is calculated using the delta method (see online appendix B). Panel A reports the results for the full
sample, while Panel B reports results for the sample 1990:02-2013:08 when variance swap data are available.
Interpretation: The dollar-neutral and dollar components of the EQ strategy, EQ-0$ and EQ-$, both have statistically significant mean
returns and Sharpe ratios, though the point estimates for the dollar component are slightly larger. The dynamic dollar strategy (EQ-D$) is
more volatile, but its Sharpe ratio is higher and skewness is lower than that of the EQ-$ strategy.
Panel A: 1976/02-2013/08 Panel B: 1990/2-2013/8
EQ EQ-0$ EQ-$ EQ-D$ EQ EQ-0$ EQ-$ EQ-D$
Mean Ret (% p.a.) 3.96 1.61 2.35 5.54 3.83 1.72 2.11 5.21
(0.91) (0.58) (0.66) (1.37) (1.17) (0.72) (0.92) (1.60)
Standard Deviation 5.06 3.28 3.85 8.18 5.43 3.30 4.31 7.89
(0.28) (0.16) (0.28) (0.38) (0.36) (0.21) (0.35) (0.47)
Sharpe Ratio 0.78 0.49 0.61 0.68 0.70 0.52 0.49 0.66
(0.19) (0.19) (0.18) (0.18) (0.24) (0.23) (0.23) (0.21)
Skewness -0.49 -0.47 -0.65 -0.11 -0.60 -0.47 -0.76 -0.05
(0.21) (0.19) (0.44) (0.17) (0.22) (0.28) (0.45) (0.22)
Excess Kurtosis 2.01 1.34 4.84 0.86 1.68 1.66 3.87 1.04
(0.53) (0.51) (2.00) (0.31) (0.57) (0.71) (1.86) (0.38)
Autocorrelation 0.08 0.05 0.05 0.00 0.05 0.05 0.05 -0.03
(0.07) (0.06) (0.06) (0.06) (0.08) (0.07) (0.06) (0.07)
Max (% per month) 4.78 3.28 3.83 9.03 4.60 3.28 3.80 9.03
Min (% per month) -6.01 -3.92 -6.69 -8.27 -6.01 -3.92 -6.69 -7.22
No. Positive 288 275 264 273 182 179 164 168
Kent Daniel et al
No. Negative 163 176 187 178 101 104 119 115
The Carry Trade: Risks and Drawdowns 27
Table 6: Dollar Neutral and Pure Dollar Carry Trade Risk Exposures
Description: Panels A, B and C present the results of regressions of carry trade returns
on the risk factors considered in Table 3. The reported α’s are annualized percentages.
Autocorrelation and heteroskedasticity consistent t-statistics from GMM are in square
brackets.
Interpretation: The dollar and dollar-neutral components of the EQ carry trade behave
very differently: the dollar-neutral component’s return is explained both by the equity
factors and the bond factors; the dollar component’s return is not. In particular, the
al pha’s of the dynamic-dollar strategy remain strongly statistically significant with all sets
of explanatory factors.
Carry Trade Strategy
EQ EQ-0$ EQ-$ EQ-D$
Panel A: Equity Factors, 1976:02-2013:08
α 3.39 1.03 2.36 5.41
[3.76] [1.54] [3.60] [3.70]
βMKT 0.05 0.08 -0.02 0.01
[2.46] [5.29] [-1.24] [0.21]
βSMB -0.03 0.02 -0.05 -0.05
[-0.90] [1.00] [-1.84] [-1.04]
βHML 0.07 0.05 0.02 0.06
[2.21] [2.83] [0.84] [1.08]
R2 .04 .10 .04 .01
Panel B: FX Factors, 1983:11-2013:08
α 1.47 -0.03 1.49 5.18
[1.73] [-0.06] [1.86] [3.40]
βRX 0.14 -0.02 0.15 0.52
[2.27] [-0.50] [2.80] [4.31]
βHML−FX 0.28 0.24 0.04 0.00
[8.24] [11.03] [1.58] [-0.01]
R2 .31 .41 .09 .20
Panel C: Bond Factors, 1976:01-2013:08
α 4.19 1.50 2.69 5.82
[4.51] [2.77] [3.88] [4.01]
βMKT 0.04 0.06 -0.03 -0.01
[1.81] [5.48] [-1.60] [-0.35]
β10y -0.39 -0.25 -0.14 -0.22
[-2.97] [-4.42] [-1.16] [-0.92]
β10y−2y 0.46 0.29 0.17 0.32
[2.46] [3.52] [1.02] [0.93]
R2 .05 .12 .02 .01
28 Kent Daniel et al
The second columns of Panels A and B of Table 5 report the first four
moments of the EQ-0$ portfolio returns as well as the Sharpe ratio and
the first-order autocorrelation. Panel A reports the full sample results, and
Panel B reports the results over the sample when VIX data are available.
For ease of comparison, we report the same set of statistics for the EQ
strategy in Column 1. The EQ-0$ portfolio has statistically significant mean
annual returns in both samples, 1.61% (0.58) for the full sample and 1.72%
(0.72) for the later sample. While these mean returns are lower than for
EQ strategy, the EQ-0$ volatility is also lower. However, the EQ-0$ Sharpe
ratios are nonetheless about 30% lower than the EQ Sharpe ratios in each
sample period. The negative skewness and insignificant autocorrelations of
the EQ-0$ strategy are comparable to those of the EQ strategy. Consistent
with the lower volatility, the maximum losses are smaller than those of the
EQ strategy. The next question is whether the EQ-0$ strategy is exposed to
risks.
The second column of Panels A and B in Table 6 shows that the alphas
of the EQ-0$ strategy are zero after controlling for either the three Fama
and French (1993) equity market risk factors, or the Lustig et al. (2014) FX
factors. Panel A shows that EQ-0$ loads significantly on the market return
and the HML factor, with t-statistics of 5.29 and 2.83, respectively. The
loading on the market return explains approximately 30% of the average
return, and the loading on the HML factor explains another 15% of the
average return. The resulting α has a t-statistic of 1.54, and the R2 is .10.
In comparison, the regression of EQ returns on the same equity risk factors
has an α of 3.39 with a t-statistic of 3.76 and an R2 of only .04. The
The Carry Trade: Risks and Drawdowns 29
Fama and French (1993) three factor model clearly does a better job of
explaining the average return of the EQ-0$ strategy than that of the EQ
strategy.
These results are consistent with Jurek (2014) who investigates a
shorter sample and finds marginally significant α’s in his spread-weighted
carry trade regressions when adding the Carhart (1997) momentum factor
to the three Fama and French (1993) model. He finds significant expo-
sures to the market return and HML, but smaller, insignificant α’s in his
spread-weighted, dollar-neutral, carry trade regressions on the same fac-
tors. These results suggest that, after eliminating the dollar exposure from
the EQ strategy, the average profitability of the developed currency carry
trade can be explained by commonly used equity risk factors.
where the weights on the EQ and EQ-0$ portfolios are given in equations
(9) and (12). Note that because both EQ and EQ-0$ are zero-investment
portfolios, EQ-$ is as well. The exact positions of the EQ-$ portfolio
depend on whether i t$ is below or above the median interest rate. If
i t$ < med ian i tk , then
j
0 if i t > med i tk
1 if i j = med i k
EQ-$ Nt t t
w j,t = 2 j
Ntif i t$ < i t < med i tk
j $
0 if i t ≤ i t
If i t$ > med ian i tk , then
j
0 if i t > i t$
− 1 if i j = med i k
EQ-$ Nt t
w j,t = 2
k t j
−
Nt
if med i t < i t ≤ i t$
j
0 if i t ≤ med i tk
30 Kent Daniel et al
The EQ-0$ and EQ-$ portfolios decompose the EQ carry trade into two
components: a dollar-neutral component and a dollar component. EQ-$
goes long (short) the dollar when the dollar interest rate is higher (lower)
than the median interest rate but only against currencies with interest rates
between the median and dollar interest rates. Thus, if these interest rates
are close, the EQ-$ portfolio will be concentrated in just a few currencies.
The third columns of Panels A and B in Table 5 present the first four
moments of the EQ-$ strategy. The EQ-$ mean returns are statistically
significant in both samples, and the Sharpe ratios are close to those of the
EQ-0$ strategy. Skewness of EQ-$ is negative but statistically insignificant
due to the large standard error in both samples. In terms of both Sharpe
ratio and skewness, the EQ-$ strategy appears no better than the EQ-0$
strategy. Note also that the kurtosis of EQ-$ is far higher. Nevertheless, the
EQ-$ strategy has a correlation of -.11 with the EQ-0$ strategy, and the
following results illustrate that the EQ-$ strategy also differs significantly
from the EQ-0$ strategy in its risk exposures.
Column 3 of Panel A in Table 6 presents regressions of EQ-$ returns
on the three Fama and French (1993) risk factors. Unlike EQ-0$, only the
SMB factor shows any explanatory power for the EQ-$ returns. The α is
2.36% with a t-statistic of 3.60. The R2 is .04. The equity market risks
clearly do not explain the average returns to the EQ-$ strategy.
Columns 1 to 3 of Panel B in Table 6 present regressions of the returns
of EQ and its two components, EQ-0$ and EQ-$, on the two FX risk factors.
The two factor FX model completely explains the average returns of the
EQ-0$ strategy while explaining only 25% of the average returns of EQ-$.
The α for EQ-$ is also significant with a value of 1.49% and a t-statistic of
1.86 in the FX two factor model.
Finally, Columns 1 to 3 of Panel C in Table 6 present regressions of
the returns to the EQ strategy and its two components, EQ-0$ and EQ-
minus, on the equity market excess return and two bond market risk factors.
Similar to our previous findings, the market excess return and the bond
risk factors have significant explanatory power for the returns of EQ-0$
and a relatively high R2 of .12. By comparison, none of the risk factors has
any significant explanatory power for the returns on the EQ-$ strategy, and
the resulting R2 is only .02.
In summary, these results suggest that for the G10 currency dollar-based
carry trade, the conditional dollar exposure contributes more to the carry
trade “puzzle” than does the non-dollar component. This conclusion is
The Carry Trade: Risks and Drawdowns 31
consistent with the analysis of Lustig et al. (2014) who conclude that by
conditioning investments on the level of the average forward discount U.S.
investors earn large currency excess returns that are not correlated with
traditional carry trade returns.
The EQ-$ strategy goes long (short) the dollar when the dollar interest rate
is above (below) the G10 median interest rate. It has the nice property of
complementing the EQ-0$ strategy to become the commonly studied equally
weighted carry trade EQ. However, as noted above, if the USD interest
rate is close to the median non-USD interest rate, EQ-$ takes positions in
relatively few currencies. Absence of diversification is consistent with the
large kurtosis noted in Table 5. Since the results just presented indicate
that the abnormal returns of EQ hinge on the conditional dollar exposure,
which is distinct from “carry,” we now expand the other leg of EQ-$ to
all foreign currencies. We define our diversified, dynamic dollar strategy,
EQ-D$, as: 1
EQ-D$ + N if med i tk > i t$
w j,t =
− N1 if med i tk ≤ i t$
The EQ-D$ strategy focuses on the conditional exposure of the U.S. dollar.
It goes long (short) nine foreign currencies against the dollar when the
dollar interest rate is lower (higher) than the global median interest rate.
The EQ-D$ strategy is essentially the G10 counterpart of the “dollar carry"
strategy developed by Lustig et al. (2014) who take an equal weight long
(short) position in 20 developed currencies versus a short (long) position
in the dollar if the dollar interest rate is lower (higher) that the average
foreign interest rate.
The fourth columns of Panels A and B of Table 5 present the first four
moments of the returns to the EQ-D$ strategy for the full sample and the
VIX sample. We find that EQ-D$ has statistically significant mean annual
returns of 5.54% (1.37) for the full sample and 5.21% (1.60) for the VIX
sample, substantially higher than the means of EQ-$. Although its volatility
is also higher than the EQ-$ strategy, its Sharpe ratio of 0.68 (0.18) in
the full sample and 0.66 (0.21) in the VIX sample are larger although not
significantly different from those of the EQ-$ strategy. Skewness of the
EQ-D$ strategy is lower than the other three strategies and is statistically in-
significant in both periods. Also, consistent with the greater diversification
32 Kent Daniel et al
Lettau et al. (2014) note that although portfolios of high interest rate
currencies have higher exposures (β’s) to the market return than portfolios
of low interest rate currencies, the differences in unconditional market β’s,
combined with the average return to the market, prove to be insufficiently
large to explain the magnitude of average carry trade returns. However,
Lettau et al. (2014) further observe that the conditional exposures of carry
trade returns to the return on the market when it is down are larger than
their respective unconditional exposures. Based on this observation, they
explore the ability of the downside risk model of Ang et al. (2006) to explain
the high average carry trade returns.
In their empirical analysis, Lettau et al. (2014) define the downside
market return, which we denote R−m,t , as the market return when it is one
sample standard deviation below its sample mean and zero otherwise.22
They run OLS regressions of portfolio returns, R t , on a constant and either
R m,t or R−m,t to define the risk exposures, β and β − . From these risk
exposures, they argue that the expected return on a portfolio can be written
as
E(R t ) = β E(R m,t ) + (β − − β)λ−
where the sample mean return on the market is used for E(R m,t ) and the
price of downside risk, λ− , must be estimated in a cross-sectional regression
of average returns, adjusted for their unconditional exposure to the market,
on the beta differentials.23
22
We also considered two other definitions of R−m,t based on alternative definitions of
the downstate, either R m,t < R m , where R m is the sample mean, or R m,t < 0. These results
(in Table A.2 of the online appendix) are similar to the results reported here.
23
In this section, we follow the approach of Lettau et al. (2014) even though we agree
with Burnside and Graveline (2016), who are critical of this approach. Burnside and
Graveline (2016) note that this restricted two-beta representation of downside risk cannot
34 Kent Daniel et al
Lettau et al. (2014) work with 53 currency returns sorted into six
portfolios. They find that high interest rate differential portfolios have
slightly higher β’s than low interest rate differential portfolios, but these
differential market risks, when combined with the mean return on the
market are insufficient to explain the cross-sectional differences in the
average returns on the portfolios. Nevertheless, the point estimates of
β − −β are sufficiently monotonic that a cross-sectional regression of market-
adjusted average returns, from a variety of assets including currency returns,
on beta differentials produces a large price of downside risk, λ− . Based on
this Lettau et al. (2014) conclude that the average returns on the currency
portfolios are explained by the downside risk model.
To examine this possible explanation of our carry trade returns, we
first run two univariate regressions where the dependent variable is a carry
trade return:
Rt = α + βR m,t + e t
R−t = α− + β − R−m,t + e−
t
The first regression uses all of the data; the second uses data only when
the market return is in the downstate. The downside risk theory requires
that β − be different from β. Hence, as a first step, we explicitly test the
difference between β − and β with a χ 2 (1) constructed from GMM using
Newey and West (1987) standard errors with three lags. Table 8 presents
the results for our eight carry trade portfolios.
In the basic regressions in Panel A we find estimates of unconditional
β’s that are quite small, ranging from -0.03 for the EQ-$ portfolio to 0.07
for the SPD portfolio. Only the β’s of the EQ-0$ and the SPD portfolios
are significantly different from 0 at the .05 marginal level of significance.
The EQ-$ and EQ-D$ strategies, on the other hand, have slightly negative
β’s, which are statistically insignificantly different from 0. Because the
risk factor in these first regressions is an excess return, the α’s can be
interpreted as average abnormal returns, and all of the α’s except the EQ-
0$ are strongly significantly different from zero, as the smallest Newey-West
t-statistic is 3.78.
Panel B of Table 8 examines the downside regression. The estimates of
−
β are also small, ranging from -0.14 for EQ-D$ to 0.15 for SPD. These
be derived from a stochastic discount factor (SDF) model that has the properties that one
would want to impose, including that the SDF is monotonically decreasing as R m,t crosses
the boundary from the down state into the up state.
The Carry Trade: Risks and Drawdowns 35
R t = α + β · R m,t + ε t ,
where R m,t is the CRSP value-weighted market return minus the one-month Treasury bill
return. Panel B presents results from the same regression, but where the sample includes
only months for which the excess market return R m,t was at least one sample standard
deviation below its sample mean. Panel B also reports the χ 2 (1) statistic that tests the
difference in the two slope coefficients (i.e., β and β − ), and the p-value associated with
that χ 2 statistic. Panel C calculates β − − β and uses estimates of downside risk premiums,
λ− ’s, from Lettau et al. (2014) to calculate expected returns on the carry trades from the
downside risk model. The sample period is 1976:02-2013:08 (451 observations). The α
estimates in Panels A and B and the premium estimates in Panel C are annualized, and
GMM-based autocorrelation and heteroskedasticity consistent t-statistics are given in square
brackets.
Interpretation: Strikingly, the estimates of the downside beta (β − ) and the unconditional
beta (β) are almost equal, and the p-value associated with the χ 2 statistic shows that they
are not statistically different. Panel C shows that the estimated magnitudes of the downside
risk premia are small (using either value of λ− ) compared with the estimated premium
presented in Panel A.
Panel A: EQ SPD EQ-RR SPD-RR OPT EQ-0$ EQ-$ EQ-D$
α 3.72 6.08 5.18 5.92 1.98 1.18 2.54 5.63
[4.01] [4.55] [4.50] [5.30] [4.12] [2.06] [3.78] [4.03]
β 0.03 0.07 0.04 0.04 0.02 0.06 -0.03 -0.01
[1.55] [2.13] [1.59] [1.71] [1.51] [4.78] [-1.68] [-0.36]
estimates are either equal to or only slightly larger, in absolute value, than
the corresponding β estimates. The standard errors of the estimates of β −
are sufficiently large that the largest t-statistic, other than for the EQ-0$
portfolio, is only 1.92, which coincides with a marginal level of significance
of .03. The p-values of the tests of the equality of β − and β are all larger
than .16, except for the EQ-0$ portfolio where we can reject the null
hypothesis at the .06 marginal level of significance. Notice also that the
negative estimate of β − for the EQ-D$ strategy indicates that this strategy is
inversely exposed to the market’s downside risk.24 Rather than concluding
that this downside risk model cannot explain our carry trade returns, we
confront our portfolio returns and the estimates of their risk exposures
with the prices of downside risk estimated by Lettau et al. (2014).
Because R−m,t is not a return, one cannot interpret the constants in
these regressions as abnormal returns, which is why Lettau et al. (2014)
perform a cross-sectional analysis of market-adjusted average returns on
risk exposures to estimate the additional price of downside market risk.
To determine how much our estimated exposures to downside risk could
possibly explain the average returns to our carry trades, we combine our
point estimates of β − − β with the point estimates of the price of downside
risk from Lettau et al. (2014), rather than performing our own cross-
sectional analysis on a small number of assets. Lettau et al. (2014) include
assets other than currencies in their cross-sectional analysis and find large
positive prices of downside market risk, depending on the cross-section of
assets included. When they include currencies and equities with returns
measured in percentage points per month, they estimate λ− = 1.41%, or
16.9% per annum. When they include only currencies, they estimate λ−
= 2.18%, or 26.2% per annum. The last two rows of Panel C in Table
8 multiply our estimates of β − − β by either 16.9% or 26.2%. Doing so
provides the explained part of our average carry trade returns that is due to
downside risk exposure. Compared to the α’s in Panel A, the extra return
24
Because the EQ-D$ strategy takes positions in all foreign currencies relative to the
USD based only on the position of the USD interest rate relative to the median interest rate,
it is not strictly a carry trade, and the return on the portfolio when the market is down
could be driven by movements in currencies whose interest rates are more extreme relative
to the median than the dollar interest rate. For example, if the USD interest rate is below
the median interest rate but above the JPY interest rate, the EQ-D$ strategy will go long
the JPY, which could massively appreciate in a market crash as carry trades unwind leading
to a large gain for the EQ-D$ strategy whereas the carry trade strategy would have shorted
the JPY and would have experienced a large loss when the market crashed.
The Carry Trade: Risks and Drawdowns 37
explained by downside risk exposure is minimal for the EQ, EQ-RR, and
OPT strategies. The downside risk premium explains between 0% and
10% of the CAPM α’s of these three strategies. For the SPD and SPD-RR
strategies, the downside risk premium explains between 21.5% and 42.7%
of the CAPM α’s. Notice also that the negative estimate of β − for the EQ-D$
strategy implies that the downside market risk theory cannot explain the
excess return of the EQ-D$ strategy as the additional expected return from
downside risk exposure is actually -2.12% or -3.28%, depending on the
value of λ− .
As a check on the sensitivity of our conclusions about the inability of
downside risk to explain the carry trade, we redo the above downside risk
analysis using the five interest rate sorted portfolios of Lettau et al. (2014).
Portfolio P1 (P5) contains the lowest (highest) interest rate currencies. The
results presented in Table 9 have the same format as Table 8.
Panel A of Table 9 demonstrates that the β’s of these portfolios are
also small, but they are monotonically increasing from 0.01 for P1 to 0.10
for P5. The CAPM α’s are also monotonically increasing from -1.77% [-
1.17] for P1 to 3.27% [1.73] for P5. While these α’s are not particularly
statistically significant, the P5−P1 portfolio has an α of 5.04% with a t-
statistic of 3.76. Panel B of Table 9 shows that the point estimates of β −
also monotonically increase from 0.01 for P1 to 0.22 for P5, but these five
estimates are insignificantly different from zero as the largest t-statistic
is 1.23. The p-values of the tests of the equality of β − to β for the five
portfolios are all larger than .43, and the test for the P5−P1 portfolio has a
.29 marginal level of significance. Although we do not find significant beta
differentials in Panel B, we again combine the point estimates of β − − β
with the estimated prices of downside risk from Lettau et al. (2014), as
above. Panel C of Table 9 shows that the predicted downside risk premiums
are not monotonically increasing from P1 to P5. The P1 portfolio has a
larger downside risk premium than the P2 and P3 portfolios, even though
the CAPM α of the P1 portfolio is -1.77% and the CAPM α’s of the P2 and
P3 portfolios are -0.69% and 0.96%, respectively. Nevertheless, we note
that 36% or 57% of the CAPM α of the P5−P1 portfolio can be explained
by the difference between the downside beta and the unconditional beta
using the point estimates of the annualized downside risk premium of
16.9% or 26.2%. Overall, these results highlight our concerns that the
downside betas are not reliably different from standard betas, and the
resulting differences in the two betas are not sufficiently large to account
38 Kent Daniel et al
Panel A: P1 P2 P3 P4 P5 P5-P1
α -1.77 -0.69 0.96 2.15 3.27 5.04
[-1.17] [-0.40] [0.58] [1.35] [1.73] [3.76]
β 0.01 0.05 0.05 0.06 0.11 0.10
[0.29] [1.29] [1.31] [1.72] [2.20] [3.08]
Panel B: P1 P2 P3 P4 P5 P5-P1
α− 1.16 -3.71 1.88 11.93 14.08 12.92
[0.08] [-0.26] [0.12] [1.02] [0.73] [1.27]
β− 0.01 0.02 0.05 0.15 0.22 0.21
[0.08] [0.15] [0.29] [1.23] [1.00] [1.89]
χ 2 (1) 0.00 0.03 0.00 0.62 0.32 1.11
p-value 0.99 0.85 0.99 0.43 0.57 0.29
Panel C: P1 P2 P3 P4 P5 P5-P1
β −− β 0.00 -0.02 -0.00 0.09 0.11 0.11
Downside Risk Premium β −− β × λ−
for the average returns to carry trade portfolios, even allowing for very
large downside-risk prices.
to the DRI.
The exposure to the market in the shorter sample, β̂MKT = 0.13 [5.26], is substantively
28
larger than the estimate for the full sample, 0.05 [2.46]. An anonymous referee suggests
that the stronger relation between carry trade returns and market returns in this post-
2000 sub-sample is consistent with increased synchronization across different markets,
particularly since the financial crisis. We leave an investigation of this very interesting
phenomenon to future research.
The Carry Trade: Risks and Drawdowns 45
the return to the variance swap only has explanatory power for the hedged
EQ-D$ strategies, and the α’s have reduced statistical significance. Given
the substantive differences between the coefficient estimates in the shorter
sample versus the full sample, we are unwilling to conclude that bond
market risk factors have unconditional ability to explain the unhedged and
the hedged returns to the EQ-D$ strategy.
Table 14 demonstrates that the downside risk indicator (DRI) of Jurek
and Stafford (2015) has strong significance for the EQ and SPD strategies,
both in their unhedged and hedged forms, for the shorter sample. The slope
coefficients are statistically significant, and the α’s are also insignificantly
different from zero indicating that the DRI alone has the power to explain
these carry trades. Consistent with our findings (with a longer sample) in
Section 6.2, the DRI has no ability to explain the unhedged EQ-D$ strategy
as the slope coefficient is essentially zero leaving an α of 4.58%, albeit with
a t-statistic that has a p-value of just .10. The DRI has no ability to explain
the return to the hedged EQ-D$ strategy either, and the α’s remain large
with t-statistics that are marginally significant at the .10 level. Consistent
with our findings for the unhedged strategies, the dynamic dollar strategy
does not appear to be unconditionally significantly exposed to any of the
proposed risk factors, despite its high average returns.
8 Drawdown Analysis
Carry trades are generally found to have negative skewness. The literature
has associated this negative skewness with crash risk. However, negative
skewness at the monthly level can stem from extreme negatively skewed
daily returns or from a sequence of persistent, negative daily returns that
are not negatively skewed. These two cases have different implications
for risk management and for theoretical explanations of the carry trade.
If persistent negative returns are the explanation, the early detection of
increased serial dependence could potentially be used to limit losses. While
the literature has almost exclusively focused on the characteristics of carry
trade returns at the monthly frequency, we now characterize the downside
risks of carry trade returns at the daily frequency while retaining the
monthly decision interval.29
29
While traders in foreign exchange markets can easily adjust their carry trade strategies
at the daily frequency, or even intraday, with minimal transaction costs, we choose to
The Carry Trade: Risks and Drawdowns 49
Table 14: Hedged Carry Trade Exposure to the Downside Risk Index
Description: This table presents regressions of the hedged carry trade returns of EQ, SPD,
and EQ-D$ strategies on the downside risk index (DRI) reported by Jurek and Stafford
(2015). The sample period is 2000:10-2013:08 (155 observations) and includes G10
currencies other than Swedish krona, for which we do not have option data. Results
for unhedged returns over the same sample are also reported. The α’s are annualized
percentages. Autocorrelation and heteroskedasticity consistent t-statistics from GMM are
in square brackets.
Interpretation: Over this shorter time period for which we have option data, the EQ and
SPD strategies have a statistically significant exposure to the DRI returns. This is not
affected by hedging with currency options. However, DRI has no ability to explain the
EQ-D$ strategy returns, before or after hedging.
Unhedged 10∆ 25∆ 10∆ 25∆ 10∆ 25∆
EQ SPD EQ-D$ EQ SPD EQ-D$
α 0.12 1.73 4.58 0.31 0.51 2.15 2.18 4.21 3.72
[0.10] [0.75] [1.66] [0.28] [0.48] [1.09] [1.28] [1.66] [1.65]
βDRI 0.25 0.45 0.03 0.19 0.11 0.37 0.26 0.01 -0.03
[4.97] [3.40] [0.20] [3.54] [1.94] [3.17] [2.41] [0.04] [-0.25]
R2 .16 .16 .00 .11 .05 .14 .09 .00 .00
examine the daily returns to carry trades that are rebalanced monthly to maintain consis-
tency with the academic literature and because we do not have quotes on forward rates for
arbitrary maturities that are necessary to close out positions within the month.
50 Kent Daniel et al
Dτ
Pt,τ + 1 + i t$ t
D1
r x t,τ = τ−1
− 1 + i t$ t
1 + i t$
Pt,τ−1 + Dt
Panel A: Daily Carry Trade Returns Panel B: Monthly Carry Trade Returns
EQ EQ-RR SPD SPD-RR EQ-D$ EQ EQ-RR SPD SPD-RR EQ-D$
Mean Ret (% p.a.) 3.92 5.38 6.53 6.13 5.47 3.96 5.44 6.60 6.18 5.54
(0.82) (0.91) (1.18) (0.93) (1.25) (0.91) (1.13) (1.31) (1.09) (1.37)
The Carry Trade: Risks and Drawdowns
Standard Dev. 5.06 5.54 7.25 5.65 7.68 5.06 5.90 7.62 6.08 8.18
(0.10) (0.11) (0.15) (0.15) (0.11) (0.28) (0.22) (0.41) (0.24) (0.38)
Sharpe Ratio 0.77 0.97 0.90 1.08 0.71 0.78 0.92 0.87 1.02 0.68
(0.17) (0.17) (0.17) (0.18) (0.16) (0.19) (0.20) (0.19) (0.19) (0.18)
Skewness -0.32 -1.01 -0.59 -1.62 0.01 -0.49 -0.37 -0.31 -0.44 -0.11
(0.17) (0.43) (0.33) (0.71) (0.15) (0.21) (0.11) (0.19) (0.14) (0.17)
Excess Kurt. 7.46 11.90 10.25 24.43 4.66 2.01 0.40 1.78 0.90 0.86
(1.16) (6.61) (4.05) (12.02) (1.22) (0.53) (0.21) (0.35) (0.29) (0.31)
Autocorr. 0.03 0.02 0.02 0.02 0.01 0.08 0.16 0.02 0.09 0.00
(0.02) (0.01) (0.01) (0.01) (0.01) (0.07) (0.05) (0.07) (0.05) (0.06)
Max (%) 3.12 2.13 4.52 2.58 5.08 4.78 5.71 8.07 5.96 9.03
Min (% ) -2.78 -5.64 -6.57 -6.61 -3.71 -6.01 -4.90 -7.26 -5.88 -8.27
No. Positive 5230 5230 5223 5223 4988 288 288 297 297 273
No. Negative 4342 4342 4349 4349 4584 163 163 154 154 178
Panel C: Normalized Ratios of Higher Central Moments
EQ EQ-RR SPD SPD-RR EQ-D$
Mean Ret (% p.a.) 1.0 1.0 1.0 1.0 1.0
51
8.1 Drawdowns
30
Chernov et al. (2016) use historical currency return processes and option data to
estimate stochastic volatility jump-diffusion models. We have not attempted to simulate
from these more realistic but decidedly more complex models to generate distributions of
drawdowns and maximum losses.
The Carry Trade: Risks and Drawdowns 53
9 3.2% 0.0% 6 4.4% 0.1% 7 3.1% 4.9% 4 3.4% 2.3% 7 3.9% 5.6% 4
10 3.1% 0.0% 7 4.4% 0.0% 3 3.0% 4.3% 4 3.2% 1.8% 6 3.9% 3.2% 10
The Carry Trade: Risks and Drawdowns 55
of the magnitudes of pure drawdowns stem mainly from the fact that the
consecutive negative returns tend to have larger variances than the typical
returns.31
For the risk-rebalanced strategies, EQ-RR and SPD-RR, we find that
the worst five pure drawdowns lie well within the .05 bounds, while less
extreme pure drawdowns happen more frequently than is implied by the
bootstrap’s .05 bound. For example, the fifth worst pure drawdown of
the SPD-RR strategy is 4.2%, and we observe five pure drawdowns of this
magnitude or larger in 22.1% of the bootstrap simulations. For the EQ-D$
strategy, the most severe pure drawdown was 5.7%, which occurred over
4 days. With a p-value of .23 this would not be considered a particularly
anomalous event. However, the 10-th worst pure drawdown is 3.9%,
which took 10 trading days. The likelihood of 10 pure drawdowns of this
magnitude or worse is only 3.2%, based on our simulations.
To sum up, studying these five carry trade strategies at the daily fre-
quency conveys rich information regarding downside risks. Although the
minimum daily returns are of similar size to the minimum monthly returns,
they do not occur in the same months. Maximum drawdowns occur over
substantial periods of time, often in highly volatile environments, sug-
gesting that extreme negative returns do not happen suddenly and could
possibly be avoided by traders who can re-balance daily. Drawdowns are
much larger than the daily losses, and simulations using an independent
bootstrap distribution fails to match the empirical frequencies of downside
events in most cases. Bootstrapping with a volatility forecasting model
helps to match the frequencies of the most extreme tail events in the data,
but it fails to match the frequencies of less extreme but still severe tail
events.
9 Conclusions
This paper provides some perspectives on the risks of currency carry trades
that differ from the conventional wisdom in the literature. First, it is
generally argued that exposure to the three Fama and French (1993) equity
market risk factors cannot explain the returns to the carry trade. We find
that these equity market risks do significantly explain the returns to an
31
Similarly, in unreported results, we find larger values of pure run-ups than is implied
by the simulations.
The Carry Trade: Risks and Drawdowns 57
equally weighted carry trade that has no direct exposure to the dollar. Our
second finding is also at variance with the literature. We find that our carry
trade strategies with alternative weighting schemes are not fully priced by
the HMLFX risk factor proposed by Lustig et al. (2011), which is basically a
carry trade return across a broader set of currencies. Third, we argue that
the time varying dollar exposure of the carry trade is at the core of carry
trade puzzle. A dynamic dollar strategy earns a significant abnormal return
in the presence of equity market risks, bond market risks, FX risks, and
a volatility risk factor. The dynamic dollar strategy also has insignificant
skewness, indicating that crash risk cannot explain its abnormal return. Our
fourth finding that is inconsistent with the literature is that the exposures
of our carry trades to downside market risk are not statistically significantly
different from their unconditional exposures. Thus, the downside risk
explanation of Dobrynskaya (2014) and Lettau et al. (2014) does not
explain the average returns to our strategies. We find that the downside
risk explanation of Jurek (2014) explains the non-dollar carry trade, but it
also fails to explain our dynamic dollar strategy.
We also show that both spread-weighting and risk-rebalancing the
currency positions improve the Sharpe ratios of the carry trades. It is
surprising that the returns to these strategies continue to earn significant
abnormal returns in the presence of the HMLFX risk factor proposed by
Lustig et al. (2011), which is basically a carry trade risk factor.
The choice of benchmark currency also matters. We show that equally
weighted carry trades can have a Sharpe ratio as low as 0.26 when the JPY
is chosen as the benchmark currency and as high as 0.78 when the USD
is chosen as the benchmark currency. Changing the base currency of the
carry trade dynamically changes the currency exposures which explains
the difference between these carry trade strategies.
Our conclusions about the profitability of the dynamic dollar strategy
differ from those of Lustig et al. (2014) who also investigate a ‘dollar carry’
strategy, but with a larger set of currencies. Empirically, their dollar carry
returns, like ours, have only a small correlation with the US equity market
return. Lustig et al. (2014) develop a reduced-form, multi-country affine
model, which matches many of the moments of the data, including an
unconditional correlation between dollar carry returns and world equity
market returns of .10, close to the .14 in the data. However, we conjecture
that their model would imply significant correlations between dollar carry
returns and both U.S. equity market and U.S. bond market returns, because,
58 Kent Daniel et al
References
Ackermann, F., W. Pohl, and K. Schmedders. 2012. “On the Risk and Return
of the Carry Trade”. Swiss Finance Institute Research Paper. (12-36).
Ang, A., J. Chen, and Y. Xing. 2006. “Downside Risk”. Review of Financial
Studies. 19(4): 1191–1239.
Backus, D. K., F. Gavazzoni, C. Telmer, and S. E. Zin. 2012. “Monetary
Policy and the Uncovered Interest Parity Puzzle”. NBER Working Paper
No. 16218.
Bakshi, G. S., P. P. Carr, and L. Wu. 2008. “Stochastic Risk Premiums,
Stochastic Skewness in Currency Options, and Stochastic Discount
Factors in International Economies”. Journal of Financial Economics.
87(1): 132–156.
Bakshi, G. and G. Panayotov. 2013. “Predictability of currency carry trades
and asset pricing implications”. Journal of Financial Economics. 110(1):
139–163.
Bank for International Settlements. 2014. “Triennial Central Bank Sur-
vey: Global foreign exchange market turnover in 2013”. Available at
http://www.bis.org/.
Bansal, R. and I. Shaliastovich. 2013. “A long-run risks explanation of
predictability puzzles in bond and currency markets”. Review of Financial
Studies. 26(1): 1–33.
Beber, A., F. Breedon, and A. Buraschi. 2010. “Differences in beliefs and
currency risk premiums”. Journal of Financial Economics. 98(3): 415–
438.
Bekaert, G. and G. Panayotov. 2015. “Good Carry, Bad Carry”. Columbia
Business School working paper.
Bhansali, V. 2007. “Volatility and the Carry Trade”. The Journal of Fixed
Income. 17(3): 72–84.
Breedon, F. 2001. “Market liquidity under stress: Observations from the FX
market”. Bank for International Settlements Papers No. 2, Proceedings
of a Workshop on Market Liquidity.
Brunnermeier, M. K., S. Nagel, and L. H. Pedersen. 2008. “Carry Trades
and Currency Crashes”. NBER Macroeconomics Annual. 23(313-342).
Burnside, C. 2012. “Carry trades and risk”. In: Handbook of Exchange Rates.
Ed. by J. James, I. W. Marsh, and L. Sarno. Hoboken, NJ: John Wiley
and Sons. 283–312.
60 Kent Daniel et al