Electronic copy available at: http://ssrn.

com/abstract=1262934
Crash-neutral Currency Carry Trades
Jakub W. Jurek

Abstract
Currency carry trades implemented within G10 currencies have historically delivered signi-
cant excess returns with annualized Sharpe ratios in excess of one. This paper investigates whether
these excess returns reect compensation for exposure to crash risk by analyzing the time-series
dynamics of the moments of the risk-neutral distribution extracted from currency options, and by
examining returns to crash-neutral currency carry trades in which exposure to crashes has been
hedged by combining positions in currencies and currency options. Risk-neutral and realized
skewness are shown to move in opposite directions in response to realized currency returns such
that insurance against currency crashes is cheapest precisely when it is needed most. Although
excess returns to crash-neutral strategies decline relative to their unhedged counterparts, they re-
main positive and highly statistically signicant. The results indicate that crash risk premia can
explain 30-40% of the total excess return to currency carry trades. Rationalizing the entirety of
the excess return via a crash risk premium would require implied volatilities of out-of-the-money
currency options to be roughly four times greater than those observed in the data.
First draft: October 2007
This draft: August 2008

Jurek: Princeton University, Bendheim Center for Finance, e-mail: jjurek@princeton.edu. I thank John Campbell,
Steve Edelstein, Mark Mueller, Erik Staord, and seminar participants at the Harvard Finance Lunch for providing
valuable comments. I am especially grateful to J.P. Morgan for providing the FX option data.
Electronic copy available at: http://ssrn.com/abstract=1262934
Uncovered interest parity predicts that high interest rate currencies should depreciate relative
to low interest rate currencies, such that investors would be indierent between holding the two.
Empirically, not only do high interest rate currencies not depreciate relative to their low interest
rate counterparts, they tend to appreciate. This nding, often referred to as the forward premium
anomaly, is one of the most prominent features of exchange rate data. Using data on foreign exchange
options, I show that 30-40% of the excess returns to currency carry trades exploiting this anomaly
are attributable to a crash risk premium attaching to currencies with high interest rates. However,
unlike in equity option markets, the cost of protection against large adverse currency moves appears
to be relatively cheap. In fact, a complete resolution of the forward premium anomaly via a crash
risk premium argument requires implied volatilities of deep out-of-the-money currency options that
are roughly four times greater than what is actually observed in the data. If true, this would imply
a massive mispricing in the foreign exchange option market.
The currency carry trade exploits the forward premium anomaly by borrowing funds in currencies
with low interest rates and lending them in currencies with high interest rates. This strategy captures
the interest rate dierential (carry) between the two currencies, but leaves the investor exposed to
uctuations in the exchange rate between the high interest rate currency and the low interest rate
currency. Historically, since high interest rate currencies have tended to appreciate relative to their
low interest rate counterparts, the currency exposure has actually turned out to be an additional
source of return. As a result, currency carry trades have been characterized by an extremely attractive
risk-return tradeo. For example, a simple strategy which equal-weighted nine individual carry trades
implemented in currency pairs involving the U.S. dollar and one of the remaining G10 currencies
earned an annualized excess return of 4.78% with an annualized volatility of 5.07% (Sharpe ratio =
0.91) over the period from January 1990 to March 2007.
1
In the latter part of the sample (1999-2007),
simple equal- and spread-weighted portfolios of carry trades delivered even higher Sharpe ratios of
1.26 and 1.48, respectively. To illustrate this, Figure 1 plots the total return indices for the carry
strategy vis a vis the total return indices for the three Fama-French risk factors and the momentum
portfolio. As can be readily seen from the plot, the returns to the carry strategy dominate the returns
to each of the alternatives when the strategies are scaled to have the same ex post volatility.
The returns to currency carry trades are characterized by two main features. First, the volatility
of the strategy is low and stands at roughly one half the volatility of any given X/USD exchange
rate, suggesting that exchange rate innovations are largely uncorrelated in the cross-section. Second,
although the low volatility allows the total return index to have an extremely smooth upwards
progression, the strategy is punctuated with infrequent, but severe, losses. The skewness of the
monthly returns to the equal-weighted carry strategy over the period from January 1990 to March
2007 is negative (-0.95) and its magnitude exceeds the skewness of excess returns on the U.S. equity
market and the momentum portfolio. The very high realized Sharpe ratio of the carry trade and
1
The set of G10 currencies consists of: Australian dollar (AUD), Canadian dollar (CAD), Swiss franc (CHF), Euro
(EUR), British pound (GBP), Japanese yen (JPY), Norwegian krone (NOK), New Zealand dollar (NZD), Swedish
krona (SEK), and the U.S. dollar (USD).
1
the prominence of negative skewness have prompted arguments that the excess returns earned by
currency carry trade strategies represent compensation for exposure to rare, but severe, crashes
in currencies with relatively higher interest rates.
2
The focus on rare events broadly parallels the
approach rst proposed by Rietz (1988), and recently revisited by Barro (2006), Martin (2008) and
Weitzman (2007), in the context of research on the equity premium.
3
This paper contributes to the literature by investigating whether crash risk premia can account
for the empirically observed violations of uncovered interest parity (UIP). Using data on foreign
exchange options, I extract a complete time-series of risk-neutral variance, skewness and kurtosis for
each exchange rate, facilitating an explicit comparison of option-implied and realized moments. Since
the distributions of currency returns embedded in option prices are forward-looking, they provide
a convenient approach for assessing the markets ex ante perceptions regarding the likelihood of
crashes, as well as, the cost of insuring against such events. Contrary to the predication of the
crash risk hypothesis, I nd that risk-neutral skewness does not forecast excess currency returns.
Nonetheless, the dynamics of risk-neutral skewness exhibit very interesting behavior. In particular,
I show that option-implied and realized skewness respond in opposite directions to lagged currency
returns, such that the option-implied skewness forecasts future realized skewness negatively. These
facts suggests that insurance against crashes is actually cheapest precisely when the risk of crashes
is largest.
Finally, I investigate the returns to carry trades in which the risk of currency crashes has been
completely eliminated through hedging in the option market. Consistent with the previous ndings
on the price of crash insurance, I nd that crash-neutral currency carry trades continue to deliver
positive and statistically signicant excess returns, indicating that a crash risk premium alone cannot
reconcile the violations of UIP. Simple equal- and spread-weighted portfolios of crash-neutral currency
carry trades hedged using out-of-the-money (10) options deliver excess returns of 3.18% (t-stat:
3.13) and 5.31% (t-stat: 3.69), respectively. Although mean excess returns decline as the strategies
are hedged with options that are progressively closer to at-the-money, they continue to remain
statistically signicant. When compared with the excess returns to their unhedged counterparts, the
mean excess returns to crash-neutral carry trades are statistically signicantly lower. The dierence
in the mean excess returns suggests that crash risk premia account for 30-40% of the excess returns
to currency carry trades. In order to drive the excess returns to carry trades down to zero, the
implied volatilities of out-of-the-money options hedging against currency crashes would have to have
been nearly four times as large as actually observed in the data.
2
Currency carry trade returns generally appear to be unrelated to risk factors proposed by traditional asset pricing
models (Burnside, et. al (2006)), although Lustig and Verdelhan (2007a, 2007b) dispute these ndings and argue in
favor of a consumption risk factor. Brunnermeier, Gollier, and Parker (2007) argue that high returns of negatively
skewed assets may be a general phenomenon.
3
An extensive literature documents the presence of high volatility and crash risk premia in the equity option market;
see for example, Coval and Shumway (2001), Pan (2002), Bakshi and Kapadia (2003), Driessen and Maenhout (2006)
and references therein.
2
1 The Currency Carry Trade
The expectations hypothesis of exchange rates, also known as uncovered interest parity (UIP),
postulates that investors should be indierent between holding riskless deposits denominated in
various currencies. Equivalently, high interest rate currencies are expected to depreciate relative to
low interest rate currencies, such that the currency return exactly osets the interest rate dierential.
If this were the case, the forward exchange rate for a currency the rate at which one can contract
to buy/sell a foreign currency at a future date would provide an unbiased estimate of the future
exchange rate. This relationship is strongly violated in the data leading to a forward premium
anomaly, whereby high interest rate currencies actually tend to appreciate, rather than depreciate,
against their low interest rate counterparts.
The currency carry trade is designed to exploit this anomaly and involves borrowing funds in
a currency with a low interest rate and lending them in a currency with a high interest rate. At
some future date the proceeds from lending in the high-interest rate currency are converted back
into the funding currency, and used to cover the low-interest rate loan. The balance of the proceeds
constitutes the prot/loss from the carry trade, and can be thought of as a combination of the interest
rate dierential (carry) and the realized currency return. However, since the carry is known ex ante,
and is riskless in the absence of counterparty risk, the sole source of risk in the carry trade stems from
uncertainty regarding future exchange rates. Principally, the carry trade exposes the arbitrageur to
rapid depreciations (crashes) of the currency which he is long vis a vis the funding currency. The
next section documents the violations of UIP in the data and characterizes the historical returns to
the currency carry trade implemented in the set of G10 currencies.
1.1 Uncovered Interest Parity
Suppose we denote the exchange rate expressed as the price of one unit of foreign currency in
terms of domestic currency by S
t
, and the foreign and domestic interest rates for -period loans
as r
f,t
and r
d,t
, respectively. Then, the forward rate, F
t,
, which is the time t price for one unit of
foreign currency to be delivered periods later, is determined through no arbitrage and satises:
F
t,
= S
t
exp ((r
d,t
r
f,t
) ) (1)
This relationship is known as covered interest parity and is essentially never violated in the data
(Burnside, et al. (2006)). When the foreign interest rate, r
f,t
, exceeds the domestic interest rate,
r
d,t
, S
t
is above F
t,
and the foreign currency is said to trade at a premium to its forward. Conversely,
when r
d,t
exceeds r
f,t
, the foreign currency trades at a discount (F
t,
> S
t
). Under UIP, the forward
rate provides an unbiased estimate of the future spot exchange rate, a condition which is stated
either in levels or logs,
F
t,
= E
t
[S
t+
] or f
t,
= E
t
[s
t+
] (2)
3
If true, investors should be indierent between buying the foreign currency at time t in the forward
market and converting their domestic currency to the foreign currency, investing in the foreign riskless
bond and re-converting their investment proceeds back to the domestic currency at the future date.
Empirical work, starting with Hansen and Hodrick (1980) and Fama (1984), tests UIP by regressing
currency returns on the forward premium, dened as the dierence between the prevailing forward
and spot prices.
4
When expressed in logs the regression test takes on the following form,
s
t+1
s
t
= a
0
+a
1
(f
t
s
t
) +
t+1
H
0
: a
0
= 0, a
1
= 1 (3)
The null hypothesis under UIP is that the currency return is, in expectation, equal to the forward
premium, which is given by the interest rate dierential, (r
d,t
r
f,t
) . Although the empirical
prediction of this theory only holds in the absence of currency risk premia (or, in the presence of
risk-neutral investors), it constitutes a useful benchmark for examining the data.
Table I presents the results of UIP regressions for nine currency pairs, each containing one of the
G10 currencies and the U.S. dollar, which is assumed to be the investors domestic currency. Currency
returns are computed for 21-day rolling windows and span the period from January 1990 to March
2007. The right panel presents the results for the January 1999 to March 2007 subperiod, which
is the focus of the ensuing sections due to the simultaneous availability of foreign exchange option
data. The (log) forward spread is measured using the dierential between the U.S. eurocurrency
rate and the foreign eurocurrency rate for one-month deposits. For the full subperiod, the null of
UIP (H
0
: a
0
= 0, a
1
= 1) is rejected at the 5% signicance level for four of the nine countries. Of
the remaining ve, another two have negative slope coecients, a
1
, indicating that high-interest rate
currencies tend to appreciate relative to their low-interest rate counterparts. For three currencies
the British pound, the Norwegian krone and and Swedish krona UIP cannot be rejected at
conventional signicance levels. Consequently, when UIP is evaluated in the context of a panel
regression with xed country eects, the null cannot be rejected, albeit the slope coecient once
again has the wrong sign. UIP fares considerably worse in the second part of the sample (1999-
2007). The null hypothesis is rejected at the 5% (10%) signicance level in six (all) of the G10
countries, with negative point estimates for the slope coecients in all countries. Unsurprisingly, the
panel regression, resoundingly rejects UIP in this subperiod.
Cross-sectional regressions also consistently indicate that the further a currencys lending rate is
above (below) the U.S. lending rate, the greater the anticipated appreciation (depreciation). Unlike
in the time series regressions, in which the adjusted R
2
of the forecasting regression is rarely above
1%, the cross-sectional R
2
is an order of magnitude higher. This suggests that although a trade
aimed at exploiting violations of UIP in a single currency may experience quite variable performance
over time, a portfolio trade exploiting the entire cross section of currency pairs is likely to be quite
lucrative. Moreover, a portfolio of carry trades in which the weights of the individual currencies
4
Froot and Thaler (1990), Lewis (1995) and Engel (1996) survey the vast empirical literature on tests of UIP.
4
are set in proportion to the interest rate dierential is predicted to outperform an equal-weighted
strategy.
Although the amount of predictability in the foreign exchange rate return at the one-month horizon
is generally small, as evidenced by the low adjusted R
2
values, the regressions indicate that investors
can earn excess returns by borrowing funds in the relatively low interest rate currency and lending
them in the currency with the relatively high interest rate. This strategy, known as the currency
carry trade, in reference to the interest rate dierential (carry) it earns, has historically been one of
the most popular foreign exchange strategies. The next sections describe the carry trade strategy in
detail and examine its historical performance to assess the economic signicance of deviations from
UIP.
1.2 Exploiting deviations from UIP
In the standard carry trade an investor borrows money in a currency with a low interest rate
and invests the proceeds in a currency with a high interest rate. We denote the price of the foreign
currency in terms of the domestic currency taken to be the U.S. dollar by S
t
, such that an increase
(decrease) in S
t
corresponds to an appreciation (depreciation) of the foreign currency relative to the
dollar. To illustrate the payo to this strategy consider an investor aiming to exploit the -month
interest rate dierential between the domestic currency, bearing a low interest rate, and a foreign
currency, bearing a high interest rate.
The investor begins by borrowing S
t
dollars at a rate of r
d,t
for a period of months in order to
nance the purchase of one unit of the foreign currency. After converting the funds to the foreign
currency he lends them out for a period of months at the then prevailing foreign interest rate
r
f,t
. When the domestic currency is the relatively high interest rate currency, the investor employs
a symmetric strategy and shorts one unit of the foreign currency. At time t + the payo to the
strategy is given by:

CT
t+1
=
_
r
f,t
> r
d,t
: exp (r
f,t
)

S
t+
exp (r
d,t
) S
t
r
d,t
> r
f,t
: exp (r
d,t
) S
t
exp (r
f,t
)

S
t+
(4)
If uncovered interest parity (UIP) held, the expected payo to the carry trade would be zero, since
S
t+
= F
t,
, and the change in the exchange rate would exactly oset the interest dierential (carry).
However, in the presence of the empirically observed violations of UIP, the expected excess return
to the currency carry trade is positive. The corresponding carry trade return can be obtained by
standardizing the payo by the funding capital in this case given by the value of one unit of foreign
currency at initiation, S
t
to obtain:

R
CT
t+1
=
_
_
_
r
f,t
> r
d,t
: exp (r
f,t
)
_

S
t+
St
_
exp (r
d,t
)
r
d,t
> r
f,t
: exp (r
d,t
) exp (r
f,t
)
_

S
t+
St
_
(5)
5
Whenever r
d,t
< r
f,t
, the carry trader is long exposure to the foreign currency, and loses money
if the foreign currency depreciates relative to the funding currency by more than the interest rate
dierential. Symmetrically, when r
f,t
< r
d,t
, the carry trade investor borrows in the foreign currency
and is short exposure, and stands to earn a negative excess return if the foreign currency appreciates.
Put dierently, if exchange rates were dened as the price of the high interest rate currency in terms
of the low interest rate currency, carry traders stand to earn a negative return in the event of a rapid
depreciation, or crash, of the high interest rate currency.
1.3 Historical performance
The historical performance of the currency carry trade is summarized in Tables II (1990-2007)
and III (1999-2007). The strategy is implemented at a monthly frequency, with positions established
at the end of month t 1 and held until the end of month t. Within each pair, to determine
which currency will be the long (short) leg of the trade I compare the one-month Eurocurrency rates
prevailing at the end of month t 1. Finally, since exchange rates are expressed in terms of units of
domestic currency (U.S. dollar) per unit of foreign currency, the excess returns computed from (5)
are interpretable as U.S. dollar returns.
Over the full sample period the carry trade delivers positive excess returns in all nine currency
pairs. For ve of the individual currency pairs the excess returns are statistically dierent from
zero at the 5% signicance level. The annualized Sharpe ratios vary from 0.16 (CHF/USD pair)
to 0.79 (SEK/USD pair). Notably, however, the carry trade returns exhibit pronounced departures
from normality, with signicant negative skewness and excess kurtosis. The minimum and maximum
monthly returns are large in magnitude and represent roughly 3.5 standard deviation moves, when
compared with the the historically realized monthly volatility. These features are broadly represen-
tative of the the carry trade and parallel the results reported in the existing literature (e.g. Burnside,
et al. (2006), Brunnermeier, Nagel and Pedersen (2008)).
Panel B of Tables II and III presents the analogous summary statistics for portfolios of the in-
dividual currency carry trades. Portfolios weighting the individual carry trades equally (EQL), or
by the absolute interest rate spread at initiation (SPR), deliver positive and statistically signicant
returns with annualized Sharpe ratios close to (1990-2007) or above (1999-2007) one. Of the total
return, between one-third and one-half comes from the carry computed as the absolute value of
the annualized interest rate dierential with the remainder attributable to the currency return.
Interestingly, the annualized volatility of the portfolio strategy is less than the mean annualized
volatility of the individual currency pairs suggesting that the currency returns from the individual
pairs are generally uncorrelated. However, the lack of correlation does not facilitate diversication of
skewness or kurtosis. In fact, the returns to portfolios of carry trades are actually even more skewed
and heavy-tailed than the underlying pairs. To summarize the data, Figure 2 plots the returns to the
portfolio strategies and the nine underlying currency pairs in mean-standard deviation space for the
1999-2007 period. The portfolios lie very close to the mean-standard deviation frontier, suggesting
6
that even a simple portfolio construction method delivers considerable benets due to diversication.
Moreover, the Sharpe ratios of the the equal- (1.26) and spread-weighted (1.48) strategies are very
close to the Sharpe ratio of the the ex post mean-standard deviation frontier (1.54), indicating that
these naive portfolio strategies are surprisingly ecient.
5
Finally, to ensure that these characteristics of the currency carry trade are not related to the net
dollar exposure of the portfolio strategies, the bottom panel also presents the results for an equal-
and spread-weighted carry trade, in which the weights are constructed such that the exposure to
the U.S. dollar nets to zero. These strategies are labeled EQL-$N and SPR-$N, respectively. The
EQL-$N strategy is constructed by assigning weights of
1
ms
to the m
s
carry trades which are short
the U.S. dollar and long the foreign currency (r
f,t
> r
d,t
), and weights of
1
m
l
to the m
l
carry trades
which are long the U.S. dollar and short the foreign currency (r
d,t
> r
f,t
). The SPR-$N strategy
is constructed analogously, but with weights that are assigned in proportion to the absolute interest
rate dierential computed within the two sets. As can be readily seen, these strategies continue to
deliver positive and statistically signicant returns, as well as, retain the high negative skewness and
kurtosis of the other non-dollar-neutral strategies. This suggests that the main characteristics of the
carry trade strategy are not attributable to net dollar exposure.
6
Notably, the return volatilities
of the dollar-neutral strategies tend to be somewhat higher that those of the standard equal- and
spread-weighted portfolio strategies. This feature may be due to excess co-movement of the relatively
high- (low-) interest rate currencies, which eectively limits the amount of attainable diversication.
Aside from its high historical excess returns, the dening characteristic of the currency carry
trade is the high negative skewness of the realized returns. Brunnermeier, Nagel and Pedersen
(2008), for example, argue that realized skewness is related to rapid unwinds of carry trade positions,
precipitated by shocks to funding liquidity. They show that funding liquidity measures predict
exchange rate movements, and that controlling for the supply of arbitrage capital helps explain
violations of UIP. In a related paper, Plantin and Shin (2008) provide a game-theoretic motivation of
how strategic complementarities, which lead to crowding in carry trades, can endogenously generate
currency crashes. In their model, the simultaneous entry of currency speculators causes currencies
with relatively high (low) interest rates to appreciate (depreciate). Consequently, the greater the
mass of speculators entering the carry trade, the more likely it is to deliver a positive excess return,
and the greater the potential for a future unwind, or crash. This prediction of their model indeed
nds empirical support, in that the change in monthly realized skewness (computed from daily carry
trade returns) is negatively related to past carry trade returns.
7
However, since the traders are
5
DeMiguel, Garlappi and Uppal (2007) nd that equal-weighted strategies involving portfolios of equities perform
favorably relative to portfolios prescribed by more sophisticated portfolio construction models,due to errors in estimating
the parameters required by those models (e.g. means and variances).
6
Carry trades constructed for investors whose domestic currency is one of the other nine G10 currencies have
features that are qualitatively comparable to those presented in Tables II and III. Results are available from the author
by request.
7
In unreported results, a panel regression of the monthly change in realized skewness on the lagged carry trade
return produces a coecient of -7.15 (t-stat: -7.07) and an adjusted R
2
of 4.89%.
7
risk-neutral in Plantin and Shins (2008) model, there are no excess returns (risk premia) to holding
high interest rate currencies. Long periods of gradual appreciation of high interest rate currencies
are followed by rapid crashes, such that traders in equilibrium are indierent between currencies
with high and low interest rates. Consequently, while their model helps explain the observed negative
skewness, it is silent on the source of the observed violations of uncovered interest parity.
2 Assessing and Hedging Exposure to Currency Crashes
The persistent protability of currency carry strategies has led to a search for underlying risk
factors responsible for the high excess returns garnered by currencies with high interest rates (Farhi
and Gabaix (2008), Lustig, Roussanov and Vedelhan (2008)). Importantly, risk factors postulated
by traditional asset pricing models (CAPM, C-CAPM, Fama-French, etc.) appear to be entirely
unrelated to the returns on carry trades (Burnside, et. al (2006)).
8
Consequently, although the
theoretical features of the currency risk premium necessary to match the empirical data on the
violation of UIP have been known since Fama (1985), the asset pricing literature has struggled with
providing a plausible rational model of this anomaly.
9
Intuitively, the challenge to rational models arises from the extremely high historical Sharpe ratio
of the carry trade, which suggests that the price of risk on the hidden underlying risk factor is
nearly twice that of the equity market. This has led rational theories to consider the importance
of rare, but extreme, crashes. For example, Farhi and Gabaix (2008) present a theoretical model in
which the forward premium anomaly is generated by a time-varying exposure to diaster risk. This
approach mirrors that taken in the equity premium literature (Rietz (1988), Barro (2006), Weitzman
(2007), Martin (2008)) and the literature seeking to reconcile the prices of deep out-of-the-money
puts with empirical return distributions (Pan (2002)). Indeed, Coval and Shumway (2001), Bakshi
and Kapadia (2003), and Driessen and Maenhout (2006) report high Sharpe ratios for various delta-
neutral option strategies, which can be interpreted as being consistent with large volatility and crash
risk premia. Consequently, exposure to a highly priced crash risk factor attaching to currencies with
relatively high interest rates provides a plausible mechanism for explaining the observed violations
of uncovered interest parity.
In order to examine the crash risk hypothesis, I turn to data on foreign exchange (FX) options and
begin by introducing methodologies used to assess and hedge exposure to currency crashes. The price
of options protecting against the risk of rapid devaluations provides valuable information regarding
the probability of currency crashes, as well as, the risk premia demanded by investors for being
8
Lustig and Verdelhan (2007a, 2007b), dispute these claims and argue that returns to carry strategies represent
compensation for exposure to real U.S. consumption risk.
9
Backus, Telmer and Foresi (2001) show that in order to account for the anomaly in an ane model, one has to either
allow for state variables to have asymmetric eects on state prices in dierent currencies or abandon the requirement
that interest rates be strictly positive. Verdelhan (2007) argues that the forward premium anomaly is consistent with
time-varying, countercyclical risk premia generated by country-specic habit processes. The model, however, requires
high eective risk aversion levels, as well as, signicant restrictions on risk sharing.
8
exposed to those risks. To examine the markets ex ante perceptions of crash risk, I rst use the
option data to extract the moments of the risk-neutral distribution. A comparison of the dynamics
of the risk-neutral moments in particular risk-neutral skewness vis a vis realized skewness,
provides surprising insights regarding the markets perceptions of tail risk. Finally, I introduce
the construction of crash-neutral currency carry trades, in which exposure to rapid depreciations
(appreciations) of the high (low) interest rate currencies has been hedged in the option market. To
assess whether violations of UIP are attributable to crash risk premia, the empirical sections examine
the excess returns to the crash-neutral strategies and compare them to the returns obtained from
the unhedged carry strategy.
2.1 The markets perception of crash risk
Breeden and Litzenberger (1978) were the rst to show that an assets entire risk-neutral distri-
bution (i.e. state price density) can be recovered from the prices of a complete set of options on that
asset. Following the logic of state-contingent pricing (Arrow (1964), Debreu (1959)), the risk-neutral
distribution, q(S), enables one to value arbitrary state contingent payos, H(S), via the following
pricing equation:
p
t
= exp(r
d,t
)
_

0
H(S
t+
) q(S
t+
)dS
t+
(6)
In particular, if we denote the continuously compounded return by, R
t,
ln S
t+
ln S
t
, the values
of the (non-central) -period moments under the risk-neutral measure can be simply computed
by setting H(S
t+
) = (R
t,
(S
t+
))
n
and removing the discounting. The corresponding discounted
values can be interpreted as the prices of contracts paying the realized (non-central) moments of
the distribution. After a few simple transformations these values can then be converted to the
prices of contracts paying the realized central moments (variance, skewness, etc.). Consequently,
this approach to deriving risk-neutral moments, developed in Bakshi, Kapadia, and Madan (2003),
can be thought of as an extension of the early results of Britten-Jones and Neuberger (2000) and
Carr and Madan (2001) on the pricing of variance swaps.
Rather than begin by extracting the entire risk-neutral distribution, q(S), Bakshi and Madan
(2000) show that any payo function with bounded expectation can be spanned by a continuum of
out-of-the-money call and put payos. This implies that the price, p
t
, of an asset paying, H(S
t+
),
can be conveniently obtained by valuing the relevant replicating portfolio of options. Specically, if
the payo function is twice-dierentiable, the assets price can be obtained from:
p
t
= exp(r
d,t
) (H(S) S) +H
S
(S) S
t
+
+
_

S
H
SS
(K) C
t
(K, )dK +
_
S
0
H
SS
(K) P
t
(K, ) (7)
9
where H
S
() and H
SS
(), denote the rst and second derivatives of the state-contingent payo, and
S is some future value of the underlying, typically taken to be the forward price. Intuitively, this
expression states that the payo H(S) can be synthesized by buying (H(S) S) units of a riskless
bond, H
S
(S) units of the underlying security and a linear combination of puts and calls with positions
given by H
SS
(K).
Since the state-contingent payos, H(S) = (R
t,
(S
t+
))
n
, satisfy the above technical conditions,
we can obtain the prices of the non-central moment swaps, by substituting the relevant derivatives
into (7). The expressions for the discounted values of the rst three non-central moments, denoted
by V
t
(), W
t
() and X
t
(), are provided in the Appendix. To x intuition regarding the associated
option portfolios, Figure 3 plots the option positions, H
SS
(), for each of the moment contracts
as a function of the option strike value. For example, replication of the second moment requires
establishing long positions in options at all moneyness levels, with sizes that scale inversely with
the square of the option strike value. As a result, the price of the second moment contract is
strictly positive, and is particularly high whenever the prices of deep out-of-the-money puts are high.
The third non-central moment is replicated by a combination of negative positions in out-of-the-
money puts and positive positions in out-of-the-money calls. Consequently, whenever the underlying
distribution is negatively (positively) skewed and the prices of puts are greater than (less than) the
prices of calls, the price of the replicating portfolio will be negative (positive). Finally, replication
of the contract paying the realized fourth moment, once again entails strictly positive positions in
options of all moneyness values, but now more heavily weighed in the tails. Once the prices of the
three non-central moment contracts have been computed, the risk-neutral variance, skewness, and
kurtosis can be obtained from:
VAR
Q
t
() = exp(r
d,t
) V
t
()
t
()
2
(8)
SKEW
Q
t
() =
exp(r
d,t
) (W
t
() 3
t
() V
t
()) + 2
t
()
3
(exp(r
d,t
) V
t
()
t
()
2
)
3
2
(9)
KURT
Q
t
() =
exp(r
d,t
)
_
X
t
() 4
t
(t) W
t
() + 6
t
()
2
V
t
()
_
3 (t, )
4
(exp(r
d,t
) V
t
()
t
()
2
)
2
(10)
where:

t
() = exp(r
d,t
)
_
V
t
()
2
+
W
t
()
6
+
X
t
()
24
_
(11)
The Q superscripts are used to denote the fact that the moments are computed under the risk-neutral
measure, in contrast to realized moments computed from historical data, which will be denoted with
P superscripts.
When applied to data on foreign exchange options, these expressions allow me to dynamically
extract the risk-neutral moments of the option-implied currency return distribution, resulting in
a daily time series of option-implied variance, skewness and kurtosis observations. Importantly,
10
the time variation in option-implied skewness provides a direct way for assessing the markets time-
varying perceptions of crash risk, and the cost of insuring against extreme currency moves. In Section
4, I characterize the behavior of risk-neutral skewness in relation to the actual realized skewness, and
relate both to measures of the attractiveness of currency carry trades (e.g. interest rate dierentials),
as well as, recent currency moves. I nd that the realized and option-implied skewness measures
exhibit dramatically dierent behavior. While future realized skewness is negatively related to past
currency appreciations, suggesting that appreciated currencies are more likely to crash, the risk-
neutral skewness is positively related to past currency appreciations, suggesting that the market
perceives appreciated currencies as less likely candidates for a crash. As a result, insurance against
crash risk becomes least expensive, precisely when it is needed most.
2.2 Crash-neutral currency carry trades
In order to provide a returns-based measure of the crash risk premium, I also construct carry
trades in which the spot currency positions of the standard currency carry trade are combined
with a position in a foreign exchange option. The option position is chosen such that the risk of
extreme negative outcomes stemming from a depreciation (appreciation) of the high (low) interest
rate currency is entirely eliminated. More precisely, whenever the foreign short-term instrument is
the long leg of the trade, an investor seeking to limit exposure to sudden depreciations purchases a
put option on the foreign currency. Conversely, if the carry trade is funded in the foreign currency, an
investor seeking to limit downside exposure purchases a call option, limiting the risk from a sudden
appreciation. I refer to these downside-protected trades as crash-neutral carry trades. They are
constructed to have two features: (1) conditional on the option protection expiring in-the-money all
currency risk exposure is eliminated; and, (2) the currency exposure of the crash-neutral portfolio
matches that of the standard carry trade (i.e. the delta exposure of the option is hedged at initiation).
Intuitively, the rst condition ensures that exposure to crash risk is entirely eliminated, while the
second, ensures that the returns from the crash-neutral carry strategies are directly comparable with
those from the standard strategy presented in Section 1. To test whether the excess returns to the
currency carry trade can be attributed to a crash risk premium, I compare the standard carry trade
to three variants of the crash-neutral carry trade, diering in the amount of downside protection
oered by the option overlay.
10
The returns to portfolio strategies combining crash-neutral trades
complement the analysis of risk-neutral moments, and indicate that 30-40% of the excess returns to
standard carry trades may indeed be interpreted as compensation for exposure to currency crashes.
Before turning to the data, however, I provide a detailed description of how the crash-neutral carry
trades are constructed.
10
Burnside, et al. (2008) construct similar currency strategies, however, their panel contains a smaller cross-section
of countries and only examines carry trades hedged using at-the-money options. Since the focus is on eliminating
exposure to extreme moves, deep out-of-the-money options are more relevant, as they provide the most direct measure
of the cost of insuring against crashes. Moreover, since their strategies do not hedge the delta exposure of the option
overlay the returns cannot be directly compared with returns from the unhedged carry trade.
11
2.2.1 Portfolio construction
First, consider the situation when the foreign interest rate, r
f,t
, exceeds the domestic interest
rate, r
d,t
. In order to take advantage of the deviation from UIP, the trader would like to establish
a long position in the foreign currency. In the standard carry trade, this long position exposes the
carry trader to losses in the event of a sudden depreciation of the foreign currency. To protect against
these losses the carry trader can purchase FX puts with a strike price of K
p
at a cost of P
t
(K
p
, )
dollars per put. If the carry trader purchases q
p
puts, he must also purchase an additional q
p
units of the foreign currency, to hedge the negative delta of the put options. Consequently, if the
trader started by buying one unit of the foreign currency as in (5) he must now buy an additional
q
p

p
units of the foreign currency. To fund this position he must borrow an additional q
p

p
S
t
in his domestic currency. Finally, we assume the purchase price of the puts is covered by borrowing
additional funds in the domestic currency. At time t + 1 the payo to this portfolio is given by:

CT
CN
t+1
(r
f,t
> r
d,t
) = exp(r
f,t
) (1 q
p

p
)

S
t+1
+q
p
max(K
p

S
t+1
, 0)
exp(r
d,t
) ((1 q
p

p
) S
t
+q
p
P
t
(K
p
, )) (12)
In order to eliminate all currency exposure below the strike price of the option, K
p
, the quantity of
puts must satisfy,
q
p
= exp(r
f,t
) (1 q
p

p
) q
p
=
exp(r
f,t
)
1 + exp(r
f,t
)
p
(13)
With the above quantity restriction, the payo equation can be re-expressed as:

CT
CN
t+1
(r
f,t
> r
d,t
) = q
p
max(K
p
,

S
t+1
) exp(r
d,t
) ((1 q
p

p
) S
t
+q
p
P
t
(K
p
, )) (14)
This expression makes transparent that the payo to the strategy is bounded from below, and that
for terminal realizations of the exchange rate that are above K
p
, the strategy payo response is
steeper than in the standard carry trade. In the standard carry trade the sensitivity to changes in

S
t+1
is equal to exp(r
f,t
), whereas in the crash-neutral strategy it is given by q
p
which is strictly
greater than exp(r
f,t
) since
p
< 0. The payo to the crash-neutral carry trade is illustrated vis a
vis the payo to the standard carry trade in the left panel of Figure 4. The crash-neutral carry trade
is assumed to include a put option struck 2.5% out-of-the-money relative to the prevailing forward
rate (
Kp
Ft
= 0.9750).
By simultaneously decreasing exposure to depreciations (crashes) of the high interest rate currency
and increasing exposure to its appreciations, the crash-neutral strategy is able to maintain the same
unconditional ex ante exposure as the standard carry trade, as desired. Moreover, as the put is struck
progressively further out-of-the-money and oers less protection, the delta of the put converges to
zero, causing the upside exposure of the crash-neutral trade, q
p
, to converge to that of the standard
carry trade. In this sense, the crash-neutral strategy nests the payo to the standard carry strategy.
12
Now consider the situation when the domestic interest rate, r
d,t
, exceeds the foreign short rate,
r
f,t
. In order to take advantage of the UIP violations, the investor borrows $1 in foreign currency
and invests the proceeds in a combination of the short-term domestic bonds and FX calls. Because
the position involves borrowing at the foreign short rate it is eectively short the foreign currency,
and the call options limit its downside exposure. Suppose the investor were to buy, q
c
calls with
a strike price of K
c
, at a price of C
t
(K
c
, ) dollars per call. In order to eliminate the additional
exposure stemming from the long position in the calls (
c
> 0), the investors shorts q
c
units of the
foreign currency in addition to the baseline short position of one unit (as in the standard currency
carry trade). Finally, the investor funds the purchase of the call options by borrowing the funds at
the domestic interest rate, r
d,t
.
11
At time t + 1 the payo to the portfolio is:

CT
CN
t+1
(r
d,t
> r
f,t
) = exp(r
d,t
) ((1 +q
c

c
) S
t
q
c
C
t
(K
c
, )) +q
c
max(

S
t+1
K
c
, 0)
exp(r
f,t
) (1 +q
c

c
)

S
t+1
(15)
Once again, the requirement that all currency exposure be eliminated above K
c
implies that the call
quantity satisfy:
q
c
= exp(r
f,t
) (1 +q
c

c
) q
c
=
exp(r
f,t
)
1 exp(r
f,t
)
c
(16)
which allows us to simplify the payo to:

CT
CN
t+1
(r
d,t
> r
f,t
) = exp(r
d,t
) ((1 +q
c

c
) S
t
q
c
C
t
(K
c
, )) q
c
min(K
c
,

S
t+1
) (17)
The payo makes clear that the portfolio is protected against appreciations of the low interest foreign
currency beyond the strike price of the option K
c
. By contrast, relative to the standard carry trade,
the crash-neutral strategy increases exposure to depreciations of the funding currency. While the
payo of the standard carry trade responds by exp(r
f,t
) to moves in the foreign exchange rate,
the crash-neutral strategy responds by q
c
, whenever

S
t+1
remains below K
c
. Since
c
> 0, one can
see that q
c
> exp(r
f,t
). The payo to this crash-neutral carry trade is illustrated vis a vis the
payo to the standard carry trade in the right panel of Figure 4. The crash-neutral carry trade is
assumed to include a call option struck 2.5% out-of-the-money relative to the prevailing forward rate
(
Kc
Ft
= 1.0250). Once again, as the call option is struck at a progressively higher price, oering less
protection, its delta converges to zero, such that the payo to the crash-neutral strategy converges
to the payo of the standard carry trade.
Finally, to compute the returns to the carry trade we divide the payos by the dollar value of the
capital necessary to establish the positions. In the case of the portfolio which includes puts and is
long the foreign currency, the funding capital is (1 q
p

p
)S
t
+q
p
P
t
(K
p
, ). For the portfolio which
11
Formally, the investor could fund the purchase of the calls at the lower, foreign interest rate. The assumption of
funding at the domestic rate is made to preserve the symmetry of the solution, which is lost when additional currency
risk is borne by the investor when he funds the purchase of the calls at the foreign rate.
13
is long the domestic currency and includes calls, the funding capital is (1 +q
c

c
) S
t
q
c
C
t
(K
c
, ).
When the quantity of options in the portfolio goes to zero, the funding capital goes to S
t
in both
cases, as in the case for the standard carry trade. Intuitively, the standard carry trade is simply
the limiting case of the crash-neutral trade as we let the strike price of the put (call) diverge to
zero (innity). In these cases, the prices of the FX options go to zero, and the two strategies oer
identical payos.
3 Data
The results in this paper are based on two datasets. The rst dataset contains information on one-
, three-, six-month and one-year Eurocurrency (LIBOR) rates and was obtained from Datastream.
Eurocurrency rates are the interest rates at which banks are willing to borrow and lend foreign
currency deposits, and constitute the eective interest rates at which a carry trade investor would
be able to borrow and lend. The LIBOR rates published by Datastream are from daily xings of
the British Bankers Association (BBA). The data are daily and cover the period from January 1990
through March 2007. In the event that BBA interest rate data are unavailable for the entire time
period, the BBA series is spliced with the corresponding interbank lending rate published by the
countrys central bank, provided through Datastream or Global Financial Data. Time series means
of the one-month LIBOR rates are reported in Panel A of Table IV. Daily exchange rates for the
nine G10 currencies versus the U.S. dollar are obtained from Reuters via Datastream.
The second dataset, the foreign exchange (FX) option dataset, is comprised of daily implied
volatility quotes at ve strikes and four maturities for options on the G10 currencies and fteen
emerging currency pairs. The exchange rate options are European and give their owners the right
to buy or sell a foreign currency at a pre-specied exchange rate measured in U.S. dollars per unit
of foreign currency. The data on exchange rate options was obtained from J.P. Morgan and covers
the period from January 1999 to March 2007. I begin by introducing some formalisms specic to the
foreign exchange option markets, characterizing the structure of implied volatilities, and describing
the procedure for converting the implied volatility data into observations of risk-neutral moments.
3.1 Foreign exchange options
FX option prices are quoted in terms of their Garman-Kohlhagen (1983) implied volatilities,
much like equity options are quoted in terms of their Black-Scholes (1973) implied volatilities. In
fact, the Garman-Kohlhagen valuation formula is equivalent to the Black-Scholes formula adjusted
for the fact that both currencies pay a continuous yield given by their respective interest rates. The
price of a call and put option can be recovered from the following formulas:
C
t
(K, ) = e
r
d,t

_
F
t,
N(d
1
) K N(d
2
)
_
(18a)
P
t
(K, ) = e
r
d,t

_
K N(d
2
) F
t,
N(d
1
)
_
(18b)
14
where:
d
1
=
ln F
t,
/K

t
(K, )

+
1
2

t
(K, )

d
2
= d
1

t
(K, )

(19)
and F
t,
is the forward rate for currency to be delivered periods forward, and r
f,t
and r
d,t
are the
foreign and domestic interest rates for -period loans, respectively. The forward rate is determined
through the covered interest parity condition, a no-arbitrage relationship which must hold at time t,
and is equal to S
t
exp {(r
d,t
r
f,t
) }. The implied volatilities necessary to match the price of the
-period options will generally depend on the options strike value, K, and are denoted by
t
(K, ).
Unlike equity options which have xed calendar expiration dates and are quoted at xed strike
prices, foreign exchange options are generally quoted at constant maturities and xed deltas. More
precisely, market makers quote prices of 0.25 and 0.10 delta risk reversals and buttery spreads,
as well as, an at-the-money delta-neutral straddle. The strike price of the straddle, for any given
maturity, is chosen such that the deltas of a put and call at that strike are equal, but of opposite
sign. From these data, one can compute implied volatilities at ve strike values. The time-series
averages of the option-implied volatilities at the ve quoted strikes are reported in Panel A of Table
IV.
The most frequently traded options have maturities of 1M, 3M, 6M an 1Y, and include at-the-
money options, as well as, calls and puts with deltas of 0.25 and 0.10 (in absolute value). The option
deltas, obtained by dierentiating the option value with respect to the spot exchange rate, S
t
, are
given by,

c
(K) = e
r
f,t

N(d
1
) (20a)

p
(K) = e
r
f,t

N(d
1
) (20b)
allowing for conversion between the strike price of an option and its corresponding delta. Specically,
the strike prices of puts and calls with delta values of
p
and
c
, respectively, are given by:
K
c
= F
t
exp
_
1
2

t
(
c
)
2

t
(
c
)

N
1
[exp(r
f,t
)
c
]
_
(21a)
K
p
= F
t
exp
_
1
2

t
(
p
)
2
+
t
(
p
)

N
1
[exp(r
f,t
)
p
]
_
(21b)
The strike price of the delta-neutral straddle is obtained by setting
c
(K) +
p
(K) = 0 and solving
for K. It is straightforward to see that the options in this portfolio must both have deltas of 0.50
(in absolute value), and the corresponding strike value is:
K
ATM
= S
t
exp
_
(r
d,t
r
f,t
)
1
2

t
(ATM)
2

_
= F
t
exp
_
1
2

t
(ATM)
2

_
(22)
15
Consequently, although the straddle volatility is described as at-the-money, the corresponding
option strike is neither equal to the spot price or the forward price. In the data, the one-month
0.25 delta options are roughly 1.5-2.5% out-of-the-money, and the one-month 0.10 delta options
are roughly 3.0-4.5% out-of-the-money (Table IV, Panel B). When normalized by the at-the-money
implied volatility (converted to monthly units), the strikes of the 0.25 delta options are 0.70 standard
deviations away from the forward price, and the 0.10 delta options are about 1.40 standard deviations
away from the forward price. Finally, in standard FX option nomenclature an option with a delta
of is typically referred to as a |100 | option. For example, a put with = 0.10, is referred to as
a 10 put. I use this convention from hereon.
3.2 Extracting the risk-neutral moments
The formulas for the risk-neutral moments derived in Section 2 assume the existence of a contin-
uum of out-of-the-money puts and calls. In reality, of course, the data are available only at a discrete
set of strikes spanning a bounded range of strike values, [K
min
, K
max
], such that any implementation
of the moment formulas provides only an approximation to the true risk-neutral moments. It is
therefore important to ensure that the available data are adequate to obtain a credible estimate of
the underlying moments.
Jiang and Tian (2005) investigate these types of approximation errors in the context of computing
estimates of the risk-neutral variance from observations of equity index option prices. To address this
issue the authors simulate data from various types of models for the underlying asset and then seek
to reconstruct the risk-neutral variance from a discrete set of observed option prices. They examine
the impact of having observations on a nite number of options with a bounded range of strikes, as
well as, the impact of various interpolation and extrapolation procedures. They conclude that the
discreteness of available strikes is not a major issue, and that estimation errors decline to 2.5% (0.5%)
of the true volatility when the most deep out-of-the-money options are struck at 1 (1.5) standard
deviations away from the forward price. With options struck at two standard deviations away from
the forward price, approximation errors essentially disappear completely. Moreover, their results
indicate that approximation errors are minimized by interpolating the option implied volatilities
within the observed range of strikes, and extrapolating the option implied volatilities below K
min
and above K
max
by appending at tails at the level of the last observed implied volatility. Consistent
with intuition, they nd that this form of extrapolation is preferred to simply truncating the range
of strikes used in the computation. Carr and Wu (2008) follow a similar protocol in their study of
variance risk premia in the equity market, and combine linear interpolation between observed implied
volatilities with appending at tails beyond the last observed strikes.
Guided by the sensitivity results in Jiang and Tian (2005), the available cross-section of foreign
exchange options is deemed to be suciently broad to ensure that the error in extracting the risk-
neutral moments is likely to be very small. The furthest out-of-the-money puts and calls are struck
at roughly 1.4 times the at-the-money implied volatility away from the prevailing forward prices.
16
Before extracting the risk-neutral moments, I augment the data by interpolating the implied volatility
functions between the observed data points, and append at tails beyond the last observed strike. I
interpolate implied volatilities using the vanna-volga method (Castagna and Mercurio (2007)), which
is the standard approach used by participants in the FX option market.
12
The resulting risk-neutral
moments turn out to be largely unaected by the precise details of the interpolation scheme, and
similar results obtain if a standard linear interpolation is used, e.g. as in Carr and Wu (2008).
The vanna-volga method is based on a static hedging argument, and essentially prices a non-traded
option by constructing and pricing a replicating portfolio, which matches all partial derivatives up
to second order. In a Black-Scholes world, only rst derivatives are matched dynamically, so the
replicating delta-neutral portfolio is comprised only of a riskless bond and the underlying. However,
in the presence of time-varying volatility, it is necessary to also hedge the vega
_
C
BS

_
, as well
as, the volga
_

2
C
BS

_
and vanna
_

2
C
BS
St
_
. In order to match these three additional moments,
the replicating portfolio must now also include an additional three traded options. Consequently,
to the extent that at least three FX options are available, the implied volatilities of the remaining
options can be obtained by constructing the relevant replicating portfolio, and then inverting its
price to obtain the corresponding implied volatility. Castagna and Mercurio (2007) show that the
interpolated implied volatility for a -period option at strike K obtained from the vanna-volga
method is approximately related to the implied volatilities of three other traded option with the
same maturity and strikes K
1
< K
2
< K
3
through:

t
(K, )
ln
K
2
K
ln
K
3
K
ln
K
2
K
1
ln
K
3
K
1

t
(K
1
, ) +
ln
K
K
1
ln
K
3
K
ln
K
2
K
1
ln
K
3
K
2

t
(K
2
, ) +
ln
K
K
1
ln
K
K
2
ln
K
3
K
1
ln
K
3
K
2

t
(K
3
, ) (23)
This formula provides a convenient shortcut for carrying out the interpolation and is known to provide
very accurate estimates of the implied volatilities whenever K is between K
1
and K
3
(Castagna and
Mercurio (2007)). Extrapolations based on this formula, however, lead to spurious results. Since the
above approximation is essentially quadratic in the log strike, it violates the technical conditions for
the existence of moments under the risk-neutral measure when extrapolated to innity (Lee (2004)).
As mentioned earlier, to avoid these issues I append at implied volatility tails beyond the last
observed strikes.
4 Results
In order to investigate the hypothesis that the empirically observed violations of uncovered in-
terest parity are attributable to the exposure of high interest rate currencies to rapid depreciations,
or crashes, I turn to data on foreign exchange options. Options provide a valuable tool for assessing
market participants perceptions of the underlying currency return distributions, as well as, the risk
12
Common approaches in the equity option literature either rely on non-parametric methods (Ait-Sahalia and Lo
(1998)) or t ad hoc functional specications to the observed data (Shimko (1993), Coval, Jurek and Staord (2008)).
17
premia that are demanded for insurance against rapid currency moves. If high interest rate curren-
cies are indeed more prone to crashes, one would expect any of the following to hold in the option
data: (a) option-implied skewness should forecast currency excess returns with a negative sign; (b)
option-implied skewness should be negatively correlated with interest rate dierentials in the time
series, as well as, the cross section; and (c) strategies in which the exposure to currency crashes has
been hedged in the option market should earn zero excess returns. I nd that: (a) option-implied
skewness does not forecast currency excess returns; (b) option-implied skewness is positively related
to interest rate dierentials in the panel in univariate specications, and is insignicant in a multi-
variate specication; and (c) carry trades in which crash risk has been hedged continue to deliver
positive, albeit signicantly smaller, excess returns. Taken together, the results for crash-hedged
strategies indicate that at most 30-40% of the excess returns to currency carry trades can be inter-
preted as compensation for exposure to currency crashes. Simply put, the price of crash insurance in
the currency markets appears to be relatively low, especially when contrasted with equity markets.
In fact, in order for the currency excess returns on crash-hedged strategies to be driven to zero, the
implied volatilities of options hedging against crashes would have to have been roughly four times
higher than what is actually observed in the data.
4.1 Implied volatility functions
Before turning to formal tests of the crash risk hypothesis, it is useful to begin with a brief
summary of the stylized features of the foreign exchange option data. Much like equity options,
the implied volatility functions of foreign exchange options the plots of volatility as a function of
strike price exhibit a pronounced smile. Unlike in equities though, where the smile is essentially
strictly downward sloping, the smile can take on a variety of shapes, suggesting that the risk-neutral
distribution can be either positively or negatively skewed. To summarize these features Figure 5
plots the time-series means of the implied volatility functions for the nine G10 currencies. The red
(blue) lines correspond to periods in which the foreign short-term interest rate was above (below)
the US short-term interest rate. Before taking means the volatilities were re-scaled by the contem-
poraneous at-the-money values to ensure a scale free representation. As can be readily seen, the
shape of the implied volatility function exhibits signicant variation across countries and time. For
example, the implied volatility functions for the Swiss franc (CHF) and the Japanese yen (JPY)
exhibit a right-skewed smile indicating a positively skewed risk-neutral distribution, suggestive of
the potential for rapid appreciations against the U.S. dollar. During the 1999-2007 period, both of
these currencies were characterized by low interest rates (time series means: 1.40% (CHF), 0.14%
(JPY)), and are anecdotally known to have been popular funding currencies for the carry trade.
Conversely, the implied volatility functions for the Australian dollar (AUD) and New Zealand dollar
(NZD), which had relatively high interest rates during the sample (time series means: 5.28% (AUD),
6.06% (NZD)) and were the target currencies for carry traders, exhibit left-skewed smiles, consistent
with a negatively skewed risk-neutral distributions. In sum, the cross-sectional evidence points to
the fact that the exchange rates of relatively high (low) interest rate currencies expose investors to
the risk of large depreciations (appreciations), consistent with the data in Brunnermeier, Nagel and
18
Pedersen (2008). Unconditionally, the risk-neutral distributions of high (low) interest rate currencies
are negatively (positively) skewed, consistent with the crash risk hypothesis.
However, theories of the forward premium puzzle invoking crash risk as the source of the risk
premium attaching to currencies with relatively higher interest rates, also make a prediction about
the time series behavior of the implied-volatility smile. In particular, the skewness of the risk-neutral
distribution should change sign conditional on the sign of the interest rate dierential. Whenever
a currency features an interest rate that is above (below) the U.S. interest rate, the risk-neutral
exchange rate distribution should be negatively (positively) skewed. If this were the case, the blue
lines would exhibit a steeper slope for moneyness values below at-the-money (depreciation) than for
moneyness values above at-the-money (appreciation). Symmetrically, the red lines would exhibit
a steeper slopes for moneyness values above at-the-money, than for values below at-the-money.
In general, the data do not seem to be supportive of the conditional prediction of the crash risk
hypothesis. Of course, one can plausibly argue that a currencys risk-neutral skewness should be a
function of the countrys interest rate dierential relative to all available currencies, rather than just
the U.S. dollar. After all, carry traders are not constrained to trade only in foreign/USD currency
pairs. Consequently, one cannot make much of this nding, beyond emphasizing that carry trades
are likely to be a global phenomenon.
4.2 Risk-neutral moments
In order to facilitate regression-based tests of the crash risk hypothesis, I summarize the currency
option data by extracting the risk-neutral moments of the option-implied distribution. To do this
I rst interpolate the implied volatility data on each day using the vanna-volga method, and then
construct and price the option portfolios replicating variance, skewness and kurtosis swaps, using
the interpolated implied volatility function. The undiscounted values of these swaps correspond
to the rst three moments of the risk-neutral distribution. Since I use currency options with a
one month maturity in the procedure, the resulting risk-neutral moments describe the one month
ahead distribution. This procedure yields a daily time series of forward looking moments, which are
plotted in Figure 6 for a representative set of currencies (AUD, CHF, EUR, GBP, JPY, NZD). The
table accompanying Figure 6 provides the time series means and standard errors of the risk-neutral
moments, computed from non-overlapping, monthly observations. The mean skewness values for
essentially all currencies are statistically distinguishable from zero at conventional signicance levels,
and range from -0.16 to 0.32. Overall these absolute magnitudes are not extreme, as one would
expect in the presence of the risk of large, but rare, crashes. Finally, the mean kurtosis values are
all greater than three, consistent with heavy-tailed risk-neutral distributions.
The top panel of Figure 6 graphs the risk-neutral volatility (
_
Var
Q
t
), and shows a clear common
trend in the riskiness of the individual currency pairs.
13
More interesting, is the behavior of risk-
13
A principal components decomposition of the risk-neutral variance time series, reveals that the rst principal
component roughly an equal-weighted combination of the individual series excluding the Canadian dollar explains
19
neutral skewness (Skew
Q
t
) which exhibits pronounced time-series and cross-sectional variation. The
skewness values for the two high interest rates (AUD, NZD) move closely in tandem and are gener-
ally negative, with a time series mean of -0.16, rejecting the null of no skewness with considerable
signicance. By contrast, the mean skewness values for the two low interest rate currencies (CHF,
JPY) are 0.10 and 0.32, respectively, both of which are statistically distinguishable from zero. The
skewness of the Swiss franc moves in tandem with the Euro, which is perhaps somewhat surprising
given the currencies dramatically dierent interest rate policies. The risk-neutral skewness of the
Japanese Yen is particularly volatile during the sample period reaching values as high as 1, and
as low as -0.5. Overall, a principal components analysis reveals that 45% of the total variation in
risk-neutral skewness is attributable to a common trend (equal-weighted portfolio). The next princi-
pal component explains roughly 30% of the variation and is essentially uniquely represented by the
skewness of the Japanese Yen, reecting its special position as a global funding currency. Finally, the
risk-neutral kurtosis of all currency pairs (Kurt
Q
t
) is plotted in the bottom panel of Figure 6. The
risk-neutral kurtosis values are consistently above three, and average to about 3.6 for all currencies
with the exception of the Japanese yen, whose mean kurtosis stands at 4.1.
4.3 Forecasting crashes and currency returns
The stochastic nature of risk-neutral skewness is suggestive of the fact that the risk of crashes
and/or the price of insuring against those events is highly time-varying. To disentangle these two
eects I turn to the data on the realized skewness of currency excess returns. The currency excess
return for time t is dened as:
xs
t
= ln
S
t
S
t1
(r
d,t
r
f,t
) (24)
and is essentially the continuously compounded return to buying a currency forward. Under UIP, the
excess return is equal to zero (in expectation). To compute the realized skewness (Skew
P
t
) for month
t, I rst construct the daily excess returns during that month and then compute their skewness. Panel
A of Table V reports the results of panel regression using monthly observations; all panel regression
specications include country xed eects.
14
Much like Brunnermeier, Nagel and Pedersen (2008),
I nd that future values of realized skewness are negatively related to past excess returns and the
lagged values of the interest rate spread (r
f,t
r
d,t
) . This indicates that currencies that have high
interest rates and have recently appreciated i.e. have been targets of successful carry trades are
more likely to experience large negative moves, or crashes. By contrast, the rst regression in Panel
B, indicates that risk-neutral skewness is strongly positively related to the realized currency return.
The existence of a positive relationship is consistent with a risk-based story in which skewness acts
65% of total variation. The second principal component which is long high interest rate currencies (AUD and NZD)
and short low interest rate currencies (CHF, JPY) explains an additional 20% of the variation.
14
Results based on panel regressions using rolling 21-day windows for all specications in Table V are available from
the author upon request. After adjusting standard errors for within time-period correlation and serial correlation,
the results are qualitatively and quantitatively indistinguishable from those obtained in the non-overlapping, monthly
panel.
20
as a proxy for a priced risk factor (e.g. as in Farhi and Gabaix (2008)). Namely, as currencies
become more negatively (positively) skewed investors charge a greater (smaller) risk premium, and
the currency experiences a contemporaneous depreciation (appreciation). However, the evidence on
future realized skewness appears to suggest that appreciated currencies become riskier, not safer. As
a result, I nd that risk-neutral skewness predicts future realized skewness with a negative sign and
is highly statistically signicant in a univariate specication. Taken together, the evidence suggests
that the cost of hedging crashes following large appreciations is low, precisely when the risk of a
crash is high.
Another result emerging from the panel regressions reported in Panel B is that risk-neutral skew-
ness is positively related to the interest rate spread in the univariate panel regression, despite a
exhibiting a strong negative relationship in the cross section. This indicates that periods of abnor-
mally high spreads are associated with more positive values of risk-neutral skewness, contrary to
the predictions of the crash risk hypothesis. Consistent with Brunnermeier, Nagel and Pedersen
(2008), however, I nd that this relationship becomes negative, but statistically insignicant in a
multivariate specication. Judging by the dramatic dierence between the values of the adjusted R
2
gross and net of the xed eects during this time period, risk-neutral skewness appears to be more
a xed feature of a country, rather than a feature of its time-varying interest rate environment.
Finally, if violations of uncovered interest parity (UIP) are attributable to crash risk premia,
the magnitude of risk-neutral skewness a proxy of the markets crash expectation would be
expected to forecast currency excess returns with a negative sign. Panel C of Table V investigates
the forecastability of one-month excess returns, using lagged excess returns, interest rate spreads, as
well as, realized and option-implied measures of skewness. I nd a minute amount of momentum
in excess returns at the monthly horizon and positive predictability on the basis of the interest rate
dierential. Not only does the option-implied skewness measure not forecast currency excess returns,
it appears in the forecasting regression with a positive, albeit insignicant, regression coecient.
4.4 Crash-neutral carry trade strategies
In order to determine whether the excess returns to carry trades can be attributed to compen-
sation for exposure to currency crashes, I turn to an analysis of crash-neutral carry trades. Since
these trades eliminate exposure to currency crashes by establishing a protective position in foreign
exchange options, their mean excess returns should be lower than those of the standard carry trade.
In the event that crash risk accounts for the entirety of the observed violations of UIP, excess returns
to crash-neutral carry trades would be statistically indistinguishable from zero. To construct a time
series of returns for the crash-neutral carry trades implemented in each of the nine individual cur-
rency pairs I proceed by analogy to the approach used in the standard carry trade. At each month
end, I compare the prevailing one-month interbank lending rates, and establish the relevant positions
in the spot markets and one-month foreign exchange options prescribed by (14) and (17). These po-
sitions are then held until the end of the following month, when the option expires. I construct
21
three variants of the crash-neutral strategy, each oering a dierent amount of crash protection, as
reected by the strike price (delta) of the included FX option. The summary statistics for these
strategies are presented in Table VI. Panel A presents the results for crash-neutral strategies using
deep out-of-the-money options (10 calls and puts), which only provide protection against moves
that are greater than roughly 1.4 times the magnitude of the monthly standard deviation. Panel B
presents results for crash-neutral strategies employing options with intermediate moneyness levels
(25 calls and puts), and nally, Panel C presents the results for strategies using at-the-money op-
tions. Crucially, note that since the crash-neutral strategies only employ options for which tradable
price data are available in the J. P. Morgan dataset, the accrued strategy returns represent returns
that were attainable in the marketplace (before transaction costs). The results in this section do not
rely on the implied volatility interpolation procedure used in the previous sections whatsoever.
4.4.1 Individual currency pairs
It is immediate from Panel A of Table VI that the addition of crash protection using 10 options
has a limited eect on the realized risk-return prole of the currency carry trade. Consistent with
intuition, the addition of the crash-protection via the FX option causes the realized returns to be more
positively skewed, and the magnitude of the minimal monthly returns to decline relative to standard
carry trades. Similarly, one also observes a minor decline in the annualized standard deviation of
the returns, although the level of volatility is still high and close to 10%. Of the four individual
currency pairs (AUD, EUR, NZD, SEK) that delivered positive and statistically signicant excess
returns in the standard carry trade strategy, three continue to do so in the crash-neutral variant, and
the t-statistic on the fourth pair (NZD) is only slightly below the requisite threshold for signicance
at the 5% level. However, tests comparing returns of the 10 crash-neutral carry trade with the
standard carry trade, also indicate a statistically signicant decline in the mean realized return. For
the four currency pairs which individually delivered positive and statistically signicant returns, the
mean returns decline by -1.36% (AUD; t-stat: -4.94), -1.28% (EUR; t-stat: -4.44), -1.03% (NZD;
t-stat: -1.92) and -1.21% (SEK; t-stat: -3.19) per year. Shifting to crash-neutral strategies with more
protection (25; Panel B) causes a further decline in the realized returns and a coincident increase in
skewness. Now all nine individual currency pairs have positively skewed excess returns, with monthly
skewness values ranging from 0.13 (NZD) to 0.73 (NOK). Eight of those continue delivering positive
excess returns, three of which are statistically signicant at the 5%-level. When compared with the
standard carry trades, the mean returns decline by -2.25% (AUD; t-stat: -2.29), -3.06% (EUR; t-stat:
-3.78), -1.97% (NZD; t-stat: -1.52) and -3.32% (SEK; t-stat: -3.78), respectively.
Finally, Panel C of Table VI presents the corresponding summary statistics for crash-neutral cur-
rency strategies constructed from positions in the spot currency markets and at-the-money options.
With ATM options, the strike price of the option is slightly above the forward exchange rate, and the
carry trader eectively pays up front for protection against all negative deviations from uncovered
interest rate parity. In other words, the investor locks in the interest rate dierential (carry) and
retains the upside from potential currency moves, in exchange for the option premium. By construc-
22
tion, the returns to these strategies are extremely positively skewed, with monthly skewness values
generally above one. However, there is a dramatic decline of excess returns in comparison to to the
standard carry trades, with excess returns declining anywhere from -1.08% (CHF) to -6.44% (SEK)
per year. Nonetheless, even with ATM hedging, the excess returns to the crash-neutral carry trade
implemented in the AUD and SEK remain statistically signicant.
4.4.2 Portfolio strategies
To summarize the ndings for crash-neutral currency carry trades, I once again construct equal-
and spread-weighted portfolios (EQL and SPR), as well as, their dollar-neutral counterparts (EQL-
$N and SPR-$N). The returns to the non-dollar-neutral strategies are reported in Panel A of Table
VII, and their dollar-neutral counterparts are reported in Panel B. This time an interesting dichotomy
emerges depending on the method of construction. While the excess returns to the non-dollar-neutral
portfolio remain positive and highly statistically signicant, even when hedged with ATM options,
the returns to the dollar-neutral portfolio are statistically indistinguishable from zero, even when
hedged with 10 options. However, this divergence in signicance appears primarily attributable to
the greater volatility of the non-dollar-neutral portfolios, which makes accurate measurement of their
mean excess returns more dicult. The proportionate declines in excess returns for dollar-neutral
and non-dollar-neutral portfolio are in fact comparable.
The excess returns on the non-dollar-neutral portfolios remain positive and highly statistically
signicant independent of the moneyness of the options used to hedge exposure to currency crashes.
For example, the equal-weighted portfolio of crash-neutral carry trades implemented in the nine
individual currency pairs delivers an annualized excess return of 3.18% (10; t-stat: 3.13). As
expected, the mean return declines with the level of protection oered by the option to 2.55% (25;
t-stat: 2.77) and 1.34% (ATM; t-stat: 1.81). As with the standard carry trades, the returns on the
spread-weighted strategies are somewhat higher and more volatile than the returns to the equal-
weighted portfolios. The combination of portfolio diversication and the benet of option protection
drive down the annualized volatilities to around 3-4% per year, such that corresponding portfolio
Sharpe ratios continue to be large and range from 0.63 to 1.28. The cumulative performance of
the three equal-weighted crash-neutral currency carry strategies is illustrated in Figure 7. As is
immediate from the plot, even though excess returns remain positive with the most aggressively
crash-hedged strategy (ATM), nearly half of the cumulative prots are eliminated, even when the
least aggressive crash-hedging (10) is employed. The declines in the mean realized returns for the
non-dollar neutral strategies range from -2.42% (t-stat: -4.73) to -6.05% (t-stat: -5.03) per year.
Consequently, while excess returns on portfolio of crash-neutral trades remain positive, providing
evidence against the hypothesis that crash risk premia can account for violations of UIP, they are also
considerably smaller. When measured relative to the excess returns earned by the corresponding non-
dollar-neutral portfolios of standard carry trades, these declines represent 30-40% of the total return
to the unhedged strategy. Therefore, while UIP violations cannot be attributed solely to exposure to
currency crashes, the crash risk explanation goes a considerable distance towards rationalizing the
23
forward premium anomaly.
4.5 Robustness
Although hedging currency crash risk using 10 options eliminates 30-40% of the excess returns to
the standard carry trade, one may wish to know by how much this result would improve once option
transaction costs were accounted for. In particular, the implied volatilities used to price the options
used in the crash-neutral portfolio represented midquotes and did not account for the existence of
a bid-ask spread. To address this question, I replicate the returns to the crash-neutral strategies
with perturbed values of the implied volatilities used to price the crash protection. In particular, I
apply a simple multiplicative transformation to the implied volatilities, thus increasing the price of
all options being purchased.
To assess the impact of transaction costs I apply a volatility multiplier of 1.1, since bid-ask spreads
in foreign exchange option markets are between 0.5 1% volatility points, and the mean option
implied volatilities in the sample are on the order of 10%. With this modication, the mean excess
returns to the 10 crash-neutral carry trades remain essentially unchanged. The equal-weighted
portfolio strategy delivers an excess return of 2.89% (t-stat: 2.81), and the spread-weighted portfolio
delivers an excess return of 5.03% (t-stat: 3.46), corresponding to a 30 basis point decline in realized
excess returns to the crash-neutral strategies versus the baseline specication without transaction
costs. Therefore, typical transaction costs are insucient to eliminate the protability of crash-
neutral currency carry trades.
Relatedly, one would also like to know, what would it have taken in terms of a shift in option
prices (implied volatilities) in order for the excess returns to the crash-neutral strategies to have
been equal to zero in the historical sample? By varying the implied volatility multiplier, I nd that
in order to eliminate the excess returns to crash-neutral carry trades requires implied volatilities
that are between twice and four times as large, as the values actually observed in the data. For
example, with a multiplier of two, excess returns on the equal- and spread-weighted portfolios are
1.38% (t-stat: 1.29) and 3.45% (t-stat: 2.32), respectively, and with a multiplier of four, they are
-1.12% (t-stat: -1.04) and 0.69 (t-stat: 0.46), respectively. Needless to say, these magnitudes for the
multiplier imply enormous mispricings in the foreign exchange option market. Consequently, while
hedging exposure to crash risk helps provide a partial resolution of the UIP puzzle, it is unlikely to
resolve it entirely.
5 Conclusion
Using option data on foreign exchange rates between the U.S. dollar and the other G10 currencies,
covering the period from January 1999 to March 2007, I show that carry trade portfolios in which the
exposure to currency crashes has been completely hedged, continue to deliver positive and statistically
signicant excess returns. The decline in the excess returns delivered by the crash-neutral portfolios
24
strategies vis a vis their unhedged counterparts, suggests that exposure to currency crashes can
account for between 30-40% of the observed excess currency returns stemming from violations of
UIP.
When implemented in the nine currency pairs involving the U.S. dollar and one of the remaining
nine G10 currencies, simple equal- and spread-weighted portfolios of crash-neutral carry trades de-
liver Sharpe ratios between 0.63 and 1.28, while simultaneously exhibiting positively skewed excess
returns. The mean excess returns to these strategies are highly statistically signicant, and the an-
nualized volatilities range from 2 to 4%. These nding are robust to various portfolio construction
methodologies (weightings, dollar neutrality), as well as, the choice of the option strike used to hedge
exposure to crash risk. As expected, the mean excess returns decline as investors hedge exposure to
currency crashes more aggressively (10 > 25 > ATM), although excess returns continue to remain
statistically signicant even when at-the-money options are used. Consequently, unlike in equities,
protection against the crashes in high interest rate currencies appears to be relatively cheap.
An analysis of the behavior of the risk-neutral and realized skeweness presents evidence in support
of the hypothesis that crash protection may be abnormally cheap. Specically, while realized skewness
tends to be negatively related to lagged currency returns, the relationship for risk-neutral skewness
is positive. Therefore, while an appreciation of a high-interest rate currency delivering positive
carry trade returns heralds an increased risk of a crash, the risk-neutral skewness simultaneously
becomes more positive, implying that the prices of puts (calls) decrease (increase). However, even
after accounting for this unusual behavior of option prices, crash risk is unlikely to be able to provide
a complete resolution of the forward premium anomaly. In order to drive excess returns on the
crash-hedged currency carry trades to zero, implied volatilities of out-of-the-money puts (calls) on
high (low) interest rate currencies would have to be two to four times their actual observed values,
implying signicant pricing errors in the foreign exchange option market.
25
A Non-central moments of the risk-neutral distribution
The expressions for the second, third and fourth non-central moments are:
V
t
() =
_

S
2
_
1 ln
K
S
_
K
2
C
t
(K, )dK +
_
S
0
2
_
1 + ln
S
K
_
K
2
P
t
(K, )dK
W
t
() =
_

S
6 ln
K
S
3
_
ln
K
S
_
2
K
2
C
t
(K, )dK

_
S
0
6 ln
S
K
+ 3
_
ln
S
K
_
2
K
2
P
t
(K, )dK
X
t
() =
_

S
12
_
ln
K
S
_
2
4
_
ln
K
S
_
3
K
2
C
t
(K, )dK +
+
_
S
0
12
_
ln
S
K
_
2
+ 4
_
ln
S
K
_
3
K
2
P
t
(K, )dK
References
[1] Ait-Sahalia, Yacine and Andrew W. Lo, 1998, Nonparametric estimation of state-price densities
implicit in nancial asset prices, Journal of Finance, 53, 499-547.
[2] Arrow, Kenneth J., 1964, The Role of Securities in the Optimal Allocation of Risk Bearing,
Review of Economic Studies 31, 91-96.
[3] Backus, David K., Silverio Foresi and Chris Telmer, 2001, Ane Term Structure Models and
the Forward Premium Anomaly, 56(1), p. 279-304.
[4] Bakshi, Gurdip and Nikunj Kapadia, 2003, Delta-Hedged Gains and the Negative Market
Volatility Risk Premium, Review of Financial Studies, 16(2), p. 527-566.
[5] Bakshi, Gurdip, Nikunj Kapadia and Dilip Madan, 2003, Stock Return Characteristics, Skew
Laws and the Dierential Pricing of Individual Equity Options, Review of Financial Studies,
16(1), p. 101-143.
[6] Barro, Robert, 2006, Rare Disasters and Asset Markets in the Twentieth Century, Quarterly
Journal of Economics, 121, p. 823-866.
[7] Black, Fischer and Myron Scholes, 1973, The pricing of options and corporate liabilities, Journal
of Political Economy, 81, p. 617-654.
[8] Breeden, Douglas and Robert Litzenberger, 1978, Prices of state-contingent claims implicit in
option prices, Journal of Business, p. 621-651.
[9] Britten-Jones, M. and A. Neuberger, 2000, Option Prices, Implied Price Processes, and Stochas-
tic Volatility, Journal of Finance, 55, p. 839-866.
[10] Brunnermeier, Markus, Christian Gollier, and Jonathan Parker, 2007, Optimal Beliefs, Asset
Prices and the Preference for Skewed Returns, American Economic Review (Paper and Proceed-
ings), 97(2), p. 159-165.
26
[11] Brunnermeier, Markus, Stefan Nagel and Lasse Pedersen, 2008, Currency Crashes and the Carry
Trade, NBER Macroeconomics Annual, 23, forthcoming.
[12] Burnside, Craig, Martin Eichenbaum, Isaac Kleshchelski and Sergio Rebelo, 2006, The Returns
to Currency Speculation, NBER working paper.
[13] Burnside, Craig, Martin Eichenbaum, Isaac Kleshchelski and Sergio Rebelo, 2008, Do Peso
Problems Explain the Returns to the Carry Trade?, NBER working paper.
[14] Campa, Jose M, Kevin Chang and Robert Reider, 1998, Implied Exchange Rate Distributions:
Evidence from OTC Option Markets, Journal of International Money and Finance, 17(1), p.
117-160.
[15] Carr, Peter and Dilip Madan, 2001, Optimal Positioning in Derivative Securities, Quantitative
Finance, 1, p. 19-37.
[16] Carr, Peter and Liuren Wu, 2007, Stochastic Skew in Currency Options, Journal of Financial
Economics, 86(1), p.213-247
[17] Carr, Peter and Liuren Wu, 2008, Variance Risk Premiums, Review of Financial Studies, forth-
coming.
[18] Coval, Joshua D. and Tyler Shumway, 2001, Expected Option Returns, Journal of Finance,
56(3), p. 983-1009.
[19] Coval, Joshua D., Jakub W. Jurek and Erik Staord, 2008, Economic Catastrophe Bonds,
American Economic Review, forthcoming.
[20] Castagna, Antonio and Fabio Brigo 2007, Consistent Pricing of FX Options, Risk
[21] Debreu, Gerard, 1959, Theory of Value, New York: Wiley.
[22] DeMiguel, V., L. Garlappi and R. Uppal, 2007, Optimal versus Naive Diversication: How
Inecient Is the 1/N Portfolio Strategy?, Review of Financial Studies, forthcoming.
[23] Driessen, Joost and Pascal Maenhout, 2006, The World Price of Jump and Volatility Risk,
working paper.
[24] Engel, Charles, 1996, The Forward Discount Anomaly and the Risk Premium: A Survey of
Recent Evidence, Journal of Empirical Finance, 3, 123192.
[25] Fama, Eugene F., Forward and Spot Exchange Rates, 1984, Journal of Monetary Economics,
14, p. 319-338.
[26] Farhi, Emmanuel and Xavier Gabaix, 2008, Rare Disasters and Exchange Rates, working paper.
[27] Froot, Kenneth.A, and Richard Thaler, 1990, Anomalies: Foreign Exchange, Journal of Eco-
nomic Perspectives, 4, p. 179-192.
[28] Garman, Mark B. and Steven W. Kohlhagen, 1983, Foreign Currency Option Values, Journal
of International Money and Finance, 2, p. 231-237.
[29] Hansen, Lars P. and Robert J. Hodrick, 1980, Forward Exchange Rates as Optimal Predictors
of Future Spot Rates: An Econometric Analysis. Journal of Political Economy, 88, p. 829-53.
27
[30] Jiang, George J. and Yisong S. Tian, 2005, The Model-Free Implied Volatility and Its Informa-
tion Content, Review of Financial Studies, 18(4), p. 1305-1342.
[31] Jorion, Philippe, 1995, Predicting Volatility in the Foreign Exchange Market, Journal of Fi-
nance, 50(2), p. 507-528.
[32] Lee, Roger. W., 2004, The Moment Formula for Implied Volatility at Extreme Strikes, Mathe-
matical Finance, 14(3), p. 469-480.
[33] Lewis, Karen K., Puzzles in International Financial Markets, in Gene Grossman and Kenneth
Rogo, eds., Handbook of International Economics, Volume 3 (Amsterdam; New York and
Oxford: Elsevier, North- Holland, 1995).
[34] Lustig, Hanno, and Adrien Verdelhan, 2007a, The Cross-Section of Foreign Currency Risk Pre-
mia and US Consumption Growth Risk, American Economics Review, 97(1), p. 89-117.
[35] Lustig, Hanno, and Adrien Verdelhan, 2007b, Note on the Cross-Section of Foreign Currency
Risk Premia and Consumption Growth Risk, working paper.
[36] Lustig, Hanno, Nick Roussanov and Adrien Verdelhan, 2008, Common Risk Factors in Currency
Markets, working paper.
[37] Martin, Ian, 2008, Consumption-Based Asset Pricing with Higher Cumulants, working paper.
[38] Pan, Jun, 2002, The Jump-Risk Premia Implicit in Options: Evidence from an Integrated
Time-Series Study, Journal of Financial Economics, 63, p. 3-50.
[39] Plantin, Guillaume, and Hyun S. Shin, 2008, Carry Trades and Speculative Dynamics, working
paper.
[40] Rietz, Thomas A., 1988, The Equity Risk Premium: A Solution, Journal of Monetary Eco-
nomics, 22, p. 117-131.
[41] Shimko, David, 1993, Bounds of probability, Risk 6, 3337.
[42] Thompson, Samuel, 2006, Simple Formulas for Standard Errors that Cluster by Both Firm and
Time, working paper.
[43] Verdelhan, Adrien, 2008, A Habit-Based Explanation of the Exchange Rate Risk Premium,
working paper.
[44] Weitzman, Martin, 2007, Subjective Expectations and Asset-Return Puzzles, American Eco-
nomic Review, 97, p. 1102-30.
28
Table I
UIP Regressions.
This table reports coecient estimates from the regression of the time t +1 log currency return on the time t log forward
premium,
s
i,t+1
s
i,t
= a
0
+ a
1
(f
i,t
s
i,t
) +
i,t+1
H
0
: a
0
= 0, a
1
= 1
Currency returns are computed using 21-day rolling windows and span the period from January 1990 to March 2007
(all exchange rates are expressed in terms of dollars per unit of foreign currency). The forward premia are measured
using the spread between one-month eurocurrency (LIBOR) rates for loans denominated in U.S. dollars and loans
denominated in the foreign currency. The table reports regression coecients, standard errors (in parentheses), and the

2
test statistic for the null hypothesis of UIP (p-values in parentheses). Standard errors in individual regressions are
adjusted for serial correlation using a Newey-West covariance matrix with 21 lags. The pooled (panel) regression is run
with country-xed eects; the reported standard errors are robust to within time-period correlation and are adjusted
for serial correlation. The pooled regression
2
statistic is computed for the null that all country xed eects are zero
and the intercept is equal to one. R
2
NFE
reports the adjusted R
2
from the panel regression net of the xed eects
(N = 40, 500). XS reports the time series means and standard errors of the regression coecients from cross-sectional
regressions performed for each t. For the cross-sectional regressions R
2
is the mean adjusted R
2
.
1990-2007 1999-2007
Currency a
0
a
1
R
2
NFE

2
test a
0
a
1
R
2
NFE

2
test
AUD -0.0025 -1.7483 0.0105 8.87 -0.0028 -3.9018 0.0310 9.34
(0.0023) (1.0522) (0.01) (0.0036) (1.9520) (0.01)
CAD -0.0001 -0.5077 0.0019 9.13 0.0027 -2.5012 0.0115 5.06
(0.0009) (0.5104) (0.01) (0.0015) (2.1091) (0.08)
CHF 0.0026 -1.2815 0.0069 5.60 0.0096 -4.5238 0.0350 9.03
(0.0024) (1.0008) (0.06) (0.0041) (1.8485) (0.01)
EUR 0.0002 -0.0320 -0.0002 1.34 0.0036 -4.4836 0.0447 11.19
(0.0016) (0.9072) (0.51) (0.0024) (1.6590) (0.00)
GBP 0.0021 0.7061 0.0020 3.65 0.0001 -1.7371 0.0061 4.55
(0.0019) (1.2755) (0.16) (0.0025) (1.7738) (0.10)
JPY 0.0058 -2.0823 0.0165 12.33 0.0048 -1.8183 0.0099 6.20
(0.0025) (0.8787) (0.00) (0.0045) (1.4003) (0.05)
NOK 0.0013 0.6255 0.0042 1.43 0.0007 -1.4005 0.0090 5.46
(0.0017) (0.6351) (0.49) (0.0026) (1.2175) (0.07)
NZD -0.0047 -2.4128 0.0147 15.46 -0.0067 -4.7728 0.0482 17.15
(0.0034) (1.1975) (0.00) (0.0045) (1.6837) (0.00)
SEK 0.0004 0.6081 0.0046 0.51 0.0026 -3.5247 0.0405 11.09
(0.0017) (0.6046) (0.77) (0.0024) (1.3764) (0.00)
Pooled FE -0.1795 0.0002 2.59 FE -3.0503 0.0248 24.17
(0.6589) (0.99) (1.1190) (0.01)
XS 0.0005 -0.1883 0.1070 - 0.0012 -0.5994 0.0966 -
(0.0003) (0.0836) (0.0005) (0.1087) -
Table IIa
Standard Currency Carry Trade (1990-2007).
Panel A of the table reports summary statistics for returns from implementing the standard carry trade in currency
pairs involving the U.S. dollar and one of the remaining nine G10 currencies. Returns are in U.S. dollars and are
computed monthly for the period from January 1990 to March 2007 (N = 207). Means, standard deviations and Sharpe
ratios (SR) are annualized. Carry is the mean absolute interest dierential between the one-month Eurocurrency rate
for country X and the one-month U.S. dollar rate, |r
f,t
r
d,t
|. Min and Max report the smallest and largest observed
monthly return. Panel B presents analogous summary statistics for portfolios of individual carry trades. The EQL
strategy equal-weights the nine underlying carry trades, whereas SPR weights them by the absolute interest rate spread
at initiation of the trade. The EQL-$N and SPR-$N are constructed analogously, by with the additional requirement
that the portfolio be neutral with respect to exposure to the U.S. dollar.
Panel A: Individual currency pairs
AUD CAD CHF EUR GBP JPY NOK NZD SEK
Mean 0.0690 0.0319 0.0168 0.0637 0.0248 0.0482 0.0519 0.0399 0.0839
t-stat 3.19 2.30 0.65 2.78 1.12 1.79 2.16 1.80 3.28
Std. dev. 0.0898 0.0577 0.1074 0.0951 0.0916 0.1120 0.1000 0.0921 0.1061
Skewness -0.30 -0.28 -0.47 -0.37 -0.68 -0.97 -0.21 -0.44 -0.73
Kurtosis 2.81 3.33 3.21 3.82 5.56 6.84 3.46 3.55 5.35
Min -0.0685 -0.0435 -0.1005 -0.1010 -0.1188 -0.1640 -0.0933 -0.0790 -0.1452
Max 0.0649 0.0464 0.0645 0.0755 0.0727 0.1054 0.0756 0.0738 0.0826
Carry 0.0206 0.0141 0.0248 0.0210 0.0219 0.0314 0.0263 0.0289 0.0304
SR 0.77 0.55 0.16 0.67 0.27 0.43 0.52 0.43 0.79
Panel B: Portfolio strategies
EQL SPR EQL-$N SPR-$N
Mean 0.0478 0.0655 0.0399 0.0709
t-stat 3.91 4.10 2.74 3.65
Std. dev. 0.0507 0.0663 0.0603 0.0807
Skewness -0.95 -0.82 -0.58 -0.73
Kurtosis 5.50 4.47 5.76 5.27
Min -0.0580 -0.0648 -0.0737 -0.0892
Max 0.0375 0.0468 0.0588 0.0651
Carry 0.0244 0.0355 0.0369 0.0514
SR 0.94 0.99 0.66 0.88
Table IIb
Standard Currency Carry Trade (1999-2007).
Panel A of the table reports summary statistics for returns from implementing the standard carry trade in currency
pairs involving the U.S. dollar and one of the remaining nine G10 currencies. Returns are in U.S. dollars and are
computed monthly for the period from January 1999 to March 2007 (N = 99). Means, standard deviations and Sharpe
ratios (SR) are annualized. Carry is the mean absolute interest dierential between the one-month Eurocurrency rate
for country X and the one-month U.S. dollar rate, |r
f,t
r
d,t
|. Min and Max report the smallest and largest observed
monthly return. Panel B presents analogous summary statistics for portfolios of individual carry trades. The EQL
strategy equal-weights the nine underlying carry trades, whereas SPR weights them by the absolute interest rate spread
at initiation of the trade. The EQL-$N and SPR-$N are constructed analogously, by with the additional requirement
that the portfolio be neutral with respect to exposure to the U.S. dollar.
Panel A: Individual currency pairs
AUD CAD CHF EUR GBP JPY NOK NZD SEK
Mean 0.0879 0.0340 -0.0019 0.0910 0.0165 0.0341 0.0496 0.0784 0.1142
t-stat 2.51 1.43 -0.06 2.94 0.62 1.07 1.43 2.01 3.38
Std. dev. 0.1005 0.0686 0.0955 0.0888 0.0766 0.0914 0.0994 0.1119 0.0970
Skewness -0.41 -0.26 -0.53 -0.20 -0.01 -0.04 -0.02 -0.56 -0.09
Kurtosis 2.72 2.71 2.63 3.24 2.64 2.94 3.04 2.96 2.82
Min -0.0685 -0.0435 -0.0723 -0.0774 -0.0533 -0.0662 -0.0700 -0.079 -0.0608
Max 0.0649 0.0464 0.0458 0.0617 0.0526 0.0681 0.0721 0.0738 0.0826
Carry 0.0186 0.0084 0.0221 0.0151 0.0140 0.0348 0.0213 0.0263 0.0190
SR 0.87 0.50 -0.02 1.02 0.22 0.37 0.50 0.70 1.18
Panel B: Portfolio strategies
EQL SPR EQL-$N SPR-$N
Mean 0.0560 0.0844 0.0434 0.0699
t-stat 3.63 4.26 2.30 2.84
Std. dev. 0.0443 0.0569 0.0543 0.0707
Skewness -0.42 -0.20 -0.52 -0.23
Kurtosis 3.73 2.97 4.31 4.33
Min -0.0368 -0.0335 -0.0569 -0.0661
Max 0.0375 0.0455 0.0369 0.0651
Carry 0.0200 0.0307 0.0345 0.0478
SR 1.26 1.48 0.80 0.99
Table III
Summary Statistics.
Panel A reports the time-series means of the one-month interbank (LIBOR) lending rates and of the implied volatilities
of one-month options on the exchange rates of individual currencies versus the U.S. dollar. The foreign exchange option
data is comprised of implied volatilities on ve options with standardized Black-Scholes deltas: puts and calls with
deltas of 0.10 and 0.25 in absolute value (denoted by 10p, 25p, 25c and 10c, respectively), and a delta-neutral
straddle (denoted by ATM). All values are annualized. Panel B presents the time series means of the corresponding
option strikes in terms of option moneyness (dened as the ratio of the option strike to the prevailing forward rate)
and a standardized moneyness (dened as the log moneyness scaled by the at-the-money implied volatility per month).
The data are daily and span the period from January 1999 to March 2007 (N = 2149).
Panel A: LIBOR and Implied Volatilities
Currency r
f,t
10p 25p ATM 25c 10c
AUD 0.0528 0.1191 0.1116 0.1071 0.1075 0.1114
CAD 0.0367 0.0799 0.0758 0.0736 0.0751 0.0785
CHF 0.0140 0.1098 0.1058 0.1051 0.1089 0.1155
EUR 0.0305 0.1062 0.1020 0.1011 0.1046 0.1109
GBP 0.0480 0.0890 0.0843 0.0824 0.0843 0.0886
JPY 0.0014 0.1068 0.1024 0.1039 0.1117 0.1238
NOK 0.0484 0.1129 0.1087 0.1079 0.1114 0.1176
NZD 0.0606 0.1322 0.1243 0.1195 0.1200 0.1241
SEK 0.0317 0.1141 0.1098 0.1088 0.1122 0.1184
USD 0.0361 - - - - -
Panel B: FX Option Strike Values
Moneyness

K
F
t

Standardized moneyness

ATM
ln
K
F
t

Currency 10p 25p ATM 25c 10c 10p 25p ATM 25c 10c
AUD 0.9580 0.9793 1.0005 1.0214 1.0423 -1.4005 -0.6824 0.0153 0.6903 1.3508
CAD 0.9714 0.9857 1.0002 1.0149 1.0295 -1.3812 -0.6830 0.0106 0.6952 1.3698
CHK 0.9611 0.9803 1.0005 1.0217 1.0437 -1.3223 -0.6625 0.0150 0.7137 1.4222
EUR 0.9623 0.9810 1.0004 1.0208 1.0420 -1.3293 -0.6637 0.0145 0.7116 1.4207
GBP 0.9682 0.9841 1.0003 1.0167 1.0335 -1.3668 -0.6745 0.0119 0.6993 1.3918
JPY 0.9624 0.9811 1.0005 1.0220 1.0464 -1.3108 -0.6528 0.0147 0.7394 1.5371
NOK 0.9600 0.9798 1.0005 1.0222 1.0446 -1.3236 -0.6611 0.0154 0.7099 1.4135
NZD 0.9532 0.9769 1.0006 1.0241 1.0475 -1.3927 -0.6789 0.0172 0.6913 1.3496
SEK 0.9596 0.9796 1.0005 1.0224 1.0450 -1.3238 -0.6622 0.0156 0.7104 1.4114
Table V
Forecasting Skewness and Currency Excess Returns.
Panel A reports panel regressions forecasting one-month ahead realized skewness (Skew
P
t
), computed from daily
currency excess returns during month t. Panel B reports the results of the corresponding panel regressions, but with
risk-neutral skewness (Skew
Q
t
) as the dependent variable. Risk-neutral skewness is extracted from one-month foreign
exchange options and is measured on the last day of month t. Panel C reports panel regressions forecasting the
one-month ahead currency excess return, (xs
t
). The dierential between the annualized foreign and domestic LIBOR
rates is denoted by r
f,t
r
d,t
and is measured at the close of business on the last day of month t. The tables report the
magnitudes of the regression coecients and the associated standard errors adjusted for within time-period correlation
(in parentheses). R
2
is the adjusted R
2
from the panel regression; R
2
NFE
is the adjusted R
2
net of the xed eects.
The data form a non-overlapping series of monthly observations spanning the period from January 1999 to March 2007
(N = 882 currency/months).
Panel A: Realized skewness
xs
t
r
f,t
r
d,t
Skew
P
t
Skew
Q
t
R
2
R
2
NFE
Skew
P
t+1
-2.9845 0.0522 0.0199
(0.7471)
Skew
P
t+1
-5.4556 0.0590 0.0269
(1.8684)
Skew
P
t+1
-0.0185 0.0382 -0.0008
(0.0483)
Skew
P
t+1
-0.4746 0.0542 0.0220
0.1557
Skew
P
t+1
-1.2211 -4.6725 -0.0281 -0.2742 0.0732 0.0416
(0.9328) (1.9204) (0.0506) (0.1597)
Panel B: Option-implied skewness
xs
t
r
f,t
r
d,t
Skew
P
t
Skew
Q
t
R
2
R
2
NFE
Skew
Q
t+1
3.7361 0.5776 0.3198
(0.4601)
Skew
Q
t+1
2.144 0.4044 0.0410
(1.0279)
Skew
Q
t+1
0.0212 0.3810 0.0032
(0.0171)
Skew
Q
t+1
0.5865 0.5912 0.3418
(0.0293)
Skew
Q
t+1
3.3864 -0.1951 0.0270 0.5606 0.7628 0.6180
(0.4922) (0.5574) (0.0078) (0.0336)
Panel C: Currency excess returns
xs
t
r
f,t
r
d,t
Skew
P
t
Skew
Q
t
R
2
R
2
NFE
xs
t+1
0.0935 0.0067 0.0076
(0.0515)
xs
t+1
0.3411 0.0446 0.0456
(0.1023)
xs
t+1
-0.0005 -0.0020 -0.0010
(0.0026)
xs
t+1
0.0129 0.0052 0.0062
(0.0085)
xs
t+1
0.0295 0.3268 0.0009 0.0038 0.0441 0.0450
(0.0540) (0.1027) (0.0024) (0.0059)
Table VI
Crash-neutral Currency Carry Trades (1999-2007): Individual currency pairs.
This table reports summary statistics for returns from implementing crash-neutral carry trades in currency pairs
involving the U.S. dollar and one of the remaining nine G10 currencies. Returns are computed monthly for the period
from January 1999 to March 2007 (N = 99). Means, standard deviations and Sharpe ratios (SR) are annualized. Panel
A presents returns for carry trades implementing crash-protection using 10 options; Panel B panel using 25 options,
and Panel C using ATM options. Min and Max report the smallest and largest observed monthly return. Mean
(di) reports the time series mean of the dierences in returns between the crash-neutral carry trade and its standard
counterpart, along with the corresponding t-statistic.
Panel A: Crash-neutral strategy using 10-options.
AUD CAD CHF EUR GBP JPY NOK NZD SEK
Mean 0.0743 0.0275 -0.0071 0.0782 0.0099 0.0180 0.0422 0.0681 0.1021
t-stat 2.20 1.22 -0.23 2.62 0.39 0.58 1.31 1.86 3.15
Std. dev. 0.0971 0.0650 0.0881 0.0858 0.0723 0.0894 0.0929 0.1050 0.0933
Skewness -0.28 -0.06 -0.28 0.00 0.24 0.06 0.32 -0.32 0.08
Kurtosis 2.47 2.41 2.17 2.73 2.36 2.79 2.68 2.51 2.62
Min -0.0557 -0.0366 -0.0581 -0.0617 -0.0433 -0.0587 -0.0507 -0.0624 -0.0598
Max 0.0634 0.0453 0.0445 0.0601 0.0513 0.0666 0.0708 0.0722 0.0805
SR 0.76 0.42 -0.08 0.91 0.14 0.20 0.45 0.65 1.10
Mean (di) -0.0136 -0.0066 -0.0052 -0.0128 -0.0067 -0.0162 -0.0074 -0.0103 -0.0121
t-stat (di) -4.94 -2.02 -0.91 -4.44 -1.77 -7.86 -1.28 -1.92 -3.19
Panel B: Crash-neutral strategy using 25-options.
AUD CAD CHF EUR GBP JPY NOK NZD SEK
Mean 0.0654 0.0230 -0.0021 0.0605 0.0046 0.0123 0.0284 0.0587 0.0812
t-stat 2.25 1.19 -0.08 2.23 0.21 0.46 0.98 1.88 2.73
Std. dev. 0.0835 0.0558 0.0710 0.0779 0.0633 0.0766 0.0834 0.0897 0.0854
Skewness 0.19 0.38 0.18 0.36 0.61 0.59 0.73 0.13 0.40
Kurtosis 2.01 2.33 2.05 2.43 2.49 2.85 2.79 2.25 2.40
Min -0.0335 -0.0239 -0.0374 -0.0380 -0.0262 -0.0377 -0.0367 -0.0412 -0.0347
Max 0.0604 0.0431 0.0417 0.0569 0.0486 0.0636 0.0679 0.0692 0.0762
SR 0.78 0.41 -0.03 0.78 0.07 0.16 0.34 0.65 0.95
Mean (di) -0.0225 -0.0110 -0.0002 -0.0305 -0.0119 -0.0219 -0.0212 -0.0197 -0.0330
t-stat (di) -2.29 -1.49 -0.01 -3.78 -1.41 -2.38 -1.92 -1.52 -3.78
Panel C: Crash-neutral strategy using ATM-options.
AUD CAD CHF EUR GBP JPY NOK NZD SEK
Mean 0.0448 0.0118 -0.0127 0.0326 0.0034 0.0013 0.0084 0.0340 0.0500
t-stat 2.08 0.84 -0.76 1.56 0.22 0.06 0.37 1.50 2.15
Std. dev. 0.0621 0.0403 0.0479 0.0602 0.0453 0.0570 0.0652 0.0650 0.0667
Skewness 0.77 1.11 0.96 1.04 1.36 1.35 1.38 0.93 0.98
Kurtosis 2.58 3.39 2.86 3.07 4.19 4.36 4.01 3.08 3.07
Min -0.0189 -0.0114 -0.0180 -0.0196 -0.0127 -0.0211 -0.0171 -0.0213 -0.0177
Max 0.0534 0.0375 0.0351 0.0494 0.0425 0.0562 0.0607 0.0619 0.0660
SR 0.72 0.29 -0.27 0.54 0.08 0.02 0.13 0.52 0.75
Mean (di) -0.0430 -0.0223 -0.0108 -0.0583 -0.0131 -0.0329 -0.0412 -0.0445 -0.0642
t-stat (di) -2.31 -1.65 -0.50 -3.76 -0.88 -1.84 -2.18 -1.95 -3.96
Table VII
Crash-neutral Currency Carry Trades (1999-2007): Portfolio strategies.
This table reports summary statistics for returns on portfolios of crash-neutral carry trades implemented in each of the
currency pairs involving the U.S. dollar and one of the remaining nine G10 currencies. The component crash neutral
strategies are denoted by CNCT (10), CNCT (25) and CNCT (ATM), to reect the level of protection demanded.
The EQL portfolio strategy equal-weights the crash-neutral carry trades in the nine underlying currency pairs, whereas
SPR weights them by the absolute interest rate spread at initiation of the trade. Panel A presents returns to portfolios
that are not constrained to be neutral with respect to dollar exposure. Panel B presents the corresponding returns for
dollar-neutral strategies, EQL-$N and SPR-$N. Returns are computed monthly for the period from January 1999 to
March 2007(N = 99). Means, standard deviations and Sharpe ratios (SR) are annualized. Min and Max report the
smallest and largest observed monthly return. Mean (di) reports the time series mean of the dierences in returns
between the crash-neutral carry trade and its standard counterpart, along with the corresponding t-statistic.
Panel A: Non-dollar-neutral portfolios
CNCT(10) CNCT(25) CNCT(ATM)
EQL SPR EQL SPR EQL SPR
Mean 0.0318 0.0532 0.0255 0.0430 0.0134 0.0239
t-stat 3.13 3.69 2.77 3.32 1.81 2.35
Std. dev. 0.0291 0.0414 0.0265 0.0372 0.0212 0.0292
Skewness -0.23 0.01 0.22 0.38 0.89 0.97
Kurtosis 3.51 3.40 3.37 3.41 3.63 3.79
Min -0.0216 -0.0251 -0.0201 -0.0248 -0.0106 -0.0142
Max 0.0248 0.0391 0.0226 0.0356 0.0200 0.0272
SR 1.09 1.28 0.97 1.16 0.63 0.82
Mean (di) -0.0242 -0.0313 -0.0305 -0.0414 -0.0426 -0.0605
t-stat (di) -4.37 -5.36 -4.08 -5.03 -4.10 -5.03
Panel B: Dollar-neutral portfolios
CNCT(10) CNCT(25) CNCT(ATM)
EQL-$N SPR-$N EQL-$N SPR-$N EQL-$N SPR-$N
Mean 0.0257 0.0490 0.0195 0.0386 0.0009 0.0106
t-stat 1.48 2.13 1.23 1.87 0.06 0.59
Std. dev. 0.0500 0.0660 0.0455 0.0593 0.0416 0.0517
Skewness -0.18 0.12 -0.06 0.31 0.32 0.57
Kurtosis 2.91 3.44 2.52 3.33 2.85 3.16
Min -0.0379 -0.0406 -0.0279 -0.0330 -0.0239 -0.0254
Max 0.0341 0.0624 0.0291 0.0571 0.0310 0.0469
SR 0.51 0.74 0.43 0.65 0.02 0.21
Mean (di) -0.0177 -0.0209 -0.0239 -0.0313 -0.0426 -0.0592
t-stat (di) -3.90 -4.48 -2.23 -2.81 -2.45 -3.14
Figure 1. Total Return Indices. This gure illustrates the total return indices for the Fama-French factors,
momentum and an equally-weighted carry trade implemented in G10 currency pairs involving the U.S. dollar. The
total return indices are computed by compounding the per period excess returns to each strategy. The returns to each
strategy are scaled ex post to match the volatility of the returns to the currency carry trade. Return are monthly and
cover the period from January 1990 to March 2007 (N = 207). The table below reports summary statistics for the
strategy returns; means, standard deviations and Sharpe ratios (SR) are annualized.
MKT SMB HML UMD FX Carry
Mean 0.0730 0.0227 0.0477 0.0985 0.0478
t-stat 2.13 0.75 1.72 2.51 3.91
St. dev. 0.1422 0.1261 0.1153 0.1630 0.0507
Skewness -0.68 0.81 0.11 -0.66 -0.95
SR 0.51 0.18 0.41 0.60 0.94
Jan90 Jul92 Jan95 Jul97 Jan00 Jul02 Jan05
0.8
1
1.2
1.4
1.6
1.8
2
2.2
2.4
P
o
r
t
f
o
l
i
o

v
a
l
u
e

(
$
)


FX Carry
RMRF
SMB
HML
UMD
Figure 2. Returns to Standard Currency Carry Trades in Mean-Standard Deviation Space (1999-2007).
This gure depicts the returns to carry trades implemented in individual currency pairs involving the U.S. dollar and
one of the nine remaining G10 currencies in mean/standard deviation space. Equal Wgt denotes a portfolio of standard
carry trades which weights each of the underlying currency pairs equally. Spread Wgt denotes a portfolio of standard
carry trades which weights the underlying currency pairs in proportion to their one-month interest rate dierential.
The dashed blue line represents the ex post mean variance frontier for zero-investment portfolios; the red-line represents
the ex post mean variance frontier for portfolios combining the tangency portfolio with a riskless asset. The plot is
constructed using monthly returns on the underlying pairs for the period from January 1999 to March 2007 (N = 99).
Means and standard deviations are expressed in annualized units.
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
0
0.02
0.04
0.06
0.08
0.1
0.12
AUDUSD
CADUSD
CHFUSD
EURUSD
GBPUSD
JPYUSD
NOKUSD
NZDUSD
SEKUSD
Standard deviation (%)
E
x
c
e
s
s

r
e
t
u
r
n

(
%
)


Equal Wgt
Spread Wgt
Equal Wgt ($ neutral)
Spread Wgt ($ neutral)
Country Pairs
MV Frontier
Figure 3. Option Portfolio Positions. This gure plots the positions of the replicating option portfolios used in
computing the (non-central) moments of the risk-neutral distribution. Portfolio positions are plotted as a function of
option moneyness (
K
F
t
). Positions for moneyness values less than (greater than) one apply to put (call) options, such
that the replicating portfolio is constructed entirely of out-of-the-money options.
0.5 0.75 1 1.25 1.5
0
2
4
6
8
10
12
14
Var
Q
Moneyness
P
o
s
i
t
i
o
n
0.5 0.75 1 1.25 1.5
25
20
15
10
5
0
5
Skew
Q
Moneyness
P
o
s
i
t
i
o
n
0.5 0.75 1 1.25 1.5
0
5
10
15
20
25
30
Kurt
Q
Moneyness
P
o
s
i
t
i
o
n
Figure 4. Carry Trade Payo Diagram. This gure plots the payo diagrams for the standard carry trade (dotted
red) and the crash-neutral carry trade (solid blue). The left panel illustrates the strategy payo when the foreign
interest rate exceeds the domestic interest rate, and the investor is long the foreign currency. The right panel illustrates
the symmetric case, when the domestic interest rate exceeds the foreign interest rate, and the investor is short the
foreign currency. The crash-neutral carry trades use foreign exchange options struck 2.5% out-of-the-money relative to
the prevailing forward rate.
0.9 0.95 1 1.05 1.1
0.08
0.06
0.04
0.02
0
0.02
0.04
0.06
0.08
0.1
Long foreign / Short domestic
S
t+1
/ F
t
E
x
c
e
s
s

r
e
t
u
r
n
0.9 0.95 1 1.05 1.1
0.08
0.06
0.04
0.02
0
0.02
0.04
0.06
0.08
0.1
Short foreign / Long domestic
S
t+1
/ F
t
E
x
c
e
s
s

r
e
t
u
r
n
Figure 5. Implied Volatility Functions. The gure illustrates the average implied volatility functions for foreign
exchange options on nine G10 currencies versus the U.S. dollar. The implied volatilities have been scaled by the
contemporaneous at-the-money implied volatilities, and are presented for ve standardized delta values (10 put, 25
put, ATM, 25 call, 10 call). The implied volatilities at the ve standardized value are actual observed volatilities.
All other volatilities were obtained by interpolating the data using the vanna-volga method. The red (blue) lines
correspond to time series means for periods in which the foreign one-month Eurocurrency rate was above (below) the
one-month U.S. interest rate. The underlying data are daily and cover the period from January 1999 to March 2007
(N = 2149).
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
AUD
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
CAD
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
CHF
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
EUR
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
GBP
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
JPY
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
NOK
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
NZD
10dp 25dp ATM 25dc 10dc
0.95
1
1.05
1.1
1.15
1.2
SEK
Figure 6. Risk-Neutral Moments of Currency Returns. The panels in this gure illustrate the time series of
one-month option-implied values of risk-neutral volatility (square root of implied variance), skewness, and kurtosis. The
currencies were selected to depict two currencies with relatively low interest rates versus the U.S. dollar (CHF, JPY)
two with relatively high interest rates (AUD, NZD), and the two other major global currencies (EUR, GBP). The data
in the gure are sampled weekly and cover the period from January 1999 to March 2007. The table reports the time
series means of the risk-neutral moments and their standard errors (in parentheses) computed from non-overlapping,
monthly observations.
Time series means and standard errors
AUD CAD CHF EUR GBP JPY NOK NZD SEK

Var
Q
0.1102 0.0761 0.1082 0.1043 0.0855 0.1079 0.1113 0.1237 0.1121
(0.0023) (0.0013) (0.0018) (0.0022) (0.0013) (0.0026) (0.0017) (0.0023) (0.0017)
Skew
Q
-0.1630 -0.0701 0.1015 0.0807 -0.0147 0.3156 0.0741 -0.1625 0.0594
(0.0149) (0.0200) (0.0156) (0.0160) (0.0163) (0.0296) (0.0147) (0.0139) (0.0143)
Kurt
Q
3.6645 3.6764 3.6425 3.6516 3.6846 4.1010 3.6204 3.6344 3.6084
(0.0154) (0.0237) (0.0199) (0.0204) (0.0208) (0.0339) (0.0205) (0.0143) (0.0167)
Jan00 Jan02 Jan04 Jan06
0
0.05
0.1
0.15
0.2
0.25
Riskneutral volatility ((Var
Q
)
0.5
)


AUD CHF EUR GBP JPY NZD
Jan00 Jan02 Jan04 Jan06
1
0.5
0
0.5
1
Riskneutral skewness (Skew
Q
)


AUD CHF EUR GBP JPY NZD
Jan00 Jan02 Jan04 Jan06
3
4
5
6
Riskneutral kurtosis (Kurt
Q
)


AUD CHF EUR GBP JPY NZD
Figure 7. Crash-neutral Currency Carry Trade Strategy Returns. The gure depicts the total return
indices for the standard carry trade and crash-neutral carry trades with various degrees of downside protection.
The strategy returns are an equally-weighted combination of the carry trade returns for nine G10 currency pairs in-
volving the U.S. dollar. Returns are computed monthly and cover the period from January 1999 to March 2007 (N = 99).
Jan00 Jan02 Jan04 Jan06
0.9
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
P
o
r
t
f
o
l
i
o

v
a
l
u
e

(
$
)


Standard
Crashneutral (10)
Crashneutral (25)
Crashneutral (ATM)