Beyond The Carry Trade:Optimal Currency Portfolios
Beyond The Carry Trade:Optimal Currency Portfolios
Beyond The Carry Trade:Optimal Currency Portfolios
Abstract
ing currency portfolios. Carry, momentum and value reversal all con-
tribute to portfolio performance, whereas the real exchange rate and the
sample returns that are not explained by risk and are valuable to diversi-
Electroniccopy
Electronic copy available
available at:
at:https://ssrn.com/abstract=2041460
http://ssrn.com/abstract=2041460
ing crash risk. We argue that besides risk, currency returns re‡ect the
Electroniccopy
Electronic copy available
available at:
at:https://ssrn.com/abstract=2041460
http://ssrn.com/abstract=2041460
I. Introduction
Currency spot rates are nearly unpredictable out of sample (Meese and Rogo¤
but with currency spot rates it is quite the opposite –it presents a challenge. Since
currencies have di¤erent interest rates, if the di¤erence in interest rates does not
forecast an o¤setting depreciation, then investors can borrow the low yielding cur-
rencies to invest in the high yielding ones (Fama (1984)). This strategy, known as
the carry trade, has performed extremely well for a long period without any funda-
mental economic explanation. Burnside, Eichenbaum, and Rebelo (2008) show that
a well-diversi…ed carry trade attains a Sharpe ratio that is more than double that of
the US stock market –itself a famous puzzle (Mehra and Prescott (1985)).
Considerable e¤ort has been devoted to explaining the returns of the carry trade
as compensation for risk. Lustig, Roussanov, and Verdelhan (2011a) show that the
risk of carry trades across currency pairs is not completely diversi…able, so there is
(HM LF X )– close in spirit to the stock market factors of Fama and French (1992)
and show that it explains the carry premium. But the HM LF X is itself a currency
strategy, so linking its returns to more fundamental risk sources has been an impor-
go¤ (2009).
Electroniccopy
Electronic copy available
available at:
at:https://ssrn.com/abstract=2041460
http://ssrn.com/abstract=2041460
Some risks of the carry trade are well known. High yielding currencies are known
to “go up by the stairs and down by the elevator,”implying that the carry trade has
substantial crash risk. Carry performs worse when there are liquidity squeezes (Brun-
nermeier, Nagel, and Pederson (2008)) and increases in foreign exchange volatility
(Menkho¤, Sarno, Schmeling, and Schrimpf (2012a)). Its risk exposure is time-
lind (2011)).
Another possible explanation of the carry premium is that there is some “peso
problem”with the carry trade –the negative event that justi…es its returns may sim-
ply have not occurred yet (Barro (2006), Fahri and Gabaix (2007), Gourio, Siemer,
and Verdelhan (2011), Burnside, Eichenbaum, Kleshchelski, and Rebelo (2011)). Us-
ing options to hedge away the “peso risk” reduces abnormal returns, lending some
support to this view, but the remaining returns depend crucially on the particular
option strategy used for hedging (Jurek (2009)). Even so, the recent …nancial cri-
sis was not the “peso event” needed to rationalize the carry trade previous returns
Despite our improved understanding of the risk of the carry trade, the fact remains
that conventional risk factors from the stock market (market, value, size, momentum)
and Rebelo (2011)). Indeed, an investor looking for signi…cant abnormal returns with
respect to, say, the Fama-French factors (1992), would do very well by just dropping
all equities from the portfolio and investing entirely in a passively managed currency
investors. But the currency market has a scarcity of pro…t-seeking capital and,
role during most of the ‡oating exchange rate era (Jylhä and Suominen (2011)). Cen-
tral banks that set domestic monetary policies and occasionally intervene in currency
markets do not seek pro…ts at all (Taylor (1982)).2 Corporate and retail participants
also a¤ect market results with hedging demands not related to pro…tability (Hafeez
The relevance of actors that do not maximize pro…ts in‡uences the pro…tability of
speculative currency strategies. For instance, while technical analysis is close to use-
less in equity markets (Fama and Blume (1966)), there is considerable evidence that
(1993), Taylor and Allen (1992)). LeBaron (1999) …nds that the e¤ectiveness of
(1994) …nds similar evidence in the cross-section of currencies. So the carry trade is
not the only strategy with puzzling returns in the currency market.
(Levich and Pojarliev (2011)). The bene…ts of combining these di¤erent approaches
2
When intervening, central banks stand ready to lose large amounts for extended
periods of time. They typically “lean against the wind”, buying a currency that is
market assumed historical proportions.3 For instance, Deutsche Bank has three
popular ETFs that track carry, value and momentum strategies with the currencies
of the G10. From August 2008 to January 2009, the carry ETF experienced a severe
crash of 32.6%, alongside the stock market, commodities and high yield bonds. But
in the same period, the momentum ETF delivered a 29.4% return and the value ETF
a 17.8% return. So while the carry trade crashed, a diversi…ed currency strategy fared
opments since 2008. Menkho¤, Sarno, Schmeling and Schrimpf (2012b) document
carry and momentum, Asness, Moskowitz, and Pederson (2012) study a combination
of value and momentum in currencies (and other asset classes), and Jordà and Taylor
(2012) combine carry, momentum and the real exchange rate. Still, the core of the
strategies.
Most of the studies on currency strategies focus on simple portfolios. This choice
Uppal (2009), Jacobs, Müller, and Weber (2010)). However, this is exactly because
3
Melvin and Taylor (2009) provide a vivid narrative of the major events in the
real time. Namely, they have to deal with the choice of what signals to use, how
to weigh each signal, and how to address measurement error and transaction costs.
To study the risk and return of currency strategies in a more realistic setting, we
use the parametric portfolio policies approach of Brandt, Santa-Clara, and Valkanov
(2009) and test the relevance of di¤erent variables in forming currency portfolios.
First, we use a pre-sample test to study which characteristics matter for invest-
ment purposes. We test the relevance of the interest rate spread (and its sign),
momentum and three proxies for value: long-term value reversal, the real exchange
rate, and the current account. Including all characteristics simultaneously in the
test allows us to see which are relevant and which are subsumed by others. Then
We …nd that the interest rate spread, momentum and value reversal create eco-
nomic value for investors whereas fundamentals such as the current account and the
real exchange rate do not. The strategy combining the relevant signals increases the
4
Though an out-of-sample exercise does not eliminate forward looking bias com-
pletely. After all, would we be conducting the same exercise in the …rst place if there
were no indications in the literature that momentum and value worked in recent
years?
sample and after transaction costs. This is a 0.29 gain, about the same as the Sharpe
account in the optimization further increases the Sharpe ratio to 1.06, a total gain of
0.49 over the equal-weighted carry benchmark. The gains in certainty equivalent are
even more impressive as the optimal diversi…ed strategy substantially reduces crash
risk.
In an online appendix we show that the risk factors recently proposed to explain
carry returns do not explain the returns of the optimized portfolio. So, while these
risk factors may have some success explaining carry returns, they struggle to justify
rency strategies with stock market factors and bonds.5 We …nd that including cur-
rency strategies in an optimized portfolio increases the Sharpe ratio by 0.51 on aver-
tails and left skewness. This contradicts crash-risk explanations for returns in the
currency market.
Finally, we regress the returns of the optimal strategy on the level of speculative
capital in the market. We …nd evidence that the expected returns of the strategy
decline as the amount of hedge fund capital increases. This suggests that the returns
5
Kroencke, Schindler, and Schrimpf (2011) show there are bene…ts of investing in
funds, as knowledge of the relevant currency characteristics spreads and more capital
is used exploiting them – a result consistent with the adaptive markets hypothesis
(Lo (2004)).
Section III.A. describes the data and the variables used in the optimization. Sections
III.B. and III.C. present the investment performance of the optimal portfolios in and
for investors holding stocks and bonds. Section V discusses possible explanations for
the abnormal returns of the strategy, including insu¢ cient speculative capital early
in the sample.
exchange market. The investor can agree at time t to buy currency i forward at time
i i
t + 1 for 1=Ft;t+1 where Ft;t+1 is the price of one USD expressed in foreign currency
units (FCU). Then at time t + 1 the investor liquidates the position selling the
i i
currency for 1/St+1 ; where St+1 is the spot price of one USD in FCU. The return (in
Therefore all returns are monthly and there are no inherited positions from month
the analysis.
We optimize the currency strategies using the parametric portfolio policies ap-
proach of Brandt, Santa-Clara, and Valkanov (2009). This method models the
that the characteristics convey all relevant information about the assets’conditional
T
(2) wi;t = xi;t =Nt
baum, Kleshchelski, and Rebelo (2011), Menkho¤, Sarno, Schmeling and Schrimpf
(2011a), (2011b).
10
(2009)). We do not place any restriction on the weights, which can be positive or
negative, re‡ecting the fact that in the forward exchange market there is no obvious
non-negativity constraint.
asset, yielding rftU S ; and a long-short portfolio in the forward exchange market.
For a given sample, uniquely determines the parametric portfolio policy, and the
X
Nt
(3) rp;t+1 = rftU S + i
wi;t rt+1
i=1
11
this utility function is that it penalizes kurtosis and skewness, as opposed to mean-
variance utility which focuses only on the …rst two moments of the distribution of
returns. So our investor dislikes crash risk and values characteristics that help reduce
The main restriction imposed on the investor’s problem is that is kept constant
across time. This substantially reduces the chances of in-sample over…tting as only a
!
X
T 1 X
Nt
(6) ^ = arg max 1 U rftU S + ( T i
xi;t =Nt )rt+1
T t=0 i=1
For statistical inference purposes, Brandt, Santa-Clara, and Valkanov (2009) show
that we can use either the asymptotic covariance matrix of ^ or bootstrap methods.8
For the interpretation of results it is important to note that (6) optimizes a utility
function and not a measure of the distance between forecasted and realized returns.
7
Bliss and Panigirtzoglou (2004) estimate empirically from risk-aversion implicit
in one-month options on the S&P and the FTSE and …nd a value very close to 4. We
adopt this value and keep it thoughout. The most important measures of economic
as these are slightly more conservative and do not rely on asymptotic results.
12
mation at all about expected returns. The characteristic may just be a predictor of
a currency’s contribution to the overall skewness or kurtosis of the portfolio, for ex-
even if it does help in forecasting returns since it may forecast both higher returns
and higher risk for a currency, o¤ering a trade-o¤ that is irrelevant for the investor’s
utility function.
egy (Lesmond et al. (2004)). So one valid concern is whether the gains of combining
momentum with carry persist after taking into consideration transaction costs. For-
tunately, parametric portfolio policies can easily incorporate transaction costs that
vary across currencies and over time. This is a particularly appealing feature of
the method, since transaction costs varied substantially as foreign exchange trading
!
X
T 1 X
Nt X
Nt
(7) ^ = arg max 1 U rftU S + ( T i
xi;t =Nt )rt+1 T
xi;t =Nt ci;t
T t=0 i=1 i=1
where ci;t is the transaction cost of currency i at time t; which we calculate as:
This is one half of the bid-ask spread as a percentage of the mid-quote. This assumes
13
For a given month and currency, transaction costs are proportional to the absolute
weight put on that particular currency. This absolute weight is a function of all the
tion between momentum and other characteristics. As Grundy and Martin (2001)
show for stocks, the way momentum portfolios are built guarantees time-varying in-
teraction with other stock characteristics. For instance, after a bear market, winners
tend to be low-beta stocks and the reverse for losers. So the momentum portfolio,
long in previous winners and short in previous losers, will have a negative beta. The
opposite holds after a bull market. The same applies for currencies, after a period
when carry experiences high returns, high yielding currencies tend to have positive
momentum. In this case, momentum reinforces the carry signal and results in larger
absolute weights and thus higher transaction costs. However, after negative carry
returns the opposite happens: high yielding currencies have negative momentum. So
momentum partially o¤sets the carry signal resulting in smaller absolute weights and
actually reduces the overall transaction costs of the portfolio. This means the trans-
document that e¤ective costs in the spot market are less than half those implied by
bid-ask quotes. Also mantained positions in the forward market can be rolled over
14
Combining value reversal and momentum with the carry trade considerably mitigated
the crash of the carry trade in the last quarter of 2008. While this is easy to point out
ex post, the relevant question is whether investors in the currency market had reasons
to believe in the virtue of diversifying their investment strategy before the 2008 crash.
For example, Levich and Pojarliev (2011) examine a sample of currency managers and
…nd that they explored carry, momentum and value strategies before the crisis but
shifted substantially across investment styles over time. In particular, right before
the height of the …nancial crisis in the last quarter of 2008, most currency managers
were heavily exposed to the carry trade, neutral on momentum and investing against
value. This raises the question of whether the bene…ts of diversi…cation were as clear
To address this issue we conduct two tests: i) a pre-sample test with the …rst
time; ii) an out-of-sample experiment since 1996 in which the investor chooses the
weight to put on each signal using only historical information available up to each
moment in time.
Section III.A. explains the data sources and the variables used in our optimization.
In section III.B. we conduct the pre-sample test with the sample from 1976:02 to
15
optimization using only the relevant variables identi…ed in the pre-sample test.
A. Data
We use data on exchange rates, the forward premium, and the real exchange rate
for the Euro zone and the 27 member countries of the Organization for Economic
Cooperation and Development (OECD). The countries in the sample are: Australia,
Austria, Belgium, Canada, the Czech Republic, Denmark, Finland, France, Germany,
Greece, Hungary, Ireland, Italy, Japan, Mexico, Netherlands, New Zealand, Norway,
Poland, Portugal, Slovakia, South Korea, Spain, Sweden, Switzerland, the UK, and
the US.
Most studies in the recent literature on currency returns use broader samples of
ski, and Rebelo (2011), Lustig, Roussanov, and Verdelhan (2011a), (2011b)). But
this raises possible issues of selection bias. It may also be hard to point out the
exact time when an emerging country currency …rst became an eligible asset to in-
vest. To avoid these issues, we restrict ourselves to OECD members –the “developed
countries club.”10
The exchange rate data are from Datastream. They include spot exchange rates
10
We also see no reason to restrict the sample further to just the three major
currencies (as some studies do) or even the G10. The assets we consider were perfectly
eligible to invest and a portfolio optimization will be of little interest if the universe
16
Rebelo (2011) we merge two datasets of forward exchange rates (against the USD
and the GBP) to have a comprehensive sample of returns in the forward market in
We calculate the real exchange rates of each currency against the USD using the
spot exchange rates and the consumer price index. The Consumer Price Index (CPI)
data come from the OECD/Main Economic Indicators (MEI) online database. In the
case of Australia, New Zealand, and Ireland (before November 1975) only quarterly
data are available. In those cases, the value of the last available period was carried
forward to the next month. In the case of the Euro, we use the Harmonized Index of
Consumer Prices (HICP) from the European Central Bank instead. The series that
starts in January 1996, was extended back to January 1988 using the weights in the
We test the economic relevance of carry, momentum, value proxies combined with
1. signi;t : The sign of the forward discount of a currency with respect to the
USD. It is 1 if the foreign currency is trading at a discount (Fi;t > Si;t ) and -1
the GBP from 1976 to 1996 and the second dataset has the same information for
17
Kleshchelski, and Rebelo (2011). Given the extensive study of this strategy we
2. fdi;t : The interest rate spread or the forward discount on the currency. We
standardize the forward discount using the cross-section mean and standard
standardized in the cross-section will have zero mean, implying that the strat-
in the previous three months because there is ample evidence for persistence
in returns for portfolios with this formation period while there are no signif-
icant gains (in fact the momentum e¤ect is often smaller) considering longer
12
Nevertheless, for a funded program, most institutional mandates for currency
funds take as the benchmark just the risk-free rate (Levich and Pojarliev (2008)).
13
Standardizing characteristics in the cross-section of assets is a usual …rst step in
18
ative performance. Even if all currencies fall relative to the USD those that
cumulative real depreciation of currency i between the basis period (h) and mo-
ment t as an index number Qi;h;t = Si;t CP Ii;h 2 CP ItU S2 =Si;h CP Ii;t 2 CP IhU S2 : We
revi;t : This is essentially the same as the notion of “currency value” used in
Asness, Moskowitz, and Pederson (2012). We just use the cumulative deviation
from purchasing power parity, instead of the cumulative return as they did, to
obtain a longer out-of-sample test period. Value reversal is positive for those
currencies that experienced the larger real depreciations against the USD in
5. qi;t : The real exchange rate standardized by its historical mean and standard
deviation. As for value reversal, we compute Qi;hi ;t with the di¤erence that here
the basis period (hi ) is the …rst month for which there is CPI and exchange rate
data available for currency i. Then we compute qi;t = (Qi;hi ;t Qi;t )= Qi;t ; where
P
t
Qi;t is the historical average Qi;hi ;j =t and Qi;t is the historical standard
j=hi
19
needed as the real exchange rate is very close to a unit root process. As such
the average distance from the historical mean each moment in time depends
ensures the optimization does not overweight the signal for currencies with
neutral in terms of the basis currency (the USD). It will tend to be positive
for all currencies when these are undervalued against the USD by historical
standards.
assumes that the previous year current account information becomes known
in April of the current year. The current account data were retrieved from
where data are available on a yearly frequency from 1960 onward. Many studies
examine the relation between the current account and exchange rates justifying
In order to be considered for the trading strategies, a currency must satisfy three
14
This is not the same for every currency as for some data starts at di¤erent periods
20
current forward and spot exchange quotes must be available; and iii) the country
must be already an OECD member in the period considered. After …ltering out
the sample. On average there are 16 currencies in the sample at each point in time.16
B. Pre-Sample Results
Table 1 shows the investment performance of the optimized strategies from 1976:02 to
1996:02. We use this pre-sample period to check which variables had strong enough
evidence supporting their relevance back in 1996, before starting the out-of-sample
experiment.
The two versions of the carry trade (sign and f d) deliver similar performance,
with high Sharpe ratios (0.96 and 0.99, respectively) but also with signi…cant crash
risk (as captured by excess kurtosis and left-skewness). Momentum provides a Sharpe
ratio of 0.56, better than the performance of the stock market of 0.40 in the same
sample. Okunev and White (2003), Burnside, Eichenbaum and Rebelo (2011), and
Menkho¤, Sarno, Schmeling, and Schrimpf (2012b) all document the presence of
Financial predictors work better in our optimization than fundamentals like the
real exchange rate and the current account. Value reversal had a Sharpe ratio of
0.36.17 But the strategies using the current account and the real exchange rate as
16
These include the euro-legacy currencies.
17
Value reversal is similar to the real exchange rate but it throws away the data
21
The seventh row shows the performance of an optimal strategy combining the
carry (both sign and f d) with momentum and value reversal – all the statistically
relevant variables. Already in 1996 there was ample evidence indicating that a strat-
egy combining di¤erent variables leads to substantial gains. The Sharpe ratio of
the optimal strategy was nearly 40% higher than the benchmark and it produced a
16.43 percentage points gain in annual certainty equivalent. Given the high Sharpe
ratio of the strategy, the optimization picks endogenously high levels of leverage that
Adding fundamentals to this strategy does not improve it: the Sharpe ratio
increases only 0.01 and the annual certainty equivalent only 13 basis points. An in-
signi…cant gain since in-sample any additional variable must always increase utility.20
We have known since Meese and Rogo¤ (1983) that currency spot rates are nearly
with more than 5 years each moment in time. We believe this is its crucial advantage
in a sample where real exchange rates are not available for all currencies and for all
periods simultaneously.
18
We also tested these variables out-of-sample (although, based on the in-sample
evidence, the investor would choose not to consider them) and found that they did
mean excess return of 5.27 percent with a standard deviation of 3.92 percent, both
annualized.
20
We provide the results on statistical signi…cance in the online appendix. They
con…rm that in the pre-sample period carry, momentum and value reversal are rele-
22
chas and Rey (2007) …nd that the current account forecasts the spot exchange rate
of the US dollar against a basket of currencies.21 But we …nd no evidence in the cross
section that the current account is relevant at all for designing a pro…table portfolio
of currencies. This does not imply that fundamentals have no e¤ect on exchange
rates. Only that expectations about future fundamentals are already embodied in
present spot rates (see Engel and West (2005) and Sarno and Schmeling (2012)), so
Concerning both carry variables (sign and f d), the correlation of their returns
was only 0.46 from 1976:02 to 1996:02, a value that has not changed much since. So
these two ways of implementing the carry trade are not identical and the investor
…nds it optimal to use both. The sign variable assigns the same weight to a currency
yielding 0.1% more than the USD as to another yielding 5% more. In contrast,
the f d variable assigns weights proportionally to the magnitude of the interest rate
di¤erential. Whenever the USD interest rate is close to the extremes of the cross
section, the sign is very exposed to variations in its value, while f d is always dollar-
neutral.
test shows that as of 1996 some of the strategies recently proposed in the literature
account information. Namely, they detrend it and also consider net foreign wealth.
23
time to build diversi…ed currency portfolios. However, this does not tell us whether
there were other investment variables that we do not test that would have seemed
C. Out-of-Sample Results
mal portfolio combining carry, momentum, and value strategies. The …rst optimal
parametric portfolio is estimated using the initial 240 months of the sample. Then
the model is re-estimated every month, using an expanding window of data until the
end of the sample. The out-of-sample returns thus obtained minimize the problem
The in-sample results also hold out of sample. Table 2 shows that the model using
interest rate variables, momentum and value reversal achieves a certainty equivalent
gain of 10.84 percent over the benchmark, with better kurtosis and skewness. Its
Sharpe ratio is 1.15, a gain of 0.45 over the benchmark sign portfolio.
strategy. For example, Jegadeesh and Titman (1993) provide compelling evidence
that there is momentum in stock prices, but Lesmond et al. (2004) …nd that after
taking transaction costs into consideration there are little to no gains to be obtained
22
Including fundamentals does not change much the results as they receive little
24
Panel B of table 2 shows the OOS performance of the strategies after taking
transaction costs into consideration. Clearly transaction costs matter. The Sharpe
ratio of the optimal strategy is reduced by 0.29, a magnitude similar to the equity
premium, and the certainty equivalent drops from 18.87 percent to just 12.15 per-
cent. Momentum and value reversal individually show no pro…tability at all after
transaction costs. This suggests a simple explanation for the new evidence on cur-
rency return predictability: investors could not exploit it due to transaction costs,
gies, as often done in the literature, is inadequate and overstates the importance of
transaction costs altogether. For example, say the stand-alone momentum strategy
is not pro…table after transaction costs, but a carry strategy is so. Then the investor
will want to follow the carry strategy. The relevant problem for the investor is not
whether stand-alone momentum is exploitable after trading costs but rather if using
momentum on top of carry is bene…cial after the increase in total transaction costs
it implies. In practice, the momentum signal reinforces the carry signal for some
currency-periods, resulting in higher trading costs, but momentum o¤sets carry for
increased costs for the investor. All depends on the interaction between signals. The
…nal row of panel B illustrates our point. The strategy using all signals (even those
that do not produce value individually after transaction costs) still results in substan-
25
g
a cost-adjusted interest rate spread variable: F Di;t = sign(F Dit )(jF Dit j cit ) and
standardize it in the cross-section to get ffdit . We use this variable instead of f di;t in
the vector of currency characteristics xi;t . We then model the parametric weight
function as:
T
(9) wi;t = I(cit < jF Dit j) xi;t =Nt
where I(:) is the indicator function, with a value of one if the condition holds and zero
otherwise. We maximize expected utility with this new portfolio policy, estimating
This method e¤ectively eliminates from the sample currencies with prohibitive
transaction costs and reduces the exposure to those that have a high ratio of cost to
forward discount. Other, more complex, rules might lead to better results, but we
refrain from this pursuit as this simple approach is enough to prove the point that
The procedure increases the Sharpe ratio of the diversi…ed strategy from 0.86 to
23
A strategy using only fd and sign achieves OOS a certainty equivalent of 6.21
percentage points. Hence momentum and reversal add value to the portfolio even
26
gain alone is higher than the momentum or value reversal certainty equivalents per
se. Indeed, the performance of the diversi…ed strategy with managed transaction
the currency strategy per dollar invested in the risk-free asset. The optimal strategy
has a mean leverage of 5.94 in the OOS period of 1996:03 to 2011:12. This means,
that for each dollar invested in the risk free rate, the investor would be long 3-dollars
worth of some set of foreign currencies and short 3-dollars worth of another set of
stable and erratic coe¢ cients OOS. Timing di¤erent investment styles is specially
challenging. For instance, Levich and Pojarliev (2008),(2011) …nd that while cur-
rency managers show some timing ability within their speci…c investment strategy,
they shift erratically across styles without any particular skill. In the online appen-
dix we show the coe¢ cients of the optimal strategy are stable, leading to consistent
exposure to the conditioning variables. So the optimal diversi…ed portfolio does not
We present and discuss the risk exposures of the optimal strategy in the online
24
In fact, some practitioners shared with us that what they are really interested in
is …nding a better method to shift across styles. For now our advice is simple: don’t!.
27
risk and innovations in transaction costs. Nevertheless, risk exposures are insu¢ -
cient to explain the mean returns of the strategy, which are close to its risk-adjusted
returns. Time-varying risk is also not relevant to explain the returns of the strategy.
Generally, the results indicate that the optimal strategy exploits market ine¢ ciencies
We assess whether the currency strategies are relevant for investors already exposed
to the major asset classes. Indeed, there is no reason a priori that investors should
restrict themselves to pure currency strategies, particularly when there are other risk
The value of currency strategies to diversi…ed investors holding bonds and stocks
is a relatively unexplored topic. Most of the literature on the currency market has
We continue to assume that the investor optimizes power utility with constant
28
as excess returns, and wi;t depends on the characteristics and the coe¢ cients that
As sets of investable factors we consider the market premium (RMRF), the Fama-
French factors (RMRF,SMB, and HML), and the Carhart factors consisting on the
strategy combines the interest rate spread, sign, momentum, and long-term value
reversal.
Figure 1 shows the OOS performance of the optimized portfolios with and without
to investors. Including currencies in the portfolio always adds to the Sharpe ratio
and raises the certainty equivalent. The OOS gains in certainty equivalent range are
specially high for a diversi…ed investment using the Carhart factors. This gain comes
mainly from the dismal performance of stock momentum in 2009, when it experienced
one of its worst crashes in history (Daniel and Moskowitz (2012), Barroso and Santa-
Clara (2012)).
These gains are far more impressive than the gains from adding factors like HML
25
In the online appendix we show the table with the descriptive statistics of the
OOS performance.
29
the certainty equivalent of the stock market OOS. Generally, the inclusion of SMB,
HML, and WML factors improves Sharpe ratios, but this increase is o¤set by higher
interest rate spread, currency momentum, and long-term value reversal to forecast
currency returns makes all conventional risk premiums seem small in comparison.
average.
One possible justi…cation for the higher Sharpe ratios obtainable by investing in
currencies is that these might entail a higher crash risk – as Brunnermeier, Nagel,
and Pedersen (2008) show for the carry trade. But diversi…ed currency strategies
portfolio policy with investments in the currency market reduces substantially the
Our results make it hard to reconcile the economic value of currency investing with
the existence of some set of risk factors that drives returns in currencies and other
asset classes. The substantial increases in Sharpe ratios combined with the lower
crash risk indicate that there is either a speci…c set of risk factors in the currency
market or that currency returns have been anomalous throughout our sample.
30
risk. The obvious alternative explanation is market ine¢ ciency. This might persist
due to insu¢ cient arbitrage capital, possibly because strategies exploring the cross
section of currency returns were not well known or because of barriers to entry
such as speci…c trading platforms, bank relationships and human capital expertise
(Levich and Pojarliev (2012)). This argument is consistent with the adaptive markets
hypothesis of Lo (2004). This hypothesis argues that it takes time for arbitrageurs to
gather enough capital to fully exploit one source of anomalous risk-adjusted returns.
As such, an anomaly can persist for some time, even if not inde…nitely.
Jylhä and Suominen (2011) …nd that carry returns explain hedge fund returns
even after controlling for the other factors proposed by Fung and Hsieh (2004) and
that growth in hedge fund speculative capital is driving carry trade pro…ts down.
Neely, Weller, and Ulrich (2009) document a similar decline over time of the prof-
We run an OLS regression of the returns of the optimal strategy on hedge fund
in their sample (AU M=M 2) and new fund ‡ows ( AU M=M 2):26 The regression
uses the out-of-sample returns, after transaction costs, of the optimal strategy from
1996:03 to 2008:12 as the dependent variable. The estimated coe¢ cients (and t-
26
We thank Matti Suominen for providing us the time series of AUM/M2. See their
31
is not signi…cant in the regression but the estimated coe¢ cient has the correct sign.
The level of hedge fund capital predicts negatively the returns of the optimal strategy.
With a t-statistic of -3.23, this provides supportive evidence that the returns of the
more hedge fund capital exploits it.27 This result supports the adaptive markets
hypothesis in the currency market and complements the existing evidence of Neely,
This opens the question of whether the large returns of the strategy are likely to
continue going forward. We note that in the last three years of our sample (2009-
2011) the strategy produces a Sharpe ratio of 0.82, lower than its historical average
but still an impressive performance (though not much di¤erent than the stock market
VI. Conclusion
ential, and value reversal, outperform the carry trade substantially. This outperfor-
27
The signi…cance of the coe¢ cient of AUM/M2 is robust to the inclusion of a time
variable in the right hand side. So this result can not be attributed to a mere trend
e¤ect.
32
reversal and momentum had large positive returns when the carry trade crashed. The
of currency returns.
Our optimal currency portfolio picks stable coe¢ cients for the relevant currency
by risk factors or time-varying risk. This suggests market ine¢ ciency or, at least,
that the right risk factors to explain currency momentum and value reversal returns
folios already exposed to stocks and bonds. So currencies either o¤er exposure to
The most plausible explanation for the returns of our optimal diversi…ed currency
away as speculative capital increases in the foreign exchange market. This is consis-
By using new optimization technology on old currency data, we show that the
puzzles in the currency market are too deep (and the economic performance of the
33
Burnside, C.; M. Eichenbaum; and S. Rebelo. “Carry Trade: The Gains of Diversi-
34
Explain the Returns to the Carry Trade?” Review of Financial Studies, 24 (2011),
853-891.
52 (1997), 57–82.
24 (2005), 1150–1175.
(2011), 1107-1125.
How Ine¢ cient is the 1/N Portfolio Strategy?” Review of Financial Studies, 22
(2009), 1915-1953.
Dornbusch, R., and S. Fischer. “Exchange Rates and the Current Account.” The
35
14 (1984), 319–338.
Fama, E. F., and M. Blume. “Filter rules and stock market trading pro…ts.”Journal
Farhi, E., and X. Gabaix. “Rare Disasters and Exchange Rates: A Theory of the
Fung, W., and D. Hsieh. “Hedge fund benchmarks: a risk based approach.”Financial
Grundy, B., and J. S. Martin. “Understanding the nature of the risks and the source
29-78.
36
Jegadeesh, N., and S. Titman. “Returns to Buying Winners and Selling Losers:
Jordà, O., and A. Taylor. “The Carry Trade and Fundamentals: Nothing to Fear
Paper (2009).
Jylhä, P., and M. Suominen. “Speculative Capital and Currency Carry Trades.”
(2011) .
Lo, Andrew W. “The adaptive markets hypothesis: Market e¢ ciency from an evo-
(2004), 15–29.
LeBaron, Blake. “Technical Trading Rule Pro…tability and Foreign Exchange Inter-
37
Levich, R., and M. Pojarliev. “Do Professional Currency Managers Beat the Bench-
Levich, R., and M. Pojarliev. “Detecting Crowded Trades in Currency Funds.” Fi-
Meese, R. A., and K. Rogo¤. “Empirical Exchange Rate Models of the Seventies :
38
Menkho¤, L.; L. Sarno; M. Schmeling; and A. Schrimpf. “Carry Trades and Global
Neely, C.; P. A. Weller; and J. M. Ulrich. “The Adaptive Markets Hypothesis: Ev-
idence from the Foreign Exchange Market.” Journal of Financial and Quantitative
Obstfeld, M., and K. Rogo¤. “Global Current Account Imbalances and Exchange
Okunev, J., and D. White. “Do Momentum-Based Strategies Still Work in For-
425— 447.
Rogo¤, K. S., and V. Stavrakeva. “The Continuing Puzzle of Short Horizon Exchange
Sarno, L., and M. Schmeling. “Do Exchange Rates Forecast Fundamentals? A Cross-
39
Taylor, D. “O¢ cial Intervention in the Foreign Exchange Market, or, Bet Against
Taylor, M. P., and H. Allen. “The use of technical analysis in the foreign exchange
40
41
42
10
0 1
-10
-20 0.5
-30
-40 0
RMRF FF Carhart S+B RMRF FF Carhart S+B
8
0
6
-0.5
4
-1
2
-1.5 0
RMRF FF Carhart S+B RMRF FF Carhart S+B
43