Professional Documents
Culture Documents
Aroop Chatterjee We have built a statistical model to predict the likelihood of a sharp unwind of FX carry trade in
aroop.chatterjee@barcap.com the near term. The model uses market risk conditions, the market price of risk, and the positioning
+1 212 412 5622 of speculators to estimate the probability of an unwind of a G10 FX carry portfolio. The model
Sen Dong has an out-of-sample hit ratio of 60% from 1999 to 2008. Using the model as a filter to trade a
sen.dong@barcap.com carry basket, where the investor is flat when the unwind probability is greater than 50%,
+1 212 412 5622 downside risk is significantly reduced and the Sharpe ratio is more than doubled. An out-of-
www.barcap.com sample back-test on trading an EM FX carry basket and equities also performs well. We plan
weekly updates on what the model is forecasting for carry trades.
Motivation
Carry-based currency strategies, designed to exploit the so-called ‘forward bias’, have
enjoyed a hugely successful run for most of this decade. A simple G10 carry strategy, of
going long the three highest-yielding currencies and short the three lowest-yielding ones
(all equally weighted) would have earned an annualised return of approximately 10% in
excess of the risk-free rate between 2000-07. However, carry trades have suffered huge
losses in the past few months as spiking volatility and increased home bias have led to
dramatic outperformance by low-yielding currencies such as the JPY and the USD. Indeed,
the drawdown associated with the above carry strategy (which we use throughout the
article and will refer to as the ‘carry portfolio’) was nearly 25% in the episode since July, by
far the largest since the 1990s (Figure 1).
A big debate among currency investors over the past few months has been whether the
rationale behind carry trades as an alpha generating strategy over the long term is still
valid. The key question is whether ways exist to mitigate potential downside risks
associated with periodic episodes of carry trade unwind. If investors can forecast these
carry trade corrections, they will be able to reduce large drawdowns, reduce the volatility of
returns and increase both absolute and risk-adjusted returns.
We believe it is possible to forecast carry trade unwinds with some degree of accuracy. In
this report we propose a model that does just that. It goes without saying that the model
cannot predict event risks (eg, the Lehman bankruptcy) but it can produce useful
Please see analyst certification(s) and important disclosures starting on page 12.
BARCAP_RESEARCH_TAG_FONDMI2NBUR7SWED
information on when carry investors should consider moving to the sideline. For the
remainder of the report, we begin by defining an unwind episode, discuss the variables
important for our forecast, and present a statistical model to aggregate them into a signal
about the probability of future carry trade unwinds. The last part of the paper focuses on
analysing the model’s performance in various out-of-sample back-tests.
100
90
80
70
Jan 93 Jul 94 Jan 96 Aug 97 Feb 99 Sep 00 Mar 02 Sep 03 Apr 05 Oct 06 May 08
1 All variables, except for the market price of risk, are expressed as 52 -week rolling z-scores.
1 2
1
-1 0
-1
-3
-2
-5 -3
Jan 94 Jan 97 Jan 00 Jan 03 Jan 06 Jan 94 Jan 97 Jan 00 Jan 03 Jan 06
Speculators’ positioning
CFTC speculators’ positioning in carry trades may reflect the sentiment of investors toward
FX carry trades. Figure 8 shows that the CFTC AUD/JPY net speculative longs tend to drop
sharply during a carry unwind and rise gradually afterwards as investors seek to rebuild
their carry positions.
⎡ Pt ,t + 4 ⎤
Log ⎢ ⎥ = α + βX t + ε t + 4
⎢⎣1 − Pt ,t + 4 ⎥⎦
On a real time basis, the logit is re-estimated using an expanding window starting with 300
weeks of data. The predicted probability at time t+4 for a carry unwind over the next four
weeks is then:
1
Pˆt ,t + 4 =
1 + exp[− (α + β X t )]
The unwind indicator is ON if the predicted probability of an unwind is greater than 50%.
)
S t ,t + 4 = 1( Pˆt ,t + 4 > 50%)
where 1(.) is an indicator variable which is equal to one if the condition in the bracket is
satisfied and zero otherwise.
2 Although each of our explanatory variables is related to the probability that carry trades are unwound, following
the intuition we discussed earlier, aggregating them provides more forecasting power and consistent performance.
We discuss more of this in later sections.
In Figure 10, we report the out-of-sample hit ratio of the unwind model calculated using
real-time data since October 1999. The unwind indicator is ON roughly 60% of the time
when an actual carry unwind is about to occur. The model yields false positives, defined as
when a carry unwind is signalled but does not actually occur, only 26% of the time. This
implies that the model is accurate in 72% of the 472 weeks of out-of-sample back-testing.
Figure 10: Hit ratio for the unwind model (since October 1999)
Carry unwind No carry unwind
Number of weeks 55 417
% indicator is ON 60.7% 26.4%
% indicator is OFF 39.3% 73.6%
Source: Bloomberg, Barclays Capital
In assessing the forecast performance, we consider two strategies (Figure 12). The long-
neutral strategy is long the carry basket in the next week if the unwind indicator is OFF and
earns Libor when it is ON. There is improvement in every performance measure in the long-
neutral strategy compared with the long-only carry strategy. The Sharpe ratio increases
from 0.51 to 1.19, due to a higher mean excess return (6.7% versus 4.7%) and lower
volatility (5.6% versus 9.1%). The downside risk is cut in half and the maximum drawdown
is -7.8%, compared with a hefty -24.6% in the long only strategy.
Figure 12: Model performance statistics (out of sample) Figure 13: Carry portfolio and forecasted unwind zones
The long-short strategy, on the other hand, goes short the carry strategy when the
indicator is ON. This filter-based strategy will get penalised more for inaccurate forecasts;
specifically, it will be less tolerant of false positives. However, even in this case the
drawdown can be reduced by half, to -12.5%, while achieving an average excess return of
8.8% and a Sharpe ratio of 0.97.
Figure 13 plots the cumulative currency returns of the carry portfolio and the forecasted
unwind zone. We find that many drawdowns would have been avoided by using the carry
unwind indicator as a filter. Further, the unwind indicator would have successfully signalled
the start of the 2007-08 period, the most significant episode. Although the model was not
able to capture the initial sell-off during the 2004 episode, it was able to reduce half of the
large drawdown of the carry portfolio. The excess return indices for the three portfolio return
strategies are shown in Figure 14. We find that much of the outperformance for the filter-
based strategies on an absolute return basis occurs post May 2008. However, they do exhibit
outperformance in terms of risk adjusted returns even before the current carry sell-off. Prior
to May 2008, the Sharpe ratio of the long-neutral was 1.23 versus 1.09 for the long only
strategy. The long-short strategy does slightly worse with a Sharpe Ratio of 0.80.
Figure 14: Excess return indices for the three carry portfolio strategies
240 G10 carry portfolio: long only
Long-neutral
220
Long-short
200
180
160
140
120
100
80
Oct 99 Oct 00 Oct 01 Oct 02 Oct 03 Oct 04 Oct 05 Oct 06 Oct 07 Oct 08
In the out-of-sample back-test for October 1999 to November 2008, the performance of the
EM carry trade is greatly improved when the forecasted unwind probability is used as a long-
neutral filter. The Sharpe ratio rises from 0.72 to 1.11, while the maximum draw-down is
decreased to -8.6% from -27.2%. For equities, the unfiltered MSCI World Index delivers a
negative 7.7% average return and a Sharpe ratio of -0.42. Here as well, filtering it with the
carry unwind signal improves the Sharpe ratio to -0.01. The improvement in maximum
drawdown (-49.2% versus -59.5%) is limited due to the fact that the forecasted unwind
probability was largely muted in the bear market of 2000-02 (Figure 9). This was a period
when equity markets underperformed other risky assets to an unusually large extent.
However, the unwind indicator successfully limited draw-downs for trading MSCI World after
2002 (-8.7% compared with -54.4% if unfiltered).
Figure 16: EM carry and equity returns filtered with unwind probability
EM carry MSCI-World
Unfiltered Long-neutral Unfiltered Long-neutral
Mean return 6.5% 6.6% -7.7% -0.1%
Std. deviation 9.1% 5.9% 18.4% 11.5%
Sharpe ratio 0.72 1.11 -0.42 -0.01
Max. drawdown -27.2% -8.6% -59.5% -49.2%
Source: Bloomberg, MSCI, Barclays Capital
Conclusion
Our empirical carry unwind indicator is informed by different variables which affect carry
trade performance. The statistical model presented in this research piece allows these
signals to be combined in a systematic manner. The model derives an estimate for the
probability of future carry trade unwinds. We show how this can be used successfully as a
filter to trade the G10 carry portfolio. The unwind indicator also shows good performance
when applied to an EM carry portfolio and equities.
Appendix
Calculation of price of risk
The price of risk refers to the premium demanded by investors for taking on any
incremental market risk. This is estimated across different asset classes as in a CAPM-type
model, as shown below. 3 In the first regression, individual asset betas to a market-wide risk
factor are calculated using time series data. The second regression is a cross-sectional one,
in which each asset’s expected returns are regressed on the estimated asset betas. The price
of risk (λ) is then given by how expected returns vary in the cross-section to the different
asset betas. As in the finance literature, we use average realised returns as an estimate for
the asset’s expected returns. The set of assets used in estimating the price of risk
encompasses currencies, equities, fixed income and commodities.
i,M
Rti − Rt f = β t ( RtM − Rt f ) + ε ti
i,M
E[ Rti − Rt f ] = β t λtM + ηti
The price of risk will be time-varying. In times of higher risk aversion, assets with high betas
would typically need to be compensated by a higher expected return, implying that the
price of risk will be higher.
∑Λ
i =1
ii
3
0.005 2
0 0
-1
-0.005 -2
-3
-0.01 -4
Jan 94 Jan 96 Jan 98 Jan 00 Jan 02 Jan 04 Jan 06 Jan 08
References
[1] V. Coudert and M. Gex, 2008. ‘Does Risk Aversion Drive Financial Crises? Testing the Predictive Power of
Empirical Indicators’, Journal of Empirical Finance.
[2] Deutsche Bundesbank, 2004. Indicators of International Investors’ Risk Aversion, Monthly Report No 57-10.
[3] M. Misina, 2006. Benchmark Index of Risk Appetite, Bank of Canada Working Paper 2006-16.
Analyst Certification: The persons named as the authors of this report hereby certify that: (i) all of the views
expressed in the research report accurately reflect the personal views of the authors about the subject securities and
issuers; and (ii) no part of their compensation was, is, or will be, directly or indirectly, related to the specific
recommendations or views expressed in the research report.
Important disclosures: On September 20, 2008, Barclays Capital Inc. acquired Lehman Brothers’ North American
investment banking, capital markets, and private investment management businesses.
During the transition period, disclosure information will be provided via the sources listed below.
For complete information, Investors are requested to consult both sources listed below.
https://ecommerce.barcap.com/research/cgi-bin/public/disclosuresSearch.pl
http://www.lehman.com/USFIdisclosures/
Clients can access Barclays Capital research produced after the acquisition date either through Barclays Capital’s
research website or through LehmanLive.
Any reference to Barclays Capital includes its affiliates.
Barclays Capital does and seeks to do business with companies covered in its research reports. As a result, investors
should be aware that Barclays Capital may have a conflict of interest that could affect the objectivity of this report.
IRS Circular 230 Prepared Materials Disclaimer: Barclays Capital and its affiliates do not provide tax advice and nothing contained herein should be
construed to be tax advice. Please be advised that any discussion of U.S. tax matters contained herein (including any attachments) (i) is not intended or
written to be used, and cannot be used, by you for the purpose of avoiding U.S. tax-related penalties; and (ii) was written to support the promotion or
marketing of the transactions or other matters addressed herein. Accordingly, you should seek advice based on your particular circumstances from an
independent tax advisor.
This publication has been prepared by Barclays Capital, Inc., a US registered broker/dealer and member of the FINRA. This publication is provided to
you for information purposes only. Prices shown in this publication are indicative and Barclays Capital is not offering to buy or sell or soliciting offers to
buy or sell any financial instrument. Other than disclosures relating to Barclays Capital, the information contained in this publication has been obtained
from sources that Barclays Capital knows to be reliable, but we do not represent or warrant that it is accurate or complete. The views in this publication
are those of Barclays Capital and are subject to change, and Barclays Capital has no obligation to update its opinions or the information in this
publication. Barclays Capital and its affiliates and their respective officers, directors, partners and employees, including persons involved in the
preparation or issuance of this document, may from time to time act as manager, co-manager or underwriter of a public offering or otherwise, in the
capacity of principal or agent, deal in, hold or act as market-makers or advisors, brokers or commercial and/or investment bankers in relation to the
securities or related derivatives which are the subject of this publication.
Neither Barclays Capital, nor any affiliate, nor any of their respective officers, directors, partners, or employees accepts any liability whatsoever for any
direct or consequential loss arising from any use of this publication or its contents. The securities discussed in this publication may not be suitable for
all investors. Barclays Capital recommends that investors independently evaluate each issuer, security or instrument discussed in this publication, and
consult any independent advisors they believe necessary. The value of and income from any investment may fluctuate from day to day as a result of
changes in relevant economic markets (including changes in market liquidity). The information in this publication is not intended to predict actual
results, which may differ substantially from those reflected.
This communication is being made available in the UK and Europe to persons who are investment professionals as that term is defined in Article 19 of
the Financial Services and Markets Act 2000 (Financial Promotion Order) 2005. It is directed at persons who have professional experience in matters
relating to investments. The investments to which it relates are available only to such persons and will be entered into only with such persons. Barclays
Capital - the investment banking division of Barclays Bank PLC, authorised and regulated by the Financial Services Authority (‘FSA’) and member of the
London Stock Exchange.
Subject to the conditions of this publication as set out above, ABSA CAPITAL, the Investment Banking Division of ABSA Bank Limited, an authorised
financial services provider (Registration No.: 1986/004794/06), is distributing this material in South Africa. Any South African person or entity wishing
to effect a transaction in any security discussed herein should do so only by contacting a representative of ABSA Capital in South Africa, ABSA
TOWERS NORTH, 180 COMMISSIONER STREET, JOHANNESBURG, 2001. ABSA CAPITAL IS AN AFFILIATE OF BARCLAYS CAPITAL.
Non-U.S. persons should contact and execute transactions through a Barclays Bank PLC branch or affiliate in their home jurisdiction unless local
regulations permit otherwise.
Barclays Bank PLC Frankfurt Branch is distributing this material in Germany under the supervision of Bundesanstalt fuer Finanzdienstleistungsaufsicht.
© Copyright Barclays Bank PLC (2008). All rights reserved. No part of this publication may be reproduced in any manner without the prior written
permission of Barclays Capital or any of its affiliates. Barclays Bank PLC is registered in England No. 1026167. Registered office 1 Churchill Place,
London, E14 5HP.
Additional information regarding this publication will be furnished upon request. [080926USD] US10919/EU11385