This article was downloaded by: [University of Illinois Chicago]

On: 20 November 2014, At: 19:25

Publisher: Routledge
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK

Applied Economics Letters

Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/rael20

The high sensitivity of pairs trading returns

a
Nicolas Huck
a
ICN Business School – CEREFIGE , 13 rue Michel Ney, 54037 , Nancy Cedex , France
Published online: 02 Jul 2013.

To cite this article: Nicolas Huck (2013) The high sensitivity of pairs trading returns, Applied Economics Letters, 20:14,
1301-1304, DOI: 10.1080/13504851.2013.802121

To link to this article: http://dx.doi.org/10.1080/13504851.2013.802121

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.

This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
 Applied Economics Letters, 2013
Vol. 20, No. 14, 1301–1304, http://dx.doi.org/10.1080/13504851.2013.802121

The high sensitivity of pairs trading

returns
Nicolas Huck
ICN Business School – CEREFIGE 13 rue Michel Ney, 54037, Nancy
Cedex, France
E-mail: nicolas.huck@icn-groupe.fr
Downloaded by [University of Illinois Chicago] at 19:25 20 November 2014

Pairs trading is a simple and popular relative value trading strategy. This
article deals with the most common implementation of the method based on
a distance criterion. It demonstrates the high sensitivity of the return to
changes in the length of the formation period and shows that a reasonable
modification of this parameter may lead to generate positive excess returns
which are robust to data snooping. This empirical result underlines the
difficulty in understanding pairs trading returns, dynamics and sources of
profitability through time.

Keywords: pairs trading; pairs selection; sensitivity

JEL Classification: G11; G14

I. Introduction GGR and Do and Faff (2010) show the continuing

Pairs trading is a popular self-financing trading strat-
egy in the financial industry (Gatev et al., 2006;
referred to as GGR henceforth) This convergence
trading strategy involves forming a portfolio of two
related stocks whose relative pricing departs from its
equilibrium value. In the literature, the distance
method, the most common form of pairs trading, is
used, among others, by Broussard and Vaihekoski
(2012), Do and Faff (2010, 2012), Baronyan et al.
(2010), Engelberg et al. (2009), and Papadakis and
Wisocky (2008). These articles propose some variants
of the GGR approach, provide a better understanding
of the strategy’s returns and examine the impact of
financial factors such as industry homogeneity, the
frequency of historical reversal in the price spread,
liquidity, attention shifts, firm-specific news or
accounting information events. This strategy is linked
to co-integration and correlation in stock prices, mean
reversion, contrarian strategies and also to the law of
one price. Other pairs trading methods include
stochastic modelling (Elliott et al., 2005), co-
integration (Vidyamurthy, 2004) and multicriteria
decision methods and neural networks (Huck, 2010).
downward trend in the profitability of pairs trading in
the US equity market. This fall could be explained by
an increase in competition between arbitrageurs who
exploit this type of opportunity or by an increase in
market efficiency. In most of the literature, authors
follow the three ad hoc parameters (formation period,
trading/eligibility period and opening trigger) pro-
posed in GGR. In this article, I will not be searching
for more profitable schemes or focusing in detail on
the determinants of the profitability of pairs trading.
Taking data-snooping biases into account, the aim is
simply to evaluate with recent data the potential sen-
sitivity of the results to changes in these parameters.
II. Design of the Application and Empirical
Results
If N stocks are under consideration, NðN  1Þ=2 dif-
ferent pairs can be defined. Initially, for each pair, I
define a measure of closeness. For each stock, I form
pairs by finding the partner which minimizes the sum
of squared differences (SSD) in the normalized daily
prices (inclusive of dividends). Both prices are scaled

# 2013 Taylor & Francis 1301

 1302
to start at $1. Among these pairs, the top 20 with the
lowest SSD become candidates to be traded.
X
T  2
SSDi;j ¼ Pit  Pjt ð1Þ
t¼1
with Pit and Pjt being the normalized prices for stock i
and stock j, respectively, on day t and T being the
number of trading days in the formation period.
Prices are once again scaled to start at $1 at the
beginning of the trading period. Then, during the
trading period, a long–short position (one dollar
short in the higher-priced stock and one dollar long
in the lower-priced stock) is initiated in a pair when-
ever its normalized price difference diverges by more
Downloaded by [University of Illinois Chicago] at 19:25 20 November 2014
than a predefined trigger, which is a multiple of the SD
from the historical spread. Positions are unwound
after convergence or automatically at the end of the
trading period. This two-step sequence (formation/
trading) is, for example, repeated every month. If the
eligibility period of each pair lasts for 6 months, the
‘entire portfolio’ is the sum of six overlapping ‘sub-
portfolios’ staggered by 1 month.
Opening and closing happen on the same day1 as the
trigger is reached. Since there are potentially several
openings and closings during the eligibility/trading
period for a given pair, (excess) portfolio return com-
putation is not a trivial issue. Following one of the two
propositions of GGR, this article considers the fully
invested return, which the authors present as more
realistic because of the flexibility of hedge funds’
funding.
Of the three parameters mentioned above, two are
of major importance for a sensitivity analysis: the
length of the formation period and the opening trig-
ger. They strongly impact the choice and the opening
of the pairs. The trading period, 6 months in general, is
chosen so that the selection process is recent and
round trips have time to occur using a reasonable
opening trigger.
The monthly returns of pairs trading strategies are
presented in Table 1 for four formation period lengths
(6 months, 1 year, 1.5 years and 2 years) and for three
levels of trigger based on the SD spread (2, 3 and 4).
These values remain reasonable for a sensitivity ana-
lysis from an economic and financial point of view.
The entire trading period with a ‘complete portfolio’
lasts from June 2004 to May 2009 (60 months). In fact,
the computation/trading with S&P 500 stocks starts
and ends 5 months before and after, and so the results
in Table 1 are always based on a portfolio defined by six
overlapping ‘sub-portfolios’.
1
A popular alternative is a 1-day delay.
N. Huck
Excess returns are computed/tested by using
Newey–West standard errors with six lags and the
consistent p-values of the bootstrap approach devel-
oped by Hansen (2005). Following Papadakis and
Wisocky (2008), a variant is also presented. If the
stocks of a pair diverge by more than 10% between
the beginning and the end of the formation period, the
pair is directly excluded from the whole process.
About 80% of the pairs are affected here. This techni-
que focuses on pairs which are more likely to have
high co-movement.
As in GGR and Engelberg et al. (2009), a Fama and
French (1993) three-factor regression augmented by
two other factors is performed in order to analyse the
risk exposure of pairs trading monthly returns. The
independent variables are standard factor returns: the

market excess return Rm  Rf where Rf is the 30-day
treasury bill return, a size factor (SMB), a value factor
(HML), a momentum factor (MOM) and a reversal
factor (REV).
The main information provided in Table 1 is as
follows:
l Simple pairs trading with ‘traditional’ para-
meters (1-year formation period) has generated
weak returns in recent years; they are not sig-
nificantly different from zero. This is in line with
GGR and Do and Faff (2010).
l The monthly excess returns of pairs trading are
highly sensitive to the length of the formation
period. For example, with a 1.5-year or a 2-year
formation period, the raw monthly excess
returns are all greater than 0.60% per month
and are likely to be economically significant.
This article does not claim that the longer the
formation period, the better the performance of
the trading system. The 6-month strategy also
provides positive excess returns.
l Returns increase slightly when the opening is
more selective (larger trigger). The restricted
approach also improves the results.
l The factors model shows, focusing on the longer
formation periods, that the five-factor alphas
are significantly positive at the 5% level and
generate returns between 5% and 7% annually
in five of the six cases. Returns were even better
during the financial crisis (market factor) and
loaded negatively on the MOM.
In this application, large ‘abnormal’ positive returns
occur mainly when using ‘long’ formation periods.
The use of several parameterizations (12) leads to
take into account a data-snooping bias. In order to
 Downloaded by [University of Illinois Chicago] at 19:25 20 November 2014

Table 1. Results

Design of the strategy

Formation period 6 months 12 months 18 months 24 months

Trigger (nb of SD) 2 3 4 2 3 4 2 3 4 2 3 4
Trigger (%) 3.43 5.15 6.86 4.38 6.56 8.75 4.83 7.24 9.65 5.01 7.51 10.02

Excess return
Excess return 0.33 0.29 0.35 0.09 0.01 –0.05 0.69 0.71 0.79 0.60 0.70 0.85
(Unrestricted, geometric mean,
in $, per $100, per month)
SE 1.22 1.14 1.40 1.23 1.34 1.50 1.47 1.67 2.02 2.32 2.19 2.96
t–Statistic (Newey–West) 2.14 2.02 1.97 0.61 0.12 –0.21 3.70 3.34 3.14 2.08 2.56 2.34
Consistent p–values (Hansen) 0.0158 0.0121 0.0132 0.2330 0.4230 1.000 0.0003 0.0008 0.0008 0.0324 0.0113 0.0181
Median 0.12 0.16 0.23 0.05 –0.04 –0.11 0.42 0.49 0.36 0.11 0.43 0.51
The high sensitivity of pairs trading returns

Skewness 0.38 0.33 0.54 0.25 0.43 0.28 1.17 1.14 1.00 0.96 0.66 0.57
Kurtosis 2.45 2.59 2.69 2.50 3.02 3.00 5.19 5.04 4.80 3.83 3.22 3.29
Min –1.89 –2.06 –2.05 –2.11 –2.43 –3.25 –1.60 –2.21 –3.00 –2.62 –2.90 –4.02
Max 2.61 2.96 3.46 2.68 3.38 3.83 5.39 6.82 9.20 7.05 6.95 8.54
Monthly returns >0 56.7% 61.7% 58.3% 50.0% 46.7% 45.0% 68.3% 63.3% 61.7% 58.3% 60.0% 61.7%
Sharpe ratio 0.08 0.05 0.08 –0.12 –0.17 –0.19 0.31 0.28 0.28 0.16 0.22 0.22
Monthly serial correlation 0.13 0.09 0.16 0.10 0.19 0.21 0.12 0.03 –0.01 0.14 0.11 0.12
Excess return (restricted, 10%) 0.33 0.28 0.34 0.11 0.03 –0.01 0.77 0.79 0.89 0.70 0.82 1.03
Trading statistics
(per pair, per 6–month period)
Traded pairs 100% 99% 92% 97% 85% 69% 91% 71% 49% 85% 59% 36%
Number of trades (SD) 2.2 (1.3) 1.6 (0.9) 1.3 (0.7) 1.7 (1.0) 1.2 (0.7) 0.9 (0.6) 1.5 (1.0) 1.0 (0.8) 0.7 (0.7) 1.3 (0.9) 0.8 (0.7) 0.6 (0.7)
Time open (SD) in months 4.5 (1.1) 3.8 (1.4) 3.2 (1.7) 3.9 (1.4) 3.1 (1.8) 2.3 (1.9) 3.4 (1.6) 2.4 (1.8) 1.7 (1.8) 3.1 (1.7) 2.0 (1.9) 1.3 (1.7)
Portfolio composition (daily)
Number of open pairs (SD) 90.4 (8.2) 77.4 (8.8) 64.9 (9.5) 79.4 (9.7) 61.6 ( 11.5) 46.6 (12.4) 68.5 (11.5) 48.1 (13.4) 33.0 (14.0) 61.9 (11.9) 40.7 (13.8) 25.1 (13.7)
Factor model
(t–statistics are given in
parentheses)
Intercept 0.10 (1.0) 0.07 (0.9) 0.12 (1.2) –0.18 (–1.6) –0.26 (–2.7) –0.31 (–2.4) 0.43 (4.6) 0.43 (3.5) 0.52 (3.1) 0.29 (1.8) 0.42 (2.3) 0.56 (2.3)
Market –0.07 (–3.5) –0.05 (–2.7) –0.07 (–3.5) –0.07 (–1.4) –0.09 (–2.3) –0.08 (–2.1) –0.17 (–2.3) –0.18 (–2.2) –0.23 (–2.2) –0.19 (–2.4) –0.19 (–2.7) –0.20 (–2.1)
SMB 0.03 (0.6) 0.03 (0.7) 0.01 (0.3) 0.08 (1.5) 0.12 (2.6) 0.06 (1.1) –0.05 (–0.4) –0.05 (–0.4) –0.01 (–0.1) 0.02 (0.2) 0.01 (0.1) 0.03 (0.2)
HML –0.11 (–2.2) –0.14 (–2.4) –0.14 (–2.1) –0.06 (–1.6) –0.10 (–2.6) –0.13 (–2.8) 0.02 (0.3) 0.03 (0.3) 0.10 (0.7) 0.08 (0.9) 0.09 (0.9) 0.13 (1.0)
MOM –0.12 (–10.0) –0.12 (–8.3) –0.15 (–8.8) –0.11 (–5.1) –0.13 (–7.4) –0.16 (–8.8) –0.10 (–4.2) –0.13 (–4.8) –0.18 (–4.6) –0.18 (–4.8) –0.20 (–5.4) –0.27 (–5.6)
REV 0.03 (1.2) 0.04 (1.5) 0.03 (0.9) –0.04 (–1.0) –0.04 (–1.0) –0.03 (–0.7) 0.02 (0.5) –0.01 (–0.1) 0.04 (0.7) –0.04 (–0.8) 0.01 (0.2) 0.00 (–0.0)
R2 0.33 0.32 0.36 0.25 0.37 0.33 0.27 0.26 0.26 0.30 0.28 0.31
1303
 1304
provide a proper adjustment, the Hansen (2005)
approach is used. This test does not change the basic
conclusions mentioned before; the consistent p-value
is very weak (0.1%) for 18-month formation period
indicating that at least one trading strategy is profit-
able. The conclusion remains unchanged and robust,
at the 5% level, to the introduction of annualized
transaction costs up to 3.5%. This level of transaction
costs is especially large and currently unrealistic with
highly liquid stocks and a very limited number of
trades per year as indicated in Table 1 (between 1.2
and 4.4 per eligible pairs).
III. Conclusion
Downloaded by [University of Illinois Chicago] at 19:25 20 November 2014
This article demonstrates, via an example using S&P
500 stocks, the high volatility of the returns of pairs
trading strategies based on the distance approach to
the length of the formation period. The main conclu-
sion of this application is not that some parameteriza-
tions seem profitable and robust to data-snooping
biases. The key point is that pairs trading returns are
highly sensitive to the choice of the length of the for-
mation period and that a data-snooping bias cannot
explain the positive excess returns obtained in some
cases.
If the adaptive market efficiency hypothesis (Lo
2004) could be an explanation for the declining profit-
ability of pairs trading using the most documented
parameterization, new questions are left open by the
sensitivity of the results to changes in the key para-
meters. In order to provide a better understanding of
pairs trading, besides studies focusing on the eco-
nomic drivers, directions of future research are
numerous and include, for example, the following
investigations:
l The length of the formation period is usually
considered as a constant in pairs trading studies.
In a framework including an update/optimiza-
tion of the length of the formation period, how
would the returns be modified? Does the best
length, obtained in the sample, lead to the ‘best’
results during the next trading period? Is it
N. Huck
possible to explain the change of the ‘optimal’
length of the formation period through time?
l The notions of length of the formation period
and memory share some connections. The
empirical patterns observed with pairs trading
strategies (sensitivity) for few decades could be
re-interpreted through a world populated with
learning agents of varying memory lengths
(short-and long-memory traders).
References
Baronyan, S., Boduroglu, I. and Sener, E. (2010)
Investigation of stochastic pairs trading strategies
under different volatility regimes, Manchester School,
78, 114–34.
Broussard, J. P. and Vaihekoski, M. (2012) Profitability of
pairs trading strategy in an illiquid market with multi-
ple share classes, Journal of International Financial
Markets, Institutions & Money, 22, 1188–201.
Do, B. and Faff, R. (2010) Does simple pairs trading still
work?, Financial Analysts Journal, 66, 1–13.
Do, B. and Faff, R. (2012) Are pairs trading profits robust to
trading costs?, Journal of Financial Research, 35,
261–87.
Elliott, R. J., Van Der Hoek, J. and Malcolm, W. P. (2005)
Pairs trading, Quantitative Finance, 5, 271–6.
Engelberg, J., Gao, P. and Jagannathan, R. (2009) An anat-
omy of pairs trading: the role of idiosyncratic news,
common information and liquidity, unpublished work-
ing paper, University of North Carolina, Chapel Hill,
NC, University of Notre Dame, Notre Dame, IN,
Northwestern University, Chicago, IL.
Fama, E. F. and French, K. R. (1993) Common risk factors
in the returns on stocks and bonds, Journal of Financial
Economics, 33, 3–56.
Gatev, E., Goetzmann, W. N. and Rouwenhorst, K. G.
(2006) Pairs trading: performance of a relative-value
arbitrage rule, Review of Financial Studies, 19, 797–827.
Hansen, P. R. (2005) A test for superior predictive ability,
Journal of Business and Economic Statistics, 23, 365–80.
Huck, N. (2010) Pairs trading and outranking: the multi-
step-ahead forecasting case, European Journal of
Operational Research, 207, 1702–16.
Lo, A. (2004) The adaptive markets hypothesis, Journal of
Portfolio Management, 30, 15–29.
Papadakis, G. and Wisocky, P. (2008) Pairs trading and
accounting information, Working Paper, MIT Sloan
School of Management, Cambridge, MA.
Vidyamurthy, G. (2004) Pairs Trading, Quantitative
Methods and Analyis, John Wiley & Sons, Canada.