See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/296332425

The profitability of pairs trading strategies:

distance, cointegration and copula methods
Article in Quantitative Finance April 2016
DOI: 10.1080/14697688.2016.1164337

CITATIONS

READS

654

3 authors, including:
Hossein Rad

Rand Kwong Yew Low

University of Queensland

Bank of America Merrill Lynch

2 PUBLICATIONS 0 CITATIONS

9 PUBLICATIONS 28 CITATIONS

SEE PROFILE

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Can commodity portfolios perform? View project

All content following this page was uploaded by Rand Kwong Yew Low on 30 April 2016.
The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.

Quantitative Finance

ISSN: 1469-7688 (Print) 1469-7696 (Online) Journal homepage: http://www.tandfonline.com/loi/rquf20

The profitability of pairs trading strategies:

distance, cointegration and copula methods
Hossein Rad, Rand Kwong Yew Low & Robert Faff
To cite this article: Hossein Rad, Rand Kwong Yew Low & Robert Faff (2016): The profitability
of pairs trading strategies: distance, cointegration and copula methods, Quantitative Finance,
DOI: 10.1080/14697688.2016.1164337
To link to this article: http://dx.doi.org/10.1080/14697688.2016.1164337

Published online: 27 Apr 2016.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=rquf20
Download by: [UQ Library]

Date: 27 April 2016, At: 16:03

Quantitative Finance, 2016

http://dx.doi.org/10.1080/14697688.2016.1164337

The profitability of pairs trading strategies:

distance, cointegration and copula methods
HOSSEIN RAD, RAND KWONG YEW LOW and ROBERT FAFF
UQ Business School, University of Queensland, Brisbane 4072, Australia
Stern School of Business, New York University, New York, NY 10012, USA

Downloaded by [UQ Library] at 16:03 27 April 2016

(Received 17 September 2015; accepted 3 March 2016; published online 27 April 2016)
We perform an extensive and robust study of the performance of three different pairs trading
strategiesthe distance, cointegration and copula methodson the entire US equity market from
1962 to 2014 with time-varying trading costs. For the cointegration and copula methods, we design a
computationally efficient two-step pairs trading strategy. In terms of economic outcomes, the distance,
cointegration and copula methods show a mean monthly excess return of 91, 85 and 43 bps (38,
33 and 5 bps) before transaction costs (after transaction costs), respectively. In terms of continued
profitability, from 2009, the frequency of trading opportunities via the distance and cointegration
methods is reduced considerably, whereas this frequency remains stable for the copula method.
Further, the copula method shows better performance for its unconverged trades compared to those of
the other methods. While the liquidity factor is negatively correlated to all strategies returns, we find
no evidence of their correlation to market excess returns. All strategies show positive and significant
alphas after accounting for various risk-factors. We also find that in addition to all strategies performing
better during periods of significant volatility, the cointegration method is the superior strategy during
turbulent market conditions.
Keywords: Pairs trading; Copula; Cointegration; Quantitative strategies; Statistical arbitrage
JEL Classication: G11, G12, G14

1. Introduction
Gatev et al. (2006) show that a simple pairs trading strategy
(PTS), namely the Distance Method (DM), generates profit
over a long period. However, Do and Faff (2010) document
that the profitability of the strategy is declining. They associate this decrease to a reduction in arbitrage opportunities
during recent years, as measured by the increase in the proportion of pairs that diverge but never converge. Do and Faff
(2012) show that the DM is largely unprofitable after 2002,
once trading costs are taken into account. Jacobs and Weber
(2015) find that the profitability of the DM is immensely timevarying. Nonetheless, there are other tools such as cointegration and copulas that can be used to implement statistical
arbitrage trading strategies. Although such concepts are cursorily introduced in the pairs trading literature, their performance has not been robustly evaluated. Accordingly, our basic
goal is to evaluate the performance of two sophisticated PTS,
namely copula and cointegration methods, using a long-term
and comprehensive data-set. We also assess if there is a decline in pairs trading profitability for these more sophisticated

methods and investigate the risk-factors that might influence

their profitability.
Pairs trading strategies are comprised of two stages: first,
the method applied to form pairs; and second, the criteria for
opening and closing positions. In the DM, securities whose
prices are closely correlated are grouped in pairs, and traded
when their prices diverge by more than a pre-specified
amount. This is the only strategy that has been tested thoroughly using extensive data-sets, a wide variety of securities
and across different financial markets (Andrade et al. 2005,
Gatev et al. 2006, Perlin 2009, Do and Faff 2010, Broussard and
Vaihekoski 2012, Do and Faff 2012, Jacobs and Weber 2015).
Cointegration can be employed in a pairs trading framework
(Vidyamurthy 2004; Lin et al. 2006).Although Lin et al. (2006)
implement a cointegration PTS, their empirical analysis only
examines two Australian shares over a short sample period of
one year. In the application of copulas in pairs trading, Xie
and Wu (2013) propose a strategy, and Wu (2013) evaluates its
performance using three pre-selected pairs. Xie et al. (2014)
explore 89 US stocks in the utility industry over a sample
period of less than a decade. Our study extends the literature
by examining the performance of a cointegration-based and a

Corresponding author. Emails: [email protected], [email protected]

2016 Informa UK Limited, trading as Taylor & Francis Group

Downloaded by [UQ Library] at 16:03 27 April 2016

H. Rad et al.

copula-based PTS, using the CRSP data-set from 1962 to 2014.

Using a comprehensive data-set spanning over five decades
and containing all US stocks, our study is a robust examination of alternative PTS. By evaluating the performance of
cointegration and copula-based trading strategies against the
DM benchmark, we establish whether these complex methods
yield better performance in the long-term.
Understanding the difference in performance outcomes between copula and cointegration PTS versus the benchmark, the
DM, will provide valuable insight into the source of pairs trading profitability and the reasons behind the observed decline
in the profitability of the DM. Has the market become more
efficient and the availability of arbitrage opportunities diminished? Or are contemporary methodologies more sophisticated
than the simple DM required to exploit market inefficiencies?
For example, the simplicity of the DM might induce increased
arbitrage activity, leading to fewer arbitrage opportunities to
exploit thereby resulting in a drop in profitability. The answer to
these questions sheds light on the direction of future research by
academics and practitioners in order to build better performing
strategies which will in turn further improve market conditions
and efficiency.
We study how increasing the sophistication of methods by
which pairs are selected and traded can affect the quality and
precision of the captured relationship within the pair and, ultimately, the performance of PTS. In theory, the presence of a
cointegration relation between two assets means that there is a
long-term relationship between them. Exploiting this relationship should allow us to accurately model the co-movement of
the pair and use that to implement a high-performance PTS.
Equities are shown to exhibit asymmetric dependence (Longin
and Solnik 1995, Patton 2004, Low et al. 2013). Using copulas
for modelling the dependence structure between two assets,
instead of restricting the framework towards the elliptical dependence structure of covariance matrix, would also possibly
lead to a superior PTS by allowing for more flexibility in capturing asymmetries in the dependence structure within pairs.
Nevertheless, more complex models can also result in inferior
performance, especially out of sample, by introducing issues
such as over-fitting. Moreover, the computational requirements
necessary to process these mathematically complex algorithms
may outweigh their relative performance improvements over
simpler strategies. This might result in weakening of motivation to adapt such strategies in practice.
The novel contributions of this paper to the relevant literature are fourfold. First, our study combines aspects of the
DM and the cointegration or copula technique to produce a
computationally efficient two-step approach to pairs trading
that can be operationalized by practitioners. When considering
In the last year of our study, 2014, there are an average of 2,377
stocks ( N ) per day, resulting in a total of 2,823,876 ( N [ N 1]/2)
unique stock pairs to be analysed for selection into the strategy. When
restricted to a single core processor, the average computation time
for selecting the best copula model and fit for each stock pair is
0.44 seconds. Thus, analysing all unique stock pairs on a single day
requires a total of 345 h for a single core processor. Performing such
an analysis within 5 h requires a minimum of 70 core processors using
parallel computing techniques. Our analysis is performed on Matlab
2014b with the Parallel Computing toolbox on a compute server with
dual Intel Xeon Processors E5-2640 (24 hyper-threaded cores, 30 MB
Cache, Max 3.00 GHz) and 128 GBs of RAM.

a trading strategy, speed and efficiency of computation is a

vital consideration (Clark 2012, Angel 2014, Brogaard et al.
2014). Stock pairs are sorted and selected by sum of squared
differences (SSD). In the copula (cointegration) strategy, for
each selected stock pair, a range of copula and marginal models
are fitted and selected based upon the AIC and BIC criterion
(the cointegration coefficient is calculated). Wu (2013) only
fits copulas and marginal models to one pair and Xie et al.
(2014) use a data sample limited in both time span and number
of stocks. We find that the Student-t copula is selected for
61% of the pairs. This highlights the fact that the dependence
structure of the stock pairs exhibits fat tails, and therefore
the classic linear correlation framework employed in simpler
methods are inadequate in modelling their relation. Second,
we perform a comprehensive evaluation of the performance
of two alternative PTS (i.e. cointegration and copulas) against
a data-set consisting of all the shares in the US market from
1962 to 2014. PTS more sophisticated than the DM, have not
been empirically tested in a robust manner, and therefore their
long-term performance remains unknown. Due to the broad
and long data sample used, this longitudinal study presents the
first extensive examination of the performance of two relatively
new PTS, using cointegration and copula methods. Third, with
various economic and risk-adjusted return metrics, we evaluate
the performance of all three PTS and show if the increased
complexity in the pairs selection and trading criteria improves
performance. With respect to studies finding a recent decline
in the performance of the DM, this comparison will lead to
understanding if arbitrage opportunities are still available in
the market, but perhaps due to increased arbitrage activity,
more complex methods such as copulas are required to take
advantage of them. Fourth, we examine the performance of
the PTS in relation to findings in the asset pricing literature
that show that momentum (Carhart 1997), liquidity (Pstor
and Stambaugh 2003), and more recently profitability and investment patterns (Fama and French 2015) explain stock
returns.
Our findings show that the cointegration method performs
as well as the DM in economic and risk-adjusted performance
measures. The two strategies also show very similar pair trade
properties and risk profiles. Based upon lower partial moment
and drawdown measures, the cointegration method performs
better than the other strategies before transaction costs are
taken into account, whereas after costs the DM is slightly superior. We find the copula methods economic and risk-adjusted
performance to be weaker than the other two methods. The
weaker performance of the copula method can be attributed to
the high proportion of unconverged trades. A positive outcome
of the copula method is that, unlike the other methods, the
frequency of its trades have not fallen in recent years, thus its
economic performance is more stable over time. We show that
the liquidity factor is negatively correlated with the return of
each strategy. No such correlation can be found with the market returns, which demonstrates the market neutrality of these
strategies. The alphas of all PTS remain large and significant
even after several asset pricing factors such as momentum,
liquidity, profitability and investment (Fama and French 2015)
are taken into account.
The remainder of this paper is structured as follows. In
section 2, we review some of the relevant literature on pairs

The protability of pairs trading strategies

trading, copulas and cointegration. A description of our dataset is in section 3. Section 4 covers the research method. And
finally, the results and conclusion are presented in sections 5
and 6, respectively.

Downloaded by [UQ Library] at 16:03 27 April 2016

2. Literature review
Research on PTS fall under the general finance banner of
statistical arbitrage. Statistical arbitrage refers to strategies
that employ some statistical model or method to take advantage of what appears to be mispricing between assets while
maintaining a level of market neutrality. Gatev et al. (2006)
is the earliest comprehensive study on pairs trading. They
test the most commonly used and simplest method of pairs
trading, the DM, against the CRSP stocks from 1962 to 2002.
Their strategy yields a monthly excess return of 1.3%, before
transaction costs, for the DMs top five unrestricted pairs, and
1.4% for its top 20. In addition, after restricting the formation
of pairs to same-industry stocks, Gatev et al. (2006) report
monthly excess returns of 1.1, 0.6, 0.8 and 0.6% on top 20
pairs in utilities, transportation, financial and industrial sectors, respectively. This study is an unbiased indication of the
strategys performance as they interpret the simplest method
that practitioners employ as PTS. To avoid any criticisms that a
profitable trading rule is data-mined and applied in their study,
the authors re-evaluate the original strategy after four years
and show that it remains profitable.
Do and Faff (2010, 2012) further examine the DM strategy of
Gatev et al. (2006) to investigate the source of its profits and the
effects of trading costs on its profitability using CRSP data from
1962 to 2009. Do and Faff (2010) find that the performance of
the DM peaks in 1970s and 1980s, and begins to decline in the
1990s. The two exceptions to the plummeting performance of
the DM strategy both occur during bear markets of 20002002
and 20072009. During these two periods, the DM shows solid
performance, which is a slight reversal in the declining trend
in the strategys profitability across the period from 1990 to
2009. Moreover, they show that increased performance during
the first bear market, i.e. 20002002, is due to higher profits of
pairs that complete more than one round-trip trade, rather than
an increase in their number. By contrast, in the second bear
market, i.e. 20072009, the increase in the number of trades
that complete more than one round-trip trade is the driver of
strategys strong profitability. After taking into account timevarying transaction costs, Do and Faff (2012) conclude that
DM on average is not profitable. However, for the duration of
the sample period, the top 4 out of the 29 portfolios constructed
show moderate monthly profits of average 28 bps or 3.37% per
annum. In addition, for the period from 1989 to 2009, the DM
is profitable, albeit the majority of its profits occur in the bear
market from 20002002.
Several studies explore pairs trading using the DM in different international markets, sample periods and asset classes
(Andrade et al. 2005, Perlin 2009, Broussard and Vaihekoski
2012). Jacobs and Weber (2015) comprehensively analyse the
Return on employed capital.
Unrestricted pairs are the pairs that have not been formed based on
specific criteria such as belonging to the same industry.

DM in 34 countries and find that although the strategy generates positive return, this return varies considerably over time.
They attribute the source of the strategys profitability towards
investors under or over-reaction to news information.
PTS may be categorized under algorithmic trading strategies
that presently dominate most markets order books. Within
this stream of literature, Bogomolov (2013) adapts technical
analysis in pairs trading. Using two Japanese charting indicators (i.e. the Renko and Kagi indicators), the study has a
non-parametric approach to pairs trading, and thus, does not
rely on modelling the equilibrium price of a pair. These indicators model the variability of the spread process within
a pair. Based on the premise that the pair expresses meanreverting behaviour, this variability is used in calculating how
much the spread should deviate, before a trade becomes potentially profitable. Thus, this strategy relies on the stability of
the statistical properties of the spread volatility. The strategy
yields a positive before-costs monthly return between 1.42 and
3.65%, when tested on the US and Australian markets. Yang et
al. (2015) use limit orders to model the trading behaviour of
different market participants in order to identify traders, and
equivalently, algorithmic traders. They do so by solving an
inverse Markov decision process using dynamic programming
and reinforcement learning. They show that this method leads
to accurate categorization of traders. All PTS apply the use
of a threshold that when crossed by the spread, triggers a
trade. Zeng and Lee (2014) aim to find the optimum value for
this threshold, given the spread follows a OrnsteinUhlenbeck
process, by defining it as an optimization problem. They focus
on maximizing the expected return per unit time. Several other
studies also focus on deriving automated trading strategies
from technical analysis or creating profitable algorithms based
on different cross-disciplinary concepts (Dempster and Jones
2001, Huck 2009, 2010, Creamer and Freund 2010).
PTS can be implemented using cointegration. Vidyamurthy
(2004) presents a theoretical framework for pairs trading using
cointegration based upon the error correction model representation of cointegrated series by Engle and Granger (1987).
Huck and Afawubo (2015) and Huck (2015) implement a PTS
using the cointegration method, and using S&P 500 and the
Nikkei 225 stocks, they show that it generates positive returns.
Bogomolov (2011) studies the performance of three different PTS. DM, cointegration and stochastic spread strategies
are implemented and tested on the Australian share market
from 1996 to 2010. The study concludes that while all three
trading strategies are profitable before transaction costs, much
of the profits are diminished after costs and liquidity issues
are taken into account. Caldeira and Moura (2013) also use
a cointegration-based trading strategy on the Sao Paolo exchange. They find that the strategy generates a 16.38% excess
return per annum with a Sharpe ratio of 1.34 from 2005 to
2012. Lin et al. (2006) also motivate the use of cointegration as
a model that can capture the long-term equilibrium of the price
spread, while addressing the deficiencies of simpler statistical
techniques used in pairs trading such as correlation and regression analysis. Using cointegration coefficient weighting, Lin
et al. (2006) implement a theoretical framework that ensures
Cointegration coefficient weighting refers to the method in which
position weights are calculated as a function of the cointegration
coefficient.

Downloaded by [UQ Library] at 16:03 27 April 2016

H. Rad et al.

some minimum nominal profit per trade (MNPPT). They proceed to introduce a five step set of trading rules to apply the
framework in pairs trading. Finally, their empirical analysis
uses a small data-set of a 20 month sample period for two
Australian bank stocks, and concludes that the MNPPT does
not put excessive constraints on trading if adapted along with
commonly-used values for trading parameters such as open
and close trade triggers. Galenko et al. (2012) implement a
PTS based on cointegration and examines its performance with
four exchange traded funds. Nevertheless, these studies share a
common shortcoming, where the empirical evidence provided
to support the cointegration-based PTS are either non-present
(being theoretical constructs) or severely limited in their analysis. For example, the strategy proposed in Vidyamurthy (2004)
is not analysed on real data and Caldeira and Moura (2013) use
data from the Sao Paulo stock exchange for a period of less than
seven years.
Due to the properties of copulas in allowing the freedom
to select marginal distributions and to flexibly model joint
distributions (particularly lower tail dependence) copulas have
also been frequently used in risk management (Siburg et al.
2015, Wei and Scheffer 2015) and asset allocation (Patton
2004, Chu 2011, Low et al. 2013, Low, Faff et al. 2016).
Okimoto (2014) studies the asymmetric dependence structure in the international equity markets including the US and
concludes that firstly, there has been an increasing trend in
the dependence within the equity markets over the last 35
years. Secondly, with the aid of copulas, the study finds strong
evidence of the asymmetry of upper and lower tail dependence
structure and points out the inadequacy of the multivariate
normal model in capturing the characteristics of equities. Given
the recent developments in the equity dependence outlined by
this study, employing traditional correlation in modelling the
joint behaviour, as done in the majority of quantitative methods
to date, would poorly represent the true relationship among
assets, and therefore, is no longer appropriate. This motivates
the use of copulas in our study, in order to overcome this
deficiency and accurately model the joint behaviour of equity
pairs.
The application of copulas in quantitative trading strategies
such as PTS is limited. Xie and Wu (2013), Wu (2013), and
Xie et al. (2014) attempt to address this limitation. Wu (2013)
points out that the main drawbacks of distance and cointegration PTS lie in the linearity restriction and symmetry that
correlation and cointegration enforce on the pairs dependence
structure. The application of copulas would be beneficial in
relaxing these restrictions. Thus, they propose a PTS that uses
copulas to measure the relative undervaluation or overvaluation of one stock against the other. Their study compares the
performance of a copula method to those of the DM and a
cointegration method, but using a limited analysis of 3 sameindustry pairs with a sample period of 36 months. Furthermore, pre-specified stock pairs with the same SIC code whose
prices are known to be related are used, and so the strategy
performance when it has the freedom to form pairs cannot be
evaluated. The results show that the copula approach yields
higher returns than the other two approaches. The copula approach also presents more trading opportunities than the distance and cointegration methods. Xie et al. (2014) employ a
similar methodology but use a broader data-set comprising of

the utility stocks (a total of 89 stocks) from 2003 to 2012.

Similarly, they show that the performance of the copula strategy is superior to that of the DM used in Gatev et al. (2006).
They also observe fewer trades with negative returns for the
copula strategy compared to the DM. Similar to the cointegration method, the main deficiency in these copula-based PTS
studies, is the limited empirical evidence to robustly measure
the performance of the strategies across a large number of
stocks over an extensive sample period.

3. Data
Our data-set consists of daily data of the stocks in CRSP from
1 July 1962 to 31 December 2014. The data-set sample period is
13216 days (630 months) and includes a total of 23 616 stocks.
In accordance with Do and Faff (2010, 2012), we restrict our
sample to ordinary shares, which are identified by share codes
10 and 11 in the CRSP database. In order to avoid relatively
high trading costs and complications, we have further restricted
our sample to liquid stocks. This is done by removing the
bottom decile stocks, in terms of market cap, in each formation
period. For the same reason, stocks with prices less than $1
in the formation period are also not considered. To increase
the robustness of our results and to replicate practical trading
environments as closely as possible, we use trading volume to
filter out stocks that have at least one day without trading in any
formation period in the respective trading period. In summary,
our data-set is consistent with that of Do and Faff (2012).

4. Research method
Pairs trading is a mean-reverting or contrarian investment strategy. It assumes a certain price relationship between two securities. Since pairs trading is a long-short strategy, modelling this
relationship would allow us to take advantage of any short-term
deviations by simultaneously buying the undervalued and selling short the overvalued security. Upon the restoration of the
price relationship, we would close, or reverse, the two opened
positions and realize the profit. We examine the performance
of three different PTS using CRSP database consisting of US
stocks from 1962 to 2014. For all strategies, we use a period
of 6 months, the trading period, to execute the strategy using
the parameters estimated in the previous 12 months, which we
call the formation period. We run the strategies each month,
without waiting six months for the current trading period to
complete. As a result, we have six overlapping portfolios,
with each portfolio associated with a trading period that has
started in a different month.
In the DM (section 4.1), potential security pairs are sorted
based on the SSD in their normalized prices during the formation period.After the pairs are formed, their spread is monitored
throughout the trading period and any deviations beyond a
certain threshold in that spread would trigger the opening of
two simultaneous long and short positions. We use this strategy
as our main benchmark to evaluate the cointegration and copula
based PTS.
By definition, cointegrated time series maintain a long-term
equilibrium and any deviation from this equilibrium is caused

Downloaded by [UQ Library] at 16:03 27 April 2016

The protability of pairs trading strategies

by white noise and will be corrected as the series evolve through
time (Vidyamurthy 2004). Using the statistical model of cointegration, we are able to incorporate the mean-reverting
attribute of this statistical property in PTS. To allow for computational efficiency, our work employs a combination of the
the cointegration framework outlined in Vidyamurthy (2004)
and the DM to develop a cointegration-based trading strategy.
After selecting nominated cointegrated pairs using the two-step
Engle-Granger method (Engle and Granger 1987), we extract
their stationary spread. Any deviation from this spread is by
definition temporary and thus can be used to open long and
short positions. We provide details of the cointegration model
in PTS in section 4.2.
Our copula strategy is specifically designed to be computationally efficient to allow it to be easily operationalized in
practice by traders. It combines aspects of the DM approach
and the copula technique. Stocks pairs are sorted and selected
by SSD, and we allow a range of marginal and copula distributions to be fitted to the resulting pair. Copulas are used (detailed
in section 4.3) to model the relationship between stocks of a pair
and to detect pairs deviations from their most probable relative
pricing (Xie et al. 2014). We define two mispriced indices,
which represent the relative under or overvaluation of stocks
of a pair, and use them as criteria to open long-short positions
when stocks move away from their relative fair prices.

4.1. The distance method

In DM, we calculate the spread between the normalized prices
of all possible combinations of stock pairs during the formation
period. The formation period is chosen to be 12 months. The
normalized price is defined as the cumulative return index,
adjusted for dividends and other corporate actions, and scaled
to $1 at the beginning of the formation period. We then select
20 of those combinations that have the least sum of squared
spreads, or SSD, to form the nominated pairs to trade in the
following trading period, that is chosen to be 6 months. The
standard deviation of the spread during the formation period
is also recorded and used as the trading criterion. A specific
stock can participate in forming more than one pair as long as
the other stock of the pair varies.
Our implementation of the DM is in accordance with Gatev
et al. (2006), Do and Faff (2010, 2012). At the beginning of
the trading period, prices are once again rescaled to $1 and the
spread is recalculated and monitored. When the spread diverges
by two or more historical standard deviation (calculated in
the formation period), we simultaneously open a long and a
short position in the pair depending on the direction of the
divergence. The two positions are closed (reversed) once the
spread converges to zero again. The pair is then monitored
for another potential divergence and therefore can complete
multiple round-trip trades during the trading period.
As the opening threshold is always set to two standard deviations, the divergence required to open positions are lower for
less volatile spreads. Therefore, such positions can converge
with a loss. Time-varying transaction costs (discussed in section 4.4) also contributes towards this effect. We analyse this
issue further in section 5.5, sensitivity analysis.
Month refers to the calender month.

4.2. The cointegration method

4.2.1. Framework. Anon-stationary time series X t is called
I (1) if its first difference forms a stationary process, i.e. I (0)
(Lin et al. 2006). Consider X 1,t and X 2,t to be two I (1) time
series. If exists a linear combination of the two time series that
is stationary, X 1,t and X 2,t are said to be cointegrated. Thus,
X 1,t and X 2,t are cointegrated if there exists a non-zero real
number such that:
X 2,t X 1,t = u t

(1)

where is the cointegration coefficient and u t is a stationary

series known as the cointegration errors. Using the Grangers
theorem (Engle and Granger 1987), the cointegration relationship can be equivalently shown in an Error Correction Model
framework (ECM) (Vidyamurthy 2004). Based on ECM, the
cointegrated series exhibits long-term equilibrium and, while
short-term deviations from this equilibrium can occur, they will
be corrected, through time, by the error term in the ECM. The
ECM representation of the cointegration relationship between
time series X 1,t and X 2,t is:
X 2,t X 2,t1 = X 2 (X 2,t1 X 1,t1 ) + X 2,t

(2)

X 1,t X 1,t1 = X 1 (X 2,t1 X 1,t1 ) + X 1,t

Equation (2) shows that the evolution of a time series, for
example X 2,t , consists of a white noise, X 2,t , and an error
correction term, X 2 (X 2,t1 X 1,t1 ), which reverts the time
series towards its long-term equilibrium as the series evolves
through time. This mean-reverting property of cointegrated
series can be used in implementing PTS.
Vidyamurthy (2004) provides a basic framework to apply
cointegration to pairs trading. We describe this framework
and extend his work to operationalize cointegration into an
executable pairs trading strategy and provide details in section
4.2.2. The error correction term from equation (2) can be split
into the rate of correction ( X 2 ) and the cointegration relation
((X 2,t1 X 1,t1 )). The cointegration relation shows the
deviation of the process from its long-term equilibrium. The
spread series is defined as the scaled difference in the price of
two stocks:
(3)
spreadt = X 2,t X 1,t
Assume that we buy one share of stock 2 and sell short share
of stock 1 at time t 1. X 1,t and X 2,t represent the price series
of stocks 1 and 2, respectively. The profit of this trade at time
t, is given by:
(X 2,t X 2,t1 ) (X 1,t X 1,t1 )

(4)

By rearranging the above equation we get:

(X 2,t X 1,t ) (X 2,t1 X 1,t1 ) = spreadt spreadt1
(5)
Thus, the profit of buying one share of stock 2 and selling
share of stock 1 for the period t is given by the change in the
spread for that period. From equation (1), the spread is stationary by definition and exhibits mean-reverting properties. We
can use this to construct a quantitative trading strategy that uses
deviations from the long-term equilibrium of a cointegrated
pair to open long and short positions. Positions are unwound
once the equilibrium is restored, as a consequence of being
stationary.

Downloaded by [UQ Library] at 16:03 27 April 2016

H. Rad et al.

4.2.2. Trading strategy. We use two criteria to implement

the pairs selection phase of the cointegration PTS. First, we
sort all possible combinations of pairs based on their SSD in
their normalized price during the formation period. Second,
we test each of the pairs with the least SSD for cointegration,
using their cumulative return series in the formation period.
Pairs that are not cointegrated are eliminated in the selection
process. Pairs that are cointegrated will have their cointegration
coefficient estimated. We continue until 20 cointegrated pairs
with minimum SSD are selected to be traded in the following
trading period.
We use the two-step Engle-Granger approach (Engle and
Granger 1987) to test for the existence of cointegration between
nominated pairs and to estimate cointegration coefficient .
In this procedure, the cointegration regression is estimated
using OLS in the first step and the error correction model
(ECM) is estimated in the second step. For each nominated
pair, we then form the spread defined in equation (3) and
calculate the spreads mean e and standard deviation e , with
the data of the formation period. These parameters are used
in the trading period as trades open and close triggers. From
equations (1) and (3) the spread is given by: spreadt = et
where et I (0), with mean e and standard deviation e .
The normalized cointegration spread as:
spread e
spreadnormalized =
(6)
e
Similar to the DM, we simultaneously open and close long and
short positions when the normalized spread diverges beyond 2.
However, the values of long and short positions vary from those
in the DM. By construction, if the spread drops below 2, we
buy 1 dollar worth of stock 2 and sell short dollar worth of
stock 1. Equivalently, we sell short 1/ dollar worth of stock 2
and buy 1 dollar worth of stock 1, when the spread moves
above the +2 threshold. We close both positions once the
spread returns to zero, which translates into the pair returning
to their long-term equilibrium. The pair is again monitored
for other potential round-trip trades for the remainder of the
trading period.

F(x1 , x2 , . . . , xn ) = C(F1 (x1 ), F2 (x2 ), . . . , Fn (xn ))

(8)

If Fi and C are differentiable 1 and n times, respectively, we

can write the joint probability density function (pdf) f as the
product of marginal density functions f i (xi ) and the copula
density function c:
f (x1 , x2 , . . . , xn ) = f 1 (x1 ) f 2 (x2 ) f n (xn )
c(F1 (x1 ), F2 (x2 ), . . . , Fn (xn )), (9)
where the copula density function c is given by differentiating
the copula function, C, n times with respect to each marginal:
n C(u 1 , u 2 , . . . , u n )
(10)
u 1 u 2 u n
Equation (9) allows us to decompose a multivariate distribution into two components, the individual marginal probability
density functions, and the copula density function. Consequently, since all the characteristics of marginal distributions
are captured in their pdfs and all the characteristics of the joint
distribution are represented by the joint pdf, the copula density
function should contain all the dependence characteristics of
the marginal distributions.
Therefore, copulas allow for higher flexibility in modelling
multivariate distributions. They allow the marginal distributions to be modelled independently from each other, and no
assumption on the joint behaviour of the marginals is required.
Moreover, the choice of copula is also not dependent on the
marginal distributions. Thus, using copulas, the linearity restriction that applies to the dependence structure of multivariate
random variables in a traditional dependence setting is relaxed.
Thus, depending on the chosen copulas, different dependence
structures can be modelled to allow for any asymmetries.
Now, let X 1 and X 2 be two random variables with probability
functions F1 (x1 ) and F2 (x2 ) and joint bivariate distribution
function F(X 1 , X 2 ). We have U1 = F1 (X 1 ) and U2 = F1 (X 2 )
where U1 , U2 U ni f or m(0, 1) and their copula function
C(u 1 , u 2 ) = P(U1 u 1 , U2 u 2 ). By definition, the partial
derivative of the copula function gives the conditional distribution function (Aas et al. 2009):
c(u 1 , u 2 , . . . , u n ) =

C(u 1 , u 2 )
(11)
u 2
C(u 1 , u 2 )
h 2 (u 2 |u 1 ) = P(U2 u 2 |U1 = u 1 ) =
u 1
Using functions h 1 and h 2 , we can estimate the probability
of outcomes where one random variable is less than a certain
value, given the other random variable has a specific value. The
application of these functions in a PTS is that we can estimate
the probability of one stock of the pair moving higher or lower
than its current price given the price of the other stock.
h 1 (u 1 |u 2 ) = P(U1 u 1 |U2 = u 2 ) =

4.3. Copula method

4.3.1. Framework. A copula is a function that links
marginal distribution functions to their joint distribution function. It captures the dependence structure between the marginal
distributions. A copula function is defined as joint multivariate
distribution function with uniform univariate marginal distributions:
C(u 1 , u 2 , . . . , u n ) = P(U1 u 1 , U2 u2, , Un u n ),
(7)
where u i [0, 1], i = 1, 2, . . . , n. Now, suppose X 1 , X 2 , . . . ,
X n are n random variables with continuous distribution functions F1 (x1 ), F2 (x2 ), . . . , Fn (xn ). Since a random variable
with arbitrary distribution can be transformed to a uniform
random variable by feeding it into its distribution function, i.e.
Ui = F(X i ) where Ui U ni f or m(0, 1), we can define the
copula function of random variables X 1 , X 2 , . . . , X n as:
This is set to 12 months to allow for consistency with the DM.
This is set to 6 months to allow for consistency with the DM.

4.3.2. Trading strategy. Similar to the DM, we sort all possible pairs based on SSD in their normalized price during
the formation period and nominate 20 pairs with the least
SSD to trade in the trading period. We continue to fit nominated pairs to copulas using the Inference for Margins method
The formation and trading periods are kept at 12 and 6 months,
respectively, to be consistent with the distance and cointegration
methods.

Downloaded by [UQ Library] at 16:03 27 April 2016

The protability of pairs trading strategies

(Joe 1997) that is a two-step process. First, for each pair,
we fit the daily returns of the formation period to marginal
distributions and find the two distributions that best fit the each
stock. Marginal distributions are selected from extreme value,
generalized extreme value, logistic and normal distribution
and independently fitted for each stock of the pair. In the
second step, with the estimated marginal models parameters
from the previous step, we nominate the copula that best fits
the uniform marginals and parameterize the copula. We allow
for a range of copulas to be employed in this step, namely
the Clayton, Rotated Clayton, Gumbel, Rotated Gumbel and
Student-t. The best fitting copula is the copula that provides
a parsimonious fit for the dependence structure between the
stocks. In quantifiable terms, the best copula is chosen by
maximizing the log likelihood of each copula density function
and calculating the corresponding AIC and BIC. The copula
associated with the highest AIC and BIC is then selected as
having a parsimonious fit.
Table 1 reports the percentage frequency of copula and
marginal models selected for the stock pairs during our empirical investigation. The Student-t is selected for 62% of all
stock pairs and the Rotated Gumbel for 10.38%. The Logistic
model is selected for 86% of the marginal models for each
individual stock.
Each day during the trading period, using the daily realizations of random variables U1 and U2 , that represent the daily
returns of two stocks of a pair, we calculate the conditional
probabilities, h 1 and h 2 functions defined in equation (11), for
each nominated pair (conditional probability functions given
in table 2). A value of 0.5 for h 1 is interpreted as 50% chance
for the random variable U1 , which is the price of stock 1, to be
below its current realization, which is its todays price, given
the current price of stock 2. The same interpretation is valid
for h 2 , which demonstrates the same conditional probability for
stock 2. Accordingly, conditional probability values above 0.5
show that chances for the stock price to fall below its current
realization is higher than they are for it to rise, while values
below 0.5 predict an increase in the stock price compared to
its current value is more probable than a decrease. Similar to
Xie et al. (2014), we define two mispriced indices:
m 1,t = h 1 (u 1 |u 2 ) 0.5 = P(U1 u 1 |U2 = u 2 ) 0.5 (12)
m 2,t = h 2 (u 2 |u 1 ) 0.5 = P(U2 u 2 |U1 = u 1 ) 0.5
The cumulative mispriced indicies M1 and M2 , which are set
to zero at the beginning of the trading period are calculated
each day:
M1,t = M1,t1 + m 1,t

(13)

M2,t = M2,t1 + m 2,t

Positive (negative) M1 and negative (positive) M2 are interpreted as stock 1 (stock 2) being overvalued relative to stock 2
(stock 1). We have arbitrarily set the strategy to open a long
short position once one of the cumulative mispriced indices is
above 0.5 and the other one is below -0.5 at the same time. The
positions are then unwound when both cumulative mispriced
indices return to zero. The pair is then monitored for other
possible trades throughout the remainder of the trading period.

4.4. Dynamic transaction costs

Transaction costs play a vital role in the profitability of PTS.
Each execution of a complete pairs trade consists of two
roundtrip trades. In addition to that, an implicit market impact
and short selling costs are also applicable. Since, the sum of
these costs can be large they can degrade the profitability of
PTS when taken into account. We use a time-varying dataset of transaction costs in line with Do and Faff (2012). The
motivation behind this is that commissions are the first element
of transaction costs to be considered. As commissions have
changed considerably over the last 50 years that we are using
as our time span, a flat commission system distorts the accuracy
of our study. As such, we use the institutional commissions that
Do and Faff (2012) calculated that starts from 70 bps in 1962
and gradually declines to 9 bps for recent years. Similar to
their study, we divided our time period into two sub-periods
and use a different market impact estimate for each sub-period:
30 bps for 19621988 and 20 bps for 1989 onward. As we
screen out stocks that have low dollar value and low market
capitalization, we assume the remaining stocks in our sample
are relatively cheap to short sell and therefore do not take into
account short selling costs explicitly. It is worth noting that we
double these costs to cover each complete pairs trade which
consists of two round-trip trades.

4.5. Performance calculation

The performance of the three PTS are recorded and compared
based on various performance measures including returns. In
accordance with Gatev et al. (2006) and Do and Faff (2010),
two types of returns are calculated: return on employed capital
(equation 14) and return on committed capital (equation 15).
Return on employed capital for month m, RmEC , is calculated
as the sum of marked-to-market returns on that months traded
pairs divided by the number of pairs that have traded during
CC ,
that month. Return on committed capital for month m, Rm
is calculated as the sum of marked-to-market returns on traded
pairs divided by the number of pairs that were nominated to
trade in that month (20 in our case), regardless of whether they
actually traded or not. In comparison to the return on employed
capital, return on committed capital is a more conservative
measure that mimics what a hedge fund might use to report
returns, as it takes into account the opportunity cost of the
capital that has been allocated for trading.
n
ri
(14)
RmEC = i=1
n
n
ri
CC
Rm
(15)
= i=1 , N P = 20
NP
We execute the PTS each month and do not wait for a trading
period to be complete, resulting in six overlapping portfolios each month. The monthly excess return of a strategy is
calculated as the equally weighted average return on these
six portfolios. As the trades neither necessarily open at the
beginning of the trading period nor close exactly at the end
of the trading period, the full capital is not always locked in a
See Do and Faff (2012) section 3 for full details on commissions
and market impact estimations.

H. Rad et al.
Table 1. Proportion of selected copulas and marginal distributions.

Panel A: Copulas
Copulas
% of Pairs

Clayton

Rotated Clayton

Gumbel

Rotated Gumbel

Student-t

6.59

11.54

9.85

10.38

61.64

Generalized extreme value

Normal

Logistic

2.61

11.10

86.14

Panel B: Marginal Distributions

Marginal distributions:
Extreme value
% of Stocks

0.15

Notes: This table shows the proportion of copula models selected for each stock pair as a parsimonious fit (Panel A), along with the proportion of marginal
models selected for the stocks as a parsimonious fit (Panel B).

Table 2. Copula conditional probability functions.

P(U1 u 1 |U2 = u 2 )

Copula

h(u 1 , u 2 ; , ) = t+1  x1 x2

Downloaded by [UQ Library] at 16:03 27 April 2016

Student-t

Clayton
Rotated Clayton

(+x22 )(1 2 )
+1

(1, 1)

xi = t1 (u i ), u i (0, 1)i = 1, 2
t : Students t-distribution cumulative distribution function with degrees of freedom.

 1 1

(+1)

u
h(u 1 , u 2 ; ) = u 2
1 + u2 1


 1 1

(+1)

(1 u 1 ) + (1 u 2 ) 1
h(u 1 , u 2 ; ) = 1 (1 u 2 )

Gumbel

Rotated Gumbel

1
h(u 1 , u 2 ; ) = C (u 1 , u 2 ) [( ln u 1 ) + ( ln u 2 ) ] ( ln u 2 )1 u12


1
C (u 1 , u 2 ) = ex p [( ln u 1 ) + ( ln u 2 ) ]

h(u 1 , u 2 ; ) = 1 C (1 u 1 , 1 u 2 ) [( ln(1 u 1 )) + ( ln(1 u 2 )) ]

1
( ln(1 u 2 ))1 1u
2


1
C (u 1 , u 2 ) = exp [( ln u 1 ) + ( ln u 2 ) ]

>0
>0
>0
>0

>0

Note: This table shows the conditional probability functions of copulas used in the copula method.

trade. In addition, there are months where no trading occurs.

As interest is not accrued to the capital when it is not involved
in a trade, the performance outcomes are underestimated.
Positions that are opened in the distance and copula methods
are $1 longshort positions. $1 longshort positions are defined
as opening a long positions worth $1 and a short positions worth
$1 simultaneously. Since the money raised from shorting a
stock can be used to buy the other stock, these positions are
self-financing and do not require any capital to trade. However, for the sake of calculating returns, we adapt the widely
used concept of using $1 as the total value of each longshort
position. In the cointegration method, by definition, long and
short positions are not valued equally. However, since we have
designed the method to ensure a $1 long position for every
trade, we assume an average $1 value for each longshort
position.
5. Results
5.1. Descriptive statistics
Table 3 reports the monthly excess return distribution for each
of the three strategies from 1962 to 2014 both before and after
transaction costs in two sections. Section 1 of the table shows
the return on employed capital, while section 2 reports return on

committed capital (see section 4.5 for details on calculations).

As both return measures achieve similar results and rankings
for the strategies, we use return on employed capital, hereafter
simply referred to as return, to report results for the remainder
of this paper, unless stated otherwise. The average monthly
excess return of the DM before transaction costs is 0.91%.
Results presented in section 1 of table 3 show that, while
the DM and cointegration method both show statistically and
economically significant and very similar average monthly
excess returns (before and after transaction costs), the copula
methods after-cost excess return is relatively small at 5 bps.
However, before transaction costs, the copula method is producing a significant 43 bps average excess return. Moreover,
all three strategies show small standard deviations, with the
lowest belonging to the copula method with 0.0067 after costs
and the highest to DM the highest with 0.0110 before costs.
The DM and the cointegration method exhibit similar Sharpe
ratios. While neither of the strategies show normally distributed
returns, the cointegration method is the only strategy whose
returns are positively skewed after transaction costs. The return on committed capital measure, presented in section 2 of
table 3, produces very similar results.
Do and Faff (2010) report a similar monthly excess return of 0.90%
before transaction costs.

The protability of pairs trading strategies

Table 3. Pairs trading strategies monthly excess return.

VaR

CVaR

JB test

Skewness

Kurtosis

(95%)

(95%)

p.value

0.3498
0.3497
0.0749

0.3491
0.3565
0.5127

13.2586
8.5718
6.9598

0.0106
0.0121
0.0107

0.0190
0.0173
0.0150

0
0
0

0.0122
0.0111
0.0071

0.7517
0.7703
0.6032

0.2231
0.8130
0.2073

9.0419
7.3561
5.7007

0.0075
0.0068
0.0066

0.0150
0.0126
0.0111

0
0
0

Section 2: Return on committed capital

Panel A: After transaction costs
Distance
0.0032
5.4941
Cointegration
0.0030
5.1963
Copula
0.0005
2.1477

0.0089
0.0085
0.0053

0.3566
0.3483
0.1008

0.3723
0.6462
0.4790

24.1712
12.3600
6.8850

0.0074
0.0091
0.0082

0.0135
0.0144
0.0115

0
0
0

Panel B: Before transaction costs

Distance
0.0068
7.0284
Cointegration
0.0068
7.0365
Copula
0.0030
7.1436

0.0102
0.0097
0.0055

0.6729
0.6969
0.5410

0.8152
1.3934
0.3170

14.6456
10.8547
6.3584

0.0050
0.0058
0.0056

0.0104
0.0105
0.0092

0
0
0

Downloaded by [UQ Library] at 16:03 27 April 2016

Strategy

Mean

t-stat

Std. Dev.

Sharpe ratio

Section 1: Return on employed capital

Panel A: After transaction costs
Distance
0.0038
5.3937
Cointegration
0.0033
5.1444
Copula
0.0005
1.5797

0.0110
0.0099
0.0067

Panel B: Before transaction costs

Distance
0.0091
7.4260
Cointegration
0.0085
7.3268
Copula
0.0043
7.1598

Notes: This table reports key distribution statistics for the monthly excess return time series of three different PTS - i.e. the distance, cointegration, and copula
methods - for July 1962 to December 2014. The formation and trading period for all strategies are set to 12 and 6 months, respectively. The column labeled
JB Test tests the null hypothesis of normality of the series using the Jarque-Bera Test.
significant at 1% level.
significant at 5% level.
significant at 10% level.

To further demonstrate the relative performance of the strategies, figure 1 compares the cumulative excess return for each
of the three strategies from 1963 to 2014. It can be seen that
the DM and cointegration methods performance are almost
identical to each other with the cointegration method slightly
underperforming compared to the DM. In contrast, the copula
method performs poorly in terms of cumulative excess return.
However, the gap between the copula method and the other
two strategies is narrower on a risk-adjusted basis, as shown
in figure 2.
The five-year rolling sample Sharpe ratio (figure 2) also
confirms the nearly identical performance of the DM and the
cointegration method. In addition, it shows the risk-adjusted
performance of all three strategies fluctuate greatly, but generally maintain an upward trend until around 1985, where they
experience their peak and the trend reverses. More importantly,
the reversal in this trend occurs in the copula, cointegration
and the DM, thus none of the strategies avoid this decline.
However, after the downward trend begins, the gap between
the risk-adjusted performance of the copula method and the
other two strategies becomes smaller than before. It appears
that as we move closer to recent years, the three strategies
show a very close risk-adjusted performance and there is no
clear winner among them.

performance measures. Table 4 reports various risk-adjusted

metrics for each of the strategies, before and after transaction
costs. The measures are divided into two main groups: lower
partial moment measures and drawdown measures. Lower
partial moment measures take into account only the negative deviations of returns from a specified minimum threshold
value. They appropriately account for downside risk compared to the Sharpe ratio that considers both positive and negative deviations equally. Omega is defined as the ratio of returns
above a threshold, to returns below that threshold. The Sortino
ratio is the ratio of average excess return, to the absolute value
of second lower partial moment (or negative standard deviation). Kappa 3, is the ratio of average excess return to the
third lower partial moment (or negative skewness). Drawdown metrics measure the magnitude of losses of a portfolio
over a period of time. Maximum drawdown is defined as the
maximum possible loss that could have occurred during a time
period. The Calmar ratio is defined as the ratio of average
excess return to the maximum drawdown. Sterling ratio is the
ratio of average excess return to the average of n most significant continuous drawdowns, thus reducing the sensitivity of
the measure to outliers. The continuous drawdown is defined
as the maximum incurred loss that is not interrupted by positive
returns. Finally, the Burke ratio is the ratio of average excess
return to the square root of sum of the squared n most significant
drawdowns.

5.2. Risk-adjusted performance

As the return series of neither of the strategies are normally
distributed, the Sharpe ratio, as the classic risk-adjusted measure, has the potential to underestimate risk thereby, overestimating the risk-adjusted performance (Eling 2008). Thus, we
further analyse the risk profile of the three PTS, using downside

The threshold value is also known as the minimum acceptable return

where we use 0%.
For detailed explanation and calculation of lower partial moment
measures, refer to Eling and Schuhmacher (2007).
For detailed explanation and calculation of drawdown measures,
refer to Schuhmacher and Eling (2011).

10

H. Rad et al.
Distance
Cointegration
Copula

Cumulative Excess Return ($)

11

1
1965

1970

1975

1980

1985

1990

1995

2000

2005

2010

2015

Year

Figure 1. Cumulative excess return.

Notes: This figure shows the evolution of wealth based upon an investment of $1 in each strategy. The return on employed capital after
transaction costs is applied for the calculation of cumulative excess returns.

5-year rolling window Sharpe ratio

Distance
Cointegration
Copula

1
0.8

Sharpe Ratio

Downloaded by [UQ Library] at 16:03 27 April 2016

1.2

0.6
0.4
0.2
0
-0.2
-0.4
1965

1970

1975

1980

1985

1990

1995

2000

2005

2010

2015

Year

Figure 2. Five-year rolling sample Sharpe ratio.

Note: This figure shows the 5-year (60-month) rolling sample Sharpe ratio for the three pairs trading strategies.
Table 4. Pairs trading strategies risk-adjusted performance.
Lower partial moments measures
Omega

Drawdown measures

Sortino ratio

Kappa 3

Max drawdown

Calmar ratio

Sterling ratio

Burke ratio

0.6419
0.6774
0.1083

0.3350
0.4093
0.0705

0.1052
0.1718
0.1845

0.0366
0.0201
0.0027

0.3434
0.3499
0.0530

0.0195
0.0180
0.0028

Panel B: Before transaction costs

Distance
9.2018
1.9965
Cointegration
9.4562
2.4345
Copula
5.0044
1.3443

0.9520
1.3344
0.7606

0.0737
0.0459
0.0418

0.1241
0.1858
0.1023

1.0415
1.0335
0.6229

0.0687
0.0826
0.0430

Panel A: After transaction costs

Distance
2.7733
Cointegration
2.6825
Copula
1.2257

Notes: This table reports key risk-adjusted performance measures for the monthly excess return time series of three different PTSi.e. the distance, cointegration,
and copula methodsfor July 1962 to December 2014. The formation and trading period for all strategies are set to 12 and 6 months, respectively.

The cointegration method exhibits the best before-cost riskadjusted performance with the best figures for all but the maximum drawdown and Sterling ratio measures. However, its
after-cost performance is similar to that of the DM. The
copula method is the poorest PTS among the three, except for
showing the least before-cost maximum drawdown. The major
contributor to the low performance of the copula method is the
insignificance of its mean returns. This can be verified by the

fact that the strategy transforms from having the best maximum drawdown to the worst when transaction costs are taken
into account. At a lower magnitude, the other two strategies
also suffer considerably when transaction costs are taken into
account. We can see the after-costs Omega ratio decreases by
70% or more for all strategies. The same decrease is observed
in the Sortino ratio of DM and the cointegration method, while
for the copula method this figure rises above 90%.

The protability of pairs trading strategies

Downloaded by [UQ Library] at 16:03 27 April 2016

5.3. Properties of pair trades

Studying the properties of converged and unconverged pairs
allows us to further investigate the sources of each PTS profitability.
Figure 3 illustrates the trade distributions after transaction
costs for each strategy. All the strategies have fatter left tails
than the right which makes extreme negative returns more
likely than extreme positives. This can be attributed to the
fact that for all strategies, we use some criteria to close a trade
once it has converged, however unconverged trades remain
open for the duration of the trading period. Albeit infrequently,
higher profits do occur when a pairs spread suddenly diverges
overnight by much more than just the triggering amount. A
similar scenario can occur upon the convergence of a pair.
Contrary to profit-making trades, trades that do not converge
can accumulate big losses before being forced to close by the
strategy at the end of their trading period, which results in fat
left tails.
Specifically, for the DM and cointegration methods, where
the criteria for opening and closing positions are directly related to the prices of stocks, the magnitude of profit for each
trade is bound to the scale of divergence that is used in the
opening trade criteria. Therefore, the profits per each trade is
generally limited thus, we do not observe fat right tails. In the
copula method, the opening and closing criteria are based on
the probability of relative mispricing within pairs, rather than
being directly related to the spread. Therefore, the observed
fatter right tail in its distribution is expected. In practice, an
additional risk-limiting criteria for closing a trade, i.e. a stoploss measure, can potentially limit the extreme losses. In that
case, strategies such as the copula method have the potential
to perform well as their profits are not bound to some specific
amount.
Table 5 reports further statistics on the converged and unconverged trades. As expected, all strategies show positive average
return for their converged and negative average return for their
unconverged trades. For converged trades, the cointegration
method has the highest average return (4.37%), but the secondhighest Sharpe ratio (1.62) after the DMs (1.79). The DM has
the highest percentage of converged trades at 62.53%, while
the cointegration method is second with 61.35%. It is in fact
this higher percentage of converged trades that puts the DM
first among the three strategies in terms of overall monthly
return, as shown in section 5.1.
Interestingly, for converged trades, the copula method also
shows a significant average return (Sharpe ratio) of 3.95%
(1.05), that are both comparable to the other two methods. The
copula method exhibits the highest mean return and Sharpe
ratio for unconverged trades among the three strategies, which
demonstrates its lower risk profile. However, the considerably
high proportion of unconverged trades for the copula method
degrades its performance to only 0.3% average return for all
trades compared to 1.2 and 1% for the DM and cointegration methods, respectively. In fact, the copula method is the
only strategy that has less converged trades than unconverged.
Therefore, the performance of the copula method can be vastly
improved if the methodology can be enhanced to reduce the
frequency of unconverged trades.

11

The DM has the lowest average number of days that takes

for its converging trades to converge (21 days), followed by
the cointegration (23 days) and the copula method (26 days).
Intuitively, the DM is less exposed to risks as the trades are
open for shorter periods. More than 98% of the converged
trades in the DM and cointegration methods generate positive
return that is slightly higher than the copula method at 94%.
Although the two-step selection method that we have used
leads to similarities in the distribution of the returns of the
three strategies, there are some notable differences in their
behaviour. In table 5, we report the difference in the number
of distinct pairs for each strategy to highlight this point. On a
monthly basis, the cointegration method leads to the highest
number of distinct pairs at 4421, while the DM and the copula
method have a lower number of distinct pairs (approximately
4100 pairs). The different sum of distinct pairs per month for
each strategy indicates that even if the same pairs are selected
in different strategies, the opening and closing characteristic of
each strategy leads to those pairs exhibiting different outcomes
for each strategy in terms of the month in which the trades
are initiated. Analysing the entire sample, the cointegration
method has the largest number of distinct pairs totaling 17307
pairs, followed by the DM and the copula method at 15026
and 13450 pairs, respectively. In other words, based on this
measure, the copula (distance) method involves more than
20% (10%) fewer distinct pairs compared to the cointegration
strategy.
When considering the converged and unconverged trades,
as shown in figure 4, the DM strategy generates the most
profitable trades (71% of all trades), followed by the cointegration method (69%). However, since the average returns of
the cointegration method are smaller in magnitude, the effect
of this high percentage is reduced, thus resulting in a slightly
lower monthly return compared to the DM as demonstrated in
section 5.1.

5.4. Risk prole

We further investigate the source of pairs trading profitability
to determine if performance is compensation for risk. Thus,
we regress the strategies monthly excess return series against
several risk factors. First, we examine the strategies against the
Fama and French (1993) three factors plus the momentum and a
liquidity factors. We added the liquidity factor since it is argued
that liquidity shock is a source of pairs trading profitability
(Engelberg et al. 2009). Similar to Do and Faff (2012), we
use the Innovations in Aggregate Liquidity series (Pstor and
Stambaugh 2003).
Results from Panels A and B of table 6 show that, the profits
of PTS are not fully explained by the risk factors. In fact, large
and significant alphas are observed for all the strategies. It
is worth mentioning that alphas are very close to their corresponding strategys mean of monthly excess return, implying that after being adjusted for these risk-factors, the PTS
profits remain unaffected. Moreover, the momentum factor is
negatively correlated with strategies profits, with significant
t-statistics both before and after transaction costs for all but
the copula methods. Similarly, liquidity has also a considerable

12

H. Rad et al.

4000

7000

3500

6000

1800
1600

2500
2000
1500

5000

Frequency

Frequency

3000

Frequency

2000

4000
3000

0
-0.8

-0.6

-0.4

-0.2

0.2

0.4

0.6

0
-1.5

1000
800
400

1000

500

1200

600

2000

1000

1400

200
-1

-0.5

Trade Returns

0.5

0
-0.8

1.5

-0.6

-0.4

-0.2

0.2

0.4

0.6

Trade Returns

Trade Returns

Figure 3. Distribution of trade returns.

Notes: This figure shows the distribution of trade returns after transaction costs for each of the three strategies.

Downloaded by [UQ Library] at 16:03 27 April 2016

Table 5. Converged and unconverged trades return series.

# Distinct

Sum of

Trade

% of

Strategy

pairs

D.P.P.M.

type

trades

Days open

Distance

4061

15087

C
U

62.53
0.0426
37.47 0.0399

Cointegration

4421

17348

C
U

Copula

4100

13463

C
U

Mean

St.D.

S.R.

Positive

Skewness

Mean

Median

trades (%)

0.0238
1.7889
0.0702 0.5691

2.1702
2.3428

21.15

14

98.42

61.35
0.0437
38.65 0.0436

0.0270
1.6173
0.0754 0.5782

5.8236
2.7620

22.65

15

98.64

39.98
0.0395
60.02 0.0215

0.0377
1.0485
0.0752 0.2855

1.7030
0.9488

26.30

17

94.41

Notes: This table reports key distribution statistics for converged and unconverged trade return series after transaction costs. The Trade Type column classifies
trades into converged (C) and unconverged (U ). # Distinct Pairs and Sum of D.P.P.M columns report the number of distinct pairs and sum of distinct
pairs per month for each trading strategy throughout the study period, respectively. The St.D and S.R columns report standard deviation and Sharpe ratio,
respectively. Two columns under the Days Open title show the mean and median number of days that a converged trade remains open. Positive Trades (%)
shows the percentage of converged trades with positive returns. Note that all calculations are based on after-transaction cost returns.

Positive return
Negative return

Positive return
Negative return
29%

Positive return
Negative return
31%
41%

59%
71%

69%

Figure 4. Proportion of positive and negative trades.

Note: This figure shows the proportion of trades with positive and negative returns after transaction costs for each of the three strategies.

negative effect on profits, but only for the DM and the cointegration method. Interestingly, liquidity is unable to explain the
profits of the copula method neither before nor after costs, as
opposed to postulations of Engelberg et al. (2009) who suggest
otherwise. This further demonstrates the unique nature of the
copula method in comparison to the other PTS. Surprisingly,
no other factor, including the market excess return in the copula method, is correlated to the strategies returns, which is
further evidence for the market-neutrality of these strategies.
From an investors perspective, this market neutrality can have
diversifying effects on investment portfolios, by reducing risk,

particularly those that have strong correlations with market

returns.
Next, we regress the monthly excess returns of the three
strategies against the recent Fama and French (2015) 5 factor
model. This model is an attempt by the authors to improve
their well-known three factor model by introducing two additional factors: profitability and investment factors. Robust
Minus Weak (RMW), a measure of profitability, is defined as
the difference between returns on robust and weak profitability portfolios, while Conservative minus Aggressive (CMA),
which represents the investment factor, is defined as the differ-

The protability of pairs trading strategies

3.5

22

20

2.5

18

16

1.5
2

25

14
1

30

2.5

25

20

20

35
Average trades per pair per 6 month period
Average days open per converged trade

Days

24

Trades

26

3.5

30
Average trades per pair per 6 month period
Average days open per converged trade

Days

Trades

4.5

Days

28
Average trades per pair per 6 month period
Average days open per converged trade

Trades

13

15
1

Opening threshold

1.5
0.2

Opening threshold

0.3

0.4

0.5

0.6

0.7

15
0.8

Opening threshold

Figure 5. Threshold sensitivity analysis.

Note: This figure shows the average number of trades per pair per six month period and the average convergence time for each strategy. The
solid lines represent Trades and the staggered lines represent Days.
2
Opening threshold = 0.8
Opening threshold = 0.5
Opening threshold = 0.2

Cumulative Excess Return ($)

1.6

1.4

1.2

0.8
1965

1970

1975

1980

1985

1990

1995

2000

2005

2010

2015

Year

Figure 6. Copula methods cumulative excess return.

Notes: This figure shows the evolution of wealth based upon an investment of $1 for the copula method using three different opening thresholds.
Return on employed capital after transaction costs is applied to calculate the cumulative excess return.

0.5

Distance
Cointegration
Copula

Mean/CVaR(95%)

0.8

0.4

0.7

Sortino Ratio

0.35
0.3
0.25
0.2
0.15

0.6
0.5
0.4
0.3

0.1

0.2

0.05

0.1

0.3

0.9
Distance
Cointegration
Copula

0.45

Sharpe Ratio

Downloaded by [UQ Library] at 16:03 27 April 2016

1.8

Crisis

Normal

Distance
Cointegration
Copula

0.25
0.2
0.15
0.1
0.05

Crisis

Normal

Crisis

Normal

Figure 7. Average monthly performance in crisis and normal periods.

Notes: This figure shows the performance of each strategy during Crisis and Normal periods. Crisis is defined as the lowest quintile of the
US S&P 500 stock market index returns on our entire sample (11 worst-performing years out of 53) and Normal consists of the remaining
annual sub-periods.

ence between the returns of conservative and aggressive portfolios. Results, reported in Panels C and D of table 6, show that
there is a negative correlation between before-cost pairs profit
and RMW for all three strategies. This relationship is statistically significant at 1, 5 and 10% for the DM, cointegration and

copula methods, respectively, albeit the size of the regression

coefficients pointing out the smallness of these correlations
economic significance. Nonetheless, RMW only affects the
after-cost profitability of the DM, which is again economically
insignificant. Notably, the size factor is related to some PTS

14

H. Rad et al.
Table 6. Monthly return risk profile.

Downloaded by [UQ Library] at 16:03 27 April 2016

Strategy

Alpha

MKT

SMB

HML

MO

LIQ

RMW

CMA

Fama French 3 Factors + Momentum + Liquidity

Panel A: After transaction costs
Distance
0.0042
0.0113
0.0174
(6.9770)
(0.3229)
(1.1739)
Cointegration
0.0038
0.0042
0.0124
(0.2326)
(0.8623)
(6.5738)
Copula
0.0005
0.0183
0.0031
(1.5958)
(2.2306)
(0.3530)

0.0095
(0.5373)
0.0093
(0.6125)
0.0026
(0.2449)

0.0341
(2.9074)
0.0311
(2.5911)
0.0133
(1.8706)

0.0395
(3.2232)
0.0391
(3.7265)
0.0053
(0.9097)

Panel B: Before transaction costs

Distance
0.0095
0.0091
(0.5870)
(12.8246)
Cointegration
0.0089
0.0013
(0.0977)
(12.3023)
Copula
0.0043
0.0156
(1.9104)
(10.9267)

0.0106
(0.6342)
0.0042
(0.2520)
0.0027
(0.2855)

0.0041
(0.2036)
0.0033
(0.1874)
0.0054
(0.5180)

0.0344
(2.3164)
0.0316
(2.6092)
0.0105
(1.4562)

0.0480
(3.6379)
0.0471
(4.1281)
0.0076
(1.1992)

Fama French 5 Factors

Panel C: After transaction costs
Distance
0.0042
0.0000
(0.1574)
(6.4904)
Cointegration
0.0037
0.0001
(0.3598)
(5.8411)
Copula
0.0005
0.0002
(1.3806)
(2.3186)

0.0004
(2.2536)
0.0003
(1.6281)
0.0001
(1.1523)

0.0001
(0.4173)
0.0000
(0.2170)
0.0000
(0.0244)

0.0006
(2.1017)
0.0003
(1.2827)
0.0003
(1.6446)

0.0002
(0.3922)
0.0001
(0.3341)
0.0002
(0.6542)

Panel D: Before transaction costs

Distance
0.0097
0.0001
(12.0386) (0.5037)
Cointegration
0.0089
0.0001
(11.3104) (0.7528)
Copula
0.0044
0.0001
(1.6047)
(10.7402)

0.0004
(1.9671)
0.0002
(1.2680)
0.0001
(0.7304)

0.0002
(0.5878)
0.0000
(0.1123)
0.0001
(0.4058)

0.0008
(2.4880)
0.0005
(1.7344)
0.0004
(2.0318)

0.0002
(0.4487)
0.0001
(0.1707)
0.0001
(0.2532)

Fama French 5 Factors + Momentum + Liquidity

Panel E: After transaction costs
Distance
0.0043
0.0085
0.0298
(0.5365)
(1.7779)
(7.0850)
Cointegration
0.0038
0.0054
0.0174
(0.3983)
(1.0756)
(6.4530)
Copula
0.0005
0.0188
0.0097
(0.9288)
(1.6229)
(2.2878)

0.0086
(0.3225)
0.0214
(0.9314)
0.0073
(0.5435)

0.0320
(2.3645)
0.0312
(2.8805)
0.0129
(1.8637)

0.0379
(3.1539)
0.0383
(3.6670)
0.0044
(0.7576)

0.0005
(1.6848)
0.0002
(0.6737)
0.0002
(1.3997)

0.0000
(-0.0771)
0.0003
(0.6717)
0.0002
(0.8898)

Panel F: Before transaction costs

Distance
0.0098
0.0050
(0.2938)
(12.8650)
Cointegration
0.0090
0.0012
(0.0830)
(12.1689)
Copula
0.0044
0.0148
(1.7631)
(10.8094)

0.0010
(0.0341)
0.0129
(0.4914)
0.0003
(0.0227)

0.0314
(2.0577)
0.0308
(2.5201)
0.0095
(1.3372)

0.0458
(3.5961)
0.0458
(4.1141)
0.0064
(1.0278)

0.0006
(2.1341)
0.0003
(1.1944)
0.0003
(1.8470)

0.0001
(0.1778)
0.0002
(0.4646)
0.0001
(0.4151)

0.0277
(1.4603)
0.0135
(0.7529)
0.0062
(0.5395)

Notes: This table shows results of regressing monthly return series (after and before transaction costs) against Fama-French 3 factors plus momentum and
liquidity factors (Panels A and B), Fama-French 5 factors (Panels C and D), and Fama-French 5 factor plus momentum and liquidity factors (Fama and French
2015) (Panels E and F). The column labeled Alpha reports the estimated regression intercept, while columns MKT, SMB, HML, MO, LIQ, RMW,
and CMA report the estimated coefficients for the following factors respectively: Market Excess Return, Small minus Big, High minus Low, Momentum,
Liquidity, Robust minus Weak, and Conservative minus Aggressive (CMA) portfolios. For the liquidity factor, the Innovations in Aggregate liquidity measure
(Pstor and Stambaugh 2003) from WRDS is applied. The t-statatistics for each regression coefficient, are given in parentheses and calculated using NeweyWest standard errors with 6 lags.
significant at 10% level.
significant at 5% level.
significant at 1% level.

return, although economically insignificant. Similar to prior

results, alphas remain large and statistically significant at 1%
for all strategies, with the exception of after-costs copula
method, which shows that the risk factors are unable to account
for the profits generated by the strategies.
Finally, by combining the two prior regressions, we attempt
to examine the effects of all mentioned factors, i.e. Fama
French five factors plus momentum and liquidity factors, on

monthly returns. We find that, even after taking into account

all the factors, the alphas for all strategies remain large and
significant, highlighting the fact that neither of these factors
can explain away the returns of the three strategies. Panels E
and F of table 6 shows that momentum and liquidity are both
negatively correlated to the DM and cointegration methods
returns as expected, though neither of the factors shows a
significant correlation with the copula methods returns.

The protability of pairs trading strategies

15

Table 7. Copula methods sensitivity analysis.

Opening threshold
Mean
Min
Max
Standard deviation

0.2
0.0001
0.0291
0.0210
0.0061

0.3
0.0004
0.0286
0.0258
0.0060

0.4
0.0006
0.0254
0.0252
0.0061

0.5

0.6

0.0005
0.0455
0.0231
0.0067

0.0006
0.0391
0.0243
0.0073

0.7

0.8

0.0007
0.0479
0.0355
0.0080

0.0007
0.0513
0.0318
0.0086

Note: This table reports statistics for sensitivity of the copula methods monthly return on employed capital after transaction cost to different opening thresholds.

Table 8. Modified threshold distance method.

Strategy

Mean

Median

Std. Dev.

Sharpe ratio

Modified threshold distance

Distance

0.0038
0.0038

0.0035
0.0032

0.0121
0.0110

0.3173
0.3498

Skewness

Kurtosis

VaR
(95%)

CVaR
(95%)

0.3806
0.3491

11.5077
13.2586

0.0132
0.0106

0.0222
0.0190

Notes: This table reports key distribution statistics for the monthly excess return time series of the MTDM and DM. MTDM is a form of DM where the opening
threshold is increased from two standard deviations to two standard deviations plus transaction costs.

Downloaded by [UQ Library] at 16:03 27 April 2016

5.5. Sensitivity analysis

In prior analysis, for DM and the cointegration method we
apply a constant threshold of 2 standard deviations, and for
the copula method a value of 0.5. We analyse the sensitivity of
the PTS outcomes when the thresholds are changed. Figure 5
shows how various opening thresholds affect the behaviour of
the PTS. For all strategies, as the opening threshold increases,
the average number of days for a pair to converge also increases. This is due to the fact that when a larger divergence is
required to open a trade, it takes more time for this divergence
to reverse. Moreover, again as expected, average trades per
pair per six month period decreases with the increase in the
opening threshold for all strategies.
The robustness of the copula methods performance is studied using multiple opening thresholds (0.2, 0.3, 0.4, 0.5, 0.6,
0.7, and 0.8). Table 7 demonstrates that the copula methods
performance is robust to the opening threshold with the mean
monthly return on employed capital after transaction costs
(varying from 1 to 7 bps). The minimum, maximum and standard deviation of the returns do not vary substantially across
multiple opening thresholds. Figure 6 shows that cumulative
excess return of the copula method is robust to different opening thresholds.
As stated in section 4.1, the DMs opening criterion is the
divergence of the pairs spread from the two historical standard
deviation threshold. The trades will then be closed upon the
convergence of the spread to zero. For the pairs with low
spread volatility, this method of opening and closing trades
can result in negative returns after the transaction costs are
deducted. As an alternative, we modified the opening threshold
of the DM from two historical standard deviations to two
standard deviations plus the transaction costs of the trades.
The performance of this Modified Threshold Distance Method
(MTDM) and its comparison with the DM is presented in
table 8.
Although the mean monthly return of the DM and MTDM
are equal, the DM has a higher Sharpe ratio, as the MTDM has
Opening threshold refers to the cumulative mispriced indicies M1
and M2 , formulated in section 4.3.2. These indicies are calculated by
accumulating the pairs daily conditional probability, h 1 and h 2 .

a higher standard deviation. The VaR and CVaR of the MTDM

are also slightly inferior to DM. Thus, increasing the opening
threshold of the strategy leads to increases in the volatility of
monthly returns due to a decrease in the number of trades.

5.6. Crisis vs. non-crisis

Next, we compare the performance of PTS in Crisis and
Normal periods. The Crisis period is defined as a subsample that comprises the lowest quintile of years where the US
equity market exhibited its worst performance. The analysis of trading strategies during crisis periods is increasingly
important due to risk-averse investors seeking assets that
are safe havens or hedges during market downturns (Low, Yao
et al. 2016). Normal refers to the subsample that excludes the
Crisis period. Figure 7 reports the risk-adjusted performance
of the strategies in these periods.
With the exception of the DMs Sortino ratio, all strategies
show better performance in the Crisis period compared to
the Normal period. While the cointegration method shows
a slightly inferior performance compared to the DM in the
Normal period, it clearly outperforms all strategies in the
Crisis period. This superiority is most obvious in the Sortino
ratio which rises from 0.63 in the Normal period to 0.84 in
the Crisis period. In comparison, the DM experiences a drop
in its Sortino ratio from 0.69 to 0.64 for the same periods.
These results concur with the work of Do and Faff (2010) that
motivate the use of such market neutral strategies in portfolios
as protection against turbulent market conditions. In times that
other passive investment strategies tend to perform poorly, an
investment strategy that includes a PTS such as the cointegration method, can provide the benefits of diversification.

5.7. Performance in different time periods

Figure 8 reports each strategys mean monthly excess return
over 5-year periods for the duration of the study, both before
This is a total of 11 out of 53 years consisting of 1969, 1972, 1973,
1977, 1981, 1983, 1987, 2000, 2001, 2007 and 2008.

16

H. Rad et al.

0.01

Distance
Cointegration
Copula

Mean Excess Return

0.008
0.006
0.004
0.002
0
-0.002
-0.004
1963-66

1967-71

1972-76

1977-81

1982-86

1987-91

1992-96

1997-2001

2002-06

2007-11

2012-14

Years
0.018

Distance
Cointegration
Copula

0.016

Mean Excess Return

0.012
0.01
0.008
0.006
0.004
0.002
0
1963-66

1967-71

1972-76

1977-81

1982-86

1987-91

1992-96

1997-2001

2002-06

2007-11

2012-14

Years
600

Average Yearly Trade Count

Downloaded by [UQ Library] at 16:03 27 April 2016

0.014

Distance
Cointegration
Copula

500

400

300

200

100

0
1963-66

1967-71

1972-76

1977-81

1982-86

1987-91

1992-96

1997-2001

2002-06

2007-11

2012-14

Years

Figure 8. Sub-period performance of pairs trading strategies.

Notes: This figure shows the performance metrics of each strategy in five-year periods. (a), (b) shows the mean monthly excess returns after
(before) transaction costs. (c) shows the trade count.

and after transaction costs. The strategies best performance

period is 19821986 if transaction costs are taken into account
(Figure 8(a)), but 19721976 if they are not (Figure 8(b)).
Interestingly, we can see a decrease in the trade count in the
19821986 period (Figure 8(c)). Since the magnitude of trading costs is considerable, this decrease causes the after-cost
performance to rise. For the rest of the periods, as the average
trade count remains relatively constant, the decline in beforecost performance directly affects the after-cost performance.

Notably, the DM and cointegration method have experienced

a 40 and a 35% drop in their average number of trades from
the 20072011 to the 20122014 period respectively, whereas
the copula methods average trade count has dropped by 15%
in the same period.
This shows that in recent years, pairs whose prices closely
follow each other diverge less frequently than they used to
before. However, more complex trading strategies are able to
produce trading opportunities more frequently. We propose that

The protability of pairs trading strategies

PTS can be improved by reducing the amount that pairs need
to deviate from the equilibrium before a strategy triggers a
pairs trade. This reduction on the restrictiveness of the PTS
will result in a higher number of trades and turnover, but as
trading costs have declined dramatically over the years, there
should be sufficient profit in each trade to cover transaction
costs. Admittedly, this reduction in the trade trigger amount
will reduce the average profit per trade, however the increased
frequency of trades will compensate for the smaller average
return per trade, thus enhancing the overall profit of the strategy. The reduction in the trade trigger amount will also reverse
the trend of the decreasing number of trades, as observed in
recent years.

Downloaded by [UQ Library] at 16:03 27 April 2016

6. Conclusion
Pairs trading is a popular market neutral trading strategy that
purports to profit regardless of market conditions. The strategy
was first pioneered in the 1980s by traders Gerry Bamberger
and the quantitative trading group headed by Nunzio Tartaglia
at Morgan Stanley (Bookstaber 2007). Prior to the spectacular
series of events that led to Long-Term Capital Management
(LTCM)s demise, LTCM itself engaged in pairs trading strategies (Lowenstein 2000). Although Gatev et al. (2006) find that
a simple pairs trading strategy (i.e. the DM) generates profits
from 1962 to 2002, Do and Faff (2010, 2012) report their
declining profitability, after accounting for transaction costs.
Other studies (Vidyamurthy 2004, Wu 2013) introduce more
sophisticated methodologies to improve the strategies using
cointegration and copula frameworks, respectively. However,
these studies are limited in terms of stock pairs investigated,
sample period, and robustness of analysis. Thus, in this paper
it leads us to ask, in a long-term study in the US market,
do pairs trading strategies using sophisticated divergence and
convergence models lead to an increase in profits? Are these
profits sustainable, even after transaction costs? What riskfactors are these pairs-trading strategies exposed to? What is
the performance of these pairs trading strategies during highly
turbulent market conditions?
Our study examines and compares the performance of three
pairs trading strategies using daily US stock data from July
1962 to December 2014: the distance, cointegration and copula
methods. We use a time-varying series of trading costs. We find
that the DM shows a slightly higher monthly return than the
cointegration method, but the cointegration method exhibits a
slightly higher Sharpe ratio. However, on a risk-adjusted basis,
the two strategies perform equally well over the full sample
period. The copula method does not perform as well as the other
two methods in either economic or risk-adjusted performance.
We find that the markets excess return fails to account for the
performance of the three strategies. This further demonstrates
the market neutrality of such strategies that motivates their
use by practitioners as an alternative investment strategy for
reducing exposure to market risk and support by academics
in asset allocation that pairs trading strategy is an effective
diversifier in an investment portfolio. We provide evidence
that the DM and cointegration methods economic and riskadjusted performance are higher in crisis periods. Therefore,
such strategies can be included in investment portfolios in

17

turbulent market conditions for both reducing downside risk

and capturing alpha.
Although we find the performance of the copula method to
be weaker than the DM and cointegration methods, certain
attributes of this strategy deserves further attention. First, in
recent years the DM and cointegration strategies suffer from
a decline in trading opportunities, whereas the copula method
remains stable in presenting such opportunities. Thus, the factors affecting the decline in the frequency of trades in other
methods have not affected the copula method, making it a
reliable substitute for the less sophisticated methods. It also
shows that while market participants might have traded away
the arbitrage opportunities captured by less sophisticated methods, leading to fewer such trading opportunities, the copula
method shows a steady trend in the number of trading opportunities. Second, the copula method shows returns comparable to
those of other methods in its converged trades, even though its
relatively high proportion of unconverged trades countervails
a considerable portion of such profits. Therefore, any attempt
to increase the ratio of converged trades or limit their losses
would result in enhanced performance outcomes. This can
be done by implementing a stop-loss mechanism that limits
the losses or by optimizing the formation and trading periods.
Third, the copula methods unconverged trades exhibit higher
risk-adjusted performance than those of any other strategy
which further motivates the use of such strategies. Finally, we
find that the Student-t is selected as the copula that provides
a parsimonious fit for the dependence structure across stock
pairs in pairs trading on the US market (61% of all pairs
investigated). This is due to the flexibility of the Student-t
copula in modelling correlation (positive and negative) and
fat tails (leptokurtosis). Thus, we highlight the advantage of
using copula in flexibly modelling the dependence structure
and marginal models across stock pairs.
We design a computationally feasible copula PTS that incorporates aspects of the DM and copula technique. Our approach
uses the SSD to identify the stock pairs, and from a range of
marginal and copula models, selects a suitably parsimonious fit
for each of the stock pairs. Such an approach is designed to ease
the implementation for use by practitioners where speed and
computational efficiency is an important consideration when
implementing a trading strategy (Clark 2012, Angel 2014, Brogaard et al. 2014). Further studies in the application of copulas
in pairs trading should involve investigating the out-of-sample
performance of a trading strategy that is solely relies on copulas
for the selection of its pairs. This investigation will lead to
further understanding on whether the application of copulas
sufficiently enhances PTS to justify the cost of the additional
resources required by its computational complexity.
In conclusion, pairs trading strategies still continue to be a
profitable trading strategy that remains robust in highly volatile
market conditions. Although the profitability of strategies
based on distance and cointegration is declining in recent years,
an application of copulas to pairs trading strategy has been
stable. In this respect, the copula method is promising and is
deserving of further attention by practitioners and the academic
community.
Disclosure statement
No potential conflict of interest was reported by the authors.

18

H. Rad et al.

Funding
The authors acknowledge funding from the (1) Australia
AwardsEndeavour Research Fellowship (2) Australian
Institute for Business and Economics (AIBE) in support of
Dr Rand K.Y. Low at the Stern School of Business, New York
University. The authors also acknowledge financial support for
this research from the following funding bodies (1)
Accounting & Finance Association of Australia and New
Zealand (AFAANZ) 2013/2014 research grants (2) UQ Postdoctoral Research Grant entitled Portfolio optimization and
risk management techniques for financial crisis.

Downloaded by [UQ Library] at 16:03 27 April 2016

References
Aas, K., Czado, C., Frigessi, A. and Bakken, H., Pair-copula
constructions of multiple dependence. Insur.: Math. Econ. 2009,
44(2), 182198.
Andrade, S., Di Pietro, V. and Seasholes, M., Understanding the
profitability of pairs trading. Unpublished working paper, UC
Berkeley, Northwestern University, 2005.
Angel, J., When finance meets physics: The impact of the speed of
light on financial markets and their regulation. Financ. Rev., 2014,
49(2), 271281.
Bogomolov, T., Pairs trading in the land down under. In Finance and
Corporate Governance Conference, Melbourne, Australia, 2011.
Bogomolov, T., Pairs trading based on statistical variability of the
spread process. Quant. Finance, 2013, 13(9), 14111430.
Bookstaber, R., A Demon of Our Own Design, 2007 (Wiley: Hoboken,
NJ).
Brogaard, J., Hendershott, T. and Riordan, R., High-frequency trading
and price discovery. Rev. Financ. Stud., 2014, 27(8), 22672306.
Broussard, J.P. and Vaihekoski, M., Profitability of pairs trading
strategy in an illiquid market with multiple share classes. J. Int.
Financ. Markets, Inst. Money, 2012, 22(5), 11881201.
Caldeira, J. and Moura, G.V., Selection of a portfolio of pairs based
on cointegration: A statistical arbitrage strategy, 2013. Available at
SSRN 2196391.
Carhart, M.M., On persistence in mutual fund performance. J.
Finance, 1997, 52(1), 5782.
Chu, B., Recovering copulas from limited information and an
application to asset allocation. J. Bank. Financ., 2011, 35(7), 1824
1842.
Clark, C., How to keep markets safe in the era of high-speed trading.
Working Paper No. 303, Federal Reserve Bank of Chicago, 2012.
Creamer, G. and Freund, Y., Automated trading with boosting and
expert weighting. Quant. Finance, 2010, 10(4), 401420.
Dempster, M.A.H. and Jones, C.M., A real-time adaptive trading
system using genetic programming. Quant. Finance, 2001, 1(4),
397413.
Do, B. and Faff, R., Does simple pairs trading still work? Financ.
Anal. J., 2010, 66(4), 8395.
Do, B. and Faff, R., Are pairs trading profits robust to trading costs?
J. Financ. Res., 2012, 35(2), 261287.
Eling, M., Does the measure matter in the mutual fund industry?
Financ. Anal. J., 2008, 64(3), 5466.
Eling, M. and Schuhmacher, F., Does the choice of performance
measure influence the evaluation of hedge funds? J. Bank. Financ.,
2007, 31(9), 26322647.
Engelberg, J., Gao, P. and Jagannathan, R., An anatomy of pairs
trading: The role of idiosyncratic news, common information and
liquidity. In Third Singapore International Conference on Finance,
Singapore, 2009.
Engle, R.F. and Granger, C.W., Co-integration and error correction:
Representation, estimation, and testing. Econom. J Econ. Soc.,
1987, 55(2), 251276.

View publication stats

Fama, E.F. and French, K.R., Common risk factors in the returns on
stocks and bonds. J. Financ. Econ., 1993, 33(1), 356.
Fama, E.F. and French, K.R., A five-factor asset pricing model. J.
Financ. Econ., 2015, 116(1), 122.
Galenko, A., Popova, E. and Popova, I., Trading in the presence of
cointegration. J. Altern. Investments, 2012, 15(1), 8597.
Gatev, E., Goetzmann, W.N. and Rouwenhorst, K.G., Pairs trading:
Performance of a relative-value arbitrage rule. Rev. Financ. Stud.,
2006, 19(3), 797827.
Huck, N., Pairs selection and outranking: An application to the S&P
100 index. Eur. J. Oper. Res., 2009, 196(2), 819825.
Huck, N., Pairs trading and outranking: The multi-step-ahead
forecasting case. Eur. J. Oper. Res., 2010, 207(3), 17021716.
Huck, N., Pairs trading: Does volatility timing matter? Appl. Econ.,
2015, 118.
Huck, N. and Afawubo, K., Pairs trading and selection methods: Is
cointegration superior? Appl. Econ., 2015, 47(6), 599613.
Jacobs, H. and Weber, M., On the determinants of pairs trading
profitability. J. Financ. Markets, 2015, 23, 7597.
Joe, H., Multivariate Models and Dependence Concepts, 1997
(Chapman & Hall: London).
Lin, Y., McCrae, M. and Gulati, C., Loss protection in pairs trading
through minimum profit bounds: A cointegration approach. Adv.
Decis. Sci., 2006.
Longin, F. and Solnik, B., Is the correlation in international equity
returns constant: 19601990? J. Int. Money Finance, 1995, 14(1),
326.
Low, R.K.Y., Alcock, J., Faff, R. and Brailsford, T., Canonical vine
copulas in the context of modern portfolio management: Are they
worth it? J. Bank. Financ., 2013, 37(8), 30853099.
Low, R.K.Y., Faff, R. and Aas, K., Enhancing mean-variance portfolio
selection by modeling distributional asymmetries. J. Econ. Bus.,
2016, 43, 114.
Low, R.K.Y., Yao, Y. and Faff, R., Diamonds vs. precious metals: What
shines brightest in your investment portfolio? Int. Rev. Financ.
Anal., 2016, 43, 114.
Lowenstein, R., When Genius Failed: The Rise and Fall of Long-term
Capital Management, 2000 (Random House: New York).
Okimoto, T., Asymmetric increasing trends in dependence in
international equity markets. J. Bank. Financ., 2014, 46, 219232.
Pstor, L. and Stambaugh, R.F., Liquidity risk and expected stock
returns. J. Political Econ., 2003, 111(3), 642685.
Patton, A.J., On the out-of-sample importance of skewness and
asymmetric dependence for asset allocation. J. Financ. Econ.,
2004, 2(1), 130168.
Perlin, M.S., Evaluation of pairs-trading strategy at the Brazilian
financial market. J. Derivatives Hedge Funds, 2009, 15(2), 122
136.
Schuhmacher, F. and Eling, M., Sufficient conditions for expected
utility to imply drawdown-based performance rankings. J. Bank.
Financ., 2011, 35(9), 23112318.
Siburg, K.F., Stoimenov, P. and Wei, G.N.F., Forecasting portfolioValue-at-Risk with nonparametric lower tail dependence estimates.
J. Bank. Financ., 2015, 54, 129140.
Vidyamurthy, G., Pairs Trading: Quantitative Methods and Analysis,
2004 (Wiley: Hoboken, NJ).
Wei, G.N.F. and Scheffer, M., Mixture pair-copula-constructions. J.
Bank. Financ., 2015, 54, 175191.
Wu, Y., Pairs trading: A copula approach. J. Derivatives Hedge Funds,
2013, 19(1), 1230.
Xie, W., Liew, Q.R., Wu, Y. and Zou, X., Pairs trading with copulas,
2014. Available at SSRN 2383185.
Xie, W. and Wu, Y., Copula-based pairs trading strategy. In Asian
Finance Association (AsFA) 2013 Conference, Nanchang, Jiangxi,
China, 2013.
Yang, S.Y., Qiao, Q., Beling, P.A., Scherer, W.T. and Kirilenko, A.A.,
Gaussian process-based algorithmic trading strategy identification.
Quant. Finance, 2015, 121.
Zeng, Z. and Lee, C.G., Pairs trading: Optimal thresholds and
profitability. Quant. Finance, 2014, 14(11), 18811893.