M of a kind: A Multivariate Approach at

Pairs Trading
Marcelo Scherer Perlin
ICMA/Reading University
WORKING PAPER
Version: December/2007
M of a kind: A Multivariate Approach at Pairs Trading
Abstract: Pairs trading is a popular trading strategy that tries to take advantage of market
inefficiencies in order to obtain profit. Such approach, on its classical formulation, uses
information of only two stocks (a stock and its pairs) in the formation of the trading signals. The
objective of this paper is to suggest a multivariate version of pairs trading, which will try to
create an artificial pair for a particular stock based on the information of m assets, instead of just
one. The performance of three different versions of the multivariate approach was assessed for
the Brazilian financial market using daily data from 2000 to 2006 for 57 assets. Considering
realistic transaction costs, the analysis of performance was conducted with the calculation of raw
and excessive returns, beta and alpha calculation, and the use of bootstrap methods for comparing
performance indicators against portfolios build with random trading signals. The main conclusion
of the paper is that the proposed version was able to beat the benchmark returns and random
portfolios for the majority of the parameters. The performance is also found superior to the
classic version of the strategy, Perlin (2006b). Another information derived from the research is
that the proposed strategy picks up volatility from the data, that is, the annualized standard
deviations of the returns are quite high. But, such event is paid by high positive returns at the
long and short positions. This result is also supported by the positive annualized sharpe ratios
presented by the strategy. Regarding systematic risk, the results showed that the proposed
strategy does have a statistically significant beta, but it isnt high in value, meaning that the
relationship between return and risk for the trading rules is still attractive.
1. Introduction
The market efficiency theory has been tested by different types of research. Such concept
postulates, on its weak form, that the past trading information of a stock is reflected on its value,
meaning that historical trading data has no potential for predicting future behavior of assets
prices. The main theorical consequence of this concept is that no logical rules of trading based on
historical data should have a significant positive excessive return over some benchmark portfolio.
In opposition to the market efficiency theory, several papers have showed that past information is
able, in some extent, to explain future stock market returns. Such predictability can appear in
different ways, including time anomalies (day of the weak effect, French (1980)) and correlation
between the assets returns and others variables, Fama and French (1992). A substantial review
on the market efficiency subject can be found at the papers of Fama (1991) and Dimson e
Mussavian (1998).
A respectable amount of papers have tried to use quantitative tools in order to model the market
and build trading rules. The basic idea of this type of research is to look for some kind of pattern
in the historical stock price behavior and, using only historical information, take such pattern into
account for the creation of long and short trading positions.
With the advent of computer power in the late 90s, more sophisticated mathematical methods
could be employed in the case of trading rules. One example is the use of nearest neighbor
algorithm in trading strategies, Fernandez-Rodrigues et al (2002), Fernandez-Rodrigues et al
(1997), Fernandez-Rodrigues et al (2001) and Perlin (2006a). The NN algorithm is a non
parametric method of modelling time series that has an intuitive appealing based on chaos theory.
The main conclusion drawn from the results presented on the predictability potential of this
method is that it is able to predict correct market direction for most of the forecasted financial
observations. But its important to say that the evidence wasnt strong in all studies.
For the case of trading strategies based on parametric models, there is the work of Efetkhari
(1997) on stock market and Dueker et al (2006) at currency. Both papers based the forecasts on
the regime switching model, where the results indicated that the method can predict the financial
time series researched in each case. Others types of strategies using quantitative formulations
includes timing the market with fundamentals or statistical models, Brooks at al (2005) and
Anderson et al (2006), momentum strategies, Siganos et al (2006) and Balsara et al (2006). The
results from these papers are also positive.
A popular strategy that has made its reputation in the early 80s is the so called pairs trading.
Such methodology was designed by a team of scientists from different areas (mathematics,
computer sciences, physics, etc), which were brought together by the Wall Street quant Nunzio
Tartaglia. The main objective of such team was to use statistical methods to develop computer
based trading platforms, where the human subjectivity had no influence whatsoever in the
process of making the decision of buy or sell a particular stock. Such systems were quite
successful for a period of time, but the performance wasnt consistent after a while and the team
was dismantled after a couple periods of bad performance. More details about the origins of pairs
trading can be found at Vidyamurthy (2004) and Gatev et al (1999). The application of this
particular strategy has already been conducted for financial time series. This includes the work of
Nath (2003), Gatev et al (1999) and, more recently, Perlin (2006b).
The main objective of this research is to suggest a multivariate version of pairs trading. Such
proposed approach will be executed to the data using three different weighting schemes. The
profitability and the risk of such logical rules are going to be assessed based on the Brazilian
financial market, with daily prices from 2000 to 2006. The present paper is based on the work of
Perlin (2006b) and can naturally be seen as an extension of it since the methodologies used for
performance assessment and the researched data is the same.
The paper is organized as follows; the first part is related to the explanation of the methodology
of the research, including performance assessment and the logical rules of trading concerning
pairs strategy. Second, the results from the execution of multivariate pairs trading over the
researched database are presented. After that, the paper finishes with some concluding remarks.
2. Methodology
The methodology of this research is going to be divided in two parts: the first one is the formal
definition of classical pairs trading and the suggested multivariate pairs trading. The second part
introduces the method used for the calculation of raw and excessive returns. It should be pointed
out that the pairs trading strategy can be implemented in many ways. In this particular research
the methods chosen for execution of trading rules and performance assessment were selected
according to its simplicity.
2.1 Classical Pairs Trading
As said before, the idea of classical pairs trading is simple: find two stocks that move together
and take long and short positions in the hope for the stocks prices to move to the historical
behavior. The first step is to normalize the price series of the assets. The reason for the unit
transformation is straightforward; each stock has its own unit. After the normalization, all stocks
are brought to the same standard unit and this permits a quantitatively fair formation of pairs.
The transformation employed is the normalization of the price series based on its mean and
standard deviation,
i
it it
it
P E P
P

) (
*

= . The equation for the univariate pairs is presented next,
Equation [1].
* * *
it it it
p P + = [1]
For Equation [1], the value of
*
it
P is the normalized price series of asset i at time t and
*
it
p is the
normalized price of the pair of stock i, which is found by searching over the database using a
symmetry rule. For instance, the pair of stock i can be identified using the squared distance
rule, meaning that
*
it
p is the asset in which the historical normalized price has, among all others,
the lowest value of sum of squared error from
*
it
P . In another words, the asset
*
it
p has the most
symmetric behavior with
*
it
P .The term
*
it
is just the residue from the difference, which,
obviously, is in the same unit as
*
it
P and
*
it
p .
After the pair of each stock is identified, the trading rule is going to create a trading signal at t+1
every time that the distance between
*
it
P and
*
it
p at time t is higher than d. For instance, at a long
position, this means that the stock is bought at the closing price at time t and, if d is uncrossed at
t+1, then such assets is sold at the closing price of t+1, therefore gaining the return at time t+1.
The value of d is arbitrary, and it represents the filter for the creation of a trading signals. It cant
be very high, otherwise only a few trading signal are going to be created and it cant be to low or
the rule is going to be too flexible and it will result in too many trades and, consequently, high
value of transaction costs.
After a trading sign is created, the next step is to define the positions taken on the stocks.
According to the pairs trading strategy, if the value of
*
it
is positive (negative) then a short (long)
position is kept for asset i and a long (short) position is made for the pair of asset i. The trading
positions are closed when distance between
*
it
P and
*
it
p uncrosses d
1
. Notes that there are two
transactions here, the first for asset i and the second for the pair of asset i. Such information is
important when addressing transaction costs.

1
This may sound counter intuitive, since, using continuous price behavior, if one buys when the distance is d and
sells it when is the distance is again d, there is no profit. But remember that the prices were in discrete time, meaning
that the buying price occurs when the distance is higher (and not equal) than d, therefore the expected profit is
positive. For the case of pairs trading at approximate continuous time (eg. 5 min quotes), this can be easily adapted
by setting a gap between the threshold for buying operation and for the sell operation
The main logic behind the expected profits of classical pairs trading strategy is: if the movements
between the pairs are going to continue in the future, then when the distance between an asset and
its pair is higher than a particular threshold value (d), there is a good possibility that such prices
are going to converge in the future, and this can be explored for profit purposes. If the distance is
positive, then the value of
*
it
P , according to the logic expressed earlier, probably will reduce in
the future (short position for asset i) and the value of
*
it
p is probably going to increase (long
position for the pair of i). The same logic is true for the cases where the distance is negative.
As an example, Figure 1 shows the pairs trading strategy for weekly prices of asset TNLP4 and
its pair, TNLP3. Notes that Figure 1 is the same as Figure 1 at Perlin (2006b).
Figure 1 Example of Pairs Trading with TNLP4 and TNLP3 with d=1
-3
-2
-1
0
1
2
3
1 22 43 64 85 106 127 148 169 190 211 232 253 274 295
Time
N
o
r
m
a
l
i
z
e
d

P
r
i
c
e
s
TNLP4 Normalized Price
TNLP3 Normalized Price
Short Position
Long Position
In Figure 1, TNLP3 is the pair of TNLP4 based on the maximum correlation criteria
2
. Its
possible to see that both normalized prices have a similar behavior. On the points that have a blue
circle or red triangle the absolute difference in the normalized prices have crossed the value of d,
meaning that a trade has taken place. The blue circles (red triangles) are the short (long) positions
created. This happens every time the absolute distance is higher than 1 and the value of the
analyzed asset is higher (lower) than its pair. Every time the absolute difference uncrosses the
value of d, the positions are closed. If the assets, after the opening of a position, move back to the
historical relationship, then the one with the higher price should have a decrease in the prices and
the one with the lower price should have an increase in price. Since a short position was made for
the first asset and a long position for the second, then, if both prices revert to the historical
behavior, a profit will arise from this trading case, and thats the whole idea behind pairs trading,
making profits out of market reversions to the average behavior.
The situations where pairs trading fails to achieve profit are: a increase in the distance between
*
it
P and
*
it
p , where the market goes the opposite way of the expectation and also a decrease

2
One could also used the minimun squared distance rule.
(increase) on the price of the long (short) position. Given that, one of the expectations in the use
of classical pairs trading is that both stocks are behaving abnormally and this may not be true. Its
possible that just one of them isnt behaving as expected, so the position taken on the other one
may not be profitable. But how to know which one is badly priced and which one is not? One
possible answer is to find others stocks that also present similar historical behavior with asset i
and check, for each time t, if the behavior of i stands out when comparing to the others. This is
the framework that motivates the proposed multivariate version of pairs trading, as will be
explained next.
2.2 Multivariate Version of Pairs Trading
The idea behind this suggested multivariate version of pairs trading is to search a pair for asset i
not just with one asset, but with the information of m assets. In other words, the basic approach is
to build (and not just find) a pair to asset i. The formal explanation starts with the formula of the
classical version, given next:
* * *
it it it
p P + =
Using last equation its possible to build a conditional mean for
*
it
p , which gives:
( ) X f p
it
=
*
For the last formula, the function ( ) X f is just a generic formulation saying that the pair of asset i
is a function based on other variables, where X is a matrix with the information of everything that
can explain
*
it
P . The function f(.) can be linear or non-linear. In order to simplify, the rest of the
discussion is going to be based on a linear formulation using the information of the prices of
other assets
3
. Such approach produces:
* *
4 4
*
3 3
*
2 2
*
1 1
*
.......
mt m t t t t it
P w P w P w P w P w p + + + =
Simplifying:

=
=
m
k
kt k it
P w p
1
* *

3
Any kind of variable can be used in this general formulation, including FF variables, Fama and French (1992).
Inserting the last formula in the first equation gives the final result, Equation [3]:
*
1
* *
it
m
k
kt k it
P w P + =

=
[3]
At Equation [3], the value of
k
w is the linear weight that asset k has in explaining i, where k goes
from 1 to m. For this particular research, three different approaches are going to be used in the
weighting scheme (calculation of
k
w ). More details about the approaches will be given later. For
Equation [3], the term
*
kt
P is the normalized price of asset k. The choice of m may be arbitrary of
not. Its possible to build a dynamic approach, selecting optimal values of m that minimizes a
particular objective function, but, for sake of simplicity, the approach at selecting m is arbitrary at
this research. After defining the choice of m, the next step is to find the m assets that have highest
correlation with i in the normalized price state. The normalized price of such assets are referred
as
*
kt
P , where k goes from 1 to m.
The trading rules of this formulation are similar as in the classical version: create a trading sign
when the absolute value of
*
it
is higher than d and take long (short) positions for asset i if the
value of
*
it
is lower (higher) than zero. Its possible to take trading positions on the formed pair
of asset i, which would require the condition that 1
1
=

=
m
k
k
w and 0 >
k
w in order to form such
portfolio, but this is not suggested since it would require the creation of a portfolio with m assets
every time that a trading sign is created. Such creation would take a high number of transaction
costs, and this can easily eat up the profits from the positions. Given that, the framework tested in
this research doesnt allow for trading positions for the artificial pair.
Its important to note that, if some restrictions are made in the formulation given before, its
possible to reach the same formula as in classical trading. This can be done with m=1 and with
the restriction of 1
1
=

=
m
k
k
w . With that, Equation [3] becomes
* * *
it kt it
P P + = , and, since
*
kt
P was
chosen with the same criteria as in the classical version, it is equal to
*
it
p , Equation [1].
As can be seen from the explanation of the trading process, this proposed version can clearly be
labeled as a mean reverting strategy, since the idea is to build a conditional mean for the
normalized prices of the series and trade when the error is considered abnormal, hoping for a
reversion at the historical behavior. As showed before, the heart of the method is at defining the
weights in the formation of the artificial pairs. This can be done in many ways, including
parametric or non-parametric models. In order to keep everything simple, this research is going to
use three accessible weighting schemes in the multivariate framework. The details about each are
given next.
2.2.1 Using OLS to Estimate
k
w
As showed before, one of the main issues about this proposed approach at pairs trading is to build
a pair for asset i. In a linear framework, its possible to use least squares to find the coefficients
that present the lowest sum of quadratic error between the assets i normalized price series and
the normal price series of the artificial pair. In this framework, the conditional mean of the
normalized price of asset i (
*
it
P ) is addressed as next formula, Equation [4].
* * *
4 4
*
3 3
*
2 2
*
1 1
*
.......
it mt m t t t t it
P w P w P w P w P w P + + + + = [4]
As said before, the coefficients
k
w of [4] are going to be estimated with the minimization of the
sum of quadratic error,
( )
2
*
1
T
it
t

=
(
(

. More details about the least squares method can be found in

any undergraduate econometrics textbook, including Maddala (2001). One should notes that,
since
*
kt
P are chosen such that the correlation with
*
it
P is maximum then there is a substantial
multicolinearity problem with the model at [4]. Such problem could be solved by reduction
methods but, in order to keep everything simple, no correction for multicolinearity is performed
here. Notes that the artificial pair cannot be traded unless some constraints are made to the values
of the coefficients, which is not the case for this tested method since the artificial pair will not be
traded due to the transaction costs involved.
2.2.2 Using Equal Weights to Estimate
k
w
In this approach the weighting scheme is the simplest one. Since the choice of k (number of
assets to model
*
it
P ) is arbitrary, the artificial pair of asset i is just the average of the k chosen
assets. Such framework produces Equation [5].
*
1
* *
1
it
m
k
kt it
P
m
P + =

=
[5]
For Equation [5], its possible to see that
m
w
k
1
= , where m is the arbitrary number of assets
chosen to build
*
it
p .
2.2.3 Using a Correlation Weighting Scheme
This framework is the most flexible one since it uses the information on the correlation vector.
Defining
k
as the correlation of the normalized price series k with the normalized price of i, this
approach will calculate the weights using

=
=
m
k
k
k
k
w
1

. With this formulation, Equation [3]

becomes [6].
* * *
1
1
m
k
it kt it m
k
k
k
P P

=
=
(
(
( = +
(
(

[6]
As can be seem from [6], this approach is very flexible, using weights according to the value of
correlations between the normalized price series. If the m chosen assets present very similar
values of
k
, then this approach will be very close to the last one, which uses equal weights.
The main advantage of this framework is that it will be flexible on the weights, giving more
values to those normalized prices that have higher values of correlation and less weight to those
with low values. This in especially good for the cases where the modeled asset i is not so popular
and only a few other stocks present similar behavior. For this particular case, the weighting
scheme is going to give more importance to the assets with high correlation, as opposed to the
equal weighting scheme.
2.2 Assessing Performance of the Strategy
One of the concerns of this study is to evaluate the performance and risk of the multivariate
version of pairs trading strategy against a nave approach. For that purpose, the strategys returns
are going to be compared against a properly weighted portfolio and also against random trading
signals (bootstrap method). The details about the returns calculations are given in the next
topics.
2.2.1 Calculation of Strategys Returns
The calculation of the strategys total return is going to be executed according to the next
formula, Equation [7].
&
1 1 1 1
1
ln
1
T n T n
L S
E it it it it
t i t i
C
R R I W Tc
C
= = = =
( | | | |
= +
| | (
+
\ . \ .

[7]
Where:
it
R
Real return of asset i on time t, calculated by
|
|
.
|

\
|
1
ln
it
it
P
P
;
S L
it
I
&
Dummy variable that takes value 1 if a long position is created for asset i, value -1 if a
short position is created and 0 otherwise. When a long position is made at time t, this
variable is going to be addressed as
L
it
I and as
S
it
I for short positions;
it
W Weighting variable that controls for portfolio construction at time t. In this particular
paper the simulated portfolio is equal weight, meaning that each trading position will
have the same weight at time t, that is

=
=
n
i
S L
it
it
I
W
1
&
1
. Naturally, the sum of
it
W for all
assets is equal to 1 or zero (no trading position at time t);
it
Tc Dummy variable that takes value 1 if a transaction is made for asset i on time t and zero
otherwise
4
C
Transaction cost per operation (in percentage);
T Number of observations on the whole trading period;
For Equation [7], the basic idea is to calculate the returns from the strategy accounting for
transaction costs. The first part of [7],
&
1 1
T n
L S
it it it
t i
R I W
= =

, calculates the total raw return of the

strategy. Every time a long and short position is created for asset i, the raw return of the
simulated portfolio on time t, is

=
n
i
it
S L
it it
W I R
1
&
, that is, the prospected returns multiplied by their
corresponding weight in the portfolio. Since t goes from 1 to T, is necessary to sum such returns,
which gives the final result for the first part of [1],
&
1 1
T n
L S
it it it
t i
R I W
= =

.
The second part of Equation [7] has the objective of accounting for transaction costs. As an
example, suppose that the trading cost of buy and selling one stock is C, which is expressed as a
percentage of the transaction price. If a stock is purchased at price
B
P and sold at price
S
P , then
the real buy and sell prices, including transaction costs, are ) 1 ( C P
B
+ and ) 1 ( C P
S
. Taking the

4
Its important to distinguish the values of
S L
it
I
&
(long and short positions) from
it
Tc (transaction dummy). The
values of
it
Tc are derived from the vector
S L
it
I
&
, but they are not equal. For example, suppose a long position is
created for asset i on time t-1 and also on time t, only. The vector of
L
it
I is going to have values of 1 to time t-1 and
t, but the vector of
it
Tc has only value 1 for time t-1, since for t, the asset was already in the portfolio, so there is no
need to buy it again. The same is true for short positions.
logarithm return of the operation results on the formula
|
|
.
|

\
|
+

=
) 1 (
) 1 (
ln
C P
C P
R
B
S
. Using logarithm
properties, the previous equation becomes |
.
|

\
|
+

+
|
|
.
|

\
|
=
C
C
P
P
R
B
S
1
1
ln ln . Its possible to see from this
result that the return for this operation has two separate components, the logarithm return from
difference between selling and buying price and also the term
|
.
|

\
|
+

C
C
1
1
ln , which accounts for the
transaction cost on the whole operation. This exemplified result basically states that the
transaction cost for one operation (buy&sell) is
|
.
|

\
|
+

C
C
1
1
ln .
Returning to the analysis of the second part of Equation [1], since
|
.
|

\
|
+

C
C
1
1
ln is the transaction
cost of one operation, logically the term
1 1
T n
it
t i
Tc
= =
| |
|
\ .

is just the number of operations made by

the trading strategy. Is important to notes that, since
C
C
+

1
1
is always less than one because C is
always positive and higher than zero, then the value of
|
.
|

\
|
+

C
C
1
1
ln is always negative, meaning
that the transaction costs are going to be subtracted from the strategy returns, which is an
intuitive result.
2.2.2 Evaluation of Strategy Returns
In order to evaluate the performance of the strategy, its necessary to compare it to a nave
approach. If the strategy performs significantly better than an out-of-skill investor, then such
trading rule has value. This is the main idea that will conduct both methods used in this research
to evaluate the performance of the proposed approach. The approaches described here are
computation of excessive return over a nave buy&hold rule and the more sophisticated bootstrap
method of random trading signals.
2.2.2.1 Computation of Excessive Return of a Nave Portfolio
The calculation of excessive return is the simplest approach to evaluate a trading strategy. The
idea is quite simple: verify how does the tested strategy exceeds a nave trading rule in terms of
profitability. In this case, the nave rule is the buy&hold of a properly weighted portfolio for
comparison with the long positions and a sell&unhold for the short positions.
The return of the nave approach, over the whole number of assets, is based on the following
formula, Equation [8].

= = = =
|
.
|

\
|
+

+ + =
n
i
n
i
T
t
it
S
i
T
t
it
L
i NE
C
C
n R P R P R
1 1 1 1
1
1
ln 2 [8]
For Equation [8], the value of
L
i
P and
S
i
P is just the proportion of days, related to the whole
trading period, that the strategy created long and short positions for asset i. Formally,
T
I
P
T
t
L
it
L
i

=
=
1
and
T
I
P
T
t
S
it
S
i

=
=
1
. Notes that, in the calculation of
S
i
P , the sum of the short positions
is always negative or equal to zero, since
S
it
I takes values -1 and 0, only.
Since pairs trading strategy uses two different types of positions in the stock market, long for the
hope of a price increase and short for the hope of a price decrease, its necessary to construct a
nave portfolios that also takes use of such positions. This is the function of the terms

= =
n
i
T
t
it
L
i
R P
1 1
and

= =
n
i
T
t
it
S
i
R P
1 1
, where the first simulates a buy&hold (long positions) of a
properly weighted portfolio and the second simulates a sell&unhold (short positions) scheme
for another properly weighted portfolio. The weights in both terms are derived from the number
of long and short positions taken on each asset, as was showed before. The higher the number of
long and short signals a strategy makes for asset i, higher the weight that such stock will have on
the simulated portfolio. Its clear to see from Equation [8] that, if
L
i
S
i
P P = , which is a perfectly
hedged position for asset i in the benchmark portfolio, the terms

= =
n
i
T
t
it
L
i
R P
1 1

= =
n
i
T
t
it
S
i
R P
1 1
nulls
each other and the contribution of accumulated return for this respective asset in the benchmark
portfolio is just the transaction cost for setting up the portfolios.
It should be notes that the calculation of return at Equation [8] doesnt include
it
W variable as in
Equation [7]. This happens because the refereed equation is calculating the sum of expected
returns of a nave long and short positions for all assets, and not the return of the simulated
portfolio over time (Equation [7]).
As can be seen from Equation [8], one of the premises of the research is that the transaction cost
per operation is the same for long and short positions. The last term of [8] is the transaction costs
for opening positions (making the portfolio) and trade them at the end of the period. In this case,
the number of trades required to form and close the two portfolios is 2n, where n is the number of
researched assets.
The excessive return for the strategy is given by the difference between [7] and [8], which forms
the final formula for computing excessive return, Equation [9].
* &
1 1 1 1 1 1 1 1
1
2 ln
1
T n n T n T n T
L S L S
E it it it i it i it it
t i i t i t i t
C
R R I W P R P R Tc n
C
= = = = = = = =
( | | | |
= +
| | (
+
\ . \ .

[9]
Analyzing Equation [9], the maximization of
*
E
R , which is the objective of any trading strategy,
is given by the maximization of
&
1 1
T n
L S
it it it
t i
R I W
= =

, minimization of

= =
n
i
T
t
it
L
i
R P
1 1
and

= =
n
i
T
t
it
S
i
R P
1 1
and also minimization of |
.
|

\
|

= =
n
i
T
t
it
n Tc
1 1
2 , since
(

|
.
|

\
|
+

C
C
1
1
ln is a constant. The conclusion
about this analysis is intuitive because the strategy is only going to be successful if it efficiently
creates long and short positions on the stocks, keeping the transaction costs and the benchmark
returns at low values. Short story, make more money with less trades.
2.2.2.2 Bootstrap Method for Assessing Pairs Trading Performance
The bootstrap method represents a way to compare the trading signals of the strategy against pure
chance. The basic idea is to simulate random entries in the market, save the values of a
performance indicator for each simulation and count the percentage number of times that those
random entries were worst than the performance obtained in the tested strategy. It should be notes
that each trading strategy takes different number of long and short positions and for a different
number of days. Such information is also taken in account at the random simulations. Before
applying the algorithm, separately, for long and short position, it should be calculated the median
number of days (nDays_Long and nDays_Short) that the strategy has been trading in the market
and also the median number of assets (nAssets_Long and nDays_Short).
The steps are:
1. With the values of the nDays and nAssets for long and short, define nDays random entries
in the market for nAssets number of assets. Again, making it clear, this procedure should
be repeated for each type of trading position (long and short). The output from this step is
a trading matrix which has, only, values 1 (long position), -1 (short position) or zero (no
transaction).
2. Taking as input the trading matrix and the transaction costs, the portfolio is build with
equal weights, resulting in a vector with the returns of the trading signals over time,
1
n
RND
t it it
i
R R W
=
=

, where
it
R is the return for asset i at time t,
RND
it
W is corresponding
portfolio weight of asset i at time t, which is build with the random trading signals from
last step. Such vector is then used for calculation of the performance indicators (eg.
annualized raw return, annualized standard deviation, annualized sharpe ratio).
3. Repeat steps 1 and 2 N number of times, saving the performance indicator value for each
simulation.
After a considerable number of simulations, for example N=5000, the result for the bootstrap
method is going to be a distribution of performance indicators. The test here is to verify the
percentage of cases that the tested strategy has beaten comparing with the use of random trading.
The performance indicators used in this particular research are annualized raw return, annualized
standard deviation and the annualized sharpe ratio.
As an example, the next ilustraton is the histogram of the accumulated returns from the use of
bootstrap algorithm
5
for the daily database with options: N=5.000, nDays_Long=400,
nDays_Long=250, nAssets_Long=5, nAssets_Short=3 and with zero transaction cost (C=0).
Figure 2 Histogram of the annualized raw returns from the Random Trading signals
Figure 2 shows that, considering the options given to the algorithm, an out-of-skill investor
would earn, in average, an annualized raw return of approximately 1.5%. The best case for the
random trading signals is approximately 15% and the worst is -10%. One can also see that the
distribution can be well pictured by a normal likelihood (the line).
The next step in using this bootstrap approach is to count the number of times that the
performance indicator from the tested strategy is better
6
than the simulated performance
indicators from the random trading signal and divide that by the number of simulations. The
result is a percentage showing how many random signals the tested strategy has beaten. If such
strategy has value, it would produce percentages close to 80%. If it is just a case of chance, it
would give a percentage close to 50% and, if the strategy doesnt present any value, it would
result in a percentage close to 20%, meaning that, in this case, its possible to get higher returns

5
The algorithm used is kindly called monkey trading and can be found at authors matlabs exchange site, together
with the classical pairs trading algorithm.
6
Better could mean higher or lower, depending of which performance indicator is being calculated. For instance, a
higher annualized return is better, while a lower annualized standard deviation is preferred
by just using a random seed to select assets and days to trade. One way of analyzing the result of
the bootstrap algorithm is that it compares the selections made by the trading strategy, that is, the
days and assets to trade, against an expected value of the indicator for the same days and number
of trades over the full researched data.
2. Database for the Research
The database for this research is based on the 57
7
most liquids stocks from the Brazilian financial
market between the periods of 2000 and 2006. The training period of the tested strategies is based
on a moving window with approximately 2 year of trading data (494 days). The artificial pair for
each stock is updated at each 10 days.
As an example, for trading at time t=495 is going to be used all information from 1 to 494 in
order to find and build each stocks pairs. For time t=496 the distances are recalculated, but the
weights of the pairs are the same ones used at t=495. At time t=505 the pairs of each asset are
updated and the weights at Equation [3] are recalculated and by using the window from
observation 11 to 504. This process is repeated for the whole data. The normalization of the price
series is also made using this moving windows structure, that is, at each arrival of a new
observation, for each stock, a new mean and standard deviation of the prices is obtained and used
for calculating the normalized prices.
It should be pointed out that no future information is used to construct the trading signs. For each
trading decision at time t, the information used is based, only, on t-window+1 to t-1. The only
future information used for this research is in the data used since it was searched the most liquid
stocks from 2000 to 2006. This was a necessary procedure to avoid illiquid stocks. A possible
approach for avoiding this would be to have a time varying research database, where the stocks
are selected according to its liquidity from t-window+1 to t-1. But, this is not the method used in
this paper.
3. Results
The three weighting schemes of the proposed multivariate pairs trading were executed
8
for the
equity database, forming a dynamic portfolio which changes its composition over time (long and
short positions). Next, Table 1, the results from the profitability point of view are presented. The
values were calculated with fixed C=0.1%
9
, m=5 and for each value of the threshold d (d=0.5
2).

7
The choice for 57 assets was that those were the most liquid among the firstly screened 100 that presented 98% of
valid closing prices.
8
All the calculations needed for this research were performed using MatLab.
9
The trading cost of 0.1% per operation is a realistic value for the Brazilian market. It can be easily achieved with a
relatively small amount of 20.000 R$, which, today (November 2006), is something close to 9.000 USD.
Table 1 Profitability Analysis for the Three Versions of Multivariate Pairs Trading Strategy
Panel A - Multivariate Pairs Trading with OLS (Ordinary Least Squares)
Total Raw Return (with transaction
costs)
Total Excessive Returns (over
Benchmark)
% of
Random
Portfolios
Beaten**
Value of
d
Long
Positions
Short
Positions
Total
Long
Positions
Short
Positions
Total
% of days
in the
Market
Annualized
Raw Return*
Annualized
Raw
Return***
0.50
53.10% -135.48% -130.74% -118.46% 164.09% -2.74% 73.62% -32.78% 0.00%
0.60
88.63% -137.57% -103.59% -15.56% 76.99% 6.78% 60.38% -25.98% 0.00%
0.70
27.67% -122.24% -122.23% -36.48% 28.87% -35.27% 49.75% -30.65% 0.00%
0.80
-21.39% -105.24% -110.61% -64.30% 3.97% -44.30% 40.62% -27.74% 0.00%
0.90
-59.20% -68.13% -115.98% -89.56% 17.51% -60.71% 32.70% -29.08% 0.00%
1.00
-15.40% -73.26% -69.42% -37.96% -6.21% -24.92% 26.98% -17.41% 0.00%
1.10
5.57% -62.76% -45.05% -12.49% -9.06% -9.40% 21.97% -11.30% 0.00%
1.20
31.95% -0.53% 27.39% 17.55% 41.31% 54.83% 17.05% 6.87% 100.00%
1.30
40.03% -18.04% 9.59% 27.37% 17.61% 32.57% 14.64% 2.40% 99.60%
1.40
3.56% 15.58% 14.37% -7.79% 44.82% 32.27% 11.03% 3.60% 100.00%
1.50
11.59% 33.03% 38.45% 1.24% 57.41% 52.48% 9.53% 9.64% 100.00%
1.60
16.52% 27.74% 41.93% 7.88% 48.74% 54.28% 7.72% 10.51% 100.00%
1.70
23.34% 25.69% 43.79% 15.44% 44.52% 54.72% 6.02% 10.98% 100.00%
1.80
24.08% 20.03% 38.02% 17.78% 38.00% 49.69% 5.52% 9.53% 100.00%
1.90
14.27% 33.71% 41.18% 8.83% 50.07% 52.10% 5.02% 10.33% 100.00%
2.00
4.89% 38.08% 39.67% 0.09% 53.54% 50.33% 4.31% 9.95% 100.00%
Panel B -Multivariate Pairs Trading with Equal Weigths
Total Raw Return (with transaction
costs)
Total Excessive Returns (over
Benchmark)
% of
Random
Portfolios
Beaten**
Value of
d
Long
Positions
Short
Positions
Total
Long
Positions
Short
Positions
Total
% of days
in the
Market
Annualized
Raw Return*
Annualized
Raw
Return***
0.50
5.17% -135.38% -112.56% -689.62% 733.41% 61.45% 88.16% -28.23% 0.00%
0.60
23.31% -112.59% -95.39% -478.27% 559.19% 74.82% 82.85% -23.92% 0.00%
0.70
67.16% -112.24% -63.52% -302.11% 408.43% 87.88% 78.84% -15.93% 0.00%
0.80
74.69% -113.40% -64.33% -190.39% 288.23% 72.21% 69.61% -16.13% 0.00%
0.90
73.62% -75.30% -39.87% -121.19% 232.68% 73.31% 58.58% -10.00% 0.10%
1.00
115.82% -55.65% 37.34% -29.57% 172.30% 119.90% 46.94% 9.36% 100.00%
1.10
65.13% -50.25% 21.61% -40.54% 121.76% 87.95% 41.42% 5.42% 100.00%
1.20
80.36% -58.54% 61.04% 0.99% 71.69% 111.90% 35.01% 15.31% 100.00%
1.30
97.41% -7.19% 96.65% 41.77% 89.52% 137.72% 29.59% 24.24% 100.00%
1.40
31.37% -8.32% 56.33% -13.43% 65.30% 85.16% 22.67% 14.12% 100.00%
1.50
63.17% -30.13% 33.37% 26.89% 24.90% 52.11% 18.96% 8.37% 100.00%
1.60
30.78% -36.96% 10.17% 4.02% 7.80% 28.18% 15.85% 2.55% 98.90%
1.70
38.16% -28.29% 22.57% 19.15% 7.69% 39.53% 11.74% 5.66% 99.90%
1.80
46.16% 17.78% 42.57% 30.72% 45.56% 54.91% 10.43% 10.67% 100.00%
1.90
75.83% 1.68% 57.34% 64.99% 23.32% 68.13% 8.02% 14.38% 100.00%
2.00
63.42% 0.03% 49.81% 55.43% 16.29% 58.08% 5.62% 12.49% 100.00%
Panel C - Multivariate Pairs Trading using a Correlation Weighting Scheme
Total Raw Return (with transaction
costs)
Total Excessive Returns (over
Benchmark)
% of
Random
Portfolios
Beaten**
Value of
d
Long
Positions
Short
Positions
Total
Long
Positions
Short
Positions
Total
% of days
in the
Market
Annualized
Raw Return*
Annualized
Raw
Return***
0.50
-6.73% -138.74% -113.56% -624.19% 788.78% 196.50% 86.36% -28.48% 0.00%
0.60
-1.43% -121.17% -103.43% -429.57% 603.81% 193.41% 82.85% -25.94% 0.00%
0.70
44.75% -127.84% -76.00% -259.83% 439.51% 186.76% 79.54% -19.06% 0.00%
0.80
42.39% -117.61% -66.33% -165.46% 325.94% 169.37% 69.21% -16.63% 0.00%
0.90
23.88% -92.72% -59.04% -116.39% 247.20% 140.60% 58.68% -14.80% 0.00%
1.00
43.87% -81.08% -14.01% -50.03% 177.48% 150.64% 50.05% -3.51% 16.50%
1.10
46.04% -79.41% -5.44% -13.85% 121.63% 135.70% 43.23% -1.36% 54.70%
1.20
70.03% -46.93% 40.78% 30.34% 107.21% 155.22% 37.81% 10.22% 100.00%
1.30
93.41% -25.82% 50.79% 71.53% 93.46% 148.20% 31.09% 12.74% 100.00%
1.40
60.12% -25.30% 57.52% 48.25% 71.41% 142.36% 24.67% 14.42% 100.00%
1.50
37.88% -19.81% 23.50% 33.28% 57.41% 96.12% 21.16% 5.89% 99.80%
1.60
47.41% -35.28% 19.86% 49.08% 28.96% 85.77% 17.55% 4.98% 99.50%
1.70
82.56% -14.78% 65.62% 88.04% 39.09% 124.98% 14.24% 16.45% 100.00%
1.80
93.39% 24.63% 86.67% 101.55% 68.85% 139.06% 12.24% 21.73% 100.00%
1.90
84.87% -2.38% 59.95% 94.44% 33.57% 105.47% 9.03% 15.03% 100.00%
2.00
87.02% 9.95% 72.33% 98.20% 38.25% 111.81% 8.12% 18.14% 100.00%
* The annualized raw return is calculated by taking the total raw return, divide it by the total number of days in the
trading sample (in this case 997) and then multiplying the result by 250 (average number of business days in one
year).
** The bootstrap method was conducted with 1000 simulations for each value of d. Simple experiments showed that
this is a reasonable number of simulations (an increase in N didnt changed significantly the results).
*** The percentage showed at this column is the number of beated portfolios given the specific performance statistic
(in this case annualized raw return).
**** This column is calculated by counting the number of days where there was at least one trading position (long or
short) and dividing the result by the total number of trading days at the sample (in this case 997).
The values presented at Table 1 were constructed using the equation described at earlier topics of
this paper. For instance, the raw returns column is calculated according to Equation [2]. The
excessive returns columns are calculated with Equation [4]. The last column is calculated using
the bootstrap procedure described at the past section of the paper.
The first values to be analyzed at Table 1 are the raw returns obtained from the different
approaches (panels) of multivariate pairs trading. For d=1.2 to d=2, most of the values at Panels
A, B and C are positive, meaning that the returns of the strategy after transaction costs are mostly
positive. For the excessive returns column, the values are all positive for Panels B and C but not
for Panel A. One should also notes that the excessive returns for short positions were in great
majority positive and high in value. Partly, this is happening because the benchmark portfolio is
underperforming significantly since the data used for the research is particularly bullish
10
. Given
that the benchmark underperforms brutally, the positions from the short signals yields a high
excessive return. One should be careful when concluding performance based only on the static
benchmark method given here.

10
From the period of 2001 to 2006, Ibovespa, which is the broad market index for Brazilian Market, grew from
17.672 to 38.382 points.
Another important information in Table 1 is the number of days that the strategy was trading in
the market for each approach. For all panels, its possible to see that the percentage of days in the
market decreases as d grows. This is expected since d controls for abnormal behavior and, when
the threshold for such case increases, less and less cases are found. Comparing the percentage of
days in the market for the different approaches (panels), its clear that the OLS method, Panel 1,
had much less trades then the other ones. This is explained by the fact that the weighting scheme
used in Panel A is concerned in statistically replicating the modeled series within a calibration
framework, meaning that finding abnormal behavior, which is the core of pairs trading, should
occur less than a non calibration type of framework, Panels B and C.
Comparing the results for the different panels, one can see that the panel B (equal weight scheme)
and C (correlation weight scheme) yielded higher raw and excessive return than Panel A (OLS
weight scheme). But, at the same time, the OLS method produced less trading signals (stayed less
days in the market), therefore produced less risk. A greater view of the risk of the strategies is
given in the next section of the paper.
When comparing the results at Table 1 for Panel B against Panel C, the values for percentage of
days in the market and total raw returns are, for the different values of d, quite similar. But, when
looking at the total excessive return at Panel C, the results for long and short positions (and both
combined) are higher than Panel B, therefore showing that, when it comes to excess profitability,
a higher performance is found for the correlation weighting scheme.
The main conclusion after the profitability analysis from the trading strategies is that the
proposed version of pairs trading performs significantly better than chance and provides positive
raw and excessive returns after transactions costs. Such evidence is consistently found over the
different versions tested in this research and over different values of the threshold parameter (d).
The best case, as stated before, is for the correlation weighting scheme (Panel C).
The risks provided from the trading signals at the different methods are assessed next. The
analysis will cover the systematic risk (beta), jensens alphas, annualized standard deviation of
the returns and annualized sharpe ratio. The analysis will also cover the bootstrap method for the
performance indicators.
Table 2 Risk Analysis for Multivariate Pairs Trading
Panel A - Multivariate Pairs Trading with OLS (Ordinary Least Squares)
Beta****
Jensen's
Alpha****
% of Random Portfolios
Beaten***
Value of d
Value Prob Value Prob
Annualized
Standard
Deviation*
Annualized
Sharpe**
Annualized
Standard
Deviation
Annualized
Sharpe
0.5 -0.066 0.011 -0.001 0.006 0.227 -1.446 0.00% 7.40%
0.6 -0.076 0.012 -0.001 0.070 0.267 -0.973 0.00% 18.80%
0.7 -0.072 0.040 -0.001 0.058 0.306 -1.003 0.00% 11.70%
0.8 -0.051 0.160 -0.001 0.093 0.315 -0.880 0.00% 15.10%
0.9 -0.066 0.075 -0.001 0.089 0.323 -0.899 0.00% 12.40%
1.0 -0.047 0.216 -0.001 0.326 0.332 -0.524 0.00% 30.60%
1.1 -0.060 0.121 0.000 0.561 0.341 -0.332 0.00% 45.40%
1.2 -0.058 0.144 0.000 0.638 0.346 0.198 0.00% 74.40%
1.3 -0.063 0.121 0.000 0.829 0.353 0.068 0.00% 66.00%
1.4 -0.065 0.104 0.000 0.772 0.349 0.103 0.00% 70.60%
1.5 -0.067 0.089 0.000 0.518 0.344 0.280 0.00% 77.20%
1.6 -0.069 0.079 0.000 0.482 0.343 0.307 0.00% 74.60%
1.7 -0.077 0.046 0.001 0.453 0.338 0.325 0.00% 78.40%
1.8 -0.034 0.333 0.000 0.506 0.309 0.309 0.00% 76.80%
1.9 -0.034 0.290 0.000 0.437 0.285 0.362 0.00% 77.40%
2.0 -0.025 0.439 0.000 0.458 0.282 0.353 0.00% 96.70%
Panel B -Multivariate Pairs Trading with Equal Weigths
Beta****
Jensen's
Alpha****
% of Random Portfolios
Beaten***
Value of d
Value Prob Value Prob
Annualized
Standard
Deviation*
Annualized
Sharpe**
Annualized
Standard
Deviation
Annualized
Sharpe
0.5 0.220 0.000 -0.001 0.000 0.149 -1.895 0.00% 7.10%
0.6 0.212 0.000 -0.001 0.000 0.165 -1.446 0.00% 26.30%
0.7 0.195 0.000 -0.001 0.026 0.190 -0.838 0.00% 53.20%
0.8 0.173 0.000 -0.001 0.069 0.225 -0.717 0.00% 52.60%
0.9 0.183 0.000 -0.001 0.266 0.258 -0.388 0.00% 73.10%
1.0 0.156 0.000 0.000 0.714 0.319 0.293 0.00% 93.70%
1.1 0.178 0.000 0.000 0.936 0.352 0.154 0.00% 89.10%
1.2 0.193 0.000 0.000 0.562 0.380 0.403 0.00% 95.70%
1.3 0.236 0.000 0.001 0.321 0.385 0.630 0.00% 97.30%
1.4 0.197 0.000 0.000 0.604 0.375 0.376 0.00% 86.70%
1.5 0.175 0.000 0.000 0.804 0.357 0.234 0.00% 82.60%
1.6 0.126 0.002 0.000 0.986 0.355 0.072 0.00% 69.20%
1.7 0.078 0.048 0.000 0.822 0.346 0.164 0.00% 76.60%
1.8 0.063 0.107 0.000 0.592 0.344 0.310 0.00% 78.80%
1.9 0.023 0.533 0.001 0.389 0.321 0.448 0.00% 87.50%
2.0 -0.009 0.799 0.001 0.392 0.295 0.423 0.00% 81.30%
Panel C - Multivariate Pairs Trading using a Correlation Scheme
Beta****
Jensen's
Alpha****
% of Random Portfolios
Beaten***
Value of d
Value Prob Value Prob
Annualized
Standard
Deviation*
Annualized
Sharpe**
Annualized
Standard
Deviation
Annualized
Sharpe
0.5 0.245 0.000 -0.001 0.000 0.150 -1.895 0.00% 14.80%
0.6 0.245 0.000 -0.001 0.000 0.167 -1.556 0.00% 21.60%
0.7 0.244 0.000 -0.001 0.005 0.189 -1.010 0.00% 54.00%
0.8 0.243 0.000 -0.001 0.037 0.222 -0.750 0.00% 59.20%
0.9 0.252 0.000 -0.001 0.090 0.251 -0.591 0.00% 58.30%
1.0 0.252 0.000 0.000 0.523 0.296 -0.119 0.00% 80.80%
1.1 0.249 0.000 0.000 0.657 0.322 -0.042 0.00% 80.00%
1.2 0.256 0.000 0.000 0.795 0.348 0.294 0.00% 89.50%
1.3 0.251 0.000 0.000 0.694 0.364 0.350 0.00% 90.00%
1.4 0.219 0.000 0.000 0.602 0.367 0.393 0.00% 90.90%
1.5 0.174 0.000 0.000 0.913 0.359 0.164 0.00% 77.60%
1.6 0.152 0.000 0.000 0.931 0.361 0.138 0.00% 82.70%
1.7 0.143 0.001 0.001 0.467 0.365 0.451 0.00% 92.50%
1.8 0.078 0.052 0.001 0.260 0.354 0.614 0.00% 96.80%
1.9 0.080 0.032 0.001 0.416 0.325 0.462 0.00% 93.90%
2.0 0.088 0.014 0.001 0.308 0.316 0.574 0.00% 94.90%
* The annualized standard deviation is calculated by the multiplication of the standard deviation of the return series
over time with the square root of 250. This calculation is based on constant volatility assumption.
** The annualized sharpe is calculated by dividing the annualized return (Table 1) by the annualized standard
deviation (Table 2)
*** The bootstrap method simulated 1000 random portfolios given the method describe in previous section of the
paper. For each simulated portfolio, the annualized standard deviation and the annualized sharpe were calculated.
The percentage in the two columns shows the number of cases beated by the strategy (percentage of random cases
with lower annualized sharpe and percentage with higher annualized standard deviation).
**** The betas and alphas are obtained with a linear regression of the vector with the strategies returns over time
against the returns from Ibovespa (Broad Brazilian Market Index).
Regarding the values of systematic risk (beta) at Table 2, for all panels, its possible to see that
most of them are relatively small in absolute value, but statistically significant. For Panel A, the
betas are mostly negatives, meaning that the OLS method usually presents returns in the opposite
direction than the overall market (in this case the Ibovespa index). Also, Panel A shows lower
absolute values of beta when comparing against Panels B and C, therefore less systematic risk for
the OLS method.
For Panels B and C, again is found symmetry in the results, where the betas are positive and close
in value. Its also clear for all panels that as d grows, the absolute value of beta decreases, which
is expected since the number of trades also has a negative relation with d, meaning that less
trades is presenting less systematic risk, which is an intuitive result. But, one should also notes
that the annualized standard deviation (unsystematic risk) is not decreasing as d grows. Such
event will be explained next.
Looking at the value of jensens alpha at Table 2, its possible to see that most of them are
positive, which corroborates with the profitability analysis at Table 1. But, its important to notes
that the values of alphas arent statistically significant, meaning that the tested framework wasnt
able to produce significant positive returns after filtering for market conditions. Such information
should be taken into account for the conclusions of this paper.
The results from the bootstrap method serve as a way of assessing how good the values of
annualized standard deviations and the annualized sharpe ratio are against pure chance. A clear
information is that, independent of the value of d, the strategy is picking up large volatility from
the data, therefore the values of annualized standard deviation arent decreasing with the increase
of the threshold parameter. This conclusion is drawed from the fact that in all simulated cases,
for all panels, the strategy presented higher annualized standard deviation than all the simulated
random portfolios, meaning that the proposed trading framework is peculiarly good at trading in
days with high volatility. This peculiar characteristic of the strategy could be explained by the
fact that the core of the trading strategy is to pick up abnormal cases and trade for the hope of a
price reversion. What the results at table 2 are suggesting is that the abnormal cases are
happening at high volatile days, therefore increasing the volatility of the resulting portfolio.
When looking at the results of the bootstrap method for the sharpe ratio, the values of beated
random portfolios are all close to 80%, meaning that the strategy, on average, presents higher
relation of return and risk than a nave approach of random trading signals. The best results on
this indicator for all panels also lies between the interval d=1.2 to d=2, which is the same for the
best results at Table 1. Such information contributes to the positive performance of the strategies
over the data since the risk and return relationship given by the strategies are better than the
expected sharpe ratio for an out-of-skill investor, therefore the positive performance is given by
skill of the method, and not pure chance.
Since the proposed method is derived from the classical pairs trading framework, a natural
analysis to be pursued here is to compare the performance indicators for each case. That is done
next, Table 3, where the classical pairs trading framework is compared against the multivariate
version in the correlation weighting scheme.
Table 3 Multivariate Pairs Trading (Correlation Weighting) X Classical Pairs Trading
% of Random Portfolios Beaten****
Annualized Raw
Return*
Annualized
Standard
Deviation**
Annualized
Sharpe Ratio***
Annualized Raw
Return
Annualized
Standard
Deviation Return
Annualized
Sharpe Ratio
Value of
d
MPT UPT MPT UPT MPT UPT MPT UPT MPT UPT MPT UPT
0.5 -28.48% -59.84% 15.03% 14.22% -1.895 -4.207 0.00% 0.00% 0.00% 0.00% 13.30% 0.60%
0.6 -25.94% -56.25% 16.67% 16.39% -1.556 -3.432 0.00% 0.00% 0.00% 0.00% 23.30% 2.70%
0.7 -19.06% -46.65% 18.86% 19.15% -1.010 -2.436 0.00% 0.00% 0.00% 0.00% 55.10% 33.40%
0.8 -16.63% -24.61% 22.19% 26.10% -0.750 -0.943 0.00% 0.00% 0.00% 0.00% 61.80% 94.30%
0.9 -14.80% -21.77% 25.06% 28.40% -0.591 -0.766 0.00% 0.00% 0.00% 0.00% 62.60% 89.20%
1.0 -3.51% -36.43% 29.61% 31.48% -0.119 -1.157 19.20% 0.00% 0.00% 0.00% 80.50% 52.20%
1.1 -1.36% -24.89% 32.22% 33.55% -0.042 -0.742 54.40% 0.00% 0.00% 0.00% 82.30% 65.50%
1.2 10.22% -29.63% 34.77% 33.23% 0.294 -0.892 100.00% 0.00% 0.00% 0.00% 89.00% 36.70%
1.3 12.74% -19.06% 36.38% 35.71% 0.350 -0.534 100.00% 0.00% 0.00% 0.00% 90.60% 55.50%
1.4 14.42% -9.51% 36.72% 36.35% 0.393 -0.262 100.00% 0.00% 0.00% 0.00% 91.20% 70.00%
1.5 5.89% 6.46% 35.91% 37.83% 0.164 0.171 99.90% 100.00% 0.00% 0.00% 79.00% 86.70%
1.6 4.98% 8.25% 36.11% 37.19% 0.138 0.222 99.50% 100.00% 0.00% 0.00% 79.80% 86.30%
1.7 16.45% 12.94% 36.47% 38.43% 0.451 0.337 100.00%100.00% 0.00% 0.00% 92.50% 88.30%
1.8 21.73% 25.39% 35.37% 37.80% 0.614 0.672 100.00%100.00% 0.00% 0.00% 96.40% 95.30%
1.9 15.03% 22.50% 32.51% 36.12% 0.462 0.623 100.00%100.00% 0.00% 0.00% 92.80% 95.90%
2.0 18.14% 4.80% 31.62% 32.90% 0.574 0.146 100.00%100.00% 0.00% 0.00% 94.60% 74.40%
2.1 16.94% -5.02% 30.53% 31.33% 0.555 -0.160 100.00% 0.00% 0.00% 0.00% 94.40% 53.80%
2.2 4.09% -9.93% 26.87% 30.32% 0.152 -0.328 99.50% 0.00% 0.00% 0.00% 80.50% 40.50%
2.3 7.31% -24.11% 27.05% 28.83% 0.270 -0.836 100.00% 0.00% 0.00% 0.00% 85.90% 13.10%
2.4 8.31% -21.16% 24.12% 28.72% 0.345 -0.737 100.00% 0.00% 0.00% 0.00% 89.80% 20.70%
2.5 6.77% -24.14% 23.58% 29.59% 0.287 -0.816 100.00% 0.00% 0.00% 0.00% 85.60% 12.90%
2.6 7.05% -21.92% 22.64% 29.32% 0.311 -0.748 100.00% 0.00% 0.00% 0.00% 85.70% 18.80%
2.7 10.48% -12.58% 23.40% 28.04% 0.448 -0.448 100.00% 0.00% 0.00% 0.00% 89.60% 36.50%
2.8 10.87% -7.73% 23.13% 27.28% 0.470 -0.283 100.00% 0.00% 0.00% 0.00% 91.80% 50.40%
2.9 15.68% -9.60% 21.88% 26.49% 0.717 -0.362 100.00% 0.00% 0.00% 0.00% 97.60% 41.70%
3.0 13.85% -10.70% 23.98% 24.62% 0.578 -0.435 100.00% 0.00% 0.00% 0.00% 95.60% 37.30%
* The annualized raw return is calculated by taking the total raw return, divide it by the total number of days in the
trading sample (in this case 997 (1491 494)) and then multiplying the result by 250 (average number of business
days in one year).
** The annualized standard deviation is calculated by the multiplication between the standard deviation of the return
series over time and the square root of 250. This calculation is based on constant volatility assumption.
*** The annualized sharpe is calculated by dividing the annualized return by the annualized standard deviation
**** The bootstrap method simulated 1000 random portfolios given the method describe in previous section of the
paper. At each simulation it was calculated the raw return, standard deviation and sharpe ratio (all in annualized
unit). Those 1000 values are them compared against the values from the univariate and multivariate versions of the
strategy, which produced the percentages showed at this column
For Table 3, MPT means multivariate pairs trading and UPT means univariate pairs trading. For
the multivariate case, the values showed at Table 3 are for the correlation weighting scheme,
which was the one with best performance over the competing methods. The values of the UPT
case at Table 3 were calculated with the same algorithm as in Perlin (2006b). It should be also
pointed out that Table 3 is dealing with values of d between .5 and 3, which is a range higher than
for the rest of the tables of the papers.
From Table 3 its clear that the multivariate version is more profitable than the Univariate
version. Such conclusion is based on the annualized returns, where the MPV case presented a
higher number of positive values over the full spectrum of d. Such conclusion is also supported
by the bootstrap method, where the MPT cases presented a higher number of percentage of
beated portfolios when it comes to annualized return. This result is more explicit for high values
of d.
When analyzing the results for the annualized standard deviation at Table 3, its possible to see
that the risk of the different strategies isnt that different from one another. The simulation of
random portfolios also shows that both versions have a tendency to pick up volatility from the
data, where in all cases, for the MPT and UPT, there was no simulated portfolio with volatility
higher than the volatility from the trading rules. In the bootstrap method for the sharpe ratios at
Table 3, its also clear that the multivariate version presents higher ratio of expected return to
risk. Such evidence is supported by the bootstrap method, where the MPT case presented a large
number of beated portfolios and the UPT only presented such evidence for a smaller region of the
threshold d. This means that the MPT version, for this particular research, delivered higher units
of return per risk than the UPT version, which is a positive result towards the performance of the
model.
The results obtained at this research shows that the proposed approach at pairs trading strategy
presented a significant performance regarding return and risk. The values at Table 1 and 2 shows
that the proposed version of the strategy performs better than a benchmark portfolio and such
result is not given by chance, as one can for see for the results of the bootstrap method. When
comparing against the classical pairs trading, the superiority of the performance indicators of the
suggested method over the univariate case was also clear. The multivariate case delivered higher
returns and higher sharpe ratios. But, not all the results were positive. For Table 2, it should be
pointed out that the Jensens alphas arent statistically significant for the different cases, meaning
that, even though the quantitative trading rules performs better than chance, it still doesnt present
significant excess return after filtering for market conditions. Its important to remember that the
whole analysis of Tables 1 and 2 was made with the arbitrary choice of m=5. The same tests
showed before were conducted for values of m between 2 and 10. The results for each value of m
are very similar to those reported here. The tables regarding this investigation are not presented in
order to save space in the paper.
Conclusions
The objective of this paper was to extend the work of Perlin (200b) by suggesting a multivariate
version of the classical pairs trading strategy. The basic idea of this proposed approach is to build
a pair for each stock based on information of others stocks, instead of just finding one, as was
done in the univariate method. Such framework is carefully exposed and tested for the Brazilian
financial market by applying the trading algorithm over the researched data. The performance of
the method regarding return and risk was assessed with the execution of the trading rules to daily
observations of 57 assets of the Brazilian financial market using a database from the period of
2000 to 2006.
Using three versions of a linear framework in the multivariate approach, the main evidence of
this paper is that the trading rules from the multivariate pairs trading strategy had a respectable
performance when applied to the Brazilian Equity market, especially with the use of a correlation
weighting scheme. The evidence of positive excessive return was found at different values of the
threshold parameter, showing consistency of the performance. For the case of risk, most of the
calculated betas were statistically significant, meaning that the strategy does present systematic
risk, but, at the same time, the betas coefficients were not so high in value. It was also found that
the strategy is particularly good at picking up volatility from the data. When analyzing the return
and risk relationship (sharpe ratio) the strategies still performed better than chance, meaning that
the rules provided high volatility in the return vector, but this is compensated by high raw returns
over time.
The results found at this study motivates the application of such quantitative formulation to other
financial markets in order to check if such positive results can be replicated in a different
database. It should be pointed out that the method used here can be shaped in many different
ways in order to accommodate more complex models. For instance, instead of using thresholds
(parameter d), one can use markov switching models in order to capture an abnormal state at
the normalized distance vectors. The use of kalman filter also seems to be a promising research
path for modeling the time varying weights of the artificial pairs. Those are interesting paths for
future research.
References
ALEXANDER, C., DIMITRIU, A (2000) The Cointegration Alpha: Enhanced Index Tracking
and Long-Short Market Strategies ICMA Discussion Paper in Finance.
ALMEIDA, N., M., PEREIRA, P., V. (2000) Mudana de regime para sries financeiras: Um
estudo emprico destes modelos para a realizao de regras de mercado Finance Lab Working
Paper, n. 03.
BROOKS, C., KATSARIS, A., PERSAND, G. (2005) Timing is Everything: A Comparison and
Evaluation of Market Timing Strategies Working Paper, Available at SSRN:
http://ssrn.com/abstract=834485.
CARVALHAL DA SILVA, A. L., RIBEIRO, T. L. (2002) Estimating and Forecasting Latin
American Indexes Using Artificial Neural Networks Proceedings of BALAS.
CHEN, A., LEUNG, M. T., DAOUK, H. (2003) Application of Neural Networks to an
Emerging Financial Market: Forecasting and Trading the Taiwan Stock Index Computers &
Operations Research, v. 30, p. 901923.
DIMSON, E., MUSSAVIAN, M. (1998) A brief history of market efficiency European
Financial Management, v. 4, p. 91-193.
DUEKER, M. J., NEELY, C. J. (2006) Can Markov Switching Models Predict Excess Foreign
Exchange Returns? Working Paper, Federal Reserve Bank of St. Louis.
EFETKHARI, B. (1997) The Markov Regime Switching model as Trading Tool Working
Paper, University of Cambridge.
FAMA, E. (1991) Efficient Capital Markets: II Journal of Finance, v. 46, p. 1575-1617.
FAMA, E., FRENCH, K. (1992) The cross-section of expected stock returns Journal of
Finance, v. 47 (2), p. 427-465.
FERNNDEZ-RODRGUEZ, F., RIVERO, S. S., FELIX, J. A. (2002) Nearest Neighbor
Predictions in Foreign Exchange Markets Working Paper, n. 05, FEDEA.
FERNNDEZ-RODRGUEZ, F., SOSVILLA-RIVERO, S., GARCA-ARTILES, M. (1997)
Using nearest neighbor predictors to forecast the Spanish Stock Market Investigaciones
Econmicas, v. 21, p. 75-91.
FERNNDEZ-RODRGUEZ, F., SOSVILLA-RIVERO, S., GARCA-ARTILES, M. (2001)
An empirical evaluation of non-linear trading rules Working paper, n. 16, FEDEA.
FRENCH, K. (1980) Stock Returns and the Weekend Effect Journal of Financial Economics,
March, p. 55-69.
GATEV, E., GOETZMANN, W. N., ROUWENHORST, K. G. (1999) Pairs Trading:
Performance of a Relative Value Arbitrage Rule Working Paper, Yale School of Management.
Available at SSRN: http://ssrn.com/abstract=141615.
MADDALA, G. S. (2001) Introduction to Econometrics John Wiley & Sons.
MURPHY, J. (2001) Technical Analysis of the Financial Markets. New York Institute of
Finance.
PARK, C., IRWIN. (2004) The Profitability of Technical Analysis: A Review. AgMAS Project
Research Report, n 04.
PERLIN, M S. (2006b) Evaluation of Pairs Trading Strategy at Brazilian Financial Market,
Working Paper, http://ssrn.com/abstract=952242.
PERLIN, M. S. (2006a) Non Parametric Modelling in Major Latin America Market Indexes: An
Analysis of the Performance from the Nearest Neighbor Algorithm in Trading Strategies.
BALAS Conference, Lima, Peru.
VIDYAMURTHY, G. (2004) Pairs Trading: Quantitative Methods and Analysis John Wiley &
Sons, 2004.
NATH, P. (2003) High Frequency Pairs Trading with U.S. Treasury Securities: Risks and
Rewards for Hedge Funds, Working Paper, London Business School.
GATEV, E., GOETZMANN, W. N., ROUWENHORST, K. G. (1999) Pairs Trading:
Performance of a Relative Value Arbitrage Rule, Working Paper, Yale School of Management.
Available at SSRN: http://ssrn.com/abstract=141615.