Scribd
Upload
68 views10 pages

Final Lecture Pairs Trading

This document discusses pairs trading strategies. It begins by introducing mean reversion in stochastic processes and using an example of Amazon stock prices. It then discusses how co-movements between two stocks can be used for pairs trading. Specifically, it shows how establishing a cointegrating relationship between the stocks allows them to be combined into a single portfolio that mean reverts. The document outlines how to construct such a portfolio and explains that a pairs trading strategy would be to buy the portfolio when its value is below the mean-reverting level and sell when above.

Uploaded by

qwsx098
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
68 views10 pages

Final Lecture Pairs Trading

This document discusses pairs trading strategies. It begins by introducing mean reversion in stochastic processes and using an example of Amazon stock prices. It then discusses how co-movements between two stocks can be used for pairs trading. Specifically, it shows how establishing a cointegrating relationship between the stocks allows them to be combined into a single portfolio that mean reverts. The document outlines how to construct such a portfolio and explains that a pairs trading strategy would be to buy the portfolio when its value is below the mean-reverting level and sell when above.

Uploaded by

qwsx098
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 10

Pairs Trading1

University of Oxford
Oxford-Man Institute

Álvaro Cartea
[email protected]

December 15, 2021

1 Notes based on textbook “Algorithmic and High-Frequency Trading” with

Sebastian Jaimungal and Jose Penalva.


1 / 10
Mean reversion
Let Xt denote a stochastic process that follows

Xt+1 = α + β Xt + σ ϵt , (1)

where α and β are constants, σ is a non-negative constant, and ϵ are


i.i.d. with distribution N(0, 1). If in (1) one sets α = 0 and β = 1 we
obtain a driftless random walk.

1815
Amazon

1810
Midprice

1805

1800

0 0.2 0.4 0.6 0.8 1


Time

Figure: Midprices AMZN July 13, 2018.

2 / 10
Comovements

1425
1.004 Amazon
Google
1420
1.002
Midprice/mean

1 1415

0.998
1410
0.996
1405
0.994

1400
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Time

Figure: GOOG and AMZN on July 13, 2018 for the whole day of trading: (left
panel) midprices; (right panel) the value a portfolio which is long 1.04 shares of
GOOG and short 0.7687 shares of AMZN (i.e., the co-integration factor). The
dashed line indicates the mean-reverting level, the dash-dotted lines indicate
the 2-standard deviation bands.

3 / 10
Simple strategy

If a trader believes that the mean-reverting pattern persists, then the


strategy is
▶ buy (sell) portfolio when its value is below (above) its
mean-reverting level
▶ For example, to sell the portfolio it requires the trader to sell 1.04
shares in GOOG and buy 0.76 shares in AMZN.

4 / 10
Portfolio with two assets

▶ Let q a denote units of security A and q b denote units of security B.


▶ We denote by
εt = q a StA + q b StB (2)
the value of the portfolio
▶ In the example depicted in Figure 2 security A is GOOG and security
B is AMZN, thus q a = 1.04 and q b = −0.76.

5 / 10
Co-integration of prices

Assume that the prices (or midprices) of the securities are given by a
vector autoregression of order one, i.e., VAR(1), thus

StA = ca + b11 St−1


A B
+ b12 St−1 + ϵAt , (3a)

StB = cb + b21 St−1


A B
+ b22 St−1 + ϵBt , (3b)
where ca , cb , b˙˙ are constants, and ϵAt and ϵBt are error terms.

6 / 10
Back to one asset

If we only focus on security A, which is given by (3a), it would be


difficult to observe a pattern that could lead to a trading strategy.
However, note that if the cross term b12 is zero, then the price dynamics
of security A would exhibit mean reversion. If S a is given by

StA = ca + b11 St−1


A
+ ϵAt , (4)

then S A follows an AR(1) process.


A
Rewrite (4), by subtracting St−1 on both sides of equation (4) and
re-arrange, as
StA − St−1
A A
+ ϵAt ,

= κ1 θ1 − St−1 (5)
where κ1 = 1 − b11 is the speed of mean-reversion and θ1 = ca /(1 − b11 )
is the mean-reversion level of the prices.

7 / 10
Pairs trading

First, we write the VAR model (3) in matrix form:

S t = A + B S t−1 + ϵt , (6)

where S is the vector of prices and A a vector of constants, both with


dimension 2 × 1. The matrix B of constants is of dimension 2 × 2 and ϵ
is a 2 × 1 vector of errors.
Second, we subtract the vector of prices S t−1 on both sides of (6) and
after re-arranging we obtain

∆S t = κ (θ − S t−1 ) + ϵt , (7)

where ∆ represents the difference operator, i.e., ∆S t = S t − S t−1 , κ is a


2 × 2 matrix and θ is a 2 × 1 vector. Here κ = I − B, where I is the
−1
identity matrix, and θ = (I − B) A.
The intuition of the model that expresses changes in prices, as in (7), is
similar to that discussed in the univariate case (5).

8 / 10
Pairs trading
Third, to devise a strategy that benefits from prices that revert to a
mean level we diagonalise the matrix κ. Thus, we write

κ = U Λ U −1 , (8)

where  
κ 0
Λ= 1 (9)
0 κ2
contains the eigenvalues of κ, and U contains the eigenvectors of κ.
The next step is to premultiply (7) by U −1 and write
 
∆S̃ t = κ̃ θ̃ − S̃ t−1 + ϵ̃t , (10)

where S̃ = U −1 S, θ̃ = U −1 θ h, and κ̃
i = Λ.
1 2 ⊺
The new vector of prices S̃ = S̃ S̃ are prices of two portfolios. Each
¯
portfolio contains a linear combination of shares in asset S A and in asset
SB.

9 / 10
100 100

80 80

60 60

40 40

20 20

0 0
-20 0 20 40 60 80 0 10 20 30 40 50 60

(a) band = 0.25 × std.dev . (b) band = 0.5 × std.dev .

100 100

80 80

60 60

40 40

20 20

0 0
0 10 20 30 40 50 -10 0 10 20 30 40 50

10 / 10
(c) band = 1.0 × std.dev . (d) band = 2.0 × std.dev .

You might also like