Instructions
User Manual:
Open the PDF directly: View PDF 
.
Page Count: 2
BMO Capital Markets
Trading Products
Pairs Trading
Data Cognition Team1
Abstract
Pairs trading is a trading strategy that matches a long position with a short position. The
strategy profits from the difference between the two instruments.[1]
Keywords
Machine Learning — Statistical Methods
1BMO Capital Markets, Toronto, Canada
Feel free to make assumptions, but make sure
to explain the reasoning behind them.
Introduction
Following a machine learning approach: Design
and evaluate a pairs trading strategy for securities
ABC and XYZ.
The Data
In the data folder you will find
pairs.csv
. There are
four columns:
•DATE
: Starting from Jan 4, 2016 to May 5,
2018
•ABC: Price of ABC
•XYZ: Price of XY Z
•RATIO:XY Z/ABC
•DIFF:XY Z −ABC
Data Exploration
Perform any cleaning, exploratory analysis, and
visualization using the data provided.
Ensure to explain your approach at every step
of the process.
Figure 1. Timeline.
Modeling
Build a model to execute the pairs strategy. Discuss
why you chose your approach, what alternatives
you considered, and any concerns you have about
your model. How valid is your model? Include any
indicators of model performance.
How would you quantify the risk associated
with the trading? How would you determine when
the enter or exit a position?
For your reference data captured in 15 minutes
intervals is also provided in the data folder.
considerations
•Validate your model
•
Pay special attention to the quality and orga-
nization of your code
•
Feel free to use a notebook environment while
doing the development, but ensure you orga-
Pairs Trading — 2/2
nize your code as if it were going to produc-
tion. At a minimum, explicitly state how you
would organize the code.
Capital Allocation
Suppose you initial capital is
$10
Million. Specify
how you would utilize the capital.
Scaling
Explain how you would scale your model if instead
of one observation per day, you have tens of thou-
sands of observations per day. Also, consider the
case where instead of having only two securities,
you can choose from an universe of several thou-
sand. How often would you train and replace the
production model?
Market data can quickly add up. Storage consid-
erations and data versioning are important. Assum-
ing you have access to large cloud provider, explain
what types of technologies you would use to store
your data, model, and serve your model.
How do you launch, monitor, and maintain your
model once it is deployed?
Deliverables
Code, plots, explanations.
Package all in one file (
.zip
or
.tar.gz
),
place it in OneDrive link provided.
Questions can be send to armando.benitez@bmo.com
Thank you.
———————————————————
——————————
References
[1]
Investopedia. Guide to pairs trading, 2018. [On-
line; accessed 04-May-2017].
