MATH 60210A Case Study II: A Currency Factor Model

Anthony Sanford∗

March 24, 2025

1 Introduction
Assume that you are an analyst doing research at a major financial institution. Your boss has
asked you to use the tools you recently learned in your financial econometrics course to construct
a factor model. Your boss has given you free rein on the currencies that you can analyze and the
factors that you can use. As a result, you have no data!
The instructions that you have is that your boss wants you to construct a factor model that
might be used by the investment company. He has also asked that you motivate your model by
first conducting a principal component analysis. Luckily, your professor has offered to help...but
you’ll need to talk to him to gain some insights on how/where to get started.

2 Background
Currency investments are investments in currencies of different countries at any given time. Since
currencies are used by all kinds of people for all kinds of reasons at any given time, their valuations
are notoriously tricky. Unlike stocks, that are often used for investments purposes, currencies are
used by firms to conduct their business, used by individuals that travel, etc. As such, factors
that affect currency valuations abound, making factor models hard to specify. The “bilateralness”
of currencies further complicates their factor specifications. Since currency investments represent
moving money from a “home” country to a “foreign” country, it means that there will be compo-
nents in the model that represent changes in the home country, changes in the foreign country, and
the links between the two. For example, if you are investing from Canada, you might need to con-
vert your Canadian currency into U.S. dollars to make your investment (this is your “purchase”),
you will typically invest that money in the meantime (often in risk-free assets), and then to divest
your position, you will need to sell your U.S. currency and buy back Canadian dollars (this is your
∗
HEC Montréal, Department of Finance. Assistant Professor. Address: 3000 chemin de la Côte-Ste-Catherine ,
Montréal, QC H3T 2A7. Email: anthony.sanford@hec.ca

1
“sale”). In this simple example, we would need to use factors that affect Canada, the U.S., and the
U.S. risk-free rate.
Again, this complexity makes currencies notoriously difficult to model accurately. Simply put,
there are just too many variables that go into the analysis of currencies. As previously illus-
trated, it isn’t enough to simply say that there are macroeconomic variables that affect currency
returns/portfolios. Investors must also consider the same macroeconomic variables from different
countries as well as determine the cross-relationships between these variables. If investors were to,
for example, use the risk-free rate as a factor for a currency factor model, you would not only need
to include in your model the risk-free rates of all country pairs in your analysis but also their level
of connectedness relative to one another.

2.1 Principal Component Analysis

PCA is a statistical method that can identify the underlying principal components that explain
the maximum amount of variation in a dataset. In the context of asset pricing factor models, these
factors are the systematic sources of risk that drive the returns of the assets.
To identify these factors in a PCA analysis, you start with a dataset of historical returns for a
broad universe of assets, such as stocks, and compute the covariance matrix of the returns data.
The covariance matrix measures the degree to which the returns of the assets move together and
provides the basis for identifying the factors.
PCA then seeks to find the principal components that explain the maximum amount of variation
in the returns data. These components are linear combinations of the original assets that capture
the common variation in their returns. The first principal component represents the component
that explains the largest amount of variation in the returns data, the second principal component
explains the next largest amount of variation, and so on.
Each principal component has associated factor loadings, which represent the weights assigned
to each asset in the linear combination that defines the factor. The factor loadings indicate how
sensitive each asset is to each factor, and they can be used to interpret the economic meaning of
the factors.

2.2 Factor Models

Factor models are a class of financial models that explain the returns of a portfolio or asset in terms
of one or more underlying factors. These factors often represent the systematic sources of risk that
drive the returns of the assets, and they are typically identified using statistical techniques such as
principal component analysis (PCA) or factor analysis.
In finance, factor models are widely used for a variety of applications, such as asset pricing,
risk management, and portfolio optimization. The basic idea behind a factor model is that the

2
returns of an asset or portfolio can be decomposed into two parts: a part that is explained by the
underlying factors and a part that is unique to the asset or portfolio.
The factors in a factor model can be thought of as representing the different dimensions of
risk that affect the returns of the assets. For example, the factors might represent macroeconomic
variables such as inflation, interest rates, or GDP growth, or they might represent firm-specific
factors such as size, book-to-market ratios, or industry classifications.
Once the factors have been identified, a regression analysis can be performed to estimate the
factor loadings for each asset. The factor loadings represent the degree to which the returns of an
asset are driven by each of the factors. The estimated factor loadings can then be used to construct
a factor model to explain the returns of the assets.
One example of a factor model is the Capital Asset Pricing Model (CAPM), which explains
the returns of an asset in terms of the market risk factor. Another example is the Fama-French
three-factor model, which explains the returns of a portfolio in terms of the market risk factor, the
size factor, and the value factor.
Factor models are widely used in finance because they provide a powerful tool for understanding
the sources of risk and return in the market. By identifying the underlying factors that drive the
returns of the assets, investors can better understand the risks and opportunities in their portfolios,
and construct portfolios that are better aligned with their investment objectives.

2.3 Example
Principal Component Analysis (PCA) can be used in finance to identify the underlying factors that
drive the returns of assets. These factors can then be used to construct an asset pricing model.
One such application of PCA in finance is the construction of a Fama-French three-factor model.
The Fama-French three-factor model explains the returns of a portfolio or asset in terms of three
factors: market risk, size, and value. The market risk factor is the excess return of the market over
the risk-free rate. The size factor is the excess return of small-cap stocks over large-cap stocks, and
the value factor is the excess return of high book-to-market (value) stocks over low book-to-market
(growth) stocks.
To construct the Fama-French three-factor model using PCA, one would start by obtaining a
dataset of historical returns for a broad universe of assets. The returns data would then be used to
construct a covariance matrix, which would measure the degree to which the returns of the assets
move together.
PCA would then be applied to the covariance matrix to identify the underlying components that
drive the returns of the assets. The first principal component would correspond to the market risk
factor, the second principal component would correspond to the size factor, and the third principal
component would correspond to the value factor. Note that we know these relationships at this
point in time. However, this may not have been the case at the onset. Therefore, if we don’t know

3
the factors, we need to infer them using the results of the PCA analysis.
Once the factors have been identified, a regression analysis can be performed to estimate the
factor loadings for each asset. The factor loadings represent the degree to which the returns of an
asset are driven by each of the factors. The estimated factor loadings can then be used to construct
a Fama-French three-factor model to explain the returns of the assets.
As previously mentioned, one important thing to consider in this example, is our prior un-
derstanding of stock returns. In the case of the F-F model, the first principal component, the
market risk factor, is based on prior knowledge and common understanding in finance (e.g. from
the CAPM). If you did not have prior knowledge of what drives asset returns, you would need to
use intuition (which is what you will do in this case) and/or additional methods to interpret the
results of the PCA analysis and infer the underlying factors.

3 From PCA to Factor Models

In the context of asset pricing factor models, the broad factors identified by PCA can be used as
the basis for constructing a factor model to explain the returns of the assets. For example, in the
Fama-French three-factor model, the first factor represents the market risk factor, the second factor
represents the size factor, and the third factor represents the value factor. The factor loadings for
each asset can then be estimated using regression analysis to create a factor model that explains
the returns of the assets in terms of the underlying factors.

4 FX Data
There are multiple sources for obtaining exchange rate (FX) data. One of the easiest sources of
FX data will be WRDS. You can proceed as follows:

1. Log in to the WRDS website using your username and password.

2. Click on “Federal Reserve Bank Reports.”

3. Select the dataset “Foreign Exchange Rates” from the list of available datasets.

4. Click on “Data” under “Foreign Exchange Rates.”

5. Choose the frequency of the data, such as daily or monthly.

6. Select the time period you want to download, such as January 1, 2010, to December 31, 2022.

7. Select the exchange rate you want to download

8. Run the query and download the resulting dataset in the format you prefer, such as CSV or
Excel.

4
5 Learning Outcomes
The goal of this case study is to apply our knowledge of financial econometrics to construct a factor
model. The details and discussion questions below are deliberately kept vague. The goal is to give
you the freedom to explore PCA and factor models. In constructing this factor model, you will
need to 1) gather data, 2) assess the characteristics of the data and ensure that they are suitable
to construct your model, 3) conduct a principal component analysis, and 4) construct a factor
model. This will not be an easy case. There will be a lot of subjectivity and so part of the learning
outcomes of this case will be based on your ability to coherently and convincingly make the case
for the specifications in your factor model. Everyone in the course will have a different analysis and
for each analysis, there will be multiple correct answers. As an analyst, your objective should be to
construct a suitable factor model with factors that make intuitive sense. There are no limitations
as to what factors you can or should use so long as you can justify them in some way that uses the
insights from our statistical models and makes intuitive sense.

6 Discussion Questions
1. This first part of the case involves gathering data. Note that this is just a guide. It is very
likely that you will revisit these steps multiples times as you are working through this case.
You should do the following

(a) Download currency data for five different currency pairs (they should all be using the
same “home” country – for example, CAD to USD exchange rate, CAD to AUD exchange
rate, etc... all use Canada as the “home” currency). You are free to choose whatever
currency pairs you want. In choosing these pairs, make sure that 1) you have access
to the currency data, 2) that there is information related to the factors you might use
for this currency pair (some countries have limited amounts of macroeconomic data, for
example, available to the public). You will need to determine the data sampling period
and the frequency. One thing to consider is that if you intend to use macroeconomic data
(which you will likely need to do), the data will likely only be available at monthly or
quarterly intervals. As such, make sure that your sample is large enough to accomodate
this level of frequency in your analysis. In other words, if you select a quarterly sample
from January 1 2020 to January 1 2022, that would only be 8 data points, not enough
for this analysis. So choose your sample and justify your choice judiciously.
(b) Download factor data from various sources. The sources can be anything. Make sure,
however, that you keep track of where you obtained your data because you will need to
provide your sources in your paper.
(c) Depending on where you get the data, it may be in levels or returns. If your data is in

5
 levels, you should convert it to returns.
(d) Present and discuss descriptive statistics and time-series graphs of all of your data. If
you decide at some later date to include more data in your analysis, make sure that you
update your descriptive statistics and graphs accordingly.

2. Using the data that you collected, you will conduct a principal component analysis (PCA).
For this analysis, you may not use a built-in PCA analysis package. You must write your own
function. The only built-in package/function you can use for this analysis is the estimation of
the eigenvalues and eigenvectors. In doing your PCA, make sure that you do and/or report
the following:

(a) Conduct your PCA analysis on your currency returns. Make sure to document how
you did this and any variable transformations that you did while conducting your PCA
analysis.
(b) Report your variance-covariance matrix. Interpret your result. In particular, what does
it represent and what does it tell you about your returns?
(c) Report your eigenvectors and eigenvalues. Interpret the results.
(d) Report your PC values. Interpret your results.
(e) Report the variances and proportion of the variances explained by your principal com-
ponents. Interpret your results.
(f) Report and discuss a scree table and graph.
(g) Based on your PCA results, what might be the most likely factors? This should lead
directly to your analysis for the next discussion question.

3. Now, you use your PCA model to construct a factor model. In doing so, make sure that you
do, interpret, and/or report the following:

(a) Based on your results from the previous question, discuss what factors you decided to
include. Discuss why you opted for these factors and how they might be related to the
PCA that you did?
(b) If you need more data, get it. Make sure that you update your graphs and/or tables
from before.
(c) Run a regression testing your factors in a factor model. You will likely want to test
multiple factors and show results for your model with variations on which variables are
in the model and which ones are not.
(d) Discuss the results. In discussing your model results, make sure that you discuss how the
model changes when you add or remove some of the factors. Also, discuss the meaning
and interpretation of the factor loadings in the context of the model.

6
 (e) How good is the model? Is there data you wish you had but were not able to find? How
would that help your model?

4. This last question is a conceptual one. Outline and describe at least one real-world application
of the factor model you constructed. In answering this question, make sure that you discuss:

(a) Who is this factor model for? Meaning who might the end-user be?
(b) What can you expect to be able to do with this model?
(c) Realistically, how “good” is the model? What are its limitations? How do you assess
the “quality” of your model? How generalizable or applicable do you think your model
is?

The deliverable for this case study is a short paper. Your paper should provide discussions to the
questions in a paper that is no more than 5-pages (excluding graphs, tables, and references) single
spaced with 12-point Times New Roman font with 1-inch margins. You should also (seperately)
provide your script/code for your analysis. The focus for this exercise is on your understanding
of the statistics and their applications and not on your ability to write code. As such, the grade
will consist mostly of your synthesis of the case rather than your ability to write the code needed
to answer the questions. Assume that the person reading your report has a basic understanding
of statistics and finance. This implies that you should spend a descent amount of space discussing
more complicated concepts in the context of this case. If, for some reason, you are unable to write
the code needed to answer some or all of the questions, you may provide insights on how you would
have answered the question had you been able to write the code. Doing this will allow me to give
you partial credit.