Factor Based Investing

Factor-based Investing
Analysing market anomalies in the US

equity market
Bachelor’s Thesis
Antti Ilvonen
7.2.2019
ISM
Approved in the Department of Information and Service Management

xx.xx.20xx and awarded the grade
Author Antti Ilvonen
Title of thesis Factor-based Investing: Analysing market anomalies in the US equity market
Degree programme Information and Service Management
Thesis advisor(s) Jyrki Wallenius
Degree Bachelor’s degree
Year of approval 2019 Number of pages 38 Language English
Abstract
Passive investment strategies can be improved by statistically sorting the market based on various
metrics known as factors. By only buying top securities of the market based on the sorting factor,
the portfolio can generate higher returns while still maintaining high diversification. I found that
two factors, value (sort based on book value to market value -ratio) and momentum (sort based
on past performance), generated stable and statistically significant excess-returns based on data
from 1926 to 2018.
The aim of this study was to evaluate the factors’ past performance and to provide logical reasons
for their persistence in the future. Key challenge to using factor sorting in investing is the risk of
market pricing the excess-returns out. However, I provide evidence that release of new
information has not led to significantly smaller factor returns. Also, I note that slow rebalancing of
portfolio betas can explain part of the poor performance of momentum strategies during
economic downturns. Furthermore, I analyzed volatilities of portfolios combining both factors with
the market portfolio and found that diversifying investment portfolio over multiple sources of risk
increases risk-adjusted returns.
1
Table of Contents
Abstract
1. Introduction
1.1 Structure of the thesis
1.2 Research questions and scope of this thesis
1.3 Introduction to differences between active and passive investment strategies
1.4 Factor investing as a method to increase passive investment returns
2. Review of literature explaining the key characteristics of factors
2.1 Criterions for true factors
2.2 Differences between risk-based and behavioral explanations for mispricings
2.2.1 Understanding systematic risk as a source of higher returns
2.2.2 Why is it important to recognize the reason for factor’s existence?
3. Methodology: Constructing factor portfolios
3.1 Short positions in factor portfolios
3.2 Measuring factor performance
4. Theory behind individual factors and analysis results
4.1 Value
4.1.2 Expected returns of value stocks
4.2.2 Logical reasons for value premium’s existence
4.2 Momentum
4.2.1 Expected returns of momentum stocks
4.2.2 Logical reasons for momentum premium’s existence
5. Combining factors to increase risk-adjusted returns
5.1 Combining two factors
5.2 Combining factor risk with market equity risk
5.3 Factor risk-optimization from 2010 to 2018
6. Conclusions and topics for further research
List of references
2
1. Introduction
1.1 Structure
In this thesis I will study factor investing strategies and their reliability in the future. The goal of
this thesis is to explain why factor investing can be profitable and validate these reasons with data
analysis. First, I will explain the basic fundamental ideas behind factor investing. I will go through
different approaches to investing and explain how factor investing differs from traditional
investment strategies. In addition, I will present various general methods to evaluate investments.
Second, I will review literature that supports factor investing and explaining how factors are
detected and validated. Third, I will go through the methodology of my analysis and how I set up
my test portfolios along with explanations for my approach. Fourth, I will present my test factors
and results. Finally, I will combine the two factors to explain diversification across factor and
equity risks and present conclusions as well as topics for further research.
1.2 Research questions and the scope of this thesis
My research question is divided into a series of hypotheses. Are there mispricings in the market
than can be measured by sorting the market based on various factors and explained by logical
reasons? If yes, is there a way to implement an investment strategy to take advantage of those
mispricings? If yes, has that strategy proven to be so efficient and reliable in the past, that it could
be used also in the future? If yes, how should such strategy be implemented in practice?
The scope of this thesis is limited to the equity market (stock market) of the United States. I do
briefly refer to literature which discusses similar mispricings and strategies in other asset classes
and geographical markets as well, but my analysis is based purely on return data of US equity
market. I do not cover every factor strategy that I found reliable, but instead limit my analysis of
two factors that I found the most reliable.
1.3 Introduction to differences between active and passive investment strategies
I will begin with briefly explaining my perception of concepts related to factor investing based on
literature that I have studied. Factor investing is a specific approach to long-term investing, which
is closely related to investing in an index consisting of very large number of stocks or other
securities. Investment strategies, which involve investing in an index are also known as passive
investment strategies. Their fundamental idea is to maximize the investment portfolio’s
3
diversification in order to provide stable returns over very long periods of time, which in this case
can be as long as 10 or 20 years. When investor invests in an index, he effectively buys a small
share of all the securities within an index, which often means hundreds or even thousands of
different securities. This is made possible by different kinds of investment funds. The investors
invest in the fund and the fund then purchases the securities in the index. This way, the investor
can invest in a very large number of securities without having to pay large transaction fees for
purchasing each of the securities independently.
Active investing
The opposite of these passive strategies is active investment management. In the passive strategy
the investor buys the securities and then holds them even if the securities lose part of their value
during the investment period. The speed of economic growth varies cyclically, which means that
sometimes the growth speed is faster and other times it is slower. The changes between economic
cycle’s phases typically translate to the prices of investment securities as well. A classic example is
the stock market, where the stock prices typically increase and decrease over time.
The periods when the stock markets are rising are called bull markets and the periods when the
stock market is losing value are called bear markets. Some investors try to profit from these
changes by trying to identify securities that are increasing value and only investing in those. This is
called active investment management. Its downsides include lower diversification, higher
transaction costs and the inevitable difficulty of identifying the right moments for buying and
selling.
Investors and academics have argued for years which of the two approaches is superior, active or
passive. As David M. Blanchett and Craig L. Israelsen note in their article “Spotlighting Common
Methodological Biases in Active-Vs.-Passive Studies” (2007), the answer is not that simple. Both
types of strategies have found support over the years, even though the passive strategies are
more often found to be better.
In this thesis, I will not be studying whether active or passive investment management is the
superior choice. However, I argue that it is important to understand the difference between active
and passive investing to fully understand the benefits of factor investing. I will handle a topic that
deals with a problem regarding the passive strategies. The reason why many investors choose the
route of stock picking is simply the relatively low expected return of the passive investment
4
strategies. For example, The Helsinki Stock Market has historically returned 12,91 % annually with
sample period from 1912 to 2009 according to a study by Peter Nyberg and Mika Vaihekoski
“Equity premium in Finland and long-term performance of the Finnish equity and money markets”
(2013). This is a nominal rate that includes dividend gains but does not take inflation into account.
However, this is irrelevant when simply comparing active and passive strategies.
Although this might seem like a relatively decent rate of return, the sample period always contains
lengthy periods of much lower returns as well. If the investors choose to go for active stock
picking, they will face higher transaction costs and less diversification. If investor invests in an
actively managed mutual/investment fund, they will face high management costs as well.
Why diversification is important?
Diversification is known as the “only free lunch in investing” as it effectively decreases the
portfolio risk without lowering the expected rate of return, as mentioned by Antti Ilmanen (2011).
According to the theory of efficient markets, which was first published by Eugene Fama (1965) in
his article “Random Walks in Stock Market Prices”, all the securities should have identical risk to
return -ratio, because the market pricing of each security reflects all the information available that
is related to the security. This means that if the security poses a higher risk, it also has a higher
expected rate of return. If any security would produce higher, or so-called excess returns, the
informed investors would take advantage of this arbitrage situation and the profits would be
immediately priced out. Thus, there should be no securities in the market that would provide the
investor with excess returns compared to other securities.
The developments of prices of different securities do vary, however. This causes negative
correlation between the price developments which lowers the portfolio risk without decreasing
the expected rate of return, which is known as diversification benefit. If the investor invests in only
one security and its market value decreases 10 %, the investor loses 10 % of his total investment. If
the investor invests in 10 securities and one of them loses 10 %, the investor only loses 1 % of his
total investment. However, because all the securities have identical risk to return -ratio, both
portfolios have identical expected rate of return, even though the latter has a significantly lower
risk. The risk of a single security is often measured by its volatility or the square root of its
variance:
5
𝑛
𝜎(𝑟) = √𝜎 2 (𝑟) = √∑ 𝑝𝑖 [𝑟𝑖 − 𝐸(𝑟)]2

𝑖=1
E(r) represents the average expected return of the security, rs represents the expected return in
scenario s and ps represents the probability of scenario s. To account for negative and positive
values offsetting each other out, the values are raised to the power of two to achieve a figure
known as variance. Because variance has been raised to the power of two, it cannot be used as
such but instead the square root of variance, the volatility (also known as standard deviation), is
used.
When we add new securities to portfolio, the combined volatility of the portfolio starts
decreasing. As mentioned above, this is the fundamental idea underlying behind passive
investment strategies. Because the expected return to risk ratio is identical for all securities, the
rational investor should just simply buy all the securities to maximize his diversification and thus
minimize his portfolio volatility to achieve the best possible portfolio risk to return -ratio. There
might be times when active stock picking will produce higher returns, but in the long run, the
passive investor will always make the highest profits.
1.4 Factor investing as a method to increase passive investment returns
Simply buying all the securities in a market does not intuitively seem like the best possible
strategy, however. Eugene Fama, together with Kenneth French (1992), are responsible for
popularizing an investment approach called factor investing as an extension to the well-known
Capital Asset Pricing -model, or CAPM, which was originally created by William Sharpe in 1964.
Before diving into the fundamental ideas of factor investing, I briefly present the CAPM.
The CAPM presents the expected return of a single security by comparing its riskiness to the risk-
free rate and the expected return of the market portfolio. The risk-free rate varies, but usually
government bond yields with the highest credit scores are considered risk free. Examples of these
bonds would be the German government bonds in Europe or the US Treasuries in the United
States. The riskiness of a single security is measured with beta, which effectively compares the
covariance between the stock and the market to the variance of the entire market. Covariance is
calculated similarly to the variance:
6
𝑛
1
𝑐𝑜𝑣(𝑥, 𝑦) = ∑[𝑥𝑖 − 𝐸(𝑥)] ∗ [𝑦𝑖 − 𝐸(𝑦)]
𝑛
𝑖=1
In this case, the single security return is expressed with x and the market return is expressed with
y. The beta (β) is then calculated as follows:
𝐶𝑜𝑣(𝑟𝑖 , 𝑟𝑚 )
𝛽𝑖 =
𝑉𝑎𝑟(𝑟𝑚 )
where m refers to market and i refers to the single security. When the beta, the risk-free return (rf)
and the expected market return [E(rm)] are known, the CAPM takes the shape of:
𝐸(𝑟𝑖 ) = 𝑟𝑓 + 𝛽𝑖 [𝐸(𝑟𝑚 ) − 𝑟𝑓 ]
From this equation we can clearly see, that the expected return of the stock increases when the
beta or its riskiness increases. The beta is often seen as a measure of individual security’s riskiness
as it represents the security’s covariance with the market. Beta less than one implies a security
that is less volatile than the market, and vice versa.
The CAPM tells us that assuming efficient markets, the individual security’s expected return is
purely explained by its covariance with the market. It’s worth noting, that the CAPM does not take
into account any metrics traditionally used by market analysts, such as valuation models like price-
to-earnings -ratio or dividend yields. The theory assumes, that this information is distributed to all
investors and all the new information is priced into the security’s market price immediately
meaning that investors cannot find under- or overvalued securities just by analyzing for example
firms’ income levels or valuation metrics.
Factor investing, however, aims to identify metrics that would provide either risk-based or
behavioral reasons for certain securities to be mispriced in the long run. These mispricings could
then be used to create investment strategies generating excess-returns. In the foreword of the
book “Your Complete Guide To Factor-Based Investing” (2016) by Andrew L. Berkin and Larry E.
Swedroe, Cliff Asness defines factor investing as “defining and then systematically following a set
of rules that produce diversified portfolios”. The key is to find those metrics or factors, and then
instead of buying all the securities in the market, buy the best securities sorted by the chosen
factor while still maintaining high diversification. By doing this, we maintain the high return-to-risk
-ratio, low transaction costs and low management costs (due to no one actively managing the
7
portfolio) discussed above but attempt to find logical reasons why some of the securities should
outperform or underperform the market in the long run.
2. Review of literature explaining the key characteristics of factors
In this section, I will review literature and present findings that are common to factors in general. I
present a list of criteria which can be used to evaluate factors and then discuss differences
between behavioral and risk-based reasons for mispricing of securities.
2.1 Criteria for true factors
As noted by Berkin and Swedroe (2016), one of the key challenges regarding factor investing is
determining which factors are truly a source for returns, that exceed the returns of market
portfolio. Market portfolio is considered to be a benchmark for comparison. From this point on, I
will simply call these higher-than-market returns excess-returns. To clarify, a portfolio can be
profitable, even if it produces lower returns than the market portfolio. The objective, however, is
to create a portfolio that produces higher returns than the market portfolio in the long run. The
challenge is, that we obviously cannot predict what is going to happen in the future, and simply
data about past returns is no guarantee of future profits.
If we would take a look at the past returns exclusively, we could probably figure out hundreds of
different portfolios that would have beaten the market portfolio during past decades. But the
question remains, that was it simply coincidence or was there a logical reason why the portfolio
did outperform the market. Many academics have attempted to find out various factors, which
could be used to generate these market-outperforming portfolios, and that have been able to do
that in the past. For example, in their paper Campbell R. Harvey, Yan Liu and Heqing Zhu (2015)
attempted to find out and evaluate various factors suggested by academia, which had been
evaluated in reputable top-level journals and conferences. Even with their strict criteria, they
ended up with a total of 316 different factors, which in my opinion is an enormous number even if
some of the factors presented were highly correlated with each other.
Campbell et al. (2015) also note, that the “overwhelming majority” of the all the factors they
studied produced returns that were statistically significant at 5 % significance. However, not all of
them count as true factors, because their existence is only or mostly based on the past returns,
they have generated. Even if the past returns have overperformed the benchmark, there might
have just simply been a special period of time when the metric has worked, but that is of no use
8
for investor if it cannot generate excess-returns reliably also in the future. Berkin and Swedroe
(2016) present 5 requirements for the metric to be classified as a true factor:
1) It has to be persistent: The factor must have been able to generate stable excess-returns in
the past for a long time.
2) It has to be pervasive: True factors should apply across asset classes, and not only in stocks
for example. This criterion decreases the risk of the factor being simply a result of data-
mining, because one must study many different markets to classify a metric as a factor.
3) It has to be robust: There must be multiple metrics that can be used to measure the
fundamental phenomenon behind the factor. As mentioned, each factor should have a
logical explanation for its excess-returns, and one must be able to test that logical
explanation with various metrics.
4) It has to be investable: Many investing strategies might work in theory, but in the real
world they might be difficult to execute. For example, a strategy where the investor
attempts to buy and sell securities with very short intervals to profit from the securities’
intraday price movements may generate some profits, but it also generates massive
transaction fees, which might make the strategy not usable.
5) It has to be intuitive: As mentioned above, there needs to be a logical, risk or behavioral-
based explanation why the factor should exist. I argue that this is the key to understanding
why these factors should exist also after they have been discovered and published, and the
investors are starting to make investment decisions based on them.
These types of requirements are presented in other academic literature as well. For instance,
Koedijk, Slager and Stork (2016) also list five requirements for factors, although they emphasize
the ease of explaining the factor over its robustness.
2.2 Differences between risk-based and behavioral explanations for mispricings
I will briefly go through what these terms mean based on explanations provided for example by
Berkin and Swedroe (2016), and I will present more concrete examples later when studying the
factors in more detail. When studying the risk-based reasons for higher profits, we must first
understand the difference between systematic and unsystematic (also known as idiosyncratic)
risk. Financial theory states that if the investor carries a higher risk on his investments, he should
be awarded with higher returns, or so-called risk premium, as well. However, this only applies to
9
systematic risk. Systematic risk is the part of the investment’s risk which cannot be diversified
away. So, as explained above, if the investor decides to invest in only a single security instead of
investing in the market portfolio, he carries a higher risk. Yet, he could diversify that risk away at
any time and thus should not be awarded a higher risk premium.
After the investor has diversified as much as he can, there is little he can do to increase the
amount of systematic risk to gain higher risk premium. Some of the factors attempt to generate
higher systematic risk, so the investor could also enjoy the higher premiums. Obviously, higher
systematic risk could also lead to losing money in short perspective but in the long run, carrying
higher risk should be awarded with higher returns.
Behavioral-based reasons are caused by inefficiencies in the way human beings make investment
decisions. Most traditional financial theories assume efficient markets, which includes the
assumption that all investors do only rational choices every time they invest. In reality, many
investors, for example, like to invest in more “high risk, high return” -type of securities which
might cause them to be overvalued, yet it does not stop investors from investing to them. I argue
that this can be seen as a one type of gambling, which people do even though it is not rational.
Some factors try to abuse these behavioral anomalies to generate excess-returns.
2.2.1 Understanding systematic risk as a source of higher profits
But why does the market not price out the higher systematic risk if we know it is there? Systematic
risk of a single security is measured by its covariance with the market (beta) and not by the
security’s volatility. The underlying logic is a simple adaption of basic economic theory, but
important for understanding why systematic risk results in risk premiums. Economic theory states,
that with perfect competition, the marginal cost of any product should match the marginal utility
that the consumer experiences when consuming the product. In the world of investing, this means
that under efficient markets, price of any security should match the experienced utility of the
return investor makes from his/her investment.
This is important, because as we know, we cannot forecast any excessively high returns under
efficient markets, but the experienced utility of the same profits might vary. For example, let’s
assume two stocks, A and B, which both have the same expected rate of return. Stock A is highly
correlated with the market (high beta), and generates market beating profits during bull markets
10
and vice versa. The stock B is negatively correlated (low or negative beta) with the market and
outperforms the market during bear markets.
If the entire market is winning, the investor does not feel too special about the profits he/she
makes with stock A. However, when the markets are crashing, the joy the investor feels from
his/her positive return from stock B is a lot higher. Thus, the stock B’s return has higher marginal
utility which leads to higher price. Now the expected return of any investment is calculated by
dividing the expected return by the investment’s price.
𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑟𝑒𝑡𝑢𝑟𝑛
𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑝𝑟𝑜𝑓𝑖𝑡 𝑜𝑓 𝑖𝑛𝑣𝑒𝑠𝑡𝑚𝑒𝑛𝑡 =
𝑃𝑟𝑖𝑐𝑒
Thus, as the expected returns are equal for all stocks, the higher price of low-correlation stocks
causes them to have lower expected profits, simply because investors value their profits more. In
the end, all sources of systematic risk, including the factor risks, lead to higher profits because the
irrational investors do not experience as high utility from systematically risky investments. The
rational investor can take advantage of this, as in the long run, all the profits are just as valuable.
Only problem is, that increasing systematic risk is difficult, but that is exactly what the factors aim
to do.
I emphasize, that this applies also to factors backed by behavioral explanations, as irrational
investor behavior is explained by the differences in utilities that the investor experiences. For
example, the behavioral bias of going for “high risk, high return” is explained by the fact, that
investors feel higher utility of the chance of big profits, and lower utility of lower yet steadier
profits.
2.2.2 Why is it important to recognize the reason for factor’s existence?
I feel that being able to explain the logic behind any factor is possibly the most important thing to
do before it can be considered a true factor. When a factor is discovered and discussed in scientific
journals, it becomes increasingly well known also among investors. After some time, it would be
logical to think that these increased premiums should be priced out because an increasing number
of rational investors try to profit from them. The increasing demand increases the price of
undervalued securities and decreases the demand for overvalued securities.
It is true, that increase in the use of factor-based strategies naturally decreases their profits, but if
there is a fundamental reason for their existence, the anomalies do not simply disappear because
11
some investors try to profit from them. For example, the overvaluation of systematically risky
securities discussed above would not simply stop even if some investors started to implement
strategies trying to profit from that overvaluation. Berkin and Swedroe aim to identify more of
these reasons behind the excess-returns. Arnott, Beck, Kalesnik and West (2016) raise a similar
concern and mention that many research articles promoting new factors fail to provide enough
credible reasons why the factor should generate structural, long-lasting excess-returns also in the
future and rely mostly on just the past performance of the factor.
3. Methodology: Constructing factor portfolios
In this section, I will go through different methods of measuring factor performance based on
literature. I will also present the metrics that I used in my data analysis and the style that I used to
create the test portfolios.
When discussing the factors in more detail, I will refer to multiple sources, which may have
constructed their test-portfolios differently. Typically, when investors implement these factor-
based strategies, it is done by a technique called “factor-tilting”. This means, that the investor
begins constructing their portfolio with 100 % of the funds invested into a vast market-index. For
example, in many cases this market-index is the Standard & Poor’s 500 -index (S&P500) which is
commonly considered to be the benchmark index for US equity market. The investor then chooses
his preferred factors, and then increases the portfolio weights of the securities which should
provide the largest factor-premiums. In a way, the portfolio is “tilted” towards the chosen factor.
This type of investing is also known as “smart-beta” investing, discussed for instance by Arnott et
al. (2016)
3.1 Short positions in factor portfolios
One key difference between test-portfolios of different research papers is the use of either
long/short or long-only portfolios. I will briefly explain the investment strategy of short-selling to
explain this. If an investor wants to profit from securities being overvalued, he could consider
lending the security, then selling it instantly. After the overvalued security’s market price has
decreased, the investor buys back the security and returns it to the lender. The decrease in the
security’s market price is the investor’s profit, net of any transaction fees and interest paid for
lending the security. This is known as short-selling and it is important in factor investing as many of
the factors are actually based on overvaluation of certain securities, rather than undervaluation.
12
So, for instance, if an investor is able to identify a factor which determines part of the stock
market overvalued, he might simply just avoid buying those stocks or he could systematically also
short-sell those stocks to increase his profits.
I note that in my analysis short-selling does not actually increase absolute returns of the portfolios,
as the short positions generate some profits even if they do underperform compared to market.
However, the short-selling helps to isolate market risk from the portfolio and generate higher
factor risk, which can be useful when diversifying across different types of risk premia. This is
explained and discussed in more detail in section 5.
Differences between long-only and long/short-portfolios
A portfolio which uses a combination of buying stocks (taking the long position) and short-selling
stocks (taking the short position) is known as long/short portfolio. This type of portfolio is used by
Berkin and Swedroe (2016) for example. If the portfolio does not consist of any short positions, it
is known as a long-only portfolio, which is a preferred for example by Koedijk et al. (2016). The
choice between taking short positions or not varies rather a lot and researchers have varying
opinions over which approach should be taken.
Long/short-portfolios have the apparent benefit of being zero-cost. The cash flow generated by
selling the lent stocks short can be used to finance the long positions of the portfolio. Thus,
implementing long/short-portfolios does not require any capital, although it does increase the
investor’s financial leverage due to lending of stocks.
Which portfolio-style should be used?
Roger Clarke, Harindra de Silva and Steven Thorley (2016) discuss long-only portfolios’ capability of
capturing these factor premiums in their article and discover that long-only portfolios do better
when purchasing only several individual stocks. However, when going for a highly diversified
factor-portfolio, they capture only 40 % of the premium. Similarly, my results discussed in section
4 of this thesis imply that long/short -portfolios do capture larger premium than long-only -
portfolios. Nevertheless, finding out whether the higher costs of short positions offset this larger
premium and figuring the general superiority of these approaches is out of scope of this thesis and
topics for further research.
13
However, I will present my findings based on varying literature on why short-selling is sometimes
avoided as these are also the reason behind some factor anomalies. First, when purchasing stocks
or other similar securities, the investor may only lose his invested capital. When short-selling, the
investor may theoretically lose an infinite amount of capital, as the price of the security does not
have an upper limit. Thus, the chance of heavy losses increases compared to long positions.
Second, short positions usually have fixed loan periods for lending the security. This means that
eventually the lent asset must be returned to the owner. Long positions, on the contrary, can be
held indefinitely assuming that the asset has any market value left.
Berkin and Swedroe (2016) note that even 10 years might not be a long enough time to prove the
efficiency of factor portfolios. Thus, taking short positions is riskier, because the investor cannot
know how long he will end up holding the position. Finally, the investor must usually also pay
some interest for lending the security which decreases the profits for short positions. Berkin and
Swedroe (2016) note that these reasons have led to many investment funds not taking any short
positions at all as they are considered too speculative.
3.2 Measuring factor performance
In general, both factors, that I will discuss, do provide a long history of stable and statistically
significant excess-returns compared to a benchmark. The statistical significance in most
publications is measured with Student’s T-test and the excess-returns are measured with simply
deducting the benchmark returns from the factor-portfolio returns. This way some researchers
such as Berkin and Swedroe (2016) also calculate the historical odds of outperforming the
benchmark. One common metric for measuring risk-adjusted excess-returns of a portfolio is also
the Sharpe Ratio, which is measured by deducting the risk-free return from the historical returns
of the portfolio and then dividing the result with the portfolio’s volatility. The risk free -rate used
in all of this thesis’ calculation is the 1-month US T-bill’s interest rate as it is the shortest maturity
US government bond available and thus also the least risky.
𝐸(𝑟𝑖 − 𝑟𝑓 )
𝑆ℎ𝑎𝑟𝑝𝑒 𝑅𝑎𝑡𝑖𝑜 =
√𝑉𝑎𝑟(𝑟𝑖 − 𝑟𝑓 )
Sharpe Ratio compares the portfolio’s returns to its riskiness, higher ratio implies better portfolio
performance.
14
4. Theory behind individual factors and analysis results
In this section, I will present two different factors that I found the most credible for producing
stable excess-returns. This section is thus divided in two sub-sections 4.1 and 4.2. In both sub-
sections, I first discuss literature that is relevant uniquely for the factor and its history. Second, I
will present my analysis of the factor’s absolute historical performance and its statistical
significance. Last, I will review literature explaining logical reasons why the factor should persist to
exist also in the future and present my own arguments.
4.1 Value
Value investing has been popularized over the years by investors such as Warren Buffett. The
general idea deals with finding undervalued stocks that will generate high profits in the long run.
In the world of factor investing, however, the investor does not attempt to pick out stocks that he
considers undervalued. Instead, the investor must determine a rule which can be used to sort out
all the stocks so that he can then buy for example the best 30 % of all the stocks.
The value factor was popularized by Kenneth French and Eugene Fama (1992), although they cite
Dennis Stattman (1980) as well as Barr Rosenberg, Kenneth Reid and Ronald Lanstein (1985) for
discovering the phenomenon in the US stock market. French and Fama (1992) took the firms’ book
values and compared them to the firms’ market values and found that the firm’s that have the
highest book-to-market value -ratios (BtM) outperform the market in the long run.
So, in this case, a long-only portfolio based on the BtM-ratio would include for example the top 30
% of all stocks based on their BtM-ratios. A long/short-portfolio would also short-sell the worst 30
% based on the BtM-ration. As a side note, the number of securities bought or sold is somewhat
irrelevant as long as the portfolio is diversified enough. It could be higher or smaller than 30 % but
the 30 % is the number used, for example, by Berkin and Swedroe. To account for value factor’s
robustness, I mention that there are other metrics than BtM as well that can be used to capture
the same phenomenon. Clifford Asness, Antti Ilmanen, Ronen Israel and Tobias Moskowitz (2015)
list a number of metrics including earnings-to-price, cash flow-to-price and sales-to-price ratios.
The top 30 % of stocks are called value stocks and the worst are known as growth or “glamour”
stocks. This is because low BtM-ratio is often caused by high market value as the market is pricing
in a chance for a smaller company to suddenly generate new innovations that would raise its value
significantly. I note that growth companies are often technology-oriented and even though they
15
rarely end up being highly profitable, there are examples such as Apple or Uber where the owners
have generated massive profits.
On the contrary, value companies are often older firms that have already stabilized their positions
in their market area. They have steady cash flows and they often generate stable dividend yields
for the investors. The market is not pricing in a chance for massive future expansion anymore such
as with the growth stocks. It is intuitive, that value companies seem like the safer bet, but on the
other hand, based on the efficient-market hypothesis, the higher risk of the growth stocks should
also be reflected in their market prices.
4.1.1 Expected returns of value stocks
To demonstrate the historical performance of value premium I used the Kenneth French’s data
library from CRSP which is also referred to, for example, by Berkin and Swedroe (2016). French’s
data consists of combined returns of all the stocks in the US stock market from the New York Stock
Exchange (NYSE), American Stock Exchange (AMEX, nowadays known as NYSE American) and
Nasdaq. Their portfolio data combines the needed accounting figures to the respective company’s
return data and then sorts the return data based on the selected criterion, which in this case is the
BtM-ratio.
To account for the pervasiveness criterion mentioned earlier, I simply note that Clifford Asness,
Tobias Moskowitz and Lasse Pedersen (2013) studied value premium in 18 different countries’
equity markets and found that the premium existed in all of them. Value cannot sensibly be
measured in other asset classes as bonds or currencies, but I argue that the diverse geographical
evidence is enough to prove value premium pervasive enough to count as a true factor.
16
High vs. low annual difference in returns of value portfolios
140
120
100
80
60
40 LO
20 HI
1987
1927
1931
1935
1939
1943
1947
1951
1955
1959
1963
1967
1971
1975
1979
1983
1991
1995
1999
2003
2007
2011
2015
-20
-40
-60
-80
Chart 1. The annual returns (%) of high-BtM stocks (HI, marked with green) compared to low-BtM
stocks (LO, marked with red)
I deducted the monthly returns of the worst 30 % of stocks based on the BtM-ratio from the best
30 %. The data dates back to July 1927, but unlike French’s and Fama’s (1992) original studies, it
also contains the latest data all the way to September of 2018. This means that there were 1107
data points, which cover data from 92 years and 3 months. The combined excess-return when
comparing high-BtM -stocks to low-BtM stocks was 375,21 % or 4,07 % when annualized, which
represents the long/short-portfolio’s excess-return in this case. I also calculated the total return of
the stock market by calculating the weighted average return of the top and bottom 30 % of stocks
based on BtM and the “core” 40 %. When I compared the high-BtM returns to the returns of the
entire stock market, the premium was still 233,40 % or 2,53 % annualized. This represents the
long-only portfolio’s excess returns.
To repeat and clarify, the excess-returns do not represent the annual total returns of the
portfolios. The long/short-portfolio’s, the long-only portfolio’s and the total stock market’s
average annual returns were 16,6 %, 15,1 % and 12,6 % respectively. For the long/short-portfolios,
the total return is calculated by simply adding the excess-returns to the market portfolio’s returns.
I do this simply to illustrate the size of the premium, as I will present practical solutions to
investing with long/short-portfolios in more detail in section 5.
Are the excess-returns statistically significant?
17
A question arises that what are the odds, that the excess-returns that I’ve presented are simply
random? I will discuss shorter time-series later, but for the sake of proving that the historical
higher returns are not simply an accident, I calculated the so-called t-statistics for the entire
dataset. T-tests, that provide the t-statistics, are statistical tests which compare the average return
to a null hypothesis and the standard deviation of the difference in returns. I used so-called paired
two-sample t-tests, which are used when the same dataset is measured twice with some changes
between the two measurements. The idea is to find out, what is the probability that the factor
portfolio’s excess-returns compared to the null-hypothesis are not just a coincidence.
In this case, the null-hypothesis, which the factor-portfolio is compared, is the entire stock market,
which had standard deviation of 5,81. The BtM-sorted long/short-portfolio’s difference to the
stock market had standard deviation of 4,02. The long/short-portfolio’s total standard deviation
was 8,53. The same measurements for the long-only-portfolio were 2,16 and 7,18 respectively.
The t-statistic is calculated as follows:
𝑋̅ − 𝜇0
𝑡=
𝑠/√𝑛
Where 𝑋̅ is the factor portfolio’s average monthly return, 𝜇0 is the stock market’s average monthly
return (the null-hypothesis), s is the standard deviation of the excess-returns and n is the number
of observations (1107 with all portfolios). The t-statistic is the compared to the t-distribution to
find out the probability. As a side note, the t-distribution is used instead of normal distribution as
we are calculating the odds based on a sample rather than the entire “population” (as it is not
sensible to attempt to gather data for the complete set of all the data of the stocks’ historical
returns). T-distributions vary slightly depending on the degrees of freedom, which are calculated
as 𝑑𝑒𝑔𝑟𝑒𝑒𝑠 𝑜𝑓 𝑓𝑟𝑒𝑒𝑑𝑜𝑚 = 𝑛 − 1. However, when the degrees of freedom exceed 30, the t-
distribution highly resembles the normal distribution. In most publications, as noted by Berkin and
Swedroe (2016), the t-statistic that is found high enough for the excess-returns to be statistically
significant is 2. Anything higher than 2 indicates even stronger significance. To prove this, I can
easily compare the t-statistic of 2 with varying degrees of freedom to the t-distribution and see
that the t-statistic of 2 implies roughly 95 % statistical certainty that the excess-returns are not
random. The degrees of freedom in this case are always 1107-1 = 1106.
18
I calculated that the long/short-portfolio’s t-statistic is 2,81 (99,7 % probability) and the long-only-
portfolio’s t-statistic is 3,25 (99,9 % probability) meaning that both portfolios generate statistically
significant excess-returns.
Returns after publication of Fama and French
Before addressing logical reasons behind the value factor, I will present similar statistics as above
for time series beginning from the publication of the factor by Fama and French (1992) to this
date. The reason for this is to prove that the value factor still exists even after wider audience has
become aware of it and the market does not just simply price the value premium out. As
mentioned earlier, the value factor was presented by other researchers already before Fama and
French, but as their 1992 publication is such widely referenced (1717 citations in EBSCOhost
Business Source Complete database) article, it is a natural choice for a benchmark. The article was
released in June 1992, so the following calculations are based on data beginning from July 1992 to
September 2018.
The results were not quite as impressive as with the longer time-series though. The annual
premium of long/short-portfolio was only 0,91% with t-statistic of only 0,44. The annual premium
of long-only-portfolio was 0,64 % with t-statistic of only 0,60. Thus, neither of these premia are
statistically significant. The dataset is relatively long, yet it does contain for example the two
longest bull markets of US equities (October 1990 to March 2000 and March 2009 to August 2018)
which might cause the results to be somewhat distorted. It appears, that the value factor has
indeed underperformed compared to the total market, first total of -17,4 % from 10/1990 to
3/2000, and then total of -5,1 % from 3/2009 to present date. Despite the long bull markets lasting
roughly two thirds of the entire post-6/1992 dataset, the value premium was still positive after all,
which could indicate that it has not disappeared even though there have been multiple academic
publications related to it. However, whether the poor performance was related to value factor
simply underperforming during bull markets in general, or due to the value premium getting
priced out by the market, remains a subject for further research.
Summary of statistics
Table 1.1 and 1.2 based on data series from July 1927 to September 2018
19
Based on Book-to-Market -ratio
Sorting Total Market Hi 30 % Lo 30 % Long/Short -premium + total
Average annual returns 12,6 % 15,1 % 11,0 % 16,6 %
Variance 33,7 51,5 28,3 72,7
Standard deviation 5,81 7,18 5,32 8,53

Portfolios Long/Short Long-only
Annual excess-return 4,07 % 2,53 %
Standard deviation 4,02 2,16
t-statistic 2,81 3,25
Probability 99,7 % 99,9 %
Tables 1.3 and 1.4 based on data series from July 1992 to September 2018

Variance 17,0 23,5 17,3 33,1

4.1.2 Logical reasons for value premium’s existence
Risk-based
I will first go through risk-based reason for higher returns of value stocks. In the chart 1, one can
roughly observe that the High-BtM -stocks generate higher returns during economically good
times but also higher losses during bad times. This is also represented by the higher standard
deviations of the high-BtM -portfolios in tables 1.1 and 1.3. This already suggests that the returns
of high-BtM stocks are more volatile and thus riskier than low-BtM stocks. And as noted earlier,
higher risk should also be rewarded with higher return as well.
Nai-Fu Chen and Feng Zhan (1998) discuss the reasons for higher risk of value companies. One of
the key reasons for the higher risk is higher financial leverage used by value companies and the
higher financial distress caused by it. Value companies have more stable cash flows and they can
handle the periodical interest payments with more certainty than growth companies. Interest is
20
usually tax-deductible and due to the fact, that issuing new equity often reveals that the
management considers the stock overpriced, as mentioned by Stewart Myers and Nicholas Majluf
(1984), debt is generally preferred over external equity for financing when available. This means
that in efficient markets all companies should prefer debt over external equity, but value
companies are more likely to be able to withstand the financial distress that comes with it. Debt
causes financial distress as interest must be paid out whereas dividends are optional. Thus, value
companies generally have higher debt/equity -ratio (financial leverage) than growth companies,
and thus they are also generally riskier.
Robert Peterkort and James Nielsen (2005) go even further than Chen and Zhan (1998) and argue
that the entire BtM-premium is explained by higher leverage. They found that in firms that have
no long-standing debt there is no excess-returns associated with higher BtM-ratio. However, I
argue that from investor’s point of view, this is not highly relevant as most value companies have
at least some level of debt. They also note, that the leverage does not explain the entire BtM-
premium in companies that are at least partly financed with debt.
Another risk-based explanation by Lu Zhang (2005) deals with my earlier observation of value
stock returns decreasing more than growth stock returns during difficult economic times.
According to Zhang (2005), value companies in general have higher amounts capacity based on
long-term fixed assets. It is difficult to reduce this excess-capacity during economically bad times
because the investments to fixed assets can rarely be reversed. Thus, value companies cannot
adapt to the lower demand as well as growth companies and have higher risk. During
economically good times, it’s relatively easy for both value and growth companies to raise capital
and expand capacity, so in conclusion value companies end up being riskier. Again, higher risk
should be rewarded with higher premium.
Behavioral
Next, I will go through behavioral reasons for value premium’s existence. The basic behavioral
concept is mostly related to the overvaluation of growth stocks. Joseph Piotroski and Eric So
(2012) researched these behavioral mispricings in their study. They found that the firms with low
BtM-ratio and weak overall financial condition are generally overvalued where as the high-BtM
and fundamentally strong firms were undervalued. In growth stocks, the investors tend to
overestimate the future outlook even if the firm in reality has very weak financial state. In a similar
21
manner, value stocks might be undervalued, because investors overestimate the risk caused by
higher financial leverage and tend to be too pessimistic about them. This leads to increase in the
value premium.
Another mispricing phenomenon is known as anchoring and is described by Berkin and Swedroe
(2016). Anchoring in this case means investors’ tendency to hold on to losing investments as they
do not want to sell them for loss. Instead, they rather hold the investment waiting for it to reach
at least break-even level even if the investment does not seem profitable on its own anymore.
Growth stocks might have very high valuation levels when measured with metrics such as price-to-
earnings (P/E) simply because their earnings levels are very small compared to the future
expectations. The investors might anchor themselves to expecting the high return suggested by
the P/E even though the future expectations might decrease after a while. So even if the P/E or
other valuation metric keeps falling, the investors still hold on to their investments wanting to
believe for the higher returns suggested by the previous high P/E would ultimately realize.
4.2 Momentum
Momentum refers to a strategy where the investor simply buys securities that have performed
well in the past and optionally also short-sells securities that have performed poorly. Even though
this strategy seems logical, it does not comply well with the basic logic of the efficient market
theory. If the markets are efficient, the market price of the security should reflect all the relevant
information about the security and any price changes should only be caused by new information
becoming available. Thus, it does not really make sense that the past good (bad) performance
would imply higher (lower) returns in the future.
Yet it still is regarded as one of the most persistent factor strategies. One of the groundbreaking
articles regarding momentum was published by Narasimhan Jegadeesh and Sheridan Titman
(1992) who noticed that momentum strategies generated high excess-returns. Importantly, they
also argued that such high returns cannot be explained purely by systematic risk and that investors
must have some sort of consistent behavioral bias that causes this effect.
The momentum has then been discussed in multiple papers over the years. Mark Carhart (1997)
combines the factor created by Jegadeesh and Titman to the original model by Fama and French
from 1992 which I cited earlier in the value-section. Like the book-to-market -ratio, momentum is
22
calculated as “winners minus losers (WML)” by deducting the returns of top ranked securities from
the returns of worst ranked securities. The sorting criterion in this case is simply the past
performance of a chosen time period. Jegadeesh and Titman (1992) used past returns and holding
periods of 6 months in most of their calculation, but they also studied different time periods and
noticed that the momentum effect reached its maximum after 12 months. Most publications seem
to agree, as for example Berking and Swedroe (2016) use 12-month periods for calculating the
past performance. Nevertheless, I argue that the fact that different time periods have performed
well for different researchers accounts for the robustness criterion for classifying momentum as
true factor.
I should note that there are two distinctive ways to measure momentum. The method explained
above is known as cross-sectional momentum and the past performance of a single security is
measured relative to the entire market. As a hypothetical example, this could lead to buying
securities that have lost value over past months if the market has been losing even more on
average. Another way to measure momentum is known as time-series momentum, which means
that the absolute past performance of a security is used. This concept might be easier to grasp, as
only securities that have gained value are bought and only securities that have lost value are sold
short.
There has been some discussion about which measurement should be used, and for example, Ron
Bird, Xiaojun Gao and Danny Yeung (2017) compared the time-series momentum to the original
cross-sectional momentum of Jegadeesh and Titman (1992). They argue, that even though both
strategies generate excess-returns, the time-series momentum is the superior choice. This is
interesting as the cross-sectional momentum has been clearly the most studied factor of the two.
Despite the findings of Bird et al. (2017) I will use the cross-sectional momentum in my
calculations as it has been the consensus choice in most literature and as I will show, there is high
evidence of its strong performance.
I used the same data source as with value, and the test portfolios were created exactly the same
for easy comparison. Again, I will present long/short-portfolio which buys the top 30 % of stocks in
the US stock market based on the past performance of 12 months and short-sells the worst 30 %,
and long-only portfolio which only buys the top 30 %. The portfolios do not have specified holding
periods like some of the portfolios in articles such as the one by Jegadeesh and Titman (1992), but
23
instead the portfolios are rebalanced on a monthly basis. This again makes the comparison
between value and momentum easier.
Momentum strategies can naturally be implemented in other asset classes as well, although that is
out of scope of this thesis. For pervasiveness criterion of momentum, I refer to the same article as
with value by Asness et al. (2013) which studied momentum in 18 different countries but also in
other asset classes such as government bonds and commodities. The study found statistically
significant excess-returns across all asset classes in all markets except Japan.
4.2.1 Expected return of momentum stocks
High vs. low annual difference in returns of momentum

portfolios
40
30
20
Hi
10 Lo
5 per. Mov. Avg. (Hi)
0
5 per. Mov. Avg. (Lo)
1927
1934
1941
1948
1955
1962
1969
1976
1983
1990
1997
2004
2011
-10
-20
-30
Chart 2. The annual returns (%) of winning momentum stocks (HI, blue) compared to losing
momentum stocks (LO, orange) smoothed with 5-year moving average.
The excess-returns generated by momentum were even more clear than with value. The
long/short -portfolio generated total average annual return of 19,51 % and the average excess-
return to the total market was 8,25 %. The long-only portfolio’s return was also impressive, the
total average annual return was 15,4 % and the average excess-return was 4,15 %. It should be
noted, that the dataset that was used to calculate these returns differs slightly from the dataset
used to calculate value premiums. The criteria for including a single stock’s data in the dataset are
marginally different as the value dataset has criterion for both market equity data and book equity
data, whereas the momentum dataset has also criterion for price data. Some missing data values
might cause a small number of stocks to be excluded from either of the portfolios and the
24
momentum dataset also begins 6 months later than the value dataset (January 1927). Thus, the
total average market return is lower at 11,25 % which amplifies the nominal values of momentum
premiums. However, as the statistical significance of these results is very strong (as I will present
next), I argue that this marginal difference does not affect the credibility of the results.
The t-statistic for the long/short -portfolio’s excess-returns is 4,20 and for the long-only portfolio
the t-statistic is 4,33. The probability for the excess-returns being only a result of randomness is
less than 0,003 % for the long/short-portfolio and less than 0,001 % for the long-only portfolio.
Momentum was discovered around the same time as value so for comparability, I will use the
same starting date for the short-term calculations as with value, the July 1992. The long/short-
portfolio still performed well as the average annual return was 16,41 % and the excess-return
compared to total market was 5,55 %. The long-only portfolio generated average annual return of
13,48 % and excess-return of 2,62 %. However, as with value, neither of the excess-returns were
statistically significant at 5 % significance (t-statistic of 2) as the t-statistics were 1,52 for
long/short-portfolio and 1,54 for the long-only portfolio.
Long/short-portfolio vs. total market
40
30
20
10
0
1927
1931
1935
1939
1943
1947
1951
1955
1959
1963
1967
1971
1975
1979
1983
1987
1991
1995
1999
2003
2007
2011
2015
-10
Chart 3. Long-short portfolio’s 10-year moving average (green) and total market portfolio’s 10-year
moving average (red)
The above chart visualizes the long/short-portfolio’s performance relative to the total market. I
smoothed both time-series by calculating the 10-year moving average to clear some of the high
volatility seen in chart 2. Interestingly, we can see that the momentum performance has faded
noticeably over the past years, but the poor performance begins relatively late after the
momentum performance was discovered. In fact, the excess-return even increases towards the
25
end of the century and then diminishes just when the financial crisis started to affect the markets
in 2007.
However, before financial crisis, the momentum premium has been very stable over the years and
changes in overall market performance have not affected the size of the premium. This suggests,
that the poor performance of momentum over past 10 years would not be due to information
effects or markets pricing the premium out. Jegadeesh and Titman in their subsequent article from
2001 come to the same conclusion and argue that the stable premium over 1990’s despite
multiple academic articles regarding the subject prove, that momentum premium is not simply a
result of data mining.
Summary of statistics
Tables 2.1 and 2.2 based on data series from January 1927 to September 2018
Based on momentum
Variance 36,2 30,8 64,0 34,2
Based on momentum
Tables 2.3 and 2.4 based on data series from July 1992 to September 2018
Based on momentum
Variance 20,0 19,0 44,7 26,2
Based on momentum
4.2.2 Logical reasons for momentum premium’s existence
26
Risk-based
When comparing statistics of value and momentum portfolios, one can clearly note an interesting
difference. The high-BtM value portfolios have significantly higher volatility than the market
portfolio which logically suggests higher returns. However, the top 30 % -momentum portfolio
based on past performance has roughly the same volatility as the market portfolio. As a matter of
fact, the low 30 % -momentum portfolio has the highest volatility of all momentum portfolios
despite its relatively poor performance. I argue that this suggests that simply higher risk is not
behind the high historical premiums of momentum stocks.
One risk-based explanation to momentum is, however, the fact that momentum securities have
generally high expectations loaded on them. The good past performance might be due to
investors’ increased hopes for future profits and thus the tail-risk of these hopes not getting
fulfilled increases. This increased risk could justify the higher returns. I argue that this might cause
major losses of momentum strategies during financial downturns such as the one seen during
2007-2008 in chart 3.
Behavioral
Most academic researchers agree with this as momentum is mostly considered to be a behavioral
phenomenon, as noted by Berking and Swedroe (2016). Simple behavioral explanation is that
momentum is caused by slow reaction of the markets. To recall, efficient markets -hypothesis
states that any new relevant information about a security causes its price to react immediately. It
is true that many professional investors might react almost immediately when new information
arises but there are still many non-professional, or so-called uninformed investors, who do not
follow the news as accurately. This causes the price to react slower than what one could expect
based on efficient markets -hypothesis.
This slow reaction causes temporary over- and undervaluation of securities. The reason why the
financial markets do not reverse these anomalies is their relatively slow development. Zhi Da,
Umit Gurun and Mitch Warachka (2013) compared how investors react to slow price
developments relative to fast price movements. Their study suggests that momentum is simply
caused by uninformed investors not reacting to small and gradual information signals even though
larger moves cause appropriate price movements. This leads to short-term excess-returns
captured by momentum portfolios.
27
Is momentum still viable strategy despite the recent underperformance?
As I mentioned, it does not seem like markets have simply managed to price out the momentum
premium due to increased knowledge about it. Multiple publications have been made already in
the 1990’s about momentum stocks’ good performance yet the premium remained stable over
the decade. I analyzed a recent explanation by Kent Daniel and Tobias Moskowitz (2016). They
argue that momentum strategies tend to fail when the market volatility is high which leads to
panic in the markets and major economic downturns, and the poor performance after such
downturn is explained by relative betas of the long and short positions within the portfolio. For
example, during a bull market when stock prices are rising, the stocks that are bought to the
portfolio generally have high betas. This is because the entire stock market is rising, and the best
performers just simply rise even more than the average. Correspondingly, the stocks that are sold
short have relatively low betas.
When the stock market then crashes, the high beta stocks tend to crash even harder than the
market. When the portfolio is rebalanced after the downturn, the best performers during the
downturn were the stocks with lowest betas because they suffered the least from the market
crashing. This leads to low beta stocks getting bought to the portfolio even though they might not
be the winners during the next bull market that follows the downturn. Thus, the momentum not
only crashes harder than the stock market during downturns, but the poor performance also lasts
long afterwards until the portfolio finally gets properly rebalanced. I studied this by calculating
timeseries of portfolio betas for both top 30 % and worst 30 % portfolios based on momentum
from beginning of year 2000 to September 2018. I calculated rolling covariances and market
variances based on previous 100 days to illustrate the changes in the portfolio betas, based on
daily data. This is somewhat shorter time period than Daniel’s and Moskowitz’s 126 days, but
shorter time period illustrates the effect more clearly.
28
Portfolio betas
2
1,8
1,6
1,4
1,2
1
0,8 long
0,6 short
0,4
0,2
0
20020131
20060403
20000525
20010327
20021129
20031001
20040803
20050602
20070202
20071203
20081002
20090804
20100604
20110404
20120202
20121204
20131004
20140806
20150608
20160407
20170206
20171205
Chart 4. Rolling observations of betas of long (blue) and short (orange) portfolios (top and bottom
30 % based on momentum) from January 2000 to September 2018.
As we can see, the beta of the long portfolio has dropped significantly first during the Dotcom-
bubble in 2000-2002 and then during the financial crisis in 2007-2009. This suggests that the
rebalancing of betas might at least partly explain the momentum’s recent poor performance and
that the momentum premia might became more evident again in the future.
5. Combining factors to increase risk-adjusted returns
In this section, I will combine the two previously presented factors. My goal is to optimize
portfolios which have the highest return relative to their risk. This is the key difference between
the previous section 4 and the section 5 as previously I have only calculated the portfolios’
absolute performance. I will first combine only the factors, and second, I will combine the
optimized factor portfolio with market portfolio to create an example of a factor portfolio which
could easily be implemented also by individual investors. Last, I will examine possibilities of factor
diversification during periods when factors are performing poorly relative to the market.
5.1 Combining two factors
The previous statistics have measured absolute excess-returns of value and momentum strategies.
All factor strategies rely on good diversification as explained earlier. The factors, however, present
29
also a new type of diversification, which deals with combining different factor strategies to a single
portfolio. Traditionally, diversification has been done across different securities and asset classes
as well as geographically. Different markets react differently to varying economic conditions and
thus the portfolio’s risk can be decreased by diversifying.
Recall, that factors, both the ones backed by behavioral and risk-based explanations, are sources
of increased systematic risk. Traditional diversification across securities, asset classes and
geographical markets is also diversification across different sources of systematic risk.
And as with traditional diversification, we can observe negative correlation between factor risks,
which opens up opportunities for diversifying across the factor risks. For example, I previously
argued that value premium has been performing poorly during bull markets and momentum has
performed poorly during economic downturns. This implies, that there might be negative
correlation between the two factors.
Correlation between two datasets can be measured with Pearson’s correlation coefficient, which
is calculated by dividing the covariance of the two datasets with the product of their respective
standard deviations.
𝐶𝑜𝑣 (𝑥, 𝑦)
𝑃𝑒𝑎𝑟𝑠𝑜𝑛′ 𝑠 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 𝑐𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 =
𝜎(𝑥) ∗ 𝜎(𝑦)
The coefficient varies between -1 and 1. A positive value implies positive correlation and vice
versa. In this case, negative correlation implies that the two portfolios react differently to the
same economic events affecting the market. And indeed, the correlation coefficient between the
excess-returns of momentum and value portfolios (deducted of risk-free return) is negative -0,43.
I calculated 11 portfolios with varying weights between value and momentum to study the
development of volatility and the average annual returns. To measure the risk-adjusted returns, I
calculated the Sharpe Ratios for all portfolios. To calculate the optimal portfolio weights with the
highest risk-adjusted return, I created a two-variable optimization model and ran it with Excel’s
GRG Nonlinear solver engine. The goal was to maximize the Sharpe Ratio of the optimal portfolio.
30
Value Momentum 50/50 Optimal
Variance 16,177 29,514 6,710 6,312
Average return (%) 4,094 8,219 6,156 6,007
Sharpe Ratio 1,018 1,513 2,377 2,391
Table 3.1 Statistics for Value, Momentum, 50 %/50 % equal-weighted and optimal portfolios.
Out of the single-factor portfolios, momentum clearly produces the best risk-adjusted returns with
Sharpe Ratio of 1,513. As noted earlier, it also produces highest absolute excess-return of these
portfolios with annual average of 8,2 %. The equal-weighted and the optimal portfolio, however,
produced clearly superior Sharpe Ratios. Despite the apparent superiority of momentum, the
optimal portfolio’s weights were actually in favor of value with roughly 53,6 % value and 46,4 %
momentum.
To illustrate this, I plotted the 11 portfolios (100 % value and 0 % momentum, 90 % and 10 % and
so on) along with the highlighted optimal portfolio. The so-called efficient frontier, where the
returns cannot be increased without increasing portfolio volatility, is highlighted with red. The y-
axis shows the absolute average return and it is compared to the portfolio’s standard deviation on
the x-axis. The lowest point on the y-axis is naturally the portfolio with 100 % value as value has
the lowest absolute return. The momentum’s weight is then increased in 10 % intervals.
Efficient frontier
9
8
Average return (%)
3
1 2 3 4 5 6
Standard deviation
31
Chart 5. Plotted returns of combined value/momentum -portfolios with highlighted efficient
frontier
As the momentum weight is increased, the portfolio’s volatility decreases while the average return
increases at the same time. The effect is very similar as when comparing to simple stocks except
that now I am comparing two portfolios that are already well diversified across the equity market.
This proves that both value and momentum portfolios’ risk-adjusted returns can be improved by
diversifying across factors even if their absolute returns do not increase.
5.2 Combining factor risk with market equity risk
When I calculate the long/short-portfolio’s excess-returns, I effectively isolate the excess-risk

generated by the factor from the equity market risk. The equity market risk is simply the
systematic risk of investing in the total market and it is the source for the total market returns. The
factors’ excess-returns is based on increased systematic risk generated by the factor. This means,
that we can further diversify our portfolio across different types of risk. This is especially
emphasized by Asness et al. (2016) who mention that long/short -portfolios generate risk
premium that has low or negative correlation with market equity risk premium. When comparing
the optimized factor portfolio to market portfolio, I once again noticed low correlation coefficient
of -0,001 which implies possibility for diversification benefits.
Brief example of isolating the market equity risk from factor risk
32
Chart 6. An example of isolating factor premium (red + striped blue) from market premium (blue +
striped blue). The percentages represent the value portfolio of section 4.1
To illustrate isolation of factor risk from market risk, I use a simple bar chart with percentages
from the value portfolio of section 4.1. The market premium is marked with blue (including striped
blue), and the factor risk exposure from taking the long position with value stocks is marked with
red (2,5 %). If we would buy the value stocks and sell short the market, we would isolate the
premium marked with red. If we sell short only the low Book-to-Market stocks, we isolate the
market risk and gain increased exposure (4,1 %) to the factor risk which is marked with striped
blue. The portfolio which combines only the red and striped blue sections of the chart does not
contain any market risk and correlates negatively with market risk. After the isolation, we can then
look out for negative correlation with other types of risk premia to gain increased risk-adjusted
returns by optimizing.
Optimizing the balance of factor risk and market risk
In section 4.3 I diversified the portfolio across two different types of factor risk, value and
momentum. Now I recombine that portfolio with total market portfolio to generate the highest
possible risk-adjusted returns. I want to emphasize that this kind of diversification strategy could
very easily be implemented by any individual investor by investing in factor-based Exchange
Traded Funds with weights that I will present next.
Again, I optimized the portfolio weights with Excel and present a 50 % / 50 % portfolio as an
example. The following chart 5 plots Sharpe Ratios for all portfolios combining the factor and
market portfolio with varying weights using 10 % intervals. This time the factor portfolio is the
“Optimal” portfolio presented in section 4.3 with roughly 53,6 % value and 46,4 % momentum.
Factors Market 50/50 Market + Factor Optimum

Variance 6,311917 33,86712 9,972909 5,870644317
Standard Deviation 2,512353 5,819546 3,157991 2,422941253
Average Return 6,007304 12,51738 9,262342 7,830995662
Sharpe Ratio 2,391107 2,15092 2,932985 3,232020443
Table 3.2 Statistics for Factor, Market, 50 % / 50 % and Risk optimized portfolios
33
Sharpe ratios of combined portfolios
3,5
3,3
3,1
2,9
Sharpe Ratio
2,7
2,5
2,3
2,1
1,9
1,7
1,5
0 10 20 30 40 50 60 70 80 90 100
Market portfolio weight (%)
Chart 7. Sharpe Ratios with varying portfolio weights for portfolios combining market and factors,
optimum market with orange
As we can see, the new optimal portfolio has lower average return than the market, but it
generates highly superior risk-adjusted returns with Sharpe Ratio of 3,23 which is by far the
highest of all portfolios presented so far in this thesis. The optimum weights were around 72,0 %
factors and 28,0 % market, which leads to final weights of 38,6 % for value and 33,4 % for
momentum. This is a simple passive investment strategy which has good diversification across
stocks but also different sources of risk, a great example of strategies Asness et al. (2016)
described in their papers. And most importantly, it could very easily be replicated by any investor
even with low investment capital.
5.3 Factor risk-optimization from 2010-2018
As noted a number of times, both value and momentum have performed poorly since the financial
crisis hit in 2007. It seems clear, that diversifying across both factors and market has been the
superior strategy when comparing risk-adjusted returns in the long historical perspective. But
would similar strategy have been a good choice also, when the absolute performance of the
factors has been historically low? I studied this by optimizing all three variables (value, momentum
and market) from January 2010 to September 2018 as this represents a time period of historically
strong market performance and poor factor performance.
34
Market Optimal
Variance 13,27059 6,658824
Standard Deviation 3,642882 2,58047
Average Return 13,69458 10,96822
Sharpe Ratio 3,759272 4,250473
Table 3.3 Statistics for market portfolio and risk-optimized portfolios over 2010-2018
Surprisingly, simply buying the market has not been the best the strategy despite the poor
absolute performance of factors over this time period. Market has indeed been performing
exceptionally well, with Sharpe Ratio of 3,76 (compared to 2,15 in 1927-2018) with absolute
annual average return of roughly 13,7 %. I argue that this also partly explains the relatively poor
performance of factors. The factor risks are not correlated with market equity risk and thus the
strong period of market performance does not necessarily lead to high factor returns. In fact,
market has been performing so strongly, that the average factor performance in the long run is
worse than the recent performance of market, which leads to diminishing of factor premiums.
However, the optimal portfolio still calls for only 68,4 % of market weight and 31,6 % of
momentum weight despite the relatively poor performance of momentum seen in Chart 3, as this
has generated an outstanding Sharpe Ratio of 4,25. I argue that this is explained by the negative
correlation between momentum risk and market equity risk. The key finding here is the use of
factor risks as hedging mechanisms against market portfolio’s volatility. Even when the factors are
not outperforming the market and they are not generating absolute excess-returns, exposure to
factor risks can help to diversify the portfolio risks and thus generate higher risk-to-return ratios as
seen in Table 3.3.
6. Conclusions and topics for further research
I showed that both value and momentum factors have generated notable excess-returns in the
past and there are several logical reasons for their persistence also in the future. I also provided an
example of how diversifying across factors can improve the return to risk -ratio even with
portfolios that have already been diversified across the market. The excess-returns have been
statistically significant with very high significance levels from 1927 to this date.
35
The major concern regarding investing in factors is market pricing the premiums out as the
strategies become more popular or as informed investors attempt to arbitrage them. However,
there seems to be little evidence that the publications of literature revealing the factors would had
affected the premiums. For example, the first major crash in momentum premia happened around
15 years after the initial research publications.
Despite this, both value and momentum strategies have been performing poorly over the past
years. I found evidence, that value strategies’ poor performance is explained by general poor
performance during strong bull markets. Similarly, I argue that post-financial crisis
underperformance of momentum strategies is explained at least partly by time variability of betas
in the momentum portfolio.
Despite diminishing factor premiums, I showed that diversifying across factors can provide higher
risk-adjusted returns even when the absolute factor performance is low. Thus, diversifying across
factors can act as a hedging mechanism against market equity risk and I argue that this leads to
practical implementations of factors even if their absolute performance is lower in the future.
Using factors for hedging could be studied more deeply in future research.
Whether the underperformance will persist in the future remains to be seen. In the future, this
study could be expanded by adding more factors and especially by doing more analysis with the
performance of multi-factor portfolios. The portfolios could also be more diversified across asset
classes and geographical areas. Also, the list of behavioral explanations for these market
anomalies will most likely never be fully completed as their effect is difficult to measure
accurately.
36
List of references:
Andrew L. Berkin and Larry E. Swedroe: Your Complete Guide to Factor-based Investing, The Way
Smart Money Invests Today (2016)
Antti Ilmanen: Expected Returns, An Investor’s Guide to Harvesting Market Rewards (2011)
Barr Rosenberg, Kenneth Reid, and Ronald Lanstein: Persuasive evidence of market inefficiency,
Journal of Portfolio Management (vol. 11, 1985, p. 9-17)
Clifford Asness, Antti Ilmanen, Ronen Israel and Tobias J. Moskowitz: Investing with Style, Journal
of Investment Management (vol. 13, no. 1, 2015, p. 27-63)
Clifford S. Asness, Tobias J. Moskowitz and Lasse Heje Pedersen: Value and Momentum
Everywhere, The Journal of Finance (vol. 68, No. 3, June 2013, p. 929-985)
David M. Blanchett & Craig L. Israelsen: Spotlighting Common Methodological Biases in Active-Vs.-
Passive Studies (11/2007), Journal of Financial planning
Dennis Stattman: Book values and stock returns, The Chicago MBA: A Journal of Selected Papers
(vol. 4, 1980, p. 25-45)
Eugene Fama: Random Walks in Stock Market Prices, Financial Analysts Journal (September-
October volume 1965)
Joseph D. Piotroski and Eric C. So, Review of Financial Studies (March 2012)
Kees G. Koedijk, Alfred M.H. Slager, Philip A. Stork: Investing in Systematic Factor Premiums,
European Financial Management (vol. 22, March 2016, p. 193-234)
Kenneth French and Eugene Fama: The Cross-Section of Expected Stock Returns, The Journal of
Finance (vol. 47, June 1992, p. 427-465)
Kent Daniel and Tobias J. Moskowitz: Momentum Crashes, Journal of Financial Economics (vol.
122, 2016, p. 221-247)
Lu Zhang: The Value Premium, The Journal of Finance (vol. 60, issue 1, February 2005, p. 67-103)
Mark Carhart: On Persistence in Mutual Fund Performance, The Journal of Finance (vol. 52, Issue 1,
March 1997, p. 57-82)
37
Nai-Fu Chen and Feng Zhang: Risk and Return of Value Stocks, The Journal of Business (vol. 71, n.
4, October 1998, p. 501-535)
Narasimhan Jegadeesh and Sheridan Titman: Profitability of Momentum Strategies: An Evaluation

of Alternative Explanations, The Journal of Finance (vol. 56, Issue 2, April 2001, p. 699-720)
Narasimhan Jegadeesh and Sheridan Titman: Returns to Buying Winners and Selling Losers:
Implications for Stock Market Efficiency, The Journal of Finance (vol. 48, No. 1, March 1993, p. 65-
91)
Peter Nyberg and Mika Vaihekoski: Equity premium in Finland and long-term performance of the
Finnish equity and money markets, (3/2013)
Rob Arnott, Noah Beck, Vitali Kalesnik & John West: How Can “Smart Beta” Go Horribly Wrong?
(February 2016)
Robert Clarke, Harindra de Silva, Steven Thorley: Fundamentals of Efficient Factor Investing,
Financial Analysts Journal (vol. 72, November-December 2016, p. 9-26)
Robert F. Peterkort and James F. Nielsen: Is The Book-To-Market Ratio A Measure Of Risk? The
Journal of Financial Research (vol. 28, issue 4, December 2005, p. 487-502
Ron Bird, Xiaojun Gao and Danny Yeung: Time-series and cross-sectional momentum strategies
under alternative implementation strategies, Australian Journal of Management (vol. 42, issue 2,
May 2017, p. 230-251)
Stewart C. Myers and Nicholas S. Majluf: Corporate Financing and Investment Decisions When
Firms Have Information That Investors Do Not Have, NBER Working Paper series No. 1396 (July
1984)
Zhi Da, Umit G. Gurun, Mitch Warachka: Frog in the Pan: Continuous Information and Momentum
(December 2013)
38

Factor Based Investing

Uploaded by

Copyright:

Available Formats

Factor Based Investing

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Factor Based Investing

Uploaded by

Copyright:

Available Formats

Factor-based Investing

Analysing market anomalies in the US

Approved in the Department of Information and Service Management

1.2 Research questions and the scope of this thesis

1.3 Introduction to differences between active and passive investment strategies

Why diversification is important?

𝜎(𝑟) = √𝜎 2 (𝑟) = √∑ 𝑝𝑖 [𝑟𝑖 − 𝐸(𝑟)]2

1.4 Factor investing as a method to increase passive investment returns

2. Review of literature explaining the key characteristics of factors

2.1 Criteria for true factors

2.2 Differences between risk-based and behavioral explanations for mispricings

2.2.1 Understanding systematic risk as a source of higher profits

2.2.2 Why is it important to recognize the reason for factor’s existence?

3. Methodology: Constructing factor portfolios

3.1 Short positions in factor portfolios

Differences between long-only and long/short-portfolios

Which portfolio-style should be used?

3.2 Measuring factor performance

4.1.1 Expected returns of value stocks

Are the excess-returns statistically significant?

Returns after publication of Fama and French

Based on Book-to-Market -ratio

Based on Book-to-Market -ratio

Based on Book-to-Market -ratio

4.1.2 Logical reasons for value premium’s existence

4.2.1 Expected return of momentum stocks

High vs. low annual difference in returns of momentum

Long/short-portfolio vs. total market

4.2.2 Logical reasons for momentum premium’s existence

5. Combining factors to increase risk-adjusted returns

5.1 Combining two factors

5.2 Combining factor risk with market equity risk

When I calculate the long/short-portfolio’s excess-returns, I effectively isolate the excess-risk

Optimizing the balance of factor risk and market risk

Factors Market 50/50 Market + Factor Optimum

5.3 Factor risk-optimization from 2010-2018

6. Conclusions and topics for further research

Narasimhan Jegadeesh and Sheridan Titman: Profitability of Momentum Strategies: An Evaluation

You might also like