Portfolio Optimization Methods, Their Application and Evaluation

Tomas Hlavaty

Dissertation submitted as partial requirement for the conferral of Master in Finance

Supervisor:
Prof. Sebestyén Szabolcs, Assistant Professor, ISCTE Business School, Department of Finance

September 2018
Portfolio Optimization Methods, Their Application and Evaluation
Tomas Hlavaty
António Augusto da Silva
- Spine -
Abstract
The submitted master’s thesis focuses on practical application of quantitative portfolio
optimization in various forms. The thesis is organized in two main parts, theoretical and practical.

The theoretical part introduces the underpinnings of portfolio theory. It describes the
optimization process, introduces a number of selected optimization methods, and provides an
overview of portfolio management. As a whole, it serves as an underlying for the practical part.

The practical part of the thesis is based on an experiment that put multiple quantitative
portfolio optimization methods into a contest. Different optimizers were applied to portfolios
composed of identical assets, which were subsequently held under different portfolio management
styles over a pre-specified period of time. The performance of each portfolio was measured ex-
post, adequately evaluated in accord with the criteria of the experiment, and confronted with the
others.

The questions that this master’s thesis tried to find answers to were (1) which portfolio
optimizer, out of the selected ones, performs the best, and (2) whether it is beneficial to conduct
rather an active, or a passive portfolio management.

Keywords: Quantitative Portfolio Management, Optimization, Asset Allocation, Diversification

JEL Classification: C610, G110

Resumo
Esta dissertação de mestrado apresenta uma aplicação prática da otimização quantitativa
de um portfólio realizada de diversas formas. A tese está organizada em duas partes principais,
uma teórica e uma prática.

A parte teórica introduz os fundamentos da teoria de portfólio. Descreve o processo de

otimização, apresenta vários métodos de otimização selecionadas e fornece uma visão geral da
gestão de portfólios. Como um todo, serve como base para a parte prática.

A parte prática da tese coloca vários métodos de otimização de portfólio quantitativos em

competição. Diferentes optimizadores foram aplicados a carteiras compostas por ativos idênticos
que foram subsequentemente mantidos sob diferentes estilos de gestão ao longo de um período de
tempo pré-especificado. O desempenho de cada carteira foi medido ex-post, adequadamente
avaliado de acordo com os critérios de otimização e comparado com as demais carteiras.

As perguntas para as quais esta tese de mestrado tentou encontrar respostas foram (1) qual
é o optimizador de portfólio, dentre os selecionados, tem o melhor desempenho e (2) se é benéfico
conduzir uma gestão de portfólio muito ativa ou passiva.

Palavras-chave: Gestão Quantitativa de Portfólios, Otimização, Alocação de Ativos,

Diversificação

JEL Classificação: C610, G110

Acknowledgement

I would like to express my gratitude to professor Sebestyén Szabolcs, PhD for supervision
of this master‘s thesis. His valuable advices, suggestions, and patience significantly contributed to
successful completion of this work.

I also would like to express my deepest thankfulness to my mother, Ing. Jaroslava Alice
Hlavata, for her endless support during my entire studies.
Table of Contents

1. Introduction ............................................................................................................................. 1

2. Portfolio Theory ...................................................................................................................... 3

2.1. Time Dimension ............................................................................................................... 3

2.2. Risk Dimension ................................................................................................................ 5

2.3. Modern Portfolio Theory ................................................................................................. 9

2.4. Efficient Frontier ............................................................................................................ 13

3. Return Generating Models .................................................................................................... 16

3.1. Capital Asset Pricing Model .......................................................................................... 16

3.2. Market Model ................................................................................................................. 18

4. Portfolio Optimization .......................................................................................................... 22

4.1. Mean-Variance Optimization ......................................................................................... 22

4.2. Treynor-Black ................................................................................................................ 32

4.3. Black-Litterman ............................................................................................................. 34

4.4. Naïve Optimization ........................................................................................................ 37

5. Portfolio Performance ........................................................................................................... 40

6. Portfolio Management .......................................................................................................... 45

7. Practical Experiment ............................................................................................................. 49

7.1. Optimal Portfolios .......................................................................................................... 51

7.2. Results ............................................................................................................................ 54

7.3. Discussion ...................................................................................................................... 64

8. Conclusion ............................................................................................................................ 67

Bibliography ................................................................................................................................. 69

Annex ............................................................................................................................................ 71
Table of Figures

Figure 1: Efficient frontier for risky assets ................................................................................... 15

Figure 2: Different CALs for portfolios composed of risk-free and risky assets ......................... 16
Figure 3: Optimal portfolios for agents with different risk aversion ............................................ 30
Figure 4: Ex-ante portfolio efficiency........................................................................................... 54
Figure 5: EAR ............................................................................................................................... 55
Figure 6: Standard deviation ......................................................................................................... 56
Figure 7: Residual SD ................................................................................................................... 56
Figure 8: Jensen's alpha ................................................................................................................ 57
Figure 9: Beta ................................................................................................................................ 57
Figure 10: Sharpe ratio.................................................................................................................. 58
Figure 11: Treynor ratio ................................................................................................................ 58
Figure 12: Information ratio.......................................................................................................... 59
Figure 13: M^2.............................................................................................................................. 59
Figure 14: Final ranking................................................................................................................ 63
Figure 15: Portfolio value development ....................................................................................... 66
Table of Tables

Table 1: Values of β ...................................................................................................................... 20

Table 2: Values of A ..................................................................................................................... 28
Table 3: Optimal relative asset allocation suggested by the models ............................................ 52
Table 4: Optimal absolute asset allocation suggested by the models ........................................... 53
Table 5: EAR ................................................................................................................................ 55
Table 6: Standard deviation .......................................................................................................... 56
Table 7: Residual SD .................................................................................................................... 56
Table 8: Jensen's alpha .................................................................................................................. 57
Table 9: Beta ................................................................................................................................. 57
Table 10: Sharpe ratio ................................................................................................................... 58
Table 11: Treynor ratio ................................................................................................................. 58
Table 12: Information ratio ........................................................................................................... 59
Table 13: M^2 ............................................................................................................................... 59
Table 14: Portfolio rankings by categories ................................................................................... 62
Table 15: Final point chart ............................................................................................................ 63
Table 16: Final ranking ................................................................................................................. 63
Table 17: Final values ................................................................................................................... 66
1. Introduction
“An investment in knowledge pays the best interest”. A famous quote pronounced by
Benjamin Franklin, an American polymath and one of The Founding Fathers of the United States,
is pointing clearly to where every person should invest in the first place. Learning is a never-ending
process and shall we look at the profiles of the greatest from the greatest, regardless the field of
interest, they have never considered their education as complete. The more we invest into our
mind, the higher the return we may expect to come back to us in the future in various forms. Our
knowing can be seen as a portfolio of knowledge. Portfolio, which we build in more or less
constructive matter.

Our knowing deeply influences the way we approach, understand, handle, and reflect all
of the challenges we encounter. Based on our obtained experience and our knowing, we derive
theories and philosophies. The investment philosophy has alike origin and, as well as other
philosophies, is a subject of evolution. Swensen (2000) describes the investment philosophy as a
coherent approach being applied consistently to all aspects of portfolio management process. In
his eyes, the philosophical principals represent time-tested insights into investment matters that
eventually evolve into lasting professional convictions. The investor’s effort to find the most
effective way to generate investment returns emanates from those convictions and fundamental
beliefs. The investment returns are seen as a product of three tools of portfolio management: (1)
asset allocation, (2) security allocation, and (3) market timing. Sophisticated investor then
considers contribution of each of the portfolio management tools to costruct portfolios in a
conscious manner.

The core focus of this thesis is on the problematics of the first tool of portfolio management,
the asset allocation and the various approaches to it. Asset allocation is often understood as the
second step of the investment process with the first one being the determination of investor’s
investment objectives, time preferences, and risk profile. Various empirical studies over the time
have shown that asset allocation has the biggest influence on an overall portfolio performance.
Asset allocation represents spreading the investor’s investment capital across various asset classes
such as stocks, bonds, derivatives, properties, commodities, funds, cash etc. in order to achieve a
diversified portfolio.

1
 The thesis is primarily structured in two main parts, theoretical and practical. The
theoretical part introduces the underpinnings of portfolio theory. It describes the optimization
process, introduces a number of selected optimization methods, and provides an overview of
portfolio management. As a whole, it serves as an underlying for the practical part.

The practical part of the thesis is based on an experiment that puts multiple quantitative
portfolio optimization methods into a contest. Different optimizers are applied to portfolios
composed of identical assets, which are subsequently held under different portfolio management
styles over a pre-specified period of time. The performance of each portfolio is measured ex-post,
adequately evaluated in accord with the criteria of the experiment, and confronted with the others.

The questions that this master’s thesis tries to find answers to are (1) which portfolio
optimizer, out of the selected ones, performs the best, and (2) whether it is beneficial to conduct
rather an active, or a passive portfolio management.

2
2. Portfolio Theory
This chapter introduces the fundamentals of portfolio theory, and describes some of the
essentials regarding the portfolio optimization frameworks.

In a wide interpretation, any portfolio can be viewed as a set of various items. Those items
are acquired by the portfolio’s owner with accordance to his needs, wants, preferences,
predispositions, possibilities, and/or expectations. From the financial perspective, such portfolio is
understood as a set of financial assets1. The finance universe has two main underlying dimensions:
the time dimension and the risk dimension.

2.1. Time Dimension

The importance of time in finance is absolutely crucial and is best explained by the concept
of the time value of money (TVM). TVM concerns equivalence between cash flows occurring at
different times. One dollar today has a higher value than one dollar one year from now. However,
the same dollar invested today grows in value over time. Formally, the value of one dollar invested
today grows accordingly to its interest rate (i). An interest rate is a rate of return that reflects the
relationship between differently dated cash flows. When considering the time dimension, we
distinguish between the discrete and the continuous time. Consequently, when considering the
interest rates under the time dimension framework, we distinguish between the discrete and the
continuous interest rates. Therefore, the TVM allows us to either obtain the future value
(compounding) or the present value (discounting).

2.1.1. Discrete Time

The discrete time considers a variable to occur at distinct, separate points in time. For
instance, monthly, quarterly, yearly. The discrete interest is then being accrued to the principal2 at
those separate points in time. The discrete interest has characteristics of a discrete random variable.

 Discrete compounding
- the future value (FV) of $1 invested for n years at the interest rate I compounded
once per annum

1
Financial asset – a tangible liquid asset which gets its value from a contractual claim
2
Principal - the amount of money originally invested

3
 𝐹𝑉 = (1 + 𝑖)𝑛 (2.1)

- FV of $1 compounded m-times per annum for n years

𝑖 𝑚𝑛
𝐹𝑉 = (1 + ) (2.2)
𝑚

 Discrete discounting
- the present value (PV) of a future $1 discounted for n years

1
𝑃𝑉 = (2.3)
(1 + 𝑖)𝑛

- PV of a future $1 discounted m-times a year

1
𝑃𝑉 = (2.4)
𝑖
(1 + 𝑚)𝑚𝑛

2.1.2. Continuous Time

The continuous time also considers the variable to occur at specific points in time.
However, and in contrast to the discrete time, between two points in time there is an infinite
number of other points of time and the distance between them is converging to zero. Thus, the
continuous time sees the variable to occur continuously. The continuous interest does have
characteristics of a continuous random variable.

The effect of continuous time, when m → ∞, on compounding can be expressed formally

using limits as following:

𝑖 𝑚𝑛
lim (1 + ) = 𝑒 𝐼𝑛 (2.5)
𝑚→ ∞ 𝑚

4
  Continuous compounding
- FV of $1 continuously compounded for n years

𝐹𝑉 = 𝑒 𝑖𝑛 (2.6)

 Continuous discounting
- PV of $1 discounted with continuously compounded i

𝑃𝑉 = 𝑒 −𝑖𝑛 (2.7)

2.2. Risk Dimension

The risk dimension relates closely with the time dimension. The time dimension
distinguishes, in its fundamental property, between two points in time. For instance, between the
presence and the future. The presence is well known. However, the future is, from its very nature,
uncertain and thus risky. The riskiness, in such case, is the potential deviation from our expectation
about the future. Taking the perspective of an investor, the realized return from his investment at
the end of the period may be different from the expected one. Those deviations can be both positive
or negative. Logically, only the negative deviations are considered undesirable. The most
widespread used measure of riskiness in finance is the square root of variance, the standard
deviation (SD). The use of SD is convenient under a number of simplifying assumptions, e.g.
normal distribution of returns. Also, SD is preferred over variance due to its equality of units with
returns as they both are expressed in percentages. Another risk measures are, for example, beta,
semi-variance, value-at-risk (VaR), or expected shortfall (ES)

Having mentioned the term expectation, it is convenient to briefly introduce the concept of
utility functions, expected utility criterion, and risk aversion. These play a crucial role in the
investment decision making process.

2.2.1. Utility Functions

Utility functions are mathematical representations of investor’s attitudes toward risk and
return. It is a function that assigns a utility value to all possible investment outcomes. The utility
value measures the degree of individual satisfaction the investor receives from any specific level

5
of wealth (W). Such satisfaction may come from several sources, typically from the additional
consumption of goods and/or services that the investor can enjoy after selling the investment
portfolio. Some investors, on the other hand, may enjoy the satisfaction derived from excitement
of the investing game itself (Esch et al., 2005).

The utility functions in general have four important properties:

 Utility is increasing and is always positive

 Marginal utility is decreasing

𝑢′ (𝑊1 ) < 𝑢′ (𝑊0 ) 𝑤ℎ𝑒𝑛 𝑊1 > 𝑊0 (2.8)

 Utility functions are strictly concave

𝑈 ′′ < 0 (2.9)
 Different investors have different utility functions
 A utility function, for a given investor and for a given time, is not unique

Examples of the most common utility functions

 Log utility

𝑢(𝑊) = ln(𝑊) (2.10)

 Exponential utility

𝑢(𝑊) = 1 − 𝑒 −𝑣𝑊 (2.11)

 Power utility

𝑊 1−𝛾
𝑢(𝑊) = (2.12)
1−𝛾
 Quadratic utility

𝑏 2
𝑢 (𝑊) = 𝑊 − 𝑊 𝑏>0 (2.13)
2
Where W represents the investor’s wealth.

Special cases of utility functions incorporating the risk aversion coefficients are introduced
in the following Section 2.2.2.

6
 2.2.1. Expected Utility Criterion
The expected utility (EU) criterion provides a framework for when an individual must
make a decision under uncertainty. That being not knowing which future outcome, out of a set of
possible outcomes, is going to result from a decision made today. In such situation, one will make
decision that offers the highest expected utility. Each outcome is assigned a probability of
occurrence. Thus, the EU is a probability-weighted average of utilities over all possible outcomes
and over a specific period of time. The final decision also depends on one’s risk aversion. To
demonstrate a general case, let’s consider a lottery L(x,y,π), where outcome x has a probability of
occurrence π and y with (1-π). The expected utility is following:

𝐸[𝑈(𝐿)] = 𝜋𝑢(𝑥) + (1 − 𝜋)𝑢(𝑦) (2.14)

The above equation is referred to as the von Neumann-Morgernstern (VNM) utility

function and represents the expected utility criterion. Let’s adjust the VNM with respect to asset
returns. Let’s consider an asset A at time 𝑡0 with possible future returns 𝑟1 , 𝑟2 , … , 𝑟𝑛 and with
probabilities of occurrence 𝜋𝑟1 , 𝜋𝑟2 , … , 𝜋𝑟𝑛 at time 𝑡1 , respectively. The expected utility can then
be expressed as following:
𝑛

𝐸[𝑈(𝑟𝐴 )] = 𝜋𝑟1 𝑢(𝑟1 ) + 𝜋𝑟2 𝑢(𝑟2 ) + ⋯ + 𝜋𝑥𝑛 𝑢(𝑟𝑛 ) = ∑ 𝜋𝑥𝑖 𝑢(𝑟𝑖 ) (2.15)
𝑖=1

The expected utility criterion confirms that the individual is concerned only with the final
payoffs and the cumulative probability associated with achieving them (Levy and Post, 2005).

2.2.2. Risk Aversion

An important fundamental property of the risk dimension under the portfolio theory
framework is the investor’s relation towards risks. His risk aversion. Every investor is different
and has a different perception of risks, or in other words, has a different degree of risk aversion.
Naturally, the desire of every investor is the same - to avoid risks, i.e. to smooth their consumption
across all states of nature and to avoid variations in the value of their portfolio holdings. The
investor is considered to be weakly risk averse when

𝑢(𝑊) ≥ 𝐸[𝑢(𝑊)] (2.16)

7
 the utility of current wealth is higher or equal than the expected utility of potential wealth
from a gamble. Strict risk aversion is represented by strict inequality (Beck, 2017).

The risk aversion is measured by the risk aversion coefficients:

 Absolute risk aversion coefficient

𝑢′′ (𝑊)
𝛼 (𝑊) = − (2.17)
𝑢′ (𝑊)

 Relative risk aversion coefficient

𝑊𝑢′′ (𝑊)
𝜌 (𝑊) = 𝑊𝛼(𝑊) = − (2.18)
𝑢′′ (𝑊)

 Risk tolerance coefficient

1 𝑢′ (𝑊)
𝜏 (𝑊) = = − ′′ (2.19)
𝛼 (𝑊) 𝑢 (𝑊)

Utility functions incorporating the risk aversion coefficients:

 Constant absolute risk aversion (CARA)

- Absolute risk aversion is constant at every wealth level
- CARA is a function of exponential utility function

𝑢 (𝑊) = − 𝑒 −𝛼𝑊 (2.20)

 Constant relative risk aversion (CRRA)

- Relative risk aversion is constant at every wealth level
- CRRA is a function of power utility function

𝑊 1−𝜌
𝑢 (𝑊) = (2.21)
1−𝜌

8
 2.3. Modern Portfolio Theory
Modern Portfolio Theory (MPT) is a financial portfolio construction theory developed by
professor Harry Max Markowitz (1952), which first introduced in his paper “Portfolio Selection”
published by the Journal of Science in 1952, and for which he was awarded the Nobel Memorial
Prize in Economic Sciences in 1990.

In Markowitz’s paper, the portfolio selection process is divided in two stages. The first
stage begins with observations and ends with some expectations regarding the future performance
of observed securities. The second stage begins with those expectations and ends with the choice
of final optimal portfolio. In this thesis, only the second stage is presented.

MPT, as well as any other theoretical concept, stands upon a number of underlying
assumptions. As such, it provides the groundwork for portfolio composition under the mean-
variance framework and for an arbitrary number of risky assets, with or without a risk-free asset.

MPT assumptions:
 Markets are perfectly efficient
 No transaction costs, no taxes
 Assets are perfectly divisible
 Risk-free asset is available to all investors
 Unlimited long and short positions are allowed
 Investors are rational and risk averse
 Utility function is quadratic
 Investors possess homogeneous investment motivation, horizon, and expectation
 Distribution of returns is Gaussian
 Investors are concerned only with asset’s mean and variance
 Investors desire to maximize their expected utility
 Investors always seek the maximum portfolio return for varying levels of risk

9
 2.3.1. Canonical Portfolio Problem
When an investor faces a portfolio creation decision he needs to make a choice regarding
budgeting the asset allocation. Specifically, how much of his capital is going to be allocated to,
and spread across, risky assets and how much to risk-free asset as an alternative to risky assets.
The future payoff from risky assets is uncertain. However, the payoff from risk-free asset is always
certain. The portfolio problem thus becomes the maximization of the expected payoff. That being
done in accord with investor’s utility function.

Canonical portfolio problem is well described by Danthine and Donaldson (2015). Let’s
consider a risk-free asset 𝑟𝑓 and a number of risky assets with returns 𝑟1 , 𝑟2 , … , 𝑟𝑖 . Also, a portion
of capital 𝑤𝑓 being invested into the risk-free asset and (1 − 𝑤𝑓 ) = ∑𝑛𝑖=1 𝑤𝑖 among n risky assets
with 𝑤𝑖 being the individual weight of each risky asset.

𝑤𝑓 + ∑𝑛𝑖=1 𝑤𝑖 = 1.

The maximization problem becomes following:

𝑛
𝑚𝑎𝑥
𝐸 {𝑢 [𝑤𝑓 (1 + 𝑟𝑓 ) + ∑ 𝑤𝑖 (𝑟𝑖 − 𝑟𝑓 )]} (2.22)
(𝑤1 , 𝑤2 , … , 𝑤𝑛 )
𝑖=1

Where the term (𝑟𝑖 − 𝑟𝑓 ) represents the risk premium of each risky asset. It is intuitive to
assume that the portion of capital invested in risky assets increases with increasing risk premium
and decreases when opposite. This intuition can be formally described via the first-order condition
(FOC). Under the risk aversion 𝑈 ′ ′() < 0, the FOC of the maximization problem becomes
𝑛 𝑛

𝐸 {𝑢′ [𝑤𝑓 (1 + 𝑟𝑓 ) + ∑ 𝑤𝑖 (𝑟𝑖 − 𝑟𝑓 )] ∑(𝑟𝑖 − 𝑟𝑓 )} = 0 (2.23)

𝑖=1 𝑖=1

The FOC allows to describe the relationship between the investor’s risk aversion and his
portfolio’s consumption via the following theorem, Equations 2.24-26, regarding the problem.

10
 For simplification, let’s substitute ∑𝑛𝑖=1 𝑤𝑖 = 𝑤𝑟 where 𝑤𝑟 represents the capital allocated
to risky assets, and 𝐸(𝑟𝑅 ) = ∑𝑛𝑖=1 𝐸( 𝑟𝑖 )𝑤𝑖 where 𝐸(𝑟𝑅 ) is the expected payoff of the risky assets
combined. Let’s assume 𝑈 ′ ′() < 0 and 𝑈 ′ () > 0.

𝑤𝑟 > 0 ↔ 𝐸(𝑟𝑅 ) > 𝑟𝑓 (2.24)

𝑤𝑟 = 0 ↔ 𝐸(𝑟𝑅 ) = 𝑟𝑓 (2.25)
𝑤𝑟 < 0 ↔ 𝐸(𝑟𝑅 ) < 𝑟𝑓 (2.26)

The theorem states that a risk averse investor is willing to make a risky investment if, and
only if, the expected payoff from the risky investment exceeds the risk-free rate. Or in other words,
if the odds are favorable and there is a positive remuneration from the additional risk accepted.

2.3.2. Mean-Variance Criterion

MPT considers mean and variance as sufficient information when making a choice
regarding the investment assets. It rests on the presumption that rational investors like the asset’s
returns and dislikes the return variance. The mean-variance criterion allows the investor to
compare various assets between each other while accounting the mean-variance preferences. Let’s
assume assets A and B, with 𝜇𝐴 , 𝜎𝐴 and 𝜇𝐵 , 𝜎𝐵 , respectively.

Portfolio A dominates portfolio B if, and only if, the following conditions are satisfied:

𝜇𝐴 ≥ 𝜇𝐵 and 𝜎𝐴 < 𝜎𝐵

Or equivalently,

𝜇𝐴 > 𝜇𝐵 and 𝜎𝐴 ≤ 𝜎𝐵

The mean-variance utility function describes the risk-return trade-off when reflecting the
investor’s degree of risk aversion. The function is based on two pivotal assumptions of MPT,
the quadratic utility function and the Gaussian distribution of returns. The quadratic utility is
important because it implies mean-variance preferences. The Gaussian distribution is attractive
due to its simplicity and properties. When considering only mean and variance, the expected
utility has following form:
11
 1
𝐸[𝑈(𝑟)] = 𝑢[𝐸(𝑟)] + 𝑢′′ [𝐸(𝑟)]𝑉𝑎𝑟(𝑟) (2.27)
2

To assure consistency with the risk-aversion assumption, the 𝑢′′ must be < 0 and hence the
positive sign changes to negative. Considering u to be quadratic in form described by Equation
2.13, setting b = 1, and including the investor’s risk aversion coefficient to reflect his
perception of risks, we arrive to the standard mean-variance utility function that can be found
across the literature and has following form:

1
𝑈 = 𝐸(𝑟) − 𝐴𝜎 2 (2.28)
2
Where E(r) is the expected return, 𝜎 2 is the variance, and A is investor’s risk aversion
coefficient.

12
 2.4. Efficient Frontier
Prior to the introduction of the minimum-variance frontier (MVF) and the efficient frontier
(EF), it is convenient to define the portfolio expected return and variance/SD in a matrix form, as
they are subsequently used throughout the thesis. Let’s consider a portfolio P of n risky assets,
with expected returns 𝑟1 , 𝑟2 , … , 𝑟𝑛 forming N x 1 vector of returns and portfolio’s assets weights
𝑤1 , 𝑤2 , … , 𝑤𝑛 forming N x 1 vector of weights. The set of covariances between the assets form the
variance-covariance matrix denoted Σ.

 Vector of expected returns

𝐸(𝑟)1
𝐸(𝑟)2
𝐸(𝑟) = 𝜇 = [ ]
⋮
𝐸(𝑟)𝑛
 Vector of portfolio weights
𝑤1
𝑤2
𝑤=[ ]
⋮
𝑤𝑛
 Variance-covariance matrix
2
𝜎11 ⋯ 𝜎1𝑛
𝛴=[ ⋮ ⋱ ⋮ ]
2
𝜎𝑛1 ⋯ 𝜎𝑛𝑛
 Portfolio expected return

𝐸(𝑟𝑝 ) = 𝜇𝑝 = 𝑤 𝑇 𝜇 (2.29)

 Portfolio variance and SD

𝑉𝑎𝑟(𝑃) = 𝜎𝑃2 = 𝑤 𝑇 𝛴𝑤 (2.30)

𝑆𝐷 = 𝜎𝑃 = √𝑤 𝑇 𝛴𝑤 (2.31)
Where the subscript T stands for transposed.

The essential notations being introduced, it is now possible to procced to the introduction
of both frontiers. Let’s consider a set of 𝑛 ≥ 2 assets. An infinite number of portfolios can be
formed from such set of assets. This creates a set of feasible portfolios, the feasible set. To evaluate
efficiency and compare the portfolios, an investor considers their expected returns and variances

13
with accordance to the mean-variance criterion. The investor will choose his or her optimal
portfolio from the feasible set, that

1. Offers highest expected return for varying levels of risk, and

2. Offers lowest level of risk for varying levels of expected returns

The above-mentioned conditions are known as the efficient set theorem (Sharpe, 1995).

The set of portfolios meeting the efficient set theorem are called the efficient portfolios and
graphically they plot the EF, which is part of the MVF. EF is the set of frontier portfolios where
each portfolio represents the portfolio with the highest expected return for varying levels of risk.
Both frontiers are conventionally plotted in an expected return-risk (μ-σ) space. The shape of EF
depends whether a risk-free asset is present or not.

2.4.1. Efficient Frontier for Risky Assets

Considering only the risky assets, the graphical representation of MVF and EF is nearly
identical yet it is important to distinguish between them. MVF represents the entire curve, EF
represents only the non-dominated part of MVF, originating in the global minimum-variance
portfolio (GMV). GMV is frontier portfolio with the smallest variance. The shape of MVF is
strongly influenced by the correlation (ρ) between the assets as it directly affects the portfolio
standard deviation (SD) as descripted by Equation 2.32 for two risky assets A and B, with 𝜇𝐴 , 𝜎𝐴
and 𝜇𝐵 , 𝜎𝐵 , respectively.

𝜎𝑃 = √𝑤𝐴2 𝜎𝐴2 + 𝑤𝐵2 𝜎𝐵2 + 2𝑤𝐴 𝑤𝐵 𝜎𝐴 𝜎𝐵 𝜌𝐴,𝐵 < 𝑤𝐴 𝜎𝐴 + 𝑤𝐵 𝜎𝐵 (2.32)

The above-mentioned inequality descripts the gain from diversification3 coming from
assets with imperfect correlation. Simply put, the lower the correlation between assets, the better
the diversification effect, the more parabolic the shape of MVF. The MVF and EF are depicted on
the following Figure 1.

3
Diversification is further introduced in Section 3.2.1

14
 Figure 1: Efficient frontier for risky assets

2.4.2. Efficient Frontier for Risk-Free and Risky Assets

The portfolio of assets where one is risk-free affects the shape of EF. Let’s consider two
assets, one risky asset A and one risk-free asset, forming a portfolio P. Since the risk-free asset
carries no risk, its SD = 0. The SD of such complete portfolio is simply a linear weighted average
𝑆𝐷𝑃 = 𝑤𝐴 𝜎𝐴 + 𝑤𝑟𝑓 𝜎𝑟𝑓 = 𝑤𝐴 𝜎𝐴 + 0 where 𝑤𝐴 + 𝑤𝑟𝑓 = 1. If short positions on risk-free asset are
allowed, then 𝑤𝐴 > 1 𝑎𝑛𝑑 𝑤𝑟𝑓 < 0. The efficient frontier of such combined portfolios is a straight
line originating in the risk-free rate on axis y.

The EF for a combination of n risky assets and a risk-free asset follows the above-described
scenario. The risky assets themselves form a parabolic-shaped MVF. The combination of risky
portfolio, originally depicted on MVF, with a risk-free asset forms a straight line, referred to as
the capital allocation line (CAL) The only CAL that dominates the parabolic curve in all of its
length, as well as the other CALs, is the tangent to the MVF. This tangent CAL represents the
efficient frontier as depicted in Figure 2. The tangent point represents the only portfolio which can
be made solely of risky assets, usually called the tangency portfolio (T). The tangency portfolio is
important and plays a crucial role in the investor’s optimal portfolio choice decision process,
described in Section 4.1.3.

15
 Figure 2: Different CALs for portfolios composed of risk-free and risky assets

3. Return Generating Models

Although asset pricing is not the main subject of interest of this thesis, there are two pricing
models that can be considered as essential to the portfolio theory and thus it is convenient for these
to be briefly introduced. These models are the capital asset pricing model and the market model.
These models represent an alternative to the historical approach when estimating asset’s expected
returns, variances, and covariances.

3.1. Capital Asset Pricing Model

Capital asset pricing model (CAPM) is an equilibrium pricing model and it has played a
pivotal role in the development of quantitative investment management since its introduction.
CAPM is derived from MPT and as such shares most of the MPT’s assumptions introduced in
Section 2.3 with few specifics. As the concept of CAPM is based on the principle of equilibrium,
few of the assumptions are worthwhile to review.

16
 CAPM assumptions4
 All investors have homogeneous expectations regarding returns, variances, and
covariances for all assets
 Risk-free asset is available for an infinite lending or borrowing
 Markets are perfectly efficient
 All assets are tradable and infinitely divisible
 Unlimited short sales are allowed
 Perfect competition, i.e. an individual alone cannot affect the price

Few implications coming from the CAPM assumptions

 MVF is identical for all investors

 EF is identical for all investors as well
 As all investors share homogeneous expectations, they all demand the same assets
 Hence, the tangency portfolio is identical for everyone
 Tangency portfolio becomes the market portfolio

According to Litterman (2003), CAPM describes the market equilibrium in a sense that, if
the model is correct and any asset’s expected return differs from its equilibrium return, the market
forces come into play and restore the relationship suggested by the model. However, CAPM theory
goes bit further. As known, risk of a stock can be split between systematic and non-systematic, or
specific, risk. If portfolio is large enough, the non-systematic risk can be diversified away5. Since
every investor holds a combination of market portfolio and risk-free asset, which both theoretically
carry zero of specific risk, the specific risk no longer matters. Therefore, CAPM fundamentally
describes a relationship of any asset’s equilibrium return as a linear function of its systematic risk,
measured by β, market risk premium and a risk-free rate. The β of market portfolio is always equal
to 1. When an asset carries higher systematic risk than the market, i.e. 𝛽 > 1, it should be
remunerated by higher return. If the asset carries no systematic risk, thus no specific risk as well,
then the equilibrium return should be equal to the risk-free rate. Let’s consider an asset A with

4
Complete list of assumptions is in Section 2.3
5
See Section 3.2.1.

17
return 𝑟𝐴 , a risk-free asset with return 𝑟𝑓 , and a market portfolio M with return 𝑟𝑀 . CAPM
equation is following:

𝑟𝐴 = 𝑟𝑓 + 𝛽𝐴 (𝑟𝑀 − 𝑟𝑓 ) (3.1)
Equivalently, let’s substitute the single asset A with a complete portfolio P, then the
equation becomes

𝑟𝑃 = 𝑟𝑓 + 𝛽𝑃 (𝑟𝑀 − 𝑟𝑓 ) (3.2)
This is the standard CAPM, where β is the systematic risk measure of an asset/portfolio,
and the term (𝑟𝑀 − 𝑟𝑓 ) is the market risk premium.

The graphical representation of CAPM is the security market line (SML). In CAPM world,
all portfolios should lie on SML. SML is plotted in the μ-σ space, originating at the risk-free rate
on axis Y and going through the market portfolio M. SML is a useful tool for determining whether
an asset is overvalued, undervalued, or correctly valued on the market. This can be done
mathematically by comparing the equilibrium return suggested by CAPM and the actual return
observed on the market, or graphically plotting the asset’s return together with the SML in one
graph.

3.2. Market Model

Market model belongs to the group of so-called factor models. Factor models represent the
building stone of the Arbitrage Pricing Theory (APT), introduced by Ross (1976). The theory is
based on an assumption that all asset returns can be determined by a set of factors. It believes that
the asset returns are related to each other through their correlations with a limited set of factors.
The simplest factor model is the market model. More advanced are, for example, the 3-factor and
5-factor Fama-French models.

The market model (MM) is a one factor model. The factor is the return on market portfolio.
MM describes a relationship between the returns on asset and the returns on market portfolio
through a classical regression. Let’s assume an asset A with return 𝑟𝐴 , and the market portfolio M
with return 𝑟𝑀 . MM regression equation is following (DeFusco et al., 2007):

18
 𝑟𝐴 = 𝛼𝐴 + 𝛽𝐴 𝑟𝑀 + 𝜀𝐴 (3.3)
Where 𝛼𝐴 is the intercept representing an average return on asset A independent of the
market, and 𝜀𝐴 is the error term representing the residual risk. Alternatively, the MM equation can
be expressed in terms of excess returns as following:

𝑟𝐴 − 𝑟𝑓 = 𝛼𝐴 + 𝛽𝐴 (𝑟𝑀 − 𝑟𝑓 ) + 𝜀𝐴 (3.4)
MM stands upon following assumptions

 𝐸(𝜀𝐴 ) = 0
 𝐶𝑜𝑣(𝑟𝑀 , 𝜀𝐴 ) = 0
 𝐶𝑜𝑣(𝜀𝐴 , 𝜀𝐵 ) = 0 𝐴 ≠ 𝐵

These assumptions partially correspond to the OLS regression model. However, MM does
not assume the error term to be normally distributed, as well as the variance of error term being
identical across assets. Given these assumptions, three postulates can be made regarding the
expected returns, variances, and covariances.

 Expected return of asset A depends on the expected return of market M, A’s β towards M,
and the independent part of A’s return

𝐸(𝑟𝐴 ) = 𝛼𝐴 + 𝛽𝐴 𝐸(𝑟𝑀 ) (3.5)

 Variance of asset A depends on the variance of market M, the residual variance of A, and
A’s β towards M

𝑉𝑎𝑟(𝑟𝐴 ) = 𝛽𝐴2 𝜎𝑀
2
+ 𝜎𝜀2𝐴 (3.6)

 Covariance between the returns of asset A and asset B depends on the variance of returns
of market M, and A’s and B’s sensitivities 𝛽𝐴 , 𝛽𝐵

2
𝐶𝑜𝑣(𝑟𝐴 , 𝑟𝐵 ) = 𝛽𝐴 𝛽𝐵 𝜎𝑀 (3.7)

19
 3.2.1. Diversification
The market model is helpful in explanation of one of the core features of large financial
portfolios and that being the diversification effect. The positive effect of correlation on
diversification is introduced already in Section 2.4.1. In this section, the concept of diversification
is extended with regard to the number of assets within the portfolio.

Let’s assume an asset, e.g. a stock. Each stock’s total risk is primarily composed of two
main types of risk. The systematic risk and the specific risk.

 Systematic risk, or market risk, refers to the risks associated with the macroeconomic
events or developments impacting the entire market. Market risk impacts all market
participants equally and from its nature cannot be eliminated via diversification. However,
its impacts can be eased using an appropriate hedging or asset allocation strategy.
Systematic risk is measured by β. Beta of an asset can be interpreted as its sensitivity
towards the market. Beta of the market is always equal to 1. Let’s assume an asset A and a
market portfolio M, with returns 𝑟𝐴 and 𝑟𝑀 , respectively. β calculation is following:

𝑐𝑜𝑣(𝑟𝐴 , 𝑟𝑀 )
𝛽𝐴 = (3.8)
𝑣𝑎𝑟(𝑟𝑀 )

Beta of a portfolio is calculated as a weighted average of individual asset betas. Let’s

assume a portfolio P with n assets. β calculation is following:
𝑛

𝛽𝑃 = ∑ 𝑤𝑖 𝛽𝑖 (3.9)
𝑖=1

β<0 Asset returns move the opposite direction compared to the market. If the
market return is positive, the asset return is negative and vice versa.
β=1 Asset returns move identically with the market.
β >1 Asset returns move the same direction as the market but quicker, both up
and down. The asset is riskier than the market.
0<β<1 Asset returns move the same direction as the market but slower, both up
and down. The asset is less risky than the market.
Table 1: Values of β

20
  Specific risk, or idiosyncratic risk or residual risk, refers to the risks associated with an
individual industry, firm, or product. Specific risk can be eliminated via diversification.

The elimination of specific risk from a portfolio is a sought-after benefit. Mathematically,

the elimination can be explained by using the properties of the market model. Let’s consider the
MM’s second postulate for an entire portfolio P consisting of n assets as following (Elton et al.,
2011):

𝑉𝑎𝑟(𝑟𝑝 ) = 𝛽𝑃2 𝜎𝑀
2
+ 𝜎𝜀2𝑃 (3.10)
Where the term 𝑉𝑎𝑟(𝑟𝑝 ) represents the total risk, 𝛽𝑃2 𝜎𝑀
2
the systematic risk, and 𝜎𝜀2𝑃 the
specific risk. In terms of n individual assets, the equation can be re-written as following:
𝑛 𝑛

𝑉𝑎𝑟(𝑟𝑝 ) = 𝛽𝑃2 𝜎𝑀
2
+ 𝜎𝜀2𝑃 = ∑ 𝑤𝑖2 𝛽𝑖2 𝜎𝑀
2
+ ∑ 𝑤𝑖2 𝜎𝜀2𝑖 (3.11)
𝑖=1 𝑖=1

For evidential purposes, it is convenient to ignore the systematic risk part and focus solely
1
on the specific one. Moreover, let’s assume an equally weighted portfolio where 𝑤𝑖 = 𝑛 , the

diversification effect on the specific part is following:

𝑛 𝑛 𝑛
1 2 1 ∑𝑛𝑖=1 𝜎𝜀2𝑖
∑ 𝑤𝑖2 𝜎𝜀2𝑖 = ∑ ( ) 𝜎𝜀2𝑖 = 2 ∑ 𝜎𝜀2𝑖 = lim =0 (3.12)
𝑛 𝑛 𝑛→∞ 𝑛2
𝑖=1 𝑖=1 𝑖=1

It is easy to see that with a number of assets increasing to infinity, the specific risk
converges towards zero.

21
4. Portfolio Optimization
Since the introduction of MPT in 1952, a number of portfolio optimization techniques have
been developed. All of them, however, more or less build upon the mean-variance optimization
(MVO) model with the motivation to overcome some of its main drawbacks. The time has proven
that the MVO developed by professor Harry Markowitz has truly become the cornerstone of the
portfolio theory. This chapter regarding portfolio optimization introduces only the methods used
in the practical part of this thesis.

4.1. Mean-Variance Optimization

This section dedicated to the MVO is a direct continuation of Section 2.4 regarding the
efficient frontier. As introduced there, the EF is influenced by the parameters of individual assets.
Their means, variances, covariances, and the presence of risk-free asset.

4.1.1. MVO for Risky Assets

To compute portfolios making up the EF considering only the risky assets, the following
constrained problem must be satisfied

𝑚𝑖𝑛 𝑤 𝑇 𝛴𝑤

𝑠. 𝑡. 𝑤 𝑇 𝜇 = 𝑟𝑅

𝑤𝑇𝐼 = 1

Where I is the N x 1 column vector of ones, and 𝑟𝑅 is the required portfolio return6
demanded by the investor. No non-negativity constrains are present. The problem can be solved
by minimizing the Lagrangian

1 𝑇
min ℒ = 𝑤 𝛴𝑤 + 𝜆(𝑟𝑅 − 𝑤 𝑇 𝜇) + 𝛾(1 − 𝑤 𝑇 𝐼) (4.1)
2

6
The use of required return is convenient as the investor may desire a return different from the expected return.
Nonetheless, the expected return may be used in the computations as well. Required return is often referred to as target
return.

22
 Where 𝜆 and 𝛾 are the Lagrange multipliers. The first-order conditions to solve the
Lagrangian are following:

𝜕ℒ
= 𝛴𝑤 − 𝜆𝜇 − 𝛾𝐼 = 0 (4.2)
𝜕𝑤
𝜕ℒ
= 𝑟𝑅 − 𝑤 𝑇 𝜇 = 0 (4.3)
𝜕𝜆
𝜕ℒ
= 1 − 𝑤𝑇𝐼 = 0 (4.4)
𝜕𝛾

The FOCs with applied constrains can be re-written in terms of portfolio weights as
following:

𝑤 = 𝑤𝑃 = 𝜆𝛴 −1 𝜇 + 𝛾𝛴 −1 𝐼 (4.5)
𝑟𝑅 = 𝑤 𝑇 𝜇 = 𝜇 𝑇 𝑤 = 𝜆(𝜇 𝑇 𝛴 −1 𝜇) + 𝛾(𝜇 𝑇 𝛴 −1 𝐼) (4.6)
1 = 𝐼 𝑇 𝑤𝑃 = 𝑤𝑃𝑇 𝐼 = 𝜆(𝐼 𝑇 𝛴 −1 𝜇) + 𝛾(𝐼 𝑇 𝛴 −1 𝐼) (4.7)

For simplification purposes, Danthine and Donaldson (2015) use the following constants

𝐴 = 𝐼 𝑇 𝛴 −1 𝜇 = 𝜇 𝑇 𝛴 −1 𝐼 (4.8)
𝐵 = 𝜇 𝑇 𝛴 −1 𝜇 > 0 (4.9)
𝐶 = 𝐼 𝑇 𝛴 −1 𝐼 > 0 (4.10)
𝐷 = 𝐵𝐶 − 𝐴2 > 0 (4.11)

Solving the set of FOCs, with applied substitution, for the Lagrange multipliers, we obtain:

𝐶𝑟𝑅 − 𝐴 𝐵 − 𝐴𝑟𝑅
𝜆= 𝑎𝑛𝑑 𝛾 =
𝐷 𝐷

Finally, substituting for the Lagrange multipliers into the Equation 4.5, the solution for
portfolio weights is following:

𝐶𝑟𝑅 − 𝐴 −1 𝐵 − 𝐴𝑟𝑅 −1
𝑤𝑃 = 𝛴 𝜇+ 𝛴 𝐼 (4.12)
𝐷 𝐷

23
 Re-arranging the terms, the solution can be written in an alternative form as following:

1 1
𝑤𝑃 = [𝐵(𝛴 −1 𝐼) − 𝐴(𝛴 −1 𝜇)] + [𝐶(𝛴 −1 𝜇) − 𝐴(𝛴 −1 𝐼)]𝑟𝑅 (4.13)
𝐷 𝐷
Or,

𝑤𝑃 = 𝑔 + ℎ𝑟𝑅 (4.14)

Where g represents the weight vector for portfolio with 𝑟𝑅 = 0 , and g + h represents the
weight vector for portfolio with 𝑟𝑅 = 1

The FOCs are essential in defining the portfolio weights representing any frontier portfolio
for a given level of required return. The solution for portfolio weights is highly practical as it
delivers the weights of corresponding frontier portfolio for a chosen level of desired return.

The computation of parameters of any frontier portfolio is quite a straightforward matter.

 Expected return

𝐸(𝑟𝑃 ) = 𝜇𝑃 = 𝑤𝑃𝑇 𝜇 (4.15)

Where 𝑤𝑃 is the portfolio weights, and 𝜇 is vector of expected returns of assets.

 Variance

𝐶 𝐴 2 𝐴
𝑉𝑎𝑟(𝑃) = 𝜎𝑃2 = 𝑤𝑃𝑇 𝛴𝑤𝑃 = (𝜇𝑃 − ) + (4.16)
𝐷 𝐶 𝐶

The global minimum-variance portfolio (GMV), is the frontier portfolio with the smallest
variance. It represents a pivotal point on the MVF, as it splits the MVF between the efficient and
non-efficient frontier. The portfolio parameters calculated by Equations 2.29-31 apply for GMV
as well. However, it can be calculated in a simpler way:

24
  Expected return

𝐴
𝐸(𝑟𝑔𝑚𝑣 ) = 𝜇𝑔𝑚𝑣 = (4.17)
𝐶

 Variance

2
1
𝑉𝑎𝑟(𝑔𝑚𝑣) = 𝜎𝑔𝑚𝑣 = (4.18)
𝐶

4.1.2. MVO for Risk-free and Risky Assets

The inclusion of a risk-free asset within a portfolio of otherwise risky assets improves the
efficiency of the complete portfolio. Let’s assume a fraction of capital denoted w invested in a
vector of risky assets and (1 − 𝐼 𝑇 𝑤) in the risk-free asset denoted 𝑤𝑓 . The optimization problem
is following:

𝑚𝑖𝑛 𝑤 𝑇 𝛴𝑤

𝑇
𝑠. 𝑡. 𝑟𝑓 + (𝜇 − 𝑟𝑓 𝐼) 𝑤 = 𝑟𝑅

Where μ represents the vector of expected returns on risky assets, 𝑟𝑓 the return on risk-free
asset, and I the vector of ones.

It is worthwhile to mention that the constrain 𝑤 𝑇 𝐼 = 1 is no longer present. Thus, 𝑤𝑓 +

∑𝑛𝑖=1 𝑤𝑖 ≠ 1. No non-negativity constrains are present. The problem can be solved by minimizing
the Lagrangian:

1
min ℒ = 𝑤 𝑇 𝛴𝑤 + 𝜆(𝑟𝑅 − 𝑟𝑓 − (𝜇 − 𝑟𝑓 𝐼)𝑇 𝑤 (4.19)
2
Where 𝜆 is the Lagrange multiplier. The FOCs to solve the Lagrangian are following:

𝜕ℒ
= 𝛴𝑤 − 𝜆(𝜇 − 𝑟𝑓 𝐼) = 0 (4.20)
𝜕𝑤
𝜕ℒ 𝑇
= 𝑟𝑅 − 𝑟𝑓 − (𝜇 − 𝑟𝑓 𝐼) 𝑤 = 0 (4.21)
𝜕𝜆

25
 The FOCs can be re-written in terms of portfolio weights as following:

𝑤 = 𝜆𝛴 −1 (𝜇 − 𝑟𝑓 𝐼) (4.22)
𝑟𝑅 − 𝑟𝑓
𝑤= 𝑇 (4.23)
(𝜇 − 𝑟𝑓 𝐼)
Applying the constrain and solving for the Lagrange multiplier, we obtain

𝑟𝑅 − 𝑟𝑓
𝜆= 𝑇 (4.24)
(𝜇 − 𝑟𝑓 𝐼) 𝛴 −1 (𝜇 − 𝑟𝑓 𝐼)
For simplification purposes, a new constant H (Danthine and Donaldson, 2015) for
replacing the denominator may be used

𝑇
𝐻 = (𝜇 − 𝑟𝑓 𝐼) 𝛴 −1 (𝜇 − 𝑟𝑓 𝐼) (4.25)

𝐻 = 𝐵 − 2𝐴𝑟𝑓 + 𝐶𝑟𝑓2 > 0 (4.26)

Where A, B, C represent the constants introduced in Section 4.1.1.

The solution for the optimal portfolio weights by replacing 𝜆 is following:

𝑟𝑅 − 𝑟𝑓 𝑟𝑅 − 𝑟𝑓 −1
𝑤= 𝑇 𝛴 −1 (𝜇 − 𝑟𝑓 𝐼) = 𝛴 (𝜇 − 𝑟𝑓 𝐼) (4.27)
(𝜇 − 𝑟𝑓 𝐼) 𝛴 −1 (𝜇 − 𝑟𝑓 𝐼) 𝐻

Where w represents the vector of portfolio weights on risky assets.

This is the formula that delivers the optimal portfolio weights when considering risky assets
in combination with a risk-free asset for any level of desired return. Since short selling is allowed,
the sum of weights of risky assets may go above 1, implying a short-position on risk-free asset, in
order to achieve the desired return. The sum of weights of risky assets below 1 implies a partial
long position on risk-free asset. Formally, it can be expressed as following: ∑𝑛𝑖=1 𝑤𝑖 ≠ 1, and 𝑤𝑝 =
𝑤𝑓 + ∑𝑛𝑖=1 𝑤𝑖 = 1 where 𝑤𝑝 represents the weights of complete portfolio. 𝑤𝑃 ≠ 𝑤. If, and only if,
∑𝑛𝑖=1 𝑤𝑖 = 1 and thus 𝑤𝑃 = 𝑤 with no holdings of risk-free asset, we identify such portfolio as the
tangency portfolio. All portfolios lie on the efficient frontier. Tangency portfolio lies on both
frontiers.

The computation of parameters of any frontier portfolio combining risky and riskless assets
is, again, a straightforward matter.

26
  Expected return

𝑇
𝐸(𝑟𝑃 ) = 𝜇𝑃 = 𝑟𝑓 + (𝜇 − 𝑟𝑓 𝐼) 𝑤 = 𝑟𝑓 + 𝜎𝑃 √𝐻 (4.28)
Where 𝜇 is the vector of expected returns, I is the vector of ones, w is the vector of portfolio
risky holdings, 𝜎𝑃 is the portfolio’s SD, and H is a constant.

 Variance

2
(𝜇𝑃 − 𝑟𝑓 ) (4.29)
𝑉𝑎𝑟(𝑃) = 𝜎𝑃2 = 𝑤 𝑇 𝛴𝑤 =
𝐻

The tangency portfolio (T) is a special case of frontier portfolio. It is the only portfolio
lying on both MVF and EF and is composed entirely of risky assets. As such, it must solve for
both of the optimization problems introduced in Sections 4.1.1 and 4.1.2. It plays an important role
in the complete portfolio construction process as the investor first determines the tangency
portfolio and then adjusts it accordingly to his individual preferences. The tangency portfolio is
determined as following:

 Tangency portfolio weights

1
𝑤𝑇 = 𝛴 −1 (𝜇 − 𝑟𝑓 𝐼) (4.30)
𝐴 − 𝐶𝑟𝑓
Where ∑𝑛𝑖=1 𝑤𝑖 = 1 and 𝑤𝑇 = 𝑤𝑃 = 𝑤 implying no holdings of risk-free asset.

 Expected return

𝑇 𝐻
𝐸(𝑟𝑇 ) = 𝜇 𝑇 = 𝑤𝑇𝑇 𝜇 = 𝑟𝑓 + (𝜇 − 𝑟𝑓 𝐼) 𝑤𝑇 = 𝑟𝑓 + (4.31)
𝐴 − 𝐶𝑟𝑓

 Variance

𝑉𝑎𝑟(𝑇) = 𝜎𝑇2 = 𝑤𝑇𝑇 𝛴𝑤𝑇 (4.32)

27
 4.1.3. Portfolio Choice
The choice of a complete portfolio within the MPT framework is a subject matter under
the mean-variance utility hypothesis. MPT considers all investors to be rational and naturally risk
averse. The investor’s level of risk aversion is primarily derived from his utility function. When
constructing a portfolio, the investor faces the canonical portfolio problem. This is a two-step
process. The first step is an identification of optimal risky portfolio regardless the investor’s
preferences. The second step is allocation of capital between the optimal risky portfolio and the
risk-free asset to form the most desired portfolio. This two-step process is formally called the
Separation theorem, or Two-fund theorem, (Sharpe, 1995).

The second step is fully done with accordance to investor’s utility function and his risk
aversion. To make this simpler, MPT assumes all investors to have a quadratic utility function.
The investor’s objective is therefore same for all and that being the maximization of his mean-
variance utility. Let’s assume a complete portfolio P with expected return 𝜇𝑃 and variance 𝜎𝑃2 . The
maximization problem then becomes following:

1
max 𝑈 = 𝜇𝑃 − 𝐴𝜎𝑃2 (4.33)
2

Where 𝐴 is the risk aversion coefficient representing the degree of investor’s risk aversion.
It is defined as the additional marginal return the investor demands for accepting more risk. It is
easy to see that the value of utility function rewards higher expected return and penalizes portfolio
risk.

Potential values of A

Risk aversion A>0

Risk neutrality A = 0
Risk seeking A<0
Table 2: Values of A

28
 Alternatively, let’s consider a capital allocated to portfolio of risky assets denoted as 𝑤𝑟
and (1 − 𝑤𝑟 ) = 𝑤𝑓 allocated to risk-free asset, then the mean-variance utility equation can be re-
written as following:

1 1
max 𝑈 = 𝜇𝑃 − 𝐴𝜎𝑃2 = 𝑤𝑟𝑇 𝜇 + (1 − 𝑤𝑟𝑇 𝐼)𝑟𝑓 − 𝐴𝑤𝑟𝑇 𝛴𝑤𝑟 (4.34)
2 2

Solving the maximization problem by setting the first derivative with respect to 𝑤𝑟 equal
to zero, we obtain

𝜕𝑈
= 𝜇 − 𝑟𝑓 𝐼 − 𝐴𝛴𝑤𝑟 = 0 (4.35)
𝜕𝑤𝑟

𝜇 − 𝑟𝑓 𝐼 1 −1 𝜇𝑃 − 𝑟𝑓
𝑤𝑟 = = 𝛴 (𝜇 − 𝑟𝑓 𝐼) 𝑜𝑟 𝑤𝑟 = (4.36)
𝐴𝛴 𝐴 𝐴𝜎𝑃2

Where 𝑤𝑟 represents the capital allocation to risky assets, 𝜇 is the vector of expected returns
on risky assets, I is the vector of ones, rf is the risk-free rate, A is the investor’s risk aversion
coefficient, and Σ is the covariance matrix.

Another method of selecting the most desirable portfolio involves the use of indifference
curves.

The indifference curves are graphical representation of investor’s preferences for risk and
return, and are conventionally plotted in two dimensional, risk and return space. Each investor
possesses an infinite set of unique indifference curves creating so-called map of indifference
curves. Each indifference curve represents all combinations of portfolios that provide the investor
the desired level of satisfaction equally. All that being done with respect to investor’s utility
function. However, and with reference to the MPT assumptions presented in Section 2.3, the MPT
assumes all investors to have a quadratic utility function. Indifference curves under the quadratic
utility assumption thus too have a quadratic form of convex shape in the relevant area of the μ-σ
space. The steepness of the curve is influenced by the investor’s risk aversion coefficient. The

29
higher the coefficient, the more risk averse the investor, the steeper the curve. With accordance to
the separation theorem, the investor first finds the tangency portfolio and then adjusts the portfolio
with risk-free asset to meet the desired characteristics, i.e. to reach the point where the investor’s
indifference curve meets the efficient frontier.

Figure 3: Optimal portfolios for agents with different risk aversion

30
 4.1.4. MPT Limitations
Although the MPT has become the cornerstone of portfolio theory and as such has its
sovereign position within quantitative finance, it possesses a number of shortcomings which make
the model being criticized from today’s perspective. In a theoretical world, the MPT is correct and
performs well. However, the assumptions under which the MPT operates usually do not hold in
reality. These matters of fact have been empirically proven by a number of studies conducted over
the time in various fields of study, e.g. behavioral economics or applied econometrics. Alongside
the research, some assumptions are simply not true from its very nature, e.g. no transaction costs
or taxes. This section provides a non-exhaustive list of the most significant limitations of the mean-
variance optimization (Michaud and Michaud, 2008).

 MVO overuses statistically estimated information resulting in a high input sensitivity. Even
a small change of inputs delivers a major impact on the optimal portfolio holdings.
Consequently, it tends to maximize the estimation error7.
 MVO tends to deliver unintuitive, highly concentrated portfolios
 Return distributions in real world are rarely normal. In fact, distributions are usually
leptokurtic (excess kurtosis) and skewed.
 Under non-normality, symmetric risk measures perform poorly and asymmetric risk
measures, such as semi-deviation or value-at-risk, are more adequate
 MVO assumes a single-period framework only, while investors usually have long term,
multi-period investment horizons
 Quadratic utility function exhibits increasing absolute risk aversion (IARA) which is
unrealistic
 Investors’ expectations are not homogenous as every investor is somehow biased
 Investors being able to buy or sell any quantity of assets doesn’t hold as investors often
have a credit limit. Moreover, some assets have the minimum order size and can’t be traded
in fractions
 Transaction costs, fees, and taxes exist in real world
 Correlations across assets are never stable and fixed

7
Is the difference between the true values of parameters (mean, var, cov) and their estimated values.

31
 4.2. Treynor-Black
The Treynor-Black model (TB) is an optimization method developed by Jack Treynor and
Fischer Black (1973), and was originally published in Journal of Business in 1973. The model is
based on a presumption that only securities showcasing abnormal returns are worth adding to an
otherwise most efficient portfolio, the market portfolio. If such securities occur and are not yet
included in the market portfolio, the market portfolio is no longer efficient. The optimal portfolio
suggested by TB is thus a combination of the market portfolio and the active portfolio composed
of selected securities with positive abnormal returns. Since the number of securities within the
active portfolio is usually limited, the incorporation of the market portfolio also significantly
improves the overall diversification. The ability to predict abnormal returns is critical within the
TB framework, so to avoid any possible inconsistencies coming from using a variety of different
security analyses, the TB assumes the use of the market model characterized by the Equation 3.4.
In MM, the abnormal return is represented by non-zero alpha, i.e. 𝛼 ≠ 0. Any rational investor
desires and seeks 𝛼 > 0, which delivers superior return to the portfolio. This inequality is
important in order to maintain the positive risk-return trade-off as the security always increases
the portfolio risk through its own residual variance. The ultimate goal of TB optimization is the
maximization of the optimal portfolio’s Sharpe ratio8. The majority of MVO assumptions apply
for the TB model as well (Kane et al., 2003).

Let’s assume n+1 assets, where n is the number of securities with abnormal returns
forming an active portfolio A and +1 represents the market index as a passive portfolio M, both
together forming an optimal portfolio P. The estimates of alpha, beta, and residual variance
coefficients on portfolio level are following (Bodie et al., 2018):

𝑛+1

𝛼𝑃 = ∑ 𝑤𝑖 𝛼𝑖 ; 𝛼𝑀 = 0 (4.37)
𝑖=1
𝑛+1

𝛽𝑃 = ∑ 𝑤𝑖 𝛽𝑖 ; 𝛽𝑀 = 1 (4.38)
𝑖=1

8 𝐸(𝑟𝑃 )−𝑟𝑓
𝑆𝑅 =
𝜎𝑃

32
 𝑛+1

𝜎𝜀2𝑃 = ∑ 𝑤𝑖2 𝜎𝜀2𝑖 ; 𝜎𝜀2𝑛+1 = 𝜎𝜀2𝑀 = 0 (4.39)

𝑖=1

The formula for optimal weight allocated to the active portfolio A is following:

2
𝐸(𝑟̅𝐴 )𝜎𝑀 − 𝐸(𝑟̅̅̅)𝜎
𝑀 𝐴𝑀
𝑤𝐴 = 2 2
(4.40)
𝐸(𝑟̅𝐴 )𝜎𝑀 + 𝐸(𝑟̅̅̅)𝜎
𝑀 𝐴 − [𝐸(𝑟̅ 𝐴 ) + 𝐸(𝑟
̅̅̅)]𝜎
𝑀 𝐴𝑀

Where 𝑟̅ stands for risk premium. Therefore,

𝐸(𝑟̅𝐴 ) = 𝐸(𝑟𝐴 ) − 𝑟𝑓 = 𝛼𝐴 + 𝛽𝐴 [𝐸(𝑟𝑀 ) − 𝑟𝑓 ]

𝐸(𝑟̅̅̅)
𝑀 = 𝐸(𝑟𝑀 ) − 𝑟𝑓

2
𝜎𝐴𝑀 = 𝛽𝐴 𝜎𝑀

𝜎𝐴2 = 𝛽𝐴2 𝜎𝑀
2
+ 𝜎𝜀2𝐴

After plugging all together and proceeding algebraic simplifying manipulations, the
allocation to portfolio A gets following:

𝑤0
𝑤𝐴 = ; 𝑤𝑀 = 1 − 𝑤𝐴 (4.41)
1 + (1 − 𝛽𝐴 )𝑤0
Where

𝛼𝐴 /𝜎𝜀2𝐴
𝑤0 = 2
(4.42)
𝐸(𝑟̅̅̅)/𝜎
𝑀 𝑀

is the initial allocation to A if 𝛽𝐴 = 1.

The allocation to n individual securities within the portfolio A is following:

𝛼𝑖
𝜎𝜀2𝑖
𝑤𝑖 = 𝑤𝐴 ∗ 𝛼𝑖 (4.43)
∑𝑛𝑖=1
𝜎𝜀2𝑖
As mentioned, the end goal of TB optimization is maximization of the optimal portfolio’s
Sharpe ratio (SR). Therefore, as optimal portfolio is a combination of market portfolio and a
portfolio of securities with superior expected returns, the overall SR must exceed the one of the
market. The exact relationship is following:

33
 2 𝑛 2
2
𝛼𝐴 2
𝛼𝑖
𝑆𝑅𝑃 = √𝑆𝑅𝑀 + [ ] = √𝑆𝑅𝑀 + ∑[ ] (4.44)
𝜎𝜀𝐴 𝜎𝜀𝑖
𝑖=1

Where the ratio of alpha to its residual SD is called the information ratio.

Parameters of optimal portfolio

 Risk premium

𝐸(𝑟̅𝑃 ) = (𝑤𝑀 + 𝑤𝐴 𝛽𝐴 )𝐸(𝑟̅̅̅)

𝑀 + 𝑤𝐴 𝛼𝐴 (4.45)
 Variance

𝜎𝑃2 = (𝑤𝑀 + 𝑤𝐴 𝛽𝐴 )2 𝜎𝑀
2
+ (𝑤𝐴 𝜎𝜀𝐴 )2 (4.46)

4.3. Black-Litterman
The Black-Litterman model (BL) is an optimization method developed by Fischer Black
and Robert Litterman (1992), and was originally published in Financial Analysts Journal in 1992.
Over the time and due to the popularity of BL approach, a number of extensions to BL have been
developed. In this thesis, only the original BL model is introduced and used. BL is based on a
combination of inverse optimization and Bayesian statistics. It assumes that the optimal portfolio
asset weights are known, represented by their weighting in the market index, and then these
weights are subjects of adjustments in accord to the investor’s unique views about the future
performance of these assets. This is in contrast with MVO, in which the estimates of expected
returns are used as a starting point in derivation of optimal weights. Such approach overcomes the
major shortcomings of MVO – input sensitivity, high concentration, and estimation error
maximization. This brief introduction of BL is based upon the works of Idzorek (2002) and Walters
(2014).

The starting point of inverse optimization under BL framework is the derivation of implied
equilibrium excess returns, denoted Π, which is a N x 1 column vector resulting from following
expression:

Π = 𝜆Σ𝑤𝑚𝑘𝑡 (4.47)

34
 𝐸(𝑟𝑀 )−𝑟𝑓
Where 𝜆 = 𝜎2 is the risk-aversion coefficient of market portfolio, Σ is the
𝑜𝑓 𝑀 𝑒𝑥𝑐𝑒𝑠𝑠 𝑟𝑒𝑡𝑢𝑟𝑛𝑠

covariance matrix of excess returns, and 𝑤𝑚𝑘𝑡 is the market capitalization weight N x 1 column
vector of the assets.

As in MVO, the optimization goal of BL is the maximization of investor’s mean-variance utility

1
max 𝑈 = 𝑤 𝑇 𝜇 − 𝜆𝑤 𝑇 Σ𝑤 (4.48)
2
⋮

𝑤 = (𝜆Σ)−1 𝜇 (4.49)
Where μ is any vector of excess returns. If 𝜇 = Π, then 𝑤 = 𝑤𝑚𝑘𝑡

If an investor possesses no specific views about the future performance of the assets, he
should then hold the portfolio with weights derived from the vector of implied equilibrium returns,
i.e. 𝑤𝑚𝑘𝑡 , which is the view-neutral starting point of the BL model.

The original BL formula is:

𝐸(𝜇̅ ) = [(𝜏Σ)−1 + 𝑃𝑇 Ω−1 𝑃]−1 [(𝜏Σ)−1 Π + 𝑃𝑇 Ω−1 𝑄] (4.50)

Where
𝐸(𝜇̅ ) is the posterior combined return vector (N x 1)
𝜏 is scalar
Σ is covariance matrix of excess returns
P is a (K x N) matrix identifying the assets involved in the views, where K is the number
of views and N the number of assets
Ω is a diagonal (K x K) matrix representing the residual variance associated with the
expressed views
Q is a (K x 1) column vector of views

Idzorek (2002: 13) describes the BL model as a “complex weighted average of the implied
equilibrium return vector Π and the view vector Q, in which the relative weightings are a function
of the scalar τ and the uncertainty of the views Ω.” Although the BL model doesn’t require one to
specify any views, the possible incorporation of investor’s views within the model is perhaps the
most attractive feature of the BL model. The views can be expressed either in an absolute or

35
relative form. The absolute view expresses an idea about an absolute return on an asset, e.g. 5%.
The relative view expresses an idea about an asset under- or outperforming relatively to some other
asset, e.g. asset A outperforms asset B by 25 b.p. The views form Q (K x 1) matrix. The uncertainty
about the views is expressed in the error term vector denoted ε, where each error term
𝜀 ~ 𝑁(0, 𝜎 2 ).

𝑄1 𝜀1
𝑄+𝜀 = [ ]+[ ⋮ ]
⋮ (4.51)
𝑄𝑘 𝜀𝑘
The expressed views are linked to the assets in question via the matrix P (K x N)

𝑝1,1 ⋯ 𝑝1,𝑛
𝑃=[ ⋮ ⋱ ⋮ ] (4.52)
𝑝𝑘,1 ⋯ 𝑝𝑘,𝑛
Where each row is associated with one specific view. If the view is positive, the associated
weight has a positive sign, e.g. +1, if negative then -1. The sum of weights in each row must be
equal to 0 in case of relative views, and equal to 1 in case of absolute views. The actual weighting
used in practice is where multiple versions of the BL model differ. Some weighting schemes use
equal weighting, market capitalization weighting, or confidence level based weighting expressed
as percentage on an intuitive scale 0-1.

The error terms enter the BL formula in form of its variance, denoted ω, and expressed in
the Ω matrix

𝜏𝜔1 0 0
𝛺=[ 0 ⋱ 0 ] (4.53)
0 0 𝜏𝜔𝑘
Where

𝜔𝑘 = 𝑃𝑘 Σ𝑃𝑘𝑇 (4.54)
The scalar τ should be more or less inversely proportional to the relative weight given to
Π. However, its recommended value differs across literature and its variation is one of the ways
how to calibrate the model for specific needs. Black and Litterman recommend to use values close
to zero, such as often recommended τ=0.0025.

The last step of BL optimization is to obtain the combined return vector 𝐸(𝜇̅ ), Equation
4.50, and plug it into the Equation 4.49, which returns the optimal portfolio weights.

36
 4.4. Naïve Optimization
Naïve portfolio optimization methods are methods that may be used when a little, or none,
statistical information about the assets, e.g. their means, variances, and/or correlations, is available
to the investor. When no information is available and, at the same time, the investor possesses no
knowledge of the capital market theory, the intuitive way to create a diversified optimal portfolio
is to spread his available wealth across chosen assets equally. With increasing awareness and
available information, more naïve methods come into play. This thesis introduces three naïve
optimizers, the equal weighting, the Sharpe ratio based method, and the most diversified portfolio
method.

4.4.1. Equal Weighting

Equal weighting is a logical starting point for any investor, who desires to allocate his
available capital across multiple assets which he has no information about, and/or is unaware of
the capital market theory. Without any information, all the assets should look exactly the same to
the investor and thus he should allocate his capital across them equally. Assuming n assets, the
individual asset weighting is following:

1
𝑤𝑖 = (4.55)
𝑛

The equal weighting approach appears to be quite popular among the investors. Potentially
due to its simplicity, or since it requires no estimations of parameters, it does not suffer the
estimation error. Another strong argument in favor of equal weighting is its empirical evidence,
for instance, DeMiguel et al. (2009) and Playkha et al. (2012), showcasing that such portfolios
usually strongly outperform portfolios build under the mean-variance framework. Thus, it raises a
question whether practitioners using 1/N approach are unaware of the information regarding the
assets, or are aware of the outcomes of such empirical studies and have decided to exploit them
(Kinlaw et al., 2017).

37
 4.4.2. Sharpe Ratio Model
Sharpe ratio based asset allocation model (SRM) is slightly more sophisticated than the
equal weighting as it requires the parameter estimates, yet still quite naïve in its fundamental
nature. Sharpe ratio (SR) is a simple measure providing an information about asset’s risk premium
per one unit of total risk measured by standard deviation. Simply put, the higher the SR the better
the investment. It is fair to note that high SR doesn’t necessarily mean the highest return or the
lowest risk. SR and SR based allocation are calculated as following (Amenc and Le Sourd, 2003):

𝜇 − 𝑟𝑓
𝑆𝑅 = (4.56)
𝜎
Where the numerator represents the asset’s risk premium and the denominator its standard
deviation.

Assuming n assets within a portfolio, the SR based allocation is following:

𝑆𝑅𝑖
𝑤𝑖 = 𝑛 (4.57)
∑𝑖=1 𝑆𝑅𝑖

This naïve optimization approach based on Sharpe ratios of individual assets as a starting
point for asset allocation is slightly different than the Sharpe ratio maximization problem for
complete portfolios and defined as

𝑤 𝑇 𝜇 − 𝑟𝑓
max
√𝑤 𝑇 Σ𝑤

𝑤𝑇𝐼 = 1

Where 𝑟𝑓 is the mean risk-free rate and I is the column vector of ones.

This maximization problem seeks the highest SR of a complete portfolio. In theory, the
highest SR is guaranteed for portfolios lying on the CML, i.e. the market portfolio (M) with
possible long/short position in risk-free asset. Should one consider rather a subset of n risky assets
instead of M, the highest SR is then guaranteed for portfolios lying on the CAL, i.e. the tangency
portfolio (T) with possible long/short position in risk-free asset. Eventually, the CAL can be

38
steeper, and therefore have a higher SR, than CML9. The optimal weights of T can be easily
obtained via MVO as described in Section 4.1.2. However, such MVO weights may suffer the
shortcomings of MVO, such as extreme long/short positions or allocation only to few assets. These
shortcomings do not apply within the SRM. On the other hand, it may suffer from the cumulative
estimation error caused, for instance, by insufficient data sets.

4.4.3. Most Diversified Portfolio

Most diversified portfolio (MDP) is a naïve optimization method focused on maximization
of the diversification ratio (DR). The maximization problem for a portfolio of n assets is following
(Scherer, 2015):

∑𝑛𝑖=1 𝑤𝑖 𝜎𝑖 𝑤𝑇𝜎
max 𝐷𝑅 = max = max
√𝑤 𝑇 Σ𝑤 (4.58)
√∑𝑛𝑖=1 ∑𝑛𝑗=1 𝑤𝑖 𝑤𝑗 𝜎𝑖 𝜎𝑗 𝜌𝑖,𝑗

Where σ is a N x 1 column vector of asset volatilities. DR itself can be interpreted in a

following way: the lower the correlation between the assets, the higher the ratio. If all assets were
perfectly correlated, the ratio equals 1.

The solution to the maximization problem in order to obtain the optimal portfolio weights
is following:

Σ −1 𝜎
𝑤𝑚𝑑𝑝 = (4.59)
𝐼 𝑇 Σ −1 𝜎
Where I is the N x 1 column vector of ones, and Σ −1 is the N x N inverse covariance matrix.

The MDP represents a mean-variance portfolio, where it is assumed that all asset returns
are proportional to their standard deviations. The level of proportionality is usually defined by a
constant SR.

𝜇𝑖 = 𝑆𝑅𝜎𝑖 (4.60)

9
Beating the market in terms of performance is the ultimate goal of active portfolio management

39
5. Portfolio Performance
Performance measurement of investments is an essential part of any investment process.
Performance analysis can be done both ex-ante and ex-post. Ex-ante analysis may be of a help to
an investor before making an investment. Whether conducting a scenario analysis or analysis of
historical data, such obtained values should only be taken with reserve as they do not possess a
real predicting value but rather only orientational. Or put differently, the historical performance
never guarantees the future performance due to the risks associated with the investment. During
the holding period, the value to the portfolio is being added through a variety of sources such as
superior asset allocation, security selection, market timing, transaction executions etc. Logically,
any investor is curious and wants to know how well his investments have been doing. Performing
a periodical performance analysis may serve as an underlying evidence for potential changes in
his investment strategy. After the holding period, therefore ex-post, it is possible to conduct an
overall, exact, risk-adjusted performance assessment to see how the portfolio performed. Then, the
answers to questions such as What is the total return? or Why the portfolio performed that way?
can be answered. Portfolio’s risk-adjusted performance measures can also be used for comparing
mutually exclusive portfolios between themselves. Although there are dozens of portfolio
performance measures available, this chapter introduces only the ones that are used within its
practical part. All presented measures can be found in Bacon (2008).

5.1.1. Time-Weighted Return

Time-weighted rate of return (TWR) is a measure of a per-period compounded return on
an investment where each time period is given an equal weight regardless the money invested10. It
is calculated as a geometric mean of a series of n realized returns. TWR is useful as an average
return in a long-term perspective.

𝑛
𝑛
𝑛
𝑇𝑊𝑅 = √(1 + 𝑟1 )(1 + 𝑟2 ) … (1 + 𝑟𝑛 ) − 1 = √∏(1 + 𝑟𝑖 ) − 1 (5.1)
𝑖=1

10
Measure where each time period is weighted by the money invested is called the money-weighted return.

40
 5.1.2. Effective Annual Return
Effective annual return (EAR) is a measure of total return over a multi-period time frame
that reflects compounding11. Alongside the daily frequency, returns are often expressed on a
weekly, monthly, or yearly basis. Therefore, the average per-period return requires adjustment to
meet the criteria. EAR is calculated as following:

𝐸𝐴𝑅 = (1 + 𝑟)𝑛 − 1 (5.2)

Where r is the average per-period return, and n is the number of periods

5.1.3. Standard Deviation

Variance (𝜎 2 ) is a statistical measure of total dispersion of a random variable, such as asset
returns, around its mean. It can be calculated both ex-ante, as expected variance, and ex-post, as
historical variance. Its square root, the standard deviation (σ or SD) or volatility, is the most
common measure of asset’s risk. Historical sample SD is calculated in following way:

𝑇
1
𝑆𝐷 = √ ∑(𝑟𝑡 − 𝑟̅ )2 (5.3)
𝑇−1
𝑡=1

Alternatively, it is possible to use the properties of the market model and obtain SD as

2
𝑆𝐷 = √𝛽 2 𝜎𝑀 + 𝜎𝜀2 (5.4)

2
Where 𝜎𝑀 is the market variance and 𝜎𝜀2 is the variance of residuals.

The residual Var/SD [Var(e)/SD(e)] represents the asset’s specific risk. It is introduced in
a more detailed way in Section 3.2 regarding the market model and diversification.

SD can be adjusted to longer periods as 𝑆𝐷𝑇 = 𝑆𝐷√𝑇 where T represents the number of
periods, e.g. from daily to monthly or yearly. This adjustment is, however, only an approximation.

11
The measure of total, multi-period return that does not reflect compounding is called Annualized Percentage Rate
(APR) and is calculated 𝐴𝑃𝑅 = 𝑟 ∗ 𝑛

41
 5.1.4. Beta
Beta as a measure of asset’s systematic risk is closely introduced in Chapter 3 and its
Section 3.2.1 about diversification.

5.1.5. Sharpe Ratio

Sharpe ratio (SR), or reward-to-volatility ratio, is a risk adjusted measure of portfolio
performace. It measures the amount of portfolio’s excess return per one unit of total risk measured
by SD. SR is the slope of the capital allocation line. The larger the SR, the better.

𝑟𝑃 − 𝑟𝑓
𝑆𝑅 = (5.5)
𝜎𝑃
Where 𝑟𝑝 is the return on portfolio P, 𝑟𝑓 is the risk-free rate, and 𝜎𝑃 is the SD of P. In ex-
ante calculations, mean values are used. In ex-post, the actual realized values are used.

5.1.6. Treynor Ratio

Treynor ratio (TR) is a risk adjusted measure of portfolio performance. It is similar to SR,
but with the difference that it considers systematic risk only. It assumes that all portfolios are well
diversified, and thus the idiosyncratic risk no longer matters. It measures the amount of portfolio’s
excess return per one unit of systematic risk measured by beta. TR is the slope of the security
market line. The larger the TR, the better.

𝑟𝑃 − 𝑟𝑓
𝑇𝑅 = (5.6)
𝛽𝑃
Where 𝛽𝑃 is the systematic risk of P. In ex-ante calculations, mean values are used. In ex-
post, the actual realized values are used.

Few comments regarding the TR need to be mentioned. TR is a theoretical concept

assuming that all portfolios are well diversified, thus with no idiosyncratic risk. However, real
portfolios usually carry some idiosyncratic risk. Therefore, portfolios that have the same amount
of systematic risk but differ in total risk, will have the same TR.

5.1.7. Jensen’s Alpha

Jensen’s alpha (α) is a risk-adjusted measure of superior performace. α is the intercept of
the standard CAPM regression equation. It compares the portfolio’s realized excess return with

42
the risk-adjusted excess return. The measure adjusts for systematic risk. The regression equation
is following:

𝑟𝑃,𝑡 − 𝑟𝑓,𝑡 = 𝛼𝑃,𝑡 + 𝛽𝑃 (𝑟𝑀,𝑡 − 𝑟𝑓,𝑡 ) + 𝜀𝑃,𝑡 (5.7)

Where t represents a time point within the time series.

In ex-post analysis, the error term can be ignored and alpha is then calculated by using the
actual realized returns

𝛼𝑃 = 𝑟𝑃 − 𝑟𝑓 − 𝛽𝑃 (𝑟𝑀 − 𝑟𝑓 ) (5.8)
This is called the Jensen’s alpha, or Jensen’s measure, or Jensen’s differential return, or
ex-post alpha.

Positive α indicates a superior risk-adjusted return, i.e. the portfolio’s return is higher than
what it should be in accord to the level of undertaken risk. In CAPM universe, 𝛼 = 0. Positive α
therefore lies above the SML and is desired. Positive α can be, for instance, due to the portfolio
manager’s superior security selection or timing skills. Jensen’s α is a measure often used to
evaluate portfolio managers. However, it does not evaluate the manager’s ability to diversify as it
accounts the systematic risk only.

5.1.8. Information Ratio

Information ratio (IR) is a measure of portfolio’s excess return (er) relative to market
portfolio per one unit of tracking error, which is the standard deviation of those excess returns.
Positive IR represents a superior performance. IR is a key statistic and is extensively used as it
evaluates the portfolio manager’s abilities to generate excess returns. The higher the IR, the better.

𝑟𝑃 − 𝑟𝑀
𝐼𝑅 = (5.9)
𝜎𝑒𝑟
Where 𝑟𝑃 is the return on P, 𝑟𝑀 is the return on market index used as a benchmark, and 𝜎𝑒𝑟
is the tracking error calculated as following:

𝑇
1
𝜎𝑒𝑟 ̅̅̅)2
= √ ∑(𝑒𝑟𝑡 − 𝑒𝑟 (5.10)
𝑇
𝑡=1

Where 𝑒𝑟𝑡 = 𝑟𝑃,𝑡 − 𝑟𝑀,𝑡 , and 𝑒𝑟

̅̅̅ is the mean excess return.

43
 The tracking error can also be seen as a function of portfolio’s SD and the correlation
between the portfolio and the market, calculated as following:

2
𝜎𝑒𝑟 = 𝜎𝑃 √(1 − 𝜌𝑃,𝑀 ) (5.11)

IR can be obtained in an alternative way by regressing the standard CAPM, Equation 3.4.

𝛼𝑃
𝐼𝑅𝑃 = (5.12)
𝜎𝜀𝑃
Where 𝛼𝑃 is the Jensen’s alpha, and 𝜎𝜀𝑃 is the standard error of regression.

In ex-ante IR calculations, mean values are used. In ex-post, the realized values are used.

The IR can also be used in calculation of portfolio’s SR by using Equation 4.44.

Since IR is extensively used for comparing different portfolios, it is essential to maintain

consistency of how the individual elements are calculated, e.g. the frequency of data, overall time
period, arithmetic or geometric means and excess returns, T or T-1, ex-post or ex-ante. The IR is
often expressed on an approximate annualized basis as 𝐼𝑅 = √𝑇𝐼𝑅𝑃 .

5.1.9. M^2
M^2 is a measure of risk-adjusted return relative to the market. It compares the hypothetical
return on portfolio with adjusted SD to match the SD of market with the return on market. The
adjusted risk of portfolio is achieved by long/short positions in the risk-free asset. M^2 is a highly
useful measure for comparing portfolios with different levels of risk. The higher the M^2, the
better.

𝑀2 = 𝑟𝑃 + 𝑆𝑅𝑃 (𝜎𝑀 − 𝜎𝑃 ) (5.13)

Or, alternatively

𝜎𝑀
𝑀2 = 𝑟𝑓 + (𝑟 − 𝑟𝑓 ) (5.14)
𝜎𝑃 𝑃
In ex-ante M^2 calculations, mean values are used. In ex-post, the actual realized
values are used. M^2 of market is always equal to its return.

44
6. Portfolio Management
Portfolio management is a sophisticated discipline that can be viewed as a process
undertaken in a consistent manner to create and maintain investment portfolios that meet the
investor’s objectives. The objectives are defined beforehand and provide an underlying framework
for the portfolio management. In a professional world, the investor’s objectives are stated in a
document so-called the Investment Policy Statement (IPS). IPS contains information regarding the
investor’s return expectations, risk profile, time horizon, along with possible constrains such as
liquidity needs or tax concerns, among others.

Portfolio management consists of three main steps conducted along the way: (1) planning
step, (2) execution step, and (3) feedback step. In the planning step, the IPC is created, market
expectations are formed, and investment strategy is established. In the execution step, the
investment portfolio is constructed and managed accordingly to the investment strategy. In the
feedback step, the portfolio performance is monitored on a constant basis and compared with the
IPC. Each of the steps deserves a closer look. This thesis, however, provides only a brief
description. For more complex description, see Maginn et al. (2007).

6.1.1. Planning Step

As stated, the planning step consists of the creation of IPC, formation of market
expectations, and establishing an investment strategy. The investment strategy involves the
strategic asset allocation (SAA) reflecting both IPC and market expectations. Investment strategies
are passive, active, or semi-active

1) Passive Strategy
Passive, or not reacting, strategy represents a portfolio management that doesn’t
react anyhow to market fluctuations. Two most used forms of passive strategy management
are indexing and buy-and-hold.
 Indexing

Portfolio is designed to replicate the performance of a specific market index as accurately

as possible. The replication can be either physical or synthetic. In physical replication, the exposure
to assets is direct and is either full, i.e. holding all constituents with identical weighting as does the

45
market index, or sample, i.e. holding only some of the index constituents. In synthetic replication,
the exposure to the index constituents is indirect through derivatives.

 Buy-and-hold

A strategy under which the selected assets are bought and held long term in order to profit
from the capital gains and/or additional sources of income such as dividends.

2) Active Strategy
In contract to passive strategy, the active strategy does acknowledge the market
fluctuations and tries to construct such optimal portfolio that exploits them as much as possible in
order to achieve superior risk-adjusted portfolio performance relative to the market, i.e. to achieve
positive alpha.

3) Semi-Active Strategy
Semi-active, or risk-controlled active, or enhanced index approach, is a combination of
active and passive approach. The strategy seeks positive alpha as well, and at the same keeps tight
control over portfolio’s risk relative to the benchmark.

6.1.2. Execution Step

The execution step turns plans into reality. It consists the construction of actual portfolio
in accord to what is established in the planning step. The portfolio composition must reflect the
investor’s objectives and risk profile. Once the optimal portfolio is established, it must be managed
accordingly to the SAA. However, if the investment horizon is long-term, some deviations from
SAA can be made, usually done on purpose and for limited time. These purposefully made
deviations are called the tactical asset allocation (TAA). The prime difference between SAA and
TAA is in the time length. While SAA involves the long-term objectives, the TAA involves short-
term adjustments to SAA in order to exploit the expected market fluctuations, and, consequently,
realize additional superior returns. TAA is thus based on a constant monitoring of both the market
and the portfolio. Monitoring and rebalancing belong to the third step of the portfolio management
process, the feedback step. Therefore, the execution step and the feedback step are closely related.

46
 6.1.3. Feedback Step
The fluctuations in market values of individual assets create deviations of the current asset
allocation from the SAA. Although these deviations may not matter much in short term, their
cumulative effect may significantly impact the overall portfolio performance, and may cause
significant deviations from the investor’s long term objectives. The way how portfolio manager
approaches these deviations is the core difference between passive and active portfolio
management. Whilst the passive management fundamentally ignores the deviations in allocation,
the active management reacts to them by rebalancing the current asset allocation to make the
allocation consistent with the SAA. In order to properly rebalance, the portfolio manager must
monitor both market and portfolio on continuous basis. Monitoring, rebalancing, and performance
evaluation are core elements of the feedback step.

1) Monitoring
Monitoring is an essential part of the feedback step. Being aware, or not, about all possible
influences that may have an impact on the portfolio is the difference whether the investor’s
objectives are going to be reached or not. Therefore, the portfolio manager should keep a constant
eye over the investor’s circumstances, market and economic changes, and the portfolio itself. Any
changes must be dealt with in an appropriate manner.

2) Rebalancing
Portfolio rebalancing represents adjustments in current asset allocation as a reaction to (1)
fluctuations in market values of assets in order to be consistent with SAA, (2) changes in investor’s
objectives, constrains, or market expectations, and (3) tactical asset allocation. In this thesis, only
the scenario 1 is considered and introduced bit further. In scenario 1, the portfolio manager sells
appreciated assets and buys depreciated assets in case of long positions, or buys appreciated assets
and sells depreciated assets in case of short positions, to make the actual composition consistent
with SAA. In practice, two most common rebalancing practices are calendar rebalancing and
percentage-of-portfolio rebalancing.

47
  Calendar rebalancing
- Rebalancing happens on a periodic basis, e.g. monthly, quarterly, semi-annually, or
annually.
 Percentage-of-portfolio rebalancing (or percent range or interval rebalancing)
- Rebalancing is triggered when an asset’s weight crosses a pre-specified corridor or
tolerance band. Let’s assume three assets A, B, and C with SAA, in percentages,
40/40/20, respectively. Assets A and B have corridor ± 5%, and asset C ± 1,5%.
If the weighting of any asset exceeds the tolerance corridor, the rebalancing is
triggered and the initial weighting 40/40/20 is re-established.

Although rebalancing does have its undoubtful benefits, e.g. it reduces the present value of
expected utility loss coming from not tracking the optimum, it does have its shortcomings as well,
e.g. transaction costs and/or tax costs12 applied on sale of the appreciated assets. These
shortcomings, however, may be reduced by imposing constrains on them in the portfolio
optimization process.

3) Performance evaluation
Portfolio performance evaluation is described in Chapter 5.

12
Tax liability depends on a particular jurisdiction and on whether the investor is a subject to taxation or not.

48
7. Practical Experiment
The practical part of this master’s thesis is based on an experiment that put multiple
quantitative portfolio optimization methods into a contest. Different optimizers were applied to
portfolios composed of identical assets, which were subsequently held under different portfolio
management styles over a pre-specified period of time. The performance of each portfolio was
measured ex-post, adequately evaluated in accord with the criteria of the experiment, and
confronted with the others.

The assets included in the experiment’s portfolio P (P) are 30 US blue-chip stocks, US 4-
week Treasury bill (T-bill), and S&P 500 market index. The stocks were selected intuitively
without any equity analysis done beforehand. Nonetheless, they were picked with a sense for
diversification and represent all leading industries. The complete list of all portfolio components
with a brief description is presented in annex at the end of the thesis.

The optimization models selected for the experiment represent 3 sophisticated and 3 naïve
models. The first group includes the Mean-variance optimization model (MVO), the Treynor-
Black model (TB), and the Black-Litterman model (BL). The latter includes the Equal-weighting
model (1/N), the Sharpe ratio based model (SRM), and the Most diversified portfolio model
(MDP). The models are presented and descripted in the theoretical part. The models were exercised
on the portfolio P which resulted in 6 different suggested optimal asset allocations, thus 6 different
portfolios.

Two different portfolio management styles were used in the experiment, active and passive.
To assure fair starting point between them, each portfolio was created twice, resulting in two equal
sets of 6 portfolios. The market portfolio included in the experiment served as a benchmark to
these portfolios. Therefore, the total number of portfolios was 13, 6 active, 6 passive and 1 market
portfolio. The portfolios under the active management were rebalanced on daily basis in accord to
the asset’s daily adjusted closing prices to maintain the initial optimal allocation suggested by the
models. The portfolios under the passive management were untouched throughout the experiment
in accord to the passive portfolio management strategy buy-and-hold. Both management styles are
descripted in Chapter 6.

49
 The total time period of the experiment was from 2/1/2013 to 29/6/2018, and can be split
into two smaller periods with 31/12/2017 being the breaking point. The period from 2/1/2013 to
29/12/2017, ex-ante, served as an estimation period to obtain in-sample estimates of inputs for the
models. The period from 2/1/2018 to 29/6/2018, the holding period (HLDP), is the period over
which the portfolios were held and managed. The appropriate time series for both periods were
collected retrospectively.

After the holding period, ex-post, the performance of each portfolio was measured by the
appropriate portfolio measures presented in Chapter 5. Each measure was treated equally, and
therefore each measure does have an equal weight within the grading system used as a final
evaluation tool. The grading system is a simple methodology that gives points to each portfolio
accordingly to its ranking within the chart of each measure. The higher the ranking, the more points
the portfolio receives. The maximum number of points for each measure was 12. As 9 measures
(EAR, SD, Residual SD, α, β, SR, TR, IR, M^2) were used, the theoretical maximum number of
points a portfolio could have collected was 108. The portfolio that collected the most points won.

The stock’s and market’s daily data, i.e. the adjusted closing prices, were obtained from
http://finance.yahoo.com. The daily rates on US 4 Week Treasury bill were obtained from
http://www.quandl.com.

To assure feasibility of the experiment, the following assumptions were assumed:

 Markets are perfectly efficient

 Returns are Gaussian
 No transaction costs, no taxes
 All assets are perfectly divisible and can be bought/sold in fractions
 Unlimited long and short positions are allowed
 Agent is risk tolerant, therefore his risk aversion played no role in the experiment
 Although risk tolerant, agent does have mean-variance preferences
 Agent accepts the optimal portfolio suggested by each model
 Agent has no specific views about the future
 Agent’s initial capital for each portfolio was $ 1 000 000
 S&P 500 index monetization followed 1 index point = $ 1 parity.

50
 7.1. Optimal Portfolios
This section presents the optimal portfolios suggested by each optimization model. The
suggested relative and absolute allocation to the assets in the portfolios is presented in tables at the
end of this section.

1) MVO
The optimal portfolio suggested by MVO is the tangency portfolio, as it is the only efficient
portfolio composed only of risky assets. Market portfolio and risk-free asset are not included.

2) TB
The optimal portfolio suggested by TB is a combination of market portfolio and portfolio
P. The risk-free asset is not included.

3) BL
In accordance with the experiment’s assumptions, the agent has no views about the future13
and thus the matrices Q, P, and Ω are omitted. Consequently, the optimal portfolio suggested by
BL is the portfolio derived from the vector of implied equilibrium excess returns. The suggested
weight vector is identical to the weight vector based on market capitalizations. The market cap of
each firm considered corresponds to its value on 29/12/2017. The market caps are presented in the
annex within the description of each firm. Market portfolio and risk-free asset are not included.

4) 1/N
The optimal portfolio suggested by 1/N model is the portfolio P with equal allocation
across all stocks. Market portfolio and risk-free asset are not included.

5) SRM
The optimal portfolio suggested by SRM is the portfolio P with allocation proportional to
SR of each stock. Market portfolio and risk-free asset are not included.

13
This assumption was made to assure consistency with other models

51
 6) MDP
The optimal portfolio suggested by MDP is the portfolio P with allocation proportional to
the inverse of individual volatility of the stocks. Market portfolio and risk-free asset are not
included.

Optimal relative asset allocation suggested by the models

MVO TB BL 1/N SRM MDP
AAPL 0.06355 0.19750 0.15801 0.03333 0.04063 0.08491
AMT 0.07397 0.12338 0.01107 0.03333 0.03270 0.00292
AMZN 0.08082 0.22512 0.10155 0.03333 0.04755 0.04234
APD 0.10798 0.21117 0.00646 0.03333 0.04133 -0.03210
BA 0.40654 0.54200 0.03151 0.03333 0.06414 0.03191
C -0.34041 0.19323 0.03665 0.03333 0.02326 -0.03834
CAT 0.02037 0.00740 0.01683 0.03333 0.02797 0.03651
DWDP 0.06091 0.10465 0.01575 0.03333 0.03603 -0.00708
EA 0.16266 0.29388 0.00586 0.03333 0.05749 0.07185
EQR -0.06446 -0.01698 0.00423 0.03333 0.01928 0.08322
FSLR -0.05178 -0.01903 0.00127 0.03333 0.01319 0.04638
GS -0.02297 -0.06653 0.01782 0.03333 0.03007 -0.03868
HD 0.42196 0.63630 0.04039 0.03333 0.06276 -0.02105
INTC 0.09178 0.17357 0.03921 0.03333 0.03915 0.01536
JNJ 0.43291 0.59027 0.06779 0.03333 0.05354 -0.05837
K -0.13020 -0.04688 0.00424 0.03333 0.01719 0.08613
KIM -0.23431 -0.17522 0.00140 0.03333 0.00771 0.03471
KO -0.22498 -0.01007 0.03537 0.03333 0.02269 0.04720
MAR 0.21148 0.44607 0.00914 0.03333 0.05697 -0.03784
MCD 0.25269 0.48702 0.02520 0.03333 0.05008 0.07741
NKE 0.03792 0.22499 0.01845 0.03333 0.04151 0.06404
NUE -0.09132 0.11537 0.00367 0.03333 0.01766 0.05203
PFE -0.06458 0.01210 0.03894 0.03333 0.02747 0.09399
PG 0.06074 0.07348 0.04235 0.03333 0.02799 0.04512
REGI 0.02498 -0.00736 0.08253 0.03333 0.01190 0.07860
T -0.01267 -0.02802 0.04315 0.03333 0.02092 0.10972
TSLA 0.09298 0.12748 0.00939 0.03333 0.04222 0.05388
WBA -0.05692 0.06584 0.01325 0.03333 0.02815 0.08190
WFC 0.10855 0.01789 0.05444 0.03333 0.03357 0.00694
XOM -0.41820 -0.48442 0.06406 0.03333 0.00483 -0.01362
SP500 0 -2.39700 0 0 0 0
SUM 1 1 1 1 1 1

Table 3: Optimal relative asset allocation suggested by the models

52
 Optimal absolute asset allocation suggested by the models
MVO TB BL 1/N SRM MDP
AAPL $ 63,552.22 $ 197,500.88 $ 158,009.76 $ 33,333.33 $ 40,633.89 $ 84,906.96
AMT $ 73,974.06 $ 123,375.87 $ 11,068.33 $ 33,333.33 $ 32,703.14 $ 2,916.14
AMZN $ 80,821.10 $ 225,116.20 $ 101,552.78 $ 33,333.33 $ 47,550.59 $ 42,341.61
APD $ 107,975.82 $ 211,171.52 $ 6,464.21 $ 33,333.33 $ 41,331.87 $ (32,098.17)
BA $ 406,543.44 $ 542,003.26 $ 31,511.21 $ 33,333.33 $ 64,142.60 $ 31,914.03
C $ (340,408.94) $ (193,230.76) $ 36,646.78 $ 33,333.33 $ 23,263.73 $ (38,337.44)
CAT $ 20,373.73 $ 7,395.06 $ 16,834.78 $ 33,333.33 $ 27,968.58 $ 36,511.34
DWDP $ 60,912.73 $ 104,648.80 $ 15,750.18 $ 33,333.33 $ 36,033.43 $ (7,077.69)
EA $ 162,661.01 $ 293,879.63 $ 5,862.26 $ 33,333.33 $ 57,486.70 $ 71,846.00
EQR $ (64,462.90) $ (16,982.91) $ 4,233.55 $ 33,333.33 $ 19,283.46 $ 83,222.26
FSLR $ (51,776.13) $ (19,028.88) $ 1,274.40 $ 33,333.33 $ 13,193.82 $ 46,375.75
GS $ (22,970.83) $ (66,534.09) $ 17,816.34 $ 33,333.33 $ 30,073.55 $ (38,682.82)
HD $ 421,960.39 $ 636,300.19 $ 40,386.84 $ 33,333.33 $ 62,764.67 $ (21,050.12)
INTC $ 91,775.43 $ 173,565.81 $ 39,210.05 $ 33,333.33 $ 39,151.34 $ 15,363.25
JNJ $ 432,912.35 $ 590,266.44 $ 67,789.23 $ 33,333.33 $ 53,541.59 $ (58,365.94)
K $ (130,202.70) $ (46,882.69) $ 4,240.78 $ 33,333.33 $ 17,192.95 $ 86,129.74
KIM $ (234,309.23) $ (175,216.01) $ 1,397.32 $ 33,333.33 $ 7,710.40 $ 34,713.33
KO $ (224,980.81) $ (10,071.02) $ 35,374.19 $ 33,333.33 $ 22,693.62 $ 47,199.86
MAR $ 211,483.07 $ 446,070.05 $ 9,137.74 $ 33,333.33 $ 56,970.62 $ (37,844.13)
MCD $ 252,686.17 $ 487,024.65 $ 25,202.46 $ 33,333.33 $ 50,084.13 $ 77,413.71
NKE $ 37,916.29 $ 224,993.35 $ 18,447.22 $ 33,333.33 $ 41,511.93 $ 64,043.36
NUE $ (91,320.34) $ (115,369.39) $ 3,671.37 $ 33,333.33 $ 17,664.81 $ 52,025.72
PFE $ (64,576.11) $ 12,102.58 $ 38,938.90 $ 33,333.33 $ 27,467.74 $ 93,990.56
PG $ 60,741.35 $ 73,478.71 $ 42,353.58 $ 33,333.33 $ 27,989.78 $ 45,117.94
REGI $ 24,984.71 $ (7,355.47) $ 82,527.11 $ 33,333.33 $ 11,896.86 $ 78,601.01
T $ (12,666.17) $ (28,022.63) $ 43,152.57 $ 33,333.33 $ 20,923.34 $ 109,717.79
TSLA $ 92,977.53 $ 127,482.25 $ 9,392.62 $ 33,333.33 $ 42,223.63 $ 53,877.12
WBA $ (56,920.79) $ 65,835.33 $ 13,251.99 $ 33,333.33 $ 28,149.87 $ 81,903.29
WFC $ 108,546.43 $ 17,894.67 $ 54,439.62 $ 33,333.33 $ 33,565.07 $ 6,942.76
XOM $ (418,202.86) $ (484,416.07) $ 64,061.82 $ 33,333.33 $ 4,832.30 $ (13,617.21)
SP500 $ - $ (2,396,995.32) $ - $ - $ - $ -
SUM $ 1,000,000.00 $ 1,000,000.00 $ 1,000,000.00 $ 1,000,000.00 $ 1,000,000.00 $ 1,000,000.00

Table 4: Optimal absolute asset allocation suggested by the models

53
 The graphical standings of each portfolio’s expected per-period return and volatility
before the experiment in comparison with the Markowitz’s efficient frontier:

0.35% Ex-Ante Portfolio Efficiency

μ
MVF
0.30%
EF
0.25% MVO

0.20% TB
BL
0.15%
1/N

0.10% SRM
MDP
0.05%
SP500
SD
0.00%
0.00% 0.50% 1.00% 1.50% 2.00% 2.50%

Figure 4: Ex-ante portfolio efficiency

As this was the starting point of the experiment, there is no distinguishing between actively
and passively managed portfolios. However, from now onwards such distinguishing is necessary.
Therefore, the actively managed portfolios have an attribute (A) to their name, e.g. MVO (A).
Likewise, the passively managed portfolios have an attribute (P), e.g. MVO (P).

7.2. Results
This chapter presents the results the portfolios achieved at the end of the holding period
(HLDP). During the HLDP, active portfolios were daily rebalanced to maintain the initial optimal
allocation suggested by the models, and passive portfolios were left untouched on their own
according to the buy-and-hold strategy. After the HLDP, performance measures used within the
evaluation system, introduced in Chapter 7, were calculated. Each measure was calculated by using
the ‘annualized’ values. The ‘annualization’ was made to make the values consistent with the
holding period, which was 124 trading days. The holding period return on US 4 Week Treasury
bill, used as the risk-free rate, was 0.4611%. The holding period return on S&P 500, used as the
market return, was 0.8368%. This chapter is organized as a presentation of the individual results
with follow up commentary and is closed with general discussion regarding the final rankings.

54
 7.2.1. Presentation
In this section, each measure with individual results and corresponding portfolio rankings
is presented one by one. As risk is, generally, undesired, the charts of risk measures are sorted
from min to max to award the less risky portfolios on account of the riskier ones. As the other
measures are desired rather higher than lower, the corresponding charts are sorted from max to
min. Follow up commentary regarding the individual results and the chart with final rankings are
presented at the end.

1) EAR
Portfolios sorted by realized EAR from max to min. Max being the best.

EAR EAR
1th TB (P) 24.472%
2nd TB (A) 19.388% 25.000%
3rd MVO (P) 16.744%
4th MVO (A) 14.026% 20.000%
5th BL (P) 7.596%
15.000%
6th MDP (P) 7.155%
7th BL (A) 5.749% 10.000%
8th MDP (A) 4.995%
9th SRM (P) 3.099% 5.000%
10th 1/N (P) 1.277%
0.000%
11th SRM (A) 1.209%
12th S&P 500 0.837% -5.000%
13th 1/N (A) -0.564%
Table 5: EAR Figure 5: EAR

55
 2) Standard Deviation
Portfolios sorted by their SD from min to max. Min being the best.

Standard Deviation
1th MDP (P) 10.709% Standard Deviation
2nd MDP (A) 10.728%
25.000%
3rd 1/N (P) 11.331%
4th 1/N (A) 11.381% 20.000%
5th BL (P) 11.415%
6th BL (A) 11.462% 15.000%

7th S&P 500 11.557% 10.000%

8th SRM (P) 12.135%
9th SRM (A) 12.188% 5.000%
10th MVO (P) 20.448%
0.000%
11th MVO (A) 20.554%
12th TB (P) 22.685%
13th TB (A) 22.863%
Table 6: Standard deviation Figure 6: Standard deviation

3) Residual Standard Deviation

Portfolios sorted by their residual SD from min to max. Min being the best.

Residual SD
1th S&P 500 0.000% Residual SD
2nd 1/N (A) 2.652%
18.000%
3rd 1/N (P) 2.662%
16.000%
4th SRM (A) 2.746% 14.000%
5th SRM (P) 2.759% 12.000%
6th BL (A) 2.944% 10.000%
7th BL (P) 2.971% 8.000%
6.000%
8th MDP (A) 5.180%
4.000%
9th MDP (P) 5.199% 2.000%
10th MVO (P) 12.378% 0.000%
11th MVO (A) 12.426%
12th TB (P) 16.808%
13th TB (A) 16.975%
Table 7: Residual SD Figure 7: Residual SD

56
 4) Jensen’s Alpha
Portfolios sorted by their alpha from max to min. Max being the best.

Jensen's α Jensen's α
1th TB (P) 23.516%
2nd TB (A) 18.429% 25.000%
3rd MVO (P) 15.754%
4th MVO (A) 13.033% 20.000%
5th BL (P) 6.777%
15.000%
6th MDP (P) 6.390%
7th BL (A) 4.928% 10.000%
8th MDP (A) 4.228%
9th SRM (P) 2.253% 5.000%
10th 1/N (P) 0.457%
0.000%
11th SRM (A) 0.362%
12th S&P 500 0.000% -5.000%
13th 1/N (A) -1.385%
Table 8: Jensen's alpha Figure 8: Jensen's alpha

5) Beta
Portfolios sorted by their beta from min to max. Max being the best.

β
1th MDP (P) 0.810
β
2nd MDP (A) 0.813
1.600
3rd 1/N (P) 0.953
1.400
4th BL (P) 0.954
1.200
5th 1/N (A) 0.958
1.000
6th BL (A) 0.958
0.800
7th S&P 500 1.000
0.600
8th SRM (P) 1.022
0.400
9th SRM (A) 1.027
0.200
10th TB (P) 1.318
0.000
11th TB (A) 1.325
12th MVO (P) 1.408
13th MVO (A) 1.417
Table 9: Beta Figure 9: Beta

57
 6) Sharpe Ratio
Portfolios sorted by their Sharpe ratio from max to min. Max being the best.

Sharpe Ratio
1th TB (P) 1.058 Sharpe Ratio
2nd TB (A) 0.828
1.200
3rd MVO (P) 0.796
4th MVO (A) 0.660 1.000
5th MDP (P) 0.625 0.800
6th BL (P) 0.625
0.600
7th BL (A) 0.461
8th MDP (A) 0.423 0.400
9th SRM (P) 0.217 0.200
10th 1/N (P) 0.072
0.000
11th SRM (A) 0.061
12th S&P 500 0.033 -0.200
13th 1/N (A) -0.090
Table 10: Sharpe ratio Figure 10: Sharpe ratio

7) Treynor Ratio
Portfolios sorted by their Treynor ratio from max to min. Max being the best.

Treynor Ratio
1th TB (P) 0.182 Treynor Ratio
2nd TB (A) 0.143
0.200
3rd MVO (P) 0.116 0.180
4th MVO (A) 0.096 0.160
5th MDP (P) 0.083 0.140
6th BL (P) 0.075 0.120
0.100
7th MDP (A) 0.056
0.080
8th BL (A) 0.055 0.060
9th SRM (P) 0.026 0.040
10th 1/N (P) 0.009 0.020
11th SRM (A) 0.007 0.000
-0.020
12th S&P 500 0.004
13th 1/N (A) -0.011
Table 11: Treynor ratio
Figure 11: Treynor ratio

58
 8) Information Ratio
Portfolios sorted by their information ratio from max to min. Max being the best.

Information Ratio
1th BL (P) 2.281 Information Ratio
2nd BL (A) 1.674
2.500
3rd TB (P) 1.399
4th MVO (P) 1.273 2.000
5th MDP (P) 1.229
1.500
6th TB (A) 1.086
7th MVO (A) 1.049 1.000
8th SRM (P) 0.817
0.500
9th MDP (A) 0.816
10th 1/N (P) 0.172 0.000
11th SRM (A) 0.132
-0.500
12th S&P 500 0.000
13th 1/N (A) -0.522 -1.000
Table 12: Information ratio Figure 12: Information ratio
9) M^2
Portfolios sorted by their M^2 measure from max to min. Max being the best.

M^2
1th TB (P) 12.694%
M^2
2nd TB (A) 10.029%
14.000%
3rd MVO (P) 9.665%
4th MVO (A) 8.089% 12.000%

5th MDP (P) 7.685% 10.000%

6th BL (P) 7.685% 8.000%
7th BL (A) 5.793% 6.000%
8th MDP (A) 5.345%
4.000%
9th SRM (P) 2.973%
2.000%
10th 1/N (P) 1.293%
11th SRM (A) 1.171% 0.000%
12th S&P 500 0.837% -2.000%
13th 1/N (A) -0.580%
Table 13: M^2
Figure 13: M^2

59
 7.2.2. Commentary regarding the individual results
In this section, a brief commentary regarding the results across the performance measures
is provided. Assessment of overall rankings with discussion is presented in the next section.

It is convenient to begin with the risk measures as everything else is more or less connected
to them. The total risk of each portfolio was measured by its standard deviation. By far the most
volatile portfolios were the ones suggested by MVO and TB, regardless the management. Yet, the
difference of approximately 2 percentage points (p.p.) in favor of MVO counts for good. It is
interesting to see that this difference of ca. 2 p.p. represents an entire spread between the rest of
portfolios. Both MDP portfolios scored the lowest volatility and lived up to their name. The
portfolios on 3rd – 7th place manifested more or less identical volatility regardless the significant
differences in their allocations.

The level of carried systematic risk was measured by each portfolio’s beta. The chart of
individual betas corresponds, with small differences, to the one of SD. This is not surprising as
systematic risk is part of total risk. Interesting observation, S&P 500 with β = 1 ended up exactly
in the middle of beta chart, that is 6 portfolios with higher β and 6 with lower.

Specific risk was measured by each portfolio’s residual standard deviation. As any market
portfolio is considered to be perfectly diversified and thus to carry zero specific risk, it is logical
that S&P 500 won this category. Nonetheless, shall one compare all three risk measures, it is
interesting to see the diversification power of 1/N model as it scores high in all of them. As for the
residual SD, little bit disappointing are the results of MDP portfolios as they carry nearly double
of residual risk than other naïve portfolios plus BL. Unsurprisingly, MVO and TB portfolios ended
up dead last with residual volatility multiple times higher than its competitors. Taking a look at all
three risk measures, it is interesting to see that different portfolio management played insignificant
role as both (A) and (P) portfolios scored similar values per each method.

In the mean-variance framework, there should always be a positive trade-off between risk
and return. More risk should provide an adequate extra return. The total return of each portfolio
was measured by EAR. By far the highest return was achieved by the second riskiest portfolio TB
(P), followed by the riskiest TB (A). The absolute difference of 5.084 p.p. between them is a value

60
that nearly half of all portfolios didn’t even reach at all. 1/N (A) is the only portfolio that scored a
negative return.

Although measuring total return and risk is interesting, in order to be consistent with the
portfolio theory it is necessary to include risk-adjusted measures in the performance analysis as
they provide better informative value about how much was achieved with respect to the undertaken
risk.

Both Sharpe and Treynor ratios measure amount of excess return per one unit of risk,
differing in the risk measure used. SD and β, respectively. The winning portfolios in both measures
are TB. Since the main purpose of TB optimization is Sharpe ratio maximization, the model lived
up to its purpose. Winning in TR then comes hand in hand. TB portfolios are followed by MVO,
MDP and BL portfolios. In both ratios, 1/N (A) scored negative due to its negative excess return.

Jensen’s Alpha is a measure of superior performance that compares realized excess returns
with risk-adjusted excess returns. Positive α is what everybody seeks and the higher, the better.
The highest α was achieved by TB portfolios, followed by MVO. All portfolios achieved positive
α but one, 1/N (A). From obvious reasons, S&P 500 has α = 0.

Information ratio measures portfolio’s excess return relative to market per one unit of
tracking error. Positive IR means the portfolio has beat the market. All portfolios achieved positive
IR but one, 1/N (A). By far the winning portfolio was BL (P), followed by BL (A). TB (P) ranked
3rd, followed by MVO portfolios. From obvious reasons, S&P 500 has IR = 0

M^2 is a hypothetical measure of risk-adjusted return relative to market. It says what the
return should be if the portfolio’s SD equaled the market’s SD. Market M^2 = market return.
Therefore, portfolio desires M^2 return ≥ market return. The highest M^2 was achieved by TB
(P), followed by TB (A), MVO (P), and MVO (A), which are the portfolios that achieved the
highest EAR. The M^2 chart matches the Sharpe ratio chart. The only portfolios with M^2 below
market return is 1/N (A).

61
Portfolios sorted by their results in individual performance measures

max to min to min to max to min to max to max to max to max to

min max max min max min min min min
EAR SD Res SD α β SR TR IR M^2
1
th TB (P) MDP (P) S&P 500 TB (P) MDP (P) TB (P) TB (P) BL (P) TB (P)
2
nd TB (A) MDP (A) 1/N (A) TB (A) MDP (A) TB (A) TB (A) BL (A) TB (A)
3
rd MVO (P) 1/N (P) 1/N (P) MVO (P) 1/N (P) MVO (P) MVO (P) TB (P) MVO (P)
4 MVO
th MVO (A) 1/N (A) SRM (A) MVO (A) BL (P) MVO (A) MVO (A) (P) MVO (A)
5 MDP
th BL (P) BL (P) SRM (P) BL (P) 1/N (A) MDP (P) MDP (P) (P) MDP (P)
6
th MDP (P) BL (A) BL (A) MDP (P) BL (A) BL (P) BL (P) TB (A) BL (P)
7 MVO
th BL (A) S&P 500 BL (P) BL (A) S&P 500 BL (A) MDP (A) (A) BL (A)
8 SRM
th MDP (A) SRM (P) MDP (A) MDP (A) SRM (P) MDP (A) BL (A) (P) MDP (A)
9 MDP
th SRM (P) SRM (A) MDP (P) SRM (P) SRM (A) SRM (P) SRM (P) (A) SRM (P)
10
th 1/N (P) MVO (P) MVO (P) 1/N (P) TB (P) 1/N (P) 1/N (P) 1/N (P) 1/N (P)
11 SRM
th SRM (A) MVO (A) MVO (A) SRM (A) TB (A) SRM (A) SRM (A) (A) SRM (A)
12 S&P
th S&P 500 TB (P) TB (P) S&P 500 MVO (P) S&P 500 S&P 500 500 S&P 500
13
th 1/N (A) TB (A) TB (A) 1/N (A) MVO (A) 1/N (A) 1/N (A) 1/N (A) 1/N (A)

Table 14: Portfolio rankings by categories

62
Final ranking
EAR SD Res SD α β SR TR IR M^2
TB (P) 12 1 1 12 3 12 12 10 12
MDP (P) 7 12 4 7 12 8 8 8 8
BL (P) 8 8 6 8 9 7 7 12 7
MVO (P) 10 3 3 10 1 10 10 9 10
TB (A) 11 0 0 11 2 11 11 7 11
BL (A) 6 7 7 6 7 6 5 11 6
MDP (A) 5 11 5 5 11 5 6 4 5
MVO (A) 9 2 2 9 0 9 9 6 9
1/N (P) 3 10 10 3 10 3 3 3 3
SRM (P) 4 5 8 4 5 4 4 5 4
S&P 500 1 6 12 1 6 1 1 1 1
SRM (A) 2 4 9 2 4 2 2 2 2
1/N (A) 0 9 11 0 8 0 0 0 0
Table 15: Final point chart

SUM of Final
Points Rank Final Ranking
TB (P) 75 1st
MDP (P) 74 2nd 80

BL (P) 72 3rd 70
MVO (P) 66 4th 60
TB (A) 64 5th 50
BL (A) 61 6th 40
MDP (A) 57 7th 30
MVO (A) 55 8th 20
1/N (P) 48 9th 10
SRM (P) 43 10th 0
S&P 500 30 11th
SRM (A) 29 12th
1/N (A) 28 13th
Table 16: Final ranking Figure 14: Final ranking

63
 7.3. Discussion
This section provides a general discussion regarding the final results the portfolios
achieved in the experiment. The final results are presented in Table 16 on previous page.

The gold medal for winning the experiment goes to TB (P) portfolio, silver medal to MDP
(P), and bronze medal to BL (P). The imaginary ‘potato medal’ goes to MVO (P). It is pleasant to
see that first five positions are occupied by portfolios suggested by different models. The initial
expectation that active management leads to a superior performance has been disproved by the
experiment as the passive portfolios occupy the first four spots. This, however, is not a big surprise
should one take a closer look. In this experiment, the active management happened on daily basis
which turned out to be quite cumbersome as it showcased a number of shortcomings. Firstly,
assuming the initial optimal allocation static was wrong. The optimal allocation evolves over the
time as much as the values of assets themselves. Therefore, the rebalancing should not be
happening to match the initial optimal allocation, but rather the new optimum. Secondly, finding
new optimum and rebalancing to it on continuous basis is a challenging process with no guarantee.
Thirdly, daily rebalancing is costly. Should this experiment had considered the transaction costs,
the actively managed portfolios would have ended up probably dead last with profits completely
erased. These factors lead to a conviction that daily rebalancing makes no real sense. Rebalancing
happening at longer periods, such as quarterly, semi-annually, or annually, and with respect to the
new optimum, sounds more reasonable. Either or, the passive strategy buy-and-hold proved to be
more efficient throughout the experiment. These findings correspond to the outcomes of similar
studies regarding the issues of active versus passive portfolio management, for example Pace et
al. (2016) and Cox (2017).

The selected optimizers provided a balanced mix of sophisticated and naïve models. The
sum of points for sophisticated models is 393 versus 279 for naïve models. Nonetheless, this
victory does possess no informative value and can be considered as irrelevant. Each model
deserves an independent assessment.

All sophisticated models used in this thesis have few things in common, e.g. they require
an extensive collection of historical data. Since future is, from its very nature, unknown, any
estimation of inputs based solely on historical values is deceptive. Input estimates are therefore
almost always a pure noise but the models accept them as true and simply deliver a solution. This

64
is misleading as one may think the delivered solution is the ‘true optimum’. With this in mind, the
Black-Litterman model is the only model that somehow deals with it and allows the investor to
combine the observed market information with his or hers views about the future, which makes
the model somewhat prospective.

On the other hand, the issue of estimation error is well known and there are approaches for
its minimization. These models are usually based on Bayesian shrinkage, resampling, or robust
optimization. It would had been interesting to include these models in the experiment to see how
they would do.

Although the portfolios suggested by TB did well in the experiment, their results need to
be taken with a reserve as they are biased with the selected monetization of the index. In reality,
index cannot be bought just like that and an index tracking mutual fund, for example, is included
instead. Both TB and MVO optimizers suggested portfolios with quite large long/short positions.
This may not be an issue for institutional investors, but for individuals most certainly yes.

The most naïve optimizer, 1/N, performed well as a diversification tool. It scored high in
all three risk related measures. Nonetheless, both 1/N portfolios performed otherwise poorly which
is rather disappointing. Especially should one consider its success in other studies.

Both SRM portfolios showcased poor performance in all categories. Both delivered low
returns and carried considerable amount of risk, which, combined, resulted in low scores in all
risk-adjusted measures.

The last naïve optimizer, MDP, happened to be a pleasant surprise. Both portfolios
achieved solid scores in all risk-adjusted measures. In all but one categories MDP (P) was superior
to MDP (A). MDP (P) happened to be the least risky portfolio, but still delivered a reasonable
return of 7.155% with alpha 6.39%. Scoring relatively high in all the categories has brought the
MDP (P) to the final second place. However, keeping in mind the biasness of TB portfolios, this
silver medal has a golden glow.

The total amount of imaginary money invested into all 12 portfolios at the beginning of the
experiment was $ 12 000 000, with $ 1 000 000 each. How the portfolio values were developing
during the holding period is depicted on the following Figure 15.

65
 Portfolio Value Development
$1,400,000.00
MVO (A)
$1,350,000.00
$1,300,000.00 TB (A)

$1,250,000.00 BL (A)

$1,200,000.00 1/N (A)

$1,150,000.00 SRM (A)

$1,100,000.00 MDP (A)

$1,050,000.00 MVO (P)
$1,000,000.00 TB (P)
$950,000.00 BL (P)
$900,000.00 1/N (P)
2/1/2018 2/2/2018 2/3/2018 2/4/2018 2/5/2018 2/6/2018

Figure 15: Portfolio value development

As can be deciphered from the graph, all portfolios exhibit strong co-integration. This is
no surprise as all portfolios are composed of identical assets. The steep start of TB and MVO
portfolios is caused by their relatively higher betas in comparison to the rest of portfolios. The
effect of individual betas is also responsible for the spread variations throughout the entire holding
period. All portfolios but one, 1/N (A), show increasing trend. The final value of each portfolio
with the total return realized from the experiment is summarized in the following table.

Initial Value Final Value

MVO (A) $ 1,000,000.00 $ 1,140,262.18
TB (A) $ 1,000,000.00 $ 1,193,876.01
BL (A) $ 1,000,000.00 $ 1,057,493.73
1/N (A) $ 1,000,000.00 $ 994,355.69
SRM (A) $ 1,000,000.00 $ 1,012,094.80
MDP (A) $ 1,000,000.00 $ 1,049,947.67
MVO (P) $ 1,000,000.00 $ 1,167,443.93
TB (P) $ 1,000,000.00 $ 1,244,722.63
BL (P) $ 1,000,000.00 $ 1,075,961.58
1/N (P) $ 1,000,000.00 $ 1,012,766.18
SRM (P) $ 1,000,000.00 $ 1,030,987.77
MDP (P) $ 1,000,000.00 $ 1,071,553.88
SUM $ 12,000,000.00 $ 13,051,466.05
Profit $ 1,051,466.05
Return 8.762%
Table 17: Final values

66
8. Conclusion
The master’s thesis with a topic ‘Portfolio Optimization Methods, Their Application and
Evaluation’ focused on practical application of various quantitative portfolio optimization
methods, their performance, usefulness, pros and cons. The quantitative portfolio management is
a complete, data-driven process in which an investor builds, optimizes, holds, and adjusts
portfolios in order to achieve superior risk-adjusted returns with respect to the applied constrains,
such as his or her risk aversion, budget limitations, turnover constrains etc. The pivotal point of
the thesis was to conduct an experiment in which a number of selected optimizers were put in a
contest.

The underlying theory on which the experiment was built is presented in the first part of
the thesis. It introduces the underpinnings of portfolio theory, describes the optimization process,
introduces the selected optimization methods, and provides an overview of portfolio management.
The experiment itself, together with the results, is presented in the second part of the thesis, called
‘Practical Experiment’.

The selected optimizers represented both sophisticated and naïve models. The
sophisticated included the classical Markowitz’s mean-variance model, the Treynor-Black model,
and the Black-Litterman model. The naive included the 1/N model, the Sharpe ratio based model,
and the Most diversified portoflio model. The optimizers were applied to portfolios composed of
identical assets and held under different portfolio management styles over a pre-specified period
of time. The performance of each portfolio was measured ex-post, adequately evaluated in accord
with the criteria of the experiment, and confronted with the others.

The questions that this master’s thesis tried to find answers to were (1) which portfolio
optimizer, out of the selected ones, performs the best, and (2) whether it is beneficial to conduct
rather an active, or a passive portfolio management.

As presented in Section 7.2 devoted to the results of the experiment, the passively
management portfolios demonstrated superiority to the actively managed ones as they occupy the
first four spots in the final ranking. The expected superiority of sophisticated models was disproved
as both of the portfolios suggested by MDP ranked among them. Moreover, the passively managed
MDP portfolio left behind all portfolios but one, the Treynor-Black passive portfolio. The TB (P)

67
won the experiment with a final score of 75 points. However, the overall performance of TB
portfolios got biased by the selected monetization of the market index, and thus their results should
be taken with reserve. With respect to this bias, the logical move is to look at the second place
where shines the MDP (P) portfolio with a final score of 74 points, only 1 point behind the TB (P).

Is therefore the MDP the best optimizer and the buy-and-hold strategy the best
management? No and no.

Regarding the optimizers, the selected models created a balanced mix but represent only a
tip of an iceberg. Due to the steep development of quantitative finance since approximately the
1990’, the set of available optimizers has enlarged significantly. Models based on, for example,
resampled efficiency, robust optimization, Bayesian shrinkage, or on various downside risk
measures, have their undoubtful charm. Furthermore, should one consider the number of possible
ways for input estimations, the set of optimizers starts to grow exponentially. To make a relevant
statement, all the optimizers with all their possible adjustments would had needed to be included
in the experiment as well. Nonetheless, in order to conclude the experiment the MDP (P) approach
happened to be the most efficient one.

Regarding the management, the passive form has its undoubtful perks such as difficulty
and costs. It is easy to do and it has no turnover costs coming from the rebalancing. However, the
optimal allocation is dynamic and changes over time. Therefore, buy-and-hold strategy might
deviate too much from the optimum in the long run and can get possibly behind its potential.
Replicating an index may be a way, however full replication is expensive. Daily rebalancing is
unnecessarily complicated and expensive as well, which makes such management style ineffective.
The turnover costs can get eased by the additional income from dividends, yet way too frequent
rebalancing almost guarantees the costs > dividends inequality. Rebalancing at longer periods, e.g.
quarterly, semi-annualy, or annualy, and with respect to the new optimum, seems to be more
reasonable. Either or, the used management style should be periodically evaluated and adjusted
accordingly.

The world of quantitative portoflio management is evolving at a pace faster than ever. This
thesis has only scratched the surface, and even though this final chapter is called Conclusion, it is
rather a personal beginning.

68
Bibliography
 Amenc N, & Le Sourd V. 2003. Portfolio theory and performance analysis. Chichester:
John Wiley & Sons, Inc.
 Bacon, C. R. 2008. Practical portfolio performance: Measurement and attribution (2nd
ed.). Chichester: John Wiley & Sons, Inc.
 Beck, K. E. 2017. Asset pricing and portfolio choice theory (2nd ed.). New York: Oxford
University Press
 Black F, & Litterman R. 1992. Global Portfolio Optimization. Financial Analysts
Journal, 48: 28-43
 Bodie, Z., Kane A., & Marcus A. 2018. Investments (11th ed.). New York: McGraw-Hill
 Cox C. C. 2017. A comparison of active and passive portfolio management. Honors
thesis projects, University of Tennessee, Knoxville
 Danthine, J., & Donaldson, J. 2015. Intermediate financial theory (3rd ed.). Oxford:
Elsevier Inc.
 DeFusco, R. A., McLeavey D. W., Pinto J. E., & Runkle D. E. 2007. Quantitative
investment analysis (2nd ed.). Hoboken: John Wiley & Sons, Inc.
 DeMiguel, V., Garlappi L., & Uppal R. 2009. Optimal versus naïve diversification: How
inefficient is the (1/N) portfolio strategy? The Review of Financial Studies, vol. 22: 1915-
1953
 Elton, E. J., Gruber M. J., Brown S. J., & Goetzmann W. N. 2011. Modern portfolio theory
and investment analysis: International student edition (8th ed.). Singapore: John Wiley &
Sons (Asia) Pte Ltd.
 Esch, L., Kieffer R., & Lopez T. 2005. Asset and risk management: Risk oriented finance.
Chichester: John Wiley & Sons, Inc.
 Idzorek, T. M. 2005. A step-by-step guide to black-litterman model: Incorporating user-
specified confidence levels. Chicago: Ibbotson Associates.
https://corporate.morningstar.com/ib/documents/MethodologyDocuments/IBBAssociates/
BlackLitterman.pdf (accessed May 14, 2018)
 Kane, A., Kim T., & White H. 2003. Active portfolio management: The power of the
treynor-black model.

69
 http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.141.6284&rep=rep1&type=pdf
(accessed May 30, 2018)
 Kinlaw, W., Kritzman M. P., & Turkington D. 2017. A practitioner’s guide to asset
allocation. Hoboken: John Wiley & Sons, Inc.
 Levy, H., & Post, T. 2005. Investments. Harlow: Pearson education limited
 Litterman, B. 2003. Modern investment management: An equilibrium approach.
Hoboken: John Wiley & Sons, Inc.
 Maginn, J. L., Tuttle D. L., McLeavey D. W., & Pinto J. E. 2007. Managing investment
portfolios: A dynamic process (3rd ed.). Hoboken: John Wiley & Sons, Inc.
 Markowitz, H. 1952. Portfolio selection. The Journal of Finance, vol. 7: 77-91
 Michaud, R. O., & Michaud, R. O. 2008. Efficient asset management: A practical guide
to stock portfolio optimization and asset allocation (2nd ed.). New York: Oxford
University Press
 Pace D., Hili J., & Grima S. 2016. Active versus passive investing: An empirical study on
the US and European mutual funds and etfs. Contemporary Studies in Economic and
Financial Analysis, vol. 97: 1-35
 Plyakha, Y., Uppal R., & Vilkov G. 2012. Why does an equal-weighted portfolio
outperform value and price-weighted portfolios? Working paper, EDHEC Business
School
 Ross, S. 1976. The arbitrage theory of capital asset pricing. Journal of Economic Theory,
13: 341-360
 Scherer, B. 2015. Portfolio construction and risk budgeting (5th ed.). London: Incisive
Media Investments Ltd.
 Sharpe, W. F., Alexander G. J., & Bailey J. V. 1995. Investments (5th ed.). Englewood
Hills: Prentice Hall, Inc.
 Swensen, D. F. 2000. Pioneering portfolio management: An unconventional approach
to institutional investment. New York: The Free Press
 Treynor J. L., & Black F. 1973. How to use security analysis to improve portfolio selection.
The Journal of Business, 46: 66-86
 Walters, J. 2014. The black-litterman model in detail.
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1314585 (accessed May 15, 2018)
70
 Annex
Apple Inc. (AAPL)
Apple Inc. is an American multinational technology company with its main area of
business in designing, developing, and selling of consumer electronics, computer software, and
online services. The company was founded in 1976 and has its headquarters in Cupertino,
California. Since its foundation, Apple has become one of the largest companies in the world. It is
a component of S&P 100, S&P 500, DJIA, and Nasdaq 100 indices. The company is listed on
Nasdaq and its market capitalization on December 29, 2017 was $ 874,11bn.

Electronic Arts Inc. (EA)

Electronic Arts Inc. is an American company engaged in the video gaming industry. EA
was founded in 1982 and has its headquarters in Redwood city, California. EA is a component of
Nasdaq 100 and S&P 500 indices, is listed on Nasdaq, and its market capitalization on December
29, 2017 was $ 32,43bn.

Intel Corporation (INTC)

Intel Corporation is an American multinational technology company specialized in
invention, innovation, and production of computer hardware such as processors, motherboards,
network interface controllers and so on. Intel was founded in 1968 and has its headquarters in
Santa Clara, California. INTC is a component of the S&P 100, S&P 500, DJIA, and Nasdaq 100
indices, is listed on Nasdaq, and its market capitalization on December 29, 2017 was $ 216,91bn.

American Tower Corporation (AMT)

American Tower Corporation is a global provider of wireless and broadcast
communications infrastructure. The company was founded in 1995 and has its headquarters in
Boston, Massachusetts. AMT is a component of the S&P 500 index, is listed on NYSE, and its
market capitalization on December 29, 2017 was $ 61,23bn.

Equity Residential (EQR)

Equity Residential is an American real estate investment company. The company’s
investment model is focused on acquisition, development, and management of rental apartment
buildings. The company was founded in 1969 and is headquartered in Chicago, Illinois. EQR is a

71
component of the S&P 500 index, is listed on NYSE, and its market capitalization on December
29, 2017 was $ 23,42bn

Kimco Realty Corporation (KIM)

Kimco Realty Corporation is an American real estate investment trust with specialization
in the segment of shopping malls. The trust was founded in 1958 and is headquartered in New
Hyde Park, New York. KIM is a component of the S&P 500 index and is listed on NYSE. The
market capitalization on December 29, 2017 was $ 7,73bn.

Johnson & Johnson (JNJ)

Johnson & Johnson is an American multinational company specialized in production of
pharmaceuticals, health care products, and medical devices. The company was founded in 1886
and is headquartered in New Brunswick, New Jersey. JNJ is a component of the S&P 100, S&P
500, and DJIA indices. Its stock is listed on NYSE and its market capitalization on December 29,
2017 was $ 375,01bn.

Pfizer Inc. (PFE)

Pfizer Inc. is an American multinational pharmaceutical company and one of the global
leaders within its field. The company was founded in 1849 and has its headquarters in New York
City, New York. PFE is a component of the S&P 100, S&P 500, and DJIA indices, is listed on
NYSE, and its market capitalization on December 29, 2017 was $ 215,41bn.

Procter & Gamble Co. (PG)

Procter & Gamble Co. is an American multinational company with specialization in
production of personal care products, beauty care products, health care products, cleaning agents,
and similar. The company was founded in 1837 and has its headquarter in Cincinnati, Ohio. PG is
a component of the S&P 100, S&P 500, and DJIA indices, is listed on NYSE, and its market
capitalization on December 29, 2017 was $ 234,30bn.

Exxon Mobil Corporation (XOM)

Exxon Mobil Corporation is an American multinational company specialized in oil and
natural gas industry. The company was founded in 1999 and is headquartered in Irving, Texas.
XOM is a component of the S&P 100, S&P 500, and DJIA indices, is listed on NYSE, and its
market capitalization on December 29, 2017 was $ 354,39bn.

72
 Renewable Energy Group Inc. (REGI)
Renewable Energy Group Inc. is an American multinational company focused on
renewable energy with specialization in biofuel, biomass-based diesel, renewable chemicals, and
carbon lowering solutions. The company was founded in 1996 and is headquartered in Ames, Iowa.
REGI is listed on Nasdaq and its market capitalization on December 29, 2017 was $ 456,54bn.

First Solar Inc. (FSLR)

First Solar Inc. is an American multinational company involved in the photovoltaic
industry with focus on production of solar panels, rigid thin film modules, and similar. The
company was founded in 1999 and has its headquarters in Tempe, Arizona. FSLR is a component
of the S&P 400 index, is listed on Nasdaq, and it market capitalization on December 29, 2017 was
$ 7,05bn.

Citigroup Inc. (C)

Citigroup Inc. is an American multinational bank providing banking and financial services.
It is one of the biggest banks worldwide. It was founded in 1813 and is headquartered in New York
City, New York. C is a component of the S&P 100 and S&P 500 indices, is listed on NYSE, and
its market capitalization on December 29, 2017 was $ 202,73bn.

Wells Fargo & Company (WFC)

Wells Fargo & Company is an American multinational company providing financial
services of all kinds. The company was founded in 1852 and is headquartered in New York City,
New York. WFC is a component of the S&P 100 and S&P 500 indices, is listed on NYSE, and its
market capitalization on December 29, 2017 was $ 301,16bn.

The Goldman Sachs Group Inc. (GS)

The Goldman Sachs Group Inc. is an American multinational investment bank and
financial services provider. The company was founded in 1869 and has its headquarters in New
York City, New York. GS is a component of the S&P 100, S&P 500, and DJIA indices. Its stock
is listed on NYSE and its market capitalization on December 29, 2017 was $ 98,56bn.

Tesla Inc. (TSLA)

Tesla Inc. is an American multinational company specializing in electromobility, lithium-
ion energy battery storage systems, and solar panel production. The company was founded in 2003

73
and is headquartered in Palo Alto, California. TSLA is a component of Russell 1000 and Nasdaq
100 indices. The company is listed on Nasdaq and its market capitalization on December 29, 2018
was $ 51,96bn.

The Boeing Company (BA)

The Boeing Company is an American multinational company that specializes in
manufacturing of airplanes, rotorcraft, rockets, satellites, and missiles. The company was founded
in 1916 and is headquartered in Chicago, Illinois. BA is a component of the S&P 100, S&P 500,
and DJIA indices. Its stock is listed on NYSE and its market capitalization on December 29, 2017
was $ 174,32bn.

Caterpillar Inc. (CAT)

Caterpillar Inc. is an American multination company specializing mainly in invention and
production of heavy machinery and engines. However, it is also a provider of insurance and
financial services via its global site of subsidiaries. The company was founded in 1925 and has its
headquarter in Deerfield, Illinois. CAT is a component of the S&P 100, S&P 500, and DJIA
indices. The company is listed on NYSE and its market capitalization on December 29, 2017 was
$ 93,13bn.

The Kellogg Company (K)

The Kellogg Company is an American multinational company specializing in the food
processing industry. The company was founded in 1906 and has its headquarters in Battle Creek,
Michigan. K is a component of the S&P 500 index and is listed on NYSE. The market
capitalization on December 29, 2017 was $ 23,46bn.

The Coca Cola Company (KO)

The Coca Cola Company is an American multinational company specializing in production
of non-alcoholic beverages and syrups. The company was founded in 1886 and has its headquarters
in Atlanta, Georgia. KO is a component of the S&P 100, S&P 500, and DJIA indices. The company
is listed on NYSE and its market capitalization on December 29, 2017 was $ 195,69b.

McDonald’s (MCD)
McDonald’s is an American multinational company operating in the fast food and real
estate industry. The company was founded in 1940 and has its headquarters in Chicago, Illinois.

74
MCD is a component of the S&P 100, S&P 500, and DJIA indices. The company is listed on
NYSE and its market capitalization on December 29, 2017 was $ 139,42b.

AT&T Inc. (T)

AT&T Inc. is an American multinational company engaged in the telecommunications and
mass media industry. It is the world’s largest provider of telecommunications services and world’s
second largest provider of mobile services. The company was founded in 1983 and has its
headquarters in Dallas, Texas. T is a component of the S&P 100 and S&P 500 indices. The
company is listed on NYSE and its market capitalization on December 29, 2017 was $ 238,72b.

Amazon.com Inc. (AMZN)

Amazon.com Inc. is an American multinational company operating in the e-commerce and
cloud services business. Amazon.com is the biggest internet retailer in the world. The company
was founded in 1994 and has its headquarters in Seattle, Washington. AMZN is a component of
the S&P 100, S&P 500, and Nasdaq 100 indices. The company is listed on Nasdaq and its market
capitalization on December 29, 2017 was $ 561,79b.

Walgreens Boots Alliance Inc. (WBA)

Walgreens Boots Alliance Inc. is an American holding company created after a merge
between a US based The Walgreens Company and a Switzerland based Alliance Boots. The
company operates in the pharmaceutical industry as one of the largest drug and health care product
retail reseller. The Walgreens Company was founded in 1901 and the merge took place in 2014.
The company is headquartered in Deerfield, Illinois. WBA is a new ticker reflecting the merge but
otherwise it is a continuation of The Walgreens Company’s stock. WBA is a component of the
S&P 100, S&P 500, and DJIA indices. The company is listed on Nasdaq and its market
capitalization on December 29, 2017 was $ 73,31b.

Air Products & Chemicals Inc. (APD)

Air Products & Chemicals Inc. is an American multinational company operating within the
chemical and gas industry. The company sells its products primarily to other businesses. The
business was founded in 1940 and has its headquarters in Allentown, Pennsylvania. APD is a
component of the S&P 500 index, is listed on NYSE and its market capitalization on December
29, 2017 was $ 35,76bn.

75
 DowDuPont Inc. (DWDP)
DowDuPont Inc. is an American multinational company operating in the chemical
industry. The company is a merge product between The Dow Chemical Co. and E. I. du Pont de
Nemours and Company and the merge took place in 2017. The original companies were founded
in 1897 and 1802, respectively. The current DowDuPont Inc. has two headquarters in Midland,
Michigan and Wilmington, Delaware. DWDP is a component of the S&P 100, S&P 500, and DJIA
indices. The company is listed on NYSE and its market capitalization on December 29, 2017 was
$ 87,13b.

Nucor Corporation (NUE)

Nucor Corporation is an American company operating within the heavy metals industry
with specialization in production of steel and steel products. The company was founded in 1940
and has its headquarters in Charlotte, North Carolina. NUE is a component of the S&P 500 index,
and is listed NYSE. The market capitalization on December 29, 2017 was $ 20,31bn.

Nike Inc. (NKE)

Nike Inc. is an American multinational corporation operating within the textile industry.
Its main area of business is in design, development, manufacture, marketing, and sales of clothes
of all kinds including footwear, apparel, accessories etc. The company was founded in 1964 and
has its headquarters in Washington County, Oregon. NKE is a component of the S&P 100, S&P
500, and DJIA indices. The company is listed on NYSE and its market capitalization on December
29, 2017 was $ 102,05b.

Marriott International (MAR)

Marriott International is an American multinational company operating within the
hospitality industry. It is one of the world’s biggest operators of hotels and related lodging
facilities. The company was founded in 1927 and has its headquarters in Bethesda, Maryland.
MAR is a component of the S&P 500 and Nasdaq 100 indices. The company is listed on Nasdaq
and its market capitalization on December 29, 2017 was $ 50,55b.

The Home Depot Inc. (HD)

The Home Depot Inc. is an American multinational retail company specialized in home
improvement supplies, construction products, tools, and related services. The company was

76
founded in 1978 and has its headquarters in Cobb County, Georgia. HD is a component of the S&P
100, S&P 500, and DJIA indices. The company is listed on NYSE and its market capitalization on
December 29, 2017 was $ 223,42b.

Standard & Poor’s 500 (S&P500, ^GSPC)

Standard & Poor’s 500 is an American stock market index operated by the S&P Dow Jones
Indices LLP. The index comprises 500 US large cap public companies across all industries. S&P
500 is a market cap weighted, free float adjusted index. For the purposes of this thesis, the S&P
500 serves as a market indicator.

US 4 Week Treasury Bill (T-bill)

US 4 Week Treasury Bill is an American short-term government bond with maturity of 4
weeks. The bond has a standard denomination of $1000. The bond is rated AAA and thus is
considered risk-free. For the purposes of this thesis, the US 4 Week Treasury Bill serves as the
risk-free asset.

77