NHH Norwegian School of Economics

Bergen, Spring 2023

me
Del
Portfolio optimization of chartering
contracts
An empirical study of the chartering policies for Western Bulk Chartering ASA
in the period from 2016 to 2022

Magnus Klund Dalgaard & Øyvind Skjævestad

Supervisor: Roar Os Adland & Haiying Jia

Master thesis in Financial Economics (FIE) &

Business Analytics (BAN)

NORWEGIAN SCHOOL OF ECONOMICS

This thesis was written as a part of the Master of Science in Economics and Business
Administration at NHH. Please note that neither the institution nor the examiners are
responsible -− through the approval of this thesis -− for the theories and methods used,
or results and conclusions drawn in this work.
Abstract
This thesis looks at how the shipping operator Western Bulk can structure their chartering
decisions through a portfolio approach. This entails looking at chartering contracts as single
investments, which are part of a universal investment set. Looking at the period 2016-2022
we seek to discover how an optimal portfolio can be constructed and how it compares with
the actual operations of Western Bulk.

Traditionally portfolio optimization has been common within securities and corporate
finance. However, its applications have been extended to a wide array of industries. Our
paper builds on this and implements this approach into bulk shipping. Through the
Markowitz model we investigate what constitutes an optimal portfolio in terms of contract-
and geographical distribution. The optimal solution finds evidence for deviations to the
historic operational data of Western Bulk. These results and preceding calculations have not
been subjected to a full set of possible constraints. Thus, our findings do not provide a
complete realistic application to Western Bulk’s
Bulk's operations.

Page 1l of 41
Acknowledgements
This thesis has been written as the final assignment of the master profiles in Financial
Economics and Business Analytics. We would like to express our gratitude to our first
supervisor, Professor Roar Os Ådland for helping us along the journey. Your expertise in the
shipping business and academic writing have been invaluable for us. Moreover, we would
like to thank Egil Husby at Western Bulk for providing us with the data allowing us to
complete our thesis. Our appreciation extends to your comments along the way, as well as
providing us with valuable insights about the bulk shipping industry. Lastly, we would like to
thank Professor Haiying Jia for being our supervisor in the final stages of our thesis.

Bergen, May 2023

Magnus Klund Dalgaard Øyvind Skjævestad

Page 2 of 41
Contents
Introduction ........................................................................................................................................... 5

Literature review .................................................................................................................................. 6

Theory .................................................................................................................................................... 7
Chartering choices for a shipping operator ......................................................................................... 8
Cargo contracts ............................................................................................................................... 8
Hedging a contract .......................................................................................................................... 9
Different operational strategies ....................................................................................................... 9
Markowitz and portfolio optimization .............................................................................................. 10
Background for portfolio optimization ......................................................................................... 10
Shipping contracts available for portfolio optimization................................................................ 11
Framework of the Markowitz model ................................................................................................ 12
The universal investment set.
set ......................................................................................................... 12
Covariance .................................................................................................................................... 13
Covariance matrix ......................................................................................................................... 14
Expected return of a portfolio ....................................................................................................... 14
Portfolio variance .......................................................................................................................... 14
Sharpe ratio ................................................................................................................................... 15

Method ................................................................................................................................................. 15
Optimization problem ....................................................................................................................... 16
Portfolio efficiency ....................................................................................................................... 17

Data collection and validation............................................................................................................ 19

Data material ..................................................................................................................................... 19
Data processing and validation ......................................................................................................... 19
Descriptive statistics ......................................................................................................................... 21
Reference data - empirical analysis .................................................................................................. 22

Empirical results ................................................................................................................................. 24

Portfolio Optimization ...................................................................................................................... 25
Minimum Variance Portfolio ........................................................................................................ 27
Efficient frontier and optimal portfolio ......................................................................................... 27
Implications of results ....................................................................................................................... 32
ofresults
Robustness of data and results .......................................................................................................... 34
Simulation of random sampling .................................................................................................... 36

Page 3 of 41
 Discussion of limitations with the results ......................................................................................... 36

Concluding remarks and recommendation ...................................................................................... 38

38

References ............................................................................................................................................ 39
39

Page 4 of 41
Introduction
As a shipping company, Western Bulk is considered an "operator"
“operator” rather than a "shipowner".
“shipowner”.
This entails that Western Bulk does not own its shipping fleet, but contract in the vessels they
desire. These vessels are then put on appropriate charters, often based on decisions
concerning risks, profits, traditions and experience (Berg-Andreassen, 2011). As an operator
Western Bulk is positioned between vessel- and cargo owners but does not serve as broker.
Currently (10.02.2023) Western bulk have employment of 127 vessels on the water, serving
over 300+ customers worldwide. An operator offers two services in terms of service for
freight and trading, where trading accounts for the majority of operations. The trading
business model of Western Bulk consists of exploiting static arbitrage between chartering in
ships on a USD/day basis, before chartering out the ships on another period. To perform these
operations Western Bulk, rely on three tools: time charter (TC) contracts, cargo contracts and
FFAs (Husby, 2023).

Commonly an operator conducts different strategies depending on numerous factors.

Strategies can include chartering a ship and then either secure a cargo immediately or holding
the TC until the market is more favorable. The operator can also fix a TC in advance in hope
for a stronger market in the future. Strategies can also involve speculations on geographies
where providing vessels and/or freight to other markets could create a profitable spread.
FFAs also serve as a tool as timing and fluctuations on these contracts in different
geographies is an attractive trading opportunity for the operator (Husby, 2023).

For an operator it is important to handle the risk involved with its operations. Factors such as
chartered fleet size, exposure to different geographies and different cargo and vessels can be
important to ensure diversification. As there are known differences in freight rates between
geographies, modem
modern portfolio theory is a great tool to find a profit optimal allocation of
vessels.

Our problem statement relates to finding optimal chartering policies of Western Bulk. This is
done through an empirical study and portfolio optimization of data regarding chartering
policies of Western Bulk between 2016 -2022. The knowledge gained from this thesis
contributes to broadening the existing maritime academic literature. Further it will provide
useful insights into how an operator should allocate its capital to maximize returns. We hope
that our calculations can provide a new perspective to Western Bulk and other operators in
the dry-bulk market.

Page 5 of 41
The thesis is introduced by a literature review arguing for the relevance of our project. The
theory section that follows is divided between shipping theory and portfolio theory. The first
part describes the relevant contracts and how an operator may conduct business including
hedging. The second part involves description of our chosen theoretical modular foundation,
Markowitz portfolio theory. Key concepts such as the efficient frontier and portfolio variance
are introduced. The theoretical section is accompanied by the method section which describes
how our model operates. Further we describe our data and provide some key descriptive
statistics which will provide essential insights. This section will later serve as a benchmark
when we introduce our empirical results. Finally, we introduce our empirical results. The
thesis ends with critique of our findings and recommendations for Western Bulk.

Literature review
Our thesis seeks to explore an optimal portfolio of contracts for shipping operators. This is
inspired by the work of (Adland, Benth, & Koekebakker, 2018) and (Adland, Bjerknes, &
Herje, 2017) where they further prove that there is a freight rate premium on the cross-
Atlantic route in the dry bulk market. The purpose of this thesis is to extend the current
literature in dry bulk shipping by exploring how and if there exists an optimal portfolio of
contracts. To our knowledge there has yet been published an article with this specific topic in
mind, although there are published some articles with similar topics.

The articles of (Carisa, Tsz, & Siu, 2019) and (George Alexandridis, 2018) seek to find
optimal portfolios of vessels/ contracts. The article by Carisa et.al is particularly interesting
as this article constructs portfolios of carefully selected vessel types to find an optimal
mixture of different types and size of vessels. Although our thesis does not account for
different vessel types the mean variance approach is similar to our own methodology. The
article also approaches the market as an operator. The works of Cullinane hold great
relevance. By considering shipowners financial commitments as investments, he constructs
an optimal portfolio theory using Markowitz modem
modern portfolio theory in the dry-bulk market
(Cullinane K. , 1995). Different to our approach is that choice of corporation. Our approach is
solely used on an operator contrary to Cullinane where the company owns the vessels.
Additionally, our work does not factor in several types of vessels. Still this article holds great
relevance for our approach and is similar to our own simulations. The works of (Adland &
Jia, 2017) finds that there are some diversification benefits when increasing the fleet size due

Page 6 of 41
to the diversification on geography and time. Our model is based on a large number of trades
and will also show benefits of diversification in terms of optimal profits.

Our attention is further focused on relevant articles from other markets. Our chosen
methodology is applied to numerous fields spanning from the securities market to the
selection of military equipment (Sokri, 2012). In the following section we will only list a
select number of relevant articles displaying some of the variability in applied methods.
Articles written on markets/ companies who operate in similar way as Western Bulk will be
mentioned.

The articles of (Algarvio, Lopez, Sousa, & Lagarto, 2017) and (Atmaca & Gökgöz, 2012) are
of interest. The first articles explore how an electricity retailer can optimize its profits by
combining a portfolio of four different contracts (Algarvio, Lopez, Sousa, & Lagarto, 2017).
The second article investigates the Turkish electricity market. This article also focuses on the
allocation between contracts and how that affects its revenue potential. Both articles use
Markowitz’
Markowitz' portfolio theory and contain roughly the same approach as this thesis applied on
another market.

Theory

In the subsequent analysis it is assumed that the shipping operator seeks to find an optimal
portfolio of contracts. The process entails a maximization of return on investments, while
managing risk at an appropriate level. Consequently, the operator will be able to employ the
ships in the voyage and (or) time charter markets. At the same time contracts of affreightment
(CoA) can be applied for hedging. The chosen chartering strategy will thus determine if the
1
operator goes long or short on tonnage
tonnage1. . Given these aspects, the limitations that fleet size
and composition have on the business options of the operator is removed (Berg-Andreassen,
2011). The shipping operator will then be able to allocate capital between a wide array of
charters, where the importance of a well-constructed portfolio of contracts is vital for
company performance (Carisa, Tsz, & Siu, 2019).

1
The term shipowner is used in the literature. However, the methods and implications described are of equal
relevance to an operator. Thus, the term operator is used in accordance with the mentioned literature.

Page 7 of 41
Understanding the composition of an optimal portfolio entails assessing various practical
calculations2. This can include constraints on the
constraints in the subsequent optimization calculations2.
concentration of the portfolio in segments, which can be contract types and geographical
region. Having this constraint will prevent overexposure in one segment, reducing risk of
drawdowns if the segment proves to be wrong. Furthermore, the general structure of the
trades can also be included as a constraint, which includes the availability of time charters.
Additionally, working capital constraints will be possible. This will ensure that capital is not
depleted (locked) if an investment opportunity should arise.

These constraints are all related to practical matters of Western Bulk's

Bulk’s operations. Including
them will improve the realism and potential impact of the portfolio optimization. This will
require extensive information about the operations of Western Bulk. We will not present such
metrics and have opted for a less practical approach in the forthcoming calculations. Doing so
will still keep the theoretical significance of our results intact. Our approach will therefore
have a more theoretical approach as these constraints are not included.

Chartering choices for a shipping operator

In the following subchapter the thesis will briefly discuss the relevant contracts used in
Western Bulk’s
Bulk's operations. The first half of the operations is to hire a vessel. The two
relevant options are Voyage Charter (Trip Charter), Time Charter and Contract of
Affreightment. Where the trip charter is a contract for a specific voyage between two specific
ports the time Time Charter is leasing of a vessel for a specific period (Stopford, 1997).

Cargo contracts

On the other side of the operations is the freight contracts. These contracts are often settled
with a broker as an intermediary. The contracts are ether customized as a form of time charter
or trip charter. The specific rate is determined by the underlying contract and the price per ton
of the specific cargo. In sum the price which the cargo owner is paying is determined by the
current rate of the specific cargo multiplied by the VC/TC contract (Stopford, 1997).

2
Assessment of constraints on shorting are mentioned later in the thesis.

Page 8 of 41
Hedging a contract

In addition to the contracts described above Freight Forward Agreements serve as a useful
tool in Western Bulks operations. A typical starting point for a shipping operator is to lease a
vessel through a period. The period is most of the time set as an interval. This means, for
instance, that for a period of 3-5 months the operator is obligated to lease the vessel for a
minimum of three months but have the option of delaying the return of the vessel by an
additional two months (Husby, 2023). The uncertainty and pricing of optionality is indeed a
part of the business model for many ship operators. A TC usually holds two sources of
optionality: the extension and the redelivery area optionality (Adland & Prochazka, 2021).
The extension of the chartering period is important as it enables the charterer to fully
capitalize on a potentially strong market. The charterer extends his period if the rates lead to
net positive profits and conversely redelivers the vessel at minimum chartering period if the
market is weaker. The area optionality is also important as it can enable the charterer to take
advantage of the documented fronthaul-backhaul in the dry-bulk market (Adland &
Prochazka, 2021).

Freight Forward Agreements (FFAs) also make a suitable hedging opportunity for an
operator. As the name suggests a Freight Forward Agreement (FFA) is a forward contract for
vessels and is typically used by operators/charterers as a hedging strategy to ensure
predictable freight rates (Kasimati & Veraros, 2018). For the operator these contracts serve as
an opportunity to earn additional revenue if the market if favorable or hedge against further
volatility. Although the process might seem simple (Adland & Jia, 2017) elaborates on
several sources for deviation between a hedge revenue stream and a sport rate. Important
factors include technical specifications, actual operating speed, geographical trading patterns
and timing mismatches. Although FFAs are an important strategy for an operator they will
not serve as a specific portfolio choice in our model. For that reason, FFAs will not be
discussed any further.

Different operational strategies

To understand how a shipping operator might conduct business we will showcase some
operational trade-offs that the operators must consider. Assuming that the operator charters a
ship it has to decide on the contract depending on the speculative forward-looking view of the
operator. As the ship is chartered the operator needs to fix a cargo to freight. As different
regions specialize in a selective number of products the operator has to carefully consider

Page 9 of 41
where the ship is chartered as well as in which region the vessel operates in. Based on the
ship operator’s
operator's knowledge and experience it will try to position the ship in a region where it
expects the rates to excide expected level (Husby, 2023).

How the rates fluctuate is determined by several factors including macroeconomic

developments, geopolitics, results of harvest, climate etc. (Stopford, 1997). The operators’
operators'
profits will be a function of the difference in rates and how it's
it’s able to maximize the
utilization of the ship. A number of fixtures can be made to make profits, including chartering
ships in one region and then take on cargo with a loss in order to move the vessel to another
region with a potential profitable freight rate (Husby, 2023).

In addition to the operational decision the timing of entering the specific contracts will have
great impact on the overall results. By agreeing to transport a cargo in advance the operator
can speculate on falling chartering rates. Furthermore, one can charter a vessel in the belief
that the market will evolve positively or to move the vessel to a more profitable region to
earn profits from unmet demand. Lastly the operator can speculate with hedging in FFAs.

Markowitz and portfolio optimization

Background for portfolio optimization

Western Bulk is faced with a variety of investment opportunities in their chartering decisions.
In a perspective of investment planning and risk management, the choice of charter
investments can be determined through portfolio optimization. The concept of portfolio
optimization is common in corporate finance and its applications extend to shipping, as well
as other industries (Lorange, 2009). Often referred to as Modem
Modern Portfolio Theory (MPT), the
1952.33 He initially published the theory in the
concept was introduced by Harry Markowitz in 1952.
Journal of Finance under the title "Portfolio
“Portfolio Selection”,
Selection", where he presented his findings on
the principles of diversification in portfolio selection. Using the same methodology, we can
apply it to the case of Western Bulk.

Comparing portfolio management in traditional stock markets with shipping markets, we find
a number of similarities but also differences. The approach in the stock market is twofold

3
Harry Markowitz, "Portfolio
“Portfolio Selection”,
Selection", Journal of Finance, March 1952.

Page 10 of 41
through, a stock picking approach or an index-based portfolio management approach
(Lorange, 2009). Conversely, the strategic composition of income yielding chartering
contracts can be determined through a portfolio approach. This constitutes that shipping
operators have a given preference in their risk/return payoff, in deciding which trades to take
part in (Berg-Andreassen, 2011).

Moreover, stock market portfolio management allows for differentiation between asset
selection, risk selection and market timing. The same goes for shipping portfolios when a
shipping operators’
operators' financial commitments are considered as investments (Cullinane K. ,
1995). The shipping portfolio will then differentiate between leverage decisions (risk), asset
mix selection and chartering/trading strategy (timing). Risk selection will therefore come
from both asset selection and leverage, which in turn
tum determines the return conditional on
expected normal performance (Lorange, 2009).

As mentioned, the Markowitz model of portfolio selection includes calculations of expected

returns, risks and risk attitudes. Applying these measurements, one could derive any optimal
portfolio to given risk/return requirements for the available set of market investments.
Finally, the application of the model treats market conditions and the shipping operator’s
operator's risk
preference as inputs in the selection of an optimal portfolio (Cullinane K. , 1995).

Shipping contracts available for portfolio optimization

The available set of investments in our thesis is comprised of three strategies. Corresponding
with each strategy is the use of specific contracts, which are the following: Spot, Short Period
and Forward Cargo. A spot contract has the shortest duration and includes fixing a cargo
forward (usually 30 days), before fixing a TC to cover the cargo. Short period contracts
include fixing a vessel for a period of 4-6 months, where the aim is making generating profits
from trades. This is a long position and aims to take advantage of rising market sentiment,
while also factoring in geographical and basis risk. The contract for forward cargo is similar
to the spot contract but is considered as a pure short position. This stems from the cargo
having a laycan which is 30 days (or more) ahead in time at the time of fixing (Husby, 2023).

These three contract types are then available for a variety of trades across the world. In our
paper we have grouped these trades to three main geographical areas: Atlantic, Indian Ocean
and Pacific. All three contracts are available for each region, which constitutes 9 possible
contracts or investment opportunities. This investment set could have been extended to

Page 11 of 41
include another contract type for index vessels. As this is a relatively new contract type for
Western Bulk it has few data points relative to the other contract types. Hence, it was chosen
to omit this contract from the investment set.

In the forthcoming methodology and calculations, our model does not include the possibility
of shorting these contracts directly. Doing so does not exclude the possibility of shorting
assets the investment set. As stated above the forward contract is a pure short position.
Hence, the possibility of shorting is still present. The contracts can therefore be considered to
be a form of "market
“market neutral".
neutral”. A "market
“market neutral”
neutral" approach seeks to profit from such
mispricing of assets and creating a portfolio of said assets (Jacobs & Levy, 2005). This is
directly related to the business model of Western Bulk, which seeks to capitalize on market
inefficiencies and price movements.

Framework of the Markowitz model

The universal investment set

Applying the theory of Markowitz, a universal set of available market investments must be
present. Thus, it is necessary to identify these investments, which constitute the universal set
of investments. By allocating available funds between these individual "assets",
“assets”, the
combination will make out the portfolio for the shipping company (Cullinane K.
K . , 1995). To
illustrate the available set of investment set we can look at the figure of the minimum
variance frontier below.

Page 12 of 41
 Efficient Frontier

• •
Global • -- l_mUvidual
Minimum- Assets
½ii1am:e- M in i llflUllfl-Val'i.ance Frontier
rtfolio

Figure 1l - The minimum variance frontier of risky assets (Bodie, Kane, & Marcus, 2014)

The figure depicts a set of n individual assets (investments) available to the investor. The
efficient frontier portrays the lowest possible variance achievable for a given expected
portfolio return. Given the available individual investments, we need to consider their
properties. This includes return of assets, variance and standard deviation (SD). Additionally,
these returns are likely to differ in terms of observations. E.g., contracts will have different
lengths, making returns occur at different intervals. Thus, it will be necessary to annualize the
returns to make them comparable investments. Through diversification among these
investments, it will be possible to obtain a portfolio which is more efficient than holding a
single asset (Berk & DeMarzo, 2017).

Covariance

From the same set of investments, the covariance between the return of investment 𝑖𝑖i and 𝑗𝑗j is
measured. Through covariance we can measure the relationships between two sets of
observations, which is necessary in determining the risk of a portfolio (Cullinane K. , 1995).
The covariance is calculated using the following equation:
𝑚𝑚
m
1
𝐶𝐶𝐶𝐶𝐶𝐶
C o v (𝑥𝑥
( x𝑖𝑖i ,, 𝑥𝑥 = 𝑚𝑚
x𝑗𝑗j)) = m i L ∑(
−1
- 𝑥𝑥̅xi)(xk1-
( x k𝑥𝑥𝑘𝑘;𝑖𝑖 − 𝑖𝑖 ) (𝑥𝑥𝑘𝑘𝑗𝑗 − 𝑥𝑥
̅x1)
𝑗𝑗 )
(1.))
(1.
k=l
𝑘𝑘=1

Here the equation depicts the covariance between investment 𝑖𝑖i and 𝑗𝑗.
j.

Page 13 of 41
Covariance matrix

The full set of calculations of variances and covariances can be stored in a matrix. This
matrix contains the properties needed to calculate the risk of a portfolio, and takes the
following form (Carisa, Tsz, & Siu, 2019):

𝜎𝜎12 𝜎𝜎1,2 ⋯ 𝜎𝜎1,𝑁𝑁

𝜎𝜎 𝜎𝜎22 ⋯ 𝜎𝜎2,𝑁𝑁
𝐕𝐕𝑁𝑁∗𝑁𝑁 = 2,1
⋮ ⋮ ⋱ ⋮
𝜎𝜎 𝜎𝜎𝑁𝑁,2 ⋯ 𝜎𝜎𝑁𝑁2 ] (2.))
(2.
[ 𝑁𝑁,1

Since the set of returns from n

𝑛𝑛 possible investments can be used, the matrix is symmetrical of
2
order n* 𝑛𝑛. Thus, the main diagonal consists of the variance (𝜎𝜎
𝑛𝑛 ∗ n. (cr2) ) of the investments. Along
the diagonal we have symmetrical covariances of the returns (Sydsæter, Hammond, & Strøm,
2012).

Expected return of a portfolio

The return for each individual investment can be denoted by 𝐸𝐸[𝑟𝑟𝑖𝑖 ]. Through the process of
E[ri]-
portfolio optimization, a portfolio will be comprised of various assets with different expected
returns. Taking the weight invested in each of the assets in the portfolio will give us the
overall return of the portfolio, as shown here:

L
𝑛𝑛
n

𝐸𝐸(𝑟𝑟
E(rp)𝑃𝑃 ) = wi𝑖𝑖 ∗* 𝐸𝐸(𝑟𝑟
= ∑ 𝑤𝑤 E(ri)𝑖𝑖 )
(3.))
(3.
i=l
𝑖𝑖=1

The weight of total capital allocated in each investment is given by 𝑤𝑤

wi𝑖𝑖 (Bodie, Kane, &
Marcus, 2014).

Portfolio variance

The calculation of the portfolio variance follows the same intuition as with the portfolio
return. For the variance and covariance of assets in the portfolio, we sum the product together
to obtain the variance of the portfolio:

Page 14 of 41
 =LL
n
𝑛𝑛 n
𝑛𝑛

𝑉𝑉𝑉𝑉𝑉𝑉(𝑟𝑟
V ar(rp 𝑃𝑃 )) =
= 𝜎𝜎𝑃𝑃2 =
CJffe ∑ ∑ 𝑤𝑤
wi𝑖𝑖 ∗
* 𝑤𝑤
wj𝑗𝑗 ∗
*C𝐶𝐶𝐶𝐶𝐶𝐶(𝑟𝑟
o v t r𝑖𝑖. ,, Tj)
𝑟𝑟𝑗𝑗 )
(4.))
(4.
i = l 𝑗𝑗=1
𝑖𝑖=1 j=l

=LL
n
𝑛𝑛 n
𝑛𝑛

𝜎𝜎𝑃𝑃2 = (5.))
(5.
𝑉𝑉𝑉𝑉𝑉𝑉(𝑟𝑟
V ar(rp 𝑃𝑃 )) =
= CJffe ∑ ∑ 𝑤𝑤
wi𝑖𝑖 ∗
* 𝑤𝑤
wj𝑗𝑗 ∗ 𝜎𝜎𝑖𝑖𝑖𝑖
* CJij
i = l 𝑗𝑗=1
𝑖𝑖=1 j=l

𝜎𝜎𝑃𝑃2 (Bodie, Kane, & Marcus, 2014).

The portfolio variance is denoted by CJffe

Sharpe ratio

As seen in Figure 1J we want to hold a portfolio along the minimum variance frontier,
preferably in the northwest region. Thus, it becomes evident that the MVP does not offer the
best risk return tradeoff. Assessing this tradeoff is done through the Sharpe Ratio, which is
given below:

𝐸𝐸(𝑟𝑟
E(rp)𝑃𝑃 ) − 𝑟𝑟𝑓𝑓
- rf
𝑆𝑆ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 Ratio
Sharpe 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅 = S
𝑆𝑆𝑃𝑃p =
= SD(rp)
𝑆𝑆𝑆𝑆(𝑟𝑟𝑃𝑃 ) (6.))
(6.

By maximizing the Sharpe ratio, we obtain the optimal portfolio to hold together with the
risk-free asset (𝑟𝑟 ) . In the investment set this is represented by the capital allocation line,
( r𝑓𝑓f ).

which is tangent to the efficient frontier (Berk & DeMarzo, 2017).

Method
This section of the thesis seeks to highlight the quantitative methodology applied to solve the
problem statement. Through application of concepts described in the theory section, the
Markowitz portfolio theory can be used in this setting. Complemented by the use of
mathematical software in Excel, we can obtain the desired output of minimum variance
4
portfolio, efficient frontier and optimal portfolio4..

4
The choice of risk-free rate in this setting is described in the results section under "Efficient
“Efficient frontier and
optimal portfolio”.
portfolio".

Page 15 of 41
Optimization problem

Through portfolio optimization, the goal is often to find a mix of assets/investments which
yields the highest return for a given risk level. Conversely, the goal can be set to yield a
minimum of risk for a given level of
ofrisk return. (Carisa, Tsz, & Siu, 2019). With the mentioned
ofreturn.
investment set of n𝑛𝑛 assets/investments, it is possible to create various portfolios minimizing
risk for different levels of return. These portfolios make out the minimum variance frontier,
depicted in Figure 2, which also includes the minimum variance portfolio (MVP) (Bodie,
Kane, & Marcus, 2014). Obtaining the MVP is conditional on the following optimization
problem:

=LL
n𝑛𝑛 n
𝑛𝑛

𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀:
Minimize: 𝑉𝑉𝑉𝑉𝑉𝑉(𝑟𝑟𝑃𝑃 ) =
Var(rp) = 𝜎𝜎𝑃𝑃2 =
CJffe ∑ ∑ 𝑤𝑤
wi𝑖𝑖 ∗
* 𝑤𝑤
wj𝑗𝑗 ∗ 𝜎𝜎𝑖𝑖𝑖𝑖
* CJij (7.))
(7.
i = l 𝑗𝑗=1
𝑖𝑖=1 j=l

Subject 𝑡𝑡𝑡𝑡:
𝑆𝑆𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢𝑢 to:

I
n
𝑛𝑛
(8.))
(8.
𝐸𝐸(𝑟𝑟𝑃𝑃 ) =
E(rp) * 𝐸𝐸(𝑟𝑟
wi𝑖𝑖 ∗
= ∑ 𝑤𝑤 E(ra 𝑖𝑖 )
i=l
𝑖𝑖=1

n
𝑛𝑛

∑
I w𝑤𝑤i𝑖𝑖 =
= 1l (9.))
(9.
i=l
𝑖𝑖=1

𝑤𝑤𝑖𝑖 ≥ 0 (10.))
(10.
The goal of the minimization problem is to find the portfolio with the smallest variance,
given by equation 8. This goal is constrained by equation (13.) to (14.). Equation (12.) states
that the sum of weights invested in the assets is equal to the return of the portfolio. The sum
of these weights needs to be equal to 1,
l, as equation (13.) states. Finally, the problem does not
allow for short selling, where the weight of any asset can’t negative5.
can't be negative5.

5
As mentioned under "Shipping
“Shipping contracts available for portfolio optimization”, shorting is available through the
parifolio optimization",
forward contract which is a pure short position.

Page 16 of 41
Portfolio efficiency

Having summarized the risk return opportunities with the minimum-variance frontier, we can
determine the optimal portfolio. This portfolio will be situated in the northwestern region of
the minimum-variance frontier (Bodie, Kane, & Marcus, 2014). Together with the capital
allocation line (CAL) it is possible to locate the optimal portfolio. The CAL is a straight line
taking the following form (Elton, Gruber, Brown, & Goetzman, 2014):

𝐸𝐸(𝑟𝑟𝑃𝑃 ) −
- r𝑟𝑟1𝑓𝑓 )
𝐶𝐶𝐶𝐶𝐶𝐶
C AL= = r𝑟𝑟𝑓𝑓f +
+ CJ1
𝜎𝜎𝐼𝐼 (
( E ( r p𝜎𝜎
)𝑃𝑃
CJp ) (11.))
(11.

The slope of the CAL is equal to the Sharpe ratio and its intercept is given by the risk-free
rate 𝑟𝑟r1
𝑓𝑓 (Elton, Gruber, Brown, & Goetzman, 2014). Choosing the appropriate risk-free

measure will be directly related to the duration of contracts available. The choice differs from
traditional equity portfolios and the background for the chosen rate is given in the results
section. As the slope of the CAL increases it eventually becomes tangent to the efficient
frontier, which can be seen in Figure 2 below:

E(rl
CA!L{P)
.,, Eff lclerrt Frontier
"'

Figure 2 - The efficient portfolio (Bodie, Kane, & Marcus, 2014).

From the figure we can see that the CAL and increases conditional on the Sharpe ratio. As
mentioned, the Sharpe ratio is equal to the slope. When the slope of the CAL is tangent to the
efficient frontier, the optimal portfolio is obtained in this tangency point. Here the Sharpe

Page 17 of 41
Ratio is maximized, and no other portfolios will offer better risk-return combinations (Bodie,
Kane, & Marcus, 2014). The optimal portfolio can therefore be found by constructing the
following maximization problem:

( r 𝑃𝑃p)) −
_ ( E𝐸𝐸(𝑟𝑟 𝑟𝑟𝑓𝑓
- r1)
Maximize:
𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀: 𝑆𝑆𝑃𝑃 =
Sp - ( ) (12.))
(12.
𝜎𝜎𝑃𝑃p
CJ

Subject 𝑡𝑡𝑡𝑡:
𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆 to:

I
𝑛𝑛
n

𝐸𝐸(𝑟𝑟
E(rp)𝑃𝑃 ) = 𝑤𝑤𝑖𝑖 ∗* 𝐸𝐸(𝑟𝑟
= ∑ wi E(ra 𝑖𝑖 ) (13.))
(13.
i=l
𝑖𝑖=1

N
𝑁𝑁

I∑w 𝑤𝑤i𝑖𝑖 =
=1l (14.))
(14.
i=l
𝑖𝑖=1

𝑤𝑤𝑖𝑖 ≥ 0 f𝑓𝑓𝑓𝑓𝑓𝑓
or 𝑎𝑎𝑎𝑎𝑎𝑎
all 𝑖𝑖i (15.))
(15.

The problem described above does not allow for short sales, making the weight in each
6
asset/contract positive
positive". . The weight imposed on the assets in the optimal risky portfolio is
denoted by wi.
𝑤𝑤𝑖𝑖 . As given above, the problem is a quadratic programming problem and can be
solved through statistical computer packages (Elton, Gruber, Brown, & Goetzman, 2014).

Having found the optimal risky portfolio through this calculation, the investor (here Western
Bulk) will finally assess the allocation of funds between this portfolio and the risk-free asset.
Doing so will determine the optimal complete portfolio (Bodie, Kane, & Marcus, 2014). By
assessing the risk-return tradeoff through a utility function, it is possible to determine the
optimal portfolio with an analysis of historical freight rates, vessel types and charter types
(Carisa, Tsz, & Siu, 2019). Any utility function for an investor will likely have individual
differences in the risk return tradeoff. This is emphasized in Modem
Modern Portfolio Theory as

6
As mentioned under "Shipping
“Shipping contracts available for portfolio optimization",
optimization”, shorting is available through the
forward contract which is a pure short position.

Page 18 of 41
indifference curves represent the investor's
investor’s attitude towards risk (Bodie, Kane, & Marcus,
2014). However, this thesis will not look further into this extension, as we primarily look for
the objectively optimal portfolio.

Data collection and validation

Data material

The data set used for this thesis consists of four strategies under three main regions. The four
strategies are spot-spot, short period, index vessel and forward cargo while the main regions
are the Pacific Ocean, Atlantic Ocean and Indian Ocean. These contracts stretch from yearly
2015 till mid-February 2023. While the start-date of the contracts cannot be set after the data
was collected the termination/ end-date of the contracts stretches as far as October 2025.
Structurally the data is divided between the four strategies. Within each strategy key data
includes a start and end of contract, portfolio, net trading results (Net TC), total voyage days,
time charter equivalent earnings and TC (cargo cost). Net TC, voyage days and TC are
essential when we calculate the returns for the individual contracts. In addition to the data
provided by Western Bulk we have collected data from the underlying Baltic 10TC
l OTC
SupraMax Index from Clarkson's.
Clarkson’s. This data is paired with the index strategy to fully
calculate the costs under this strategy as the costs are reflected as a percentage of the dry bulk
index. It is assumed that all contracts under Index strategy is conducted with the similar
vessel type.

Data processing and validation

The data from Western Bulk was provided over four strategies, where they commonly differ
in duration. This is of relevance as we need to have comparable returns from between each
contract type. The spot contracts have the shortest length with an average duration of 33 days
in the period 2016-2022. Similarly, over the same period the forward contracts have an
average duration of 36 days. Contrary, the short period contracts are longer and usually last
for a period of 4-6 months. The longest contracts are held with index vessels and have a
duration for 6-12 months.

Additionally, the data arrived in a format where many contracts were not assigned
specifically to one of the three mentioned regions specifically. Each contract was therefore

Page 19 of 41
assigned to a specific region based on the fronthaul departure port. For example, a vessel
from Atlantic to the Pacific where assigned to the Atlantic-region For a spot contract the
location of the starting port determined its region, also when the travel is cross-regional. For
contracts of longer duration with multiple port entries and departures the contracts are
assigned to a region if most ports are in one geographical area.

To compare the returns between each contract, we have annualized the returns. Thus, our data
have been aggregated from the mentioned durations to a yearly perspective. We then assumed
that the returns from each contract have linear yearly returns. This assumption was chosen
over compounding returns, as that would incur significantly biased and unlikely large yearly
returns. As the contracts for spot, short period and forward are not part of a market they can
be created any day of the year. Thus, the linear return for a year is calculated in the following
manner.

365
365 ) (15.))
(15.
𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿
Linear 𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦
yearly 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟
return = = Contract 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 (( l
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 return h f . d. 'd l
engt 𝑜𝑜𝑜𝑜
𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙𝑙ℎ m wi ua 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐
o 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖 contract
)

In order to calculate the returns for the index vessels, the contracts had to be "matched"
“matched”
against the underlying Baltic 10TC
l OTC Supramax Index. It is assumed that this strategy is
entirely based in the Supramax segment. Data is downloaded from Clarksons Research and
filtered for the same period as the data from WB (Clarksons Research, 2023). Where the
index contracts are dealt with over a time period the BDI is given for every trade day. To
overcome the evident difference in time we have matched all BDI’s
BDI's within the index contract
time and then made an average. To derive the cost the BDI average is multiplied by the
percentage BDI index value and then multiplied with its respective length of contract, similar
to the calculation of the other contracts. This computational approach solves the time period
issue, while also providing a reasonable approximation of the cost that were carried by WB of
the respective contracts. Since the BDI index is part of a regulated market venue, the
possibility to place positions is limited to time of opening and closing. Hence the number of
trading days are limited, making the linear return take the following form (Odean, 1999).

252
252 (16.))
) (16.
Linear 𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦
𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 yearly return = Contract
𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 = 𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟𝑟 ∗* (( l
𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶𝐶 retrun h f . d. 'd l
engt 𝑜𝑜𝑜𝑜
𝑙𝑙𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒ℎ m wi ua 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑎𝑎𝑎𝑎𝑎𝑎
o 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣 contract
)

Page 20 of 41
Descriptive statistics

The section of descriptive statistics is meant to provide some basic insights into the data. It
will also be held as a benchmark comparison to the optimal portfolio displayed in the next
section.

Number of contracts per region Atlantic Indian Pacific

Spot-Spot •
Short Period 359 159 230 748
Index Vessel 60 15 16 91
Forward Cargo 192 51 143 386 ..
Table 1J -– overview of contract for different regions aggregated across all years.

Firstly, we glance at the overview of the contracts, its composition between different
contracts and the contracts geographic origin. For the interest of our analysis, we can clearly
see that there is high number of observations of all contracts excluding the index vessels. As
there are few observations spanning over a total of 8 years (2015-2023) the index vessel
contract type is removed in our empirical analysis. Few observations increase the impact of a
single observation and may skew the overall average and standard deviation affecting our
model. This issue will be further discussed in the empirical analysis. Further analysis reveals
that the data span from April 2015 till February 2023. Observations are quit evenly split
between all full years.

Closest to our empirical results is how the different geographies affect the overall
profitability of Western Bulk. The graph below depicts how the different areas affect the
overall profitability.

Page 21 of 41
Page
 Yearly returns per region

250
250

200
200

150
150
RETURN (%)

100
z 100
0:::
:J
I-
LU
so
50
0:::
0
0
2016 2017 2018 2019 2020 2021 2022
-50
-50

-100
-100
YEAR

Pacific Indian Atlantic

Figure 3 -– Yearly returns for each geographical region (2016-2022).

As the graph shows, the three main geographical areas are contributing differently to the
overall profits. While some trades and strategies might be profitable in one year, we can
clearly see that markets change, and different allocation could be optimal for different years
and market circumstances. The main takeaway is that all three geographies prove profitable
for most years, but all areas lose money in at least one year. When reading the graph, it is
important to recognize that this graph does not account for the composition of different
contracts or the inherent risk following those contracts. This is merely a brief overview of
how the returns are made between the main geographies.

Reference data - empirical analysis

Below, table 2 is depicting capital allocation/ investments across the 5 periods. Which
regions that are receives the largest investments vary from each period. This insight is
important to bear in mind when we later review our empirical analysis. Table 2 will serve as a
benchmark comparison for our model. As the number of datapoints varied substantially we
have compiled the 7 years of data points in to 5 periods, as depicted in table 3.

Page 22 of 41
 Period Atlantic
Atlantic Indian Pacific
Pacific

1l 48,88 % 20,80
I : I %
', 30,33 %

2 33,35 % 25,73 % 40,92 %

3 48,63 % 21,69 % 29,68

29,68%%

4 30,61 % 25,89 % 43,50 %

5 40,37 % 23,53 % 36,11 %

Table 2 --Allocated
Allocated capital in main geographical regions per period.

Choosing to structure the data in the abovementioned periods stems from the number of
observations in our data set. For each individual year in the period 2016 to 2022 the number
of observations had substantial variations. This poses an issue in terms of the robustness of
portfolio calculations and available contracts in the portfolios77.. Thus, we aggregated the data
over a longer timespan to omit this issue. By doing this we have obtained the following
periods to be used in our calculations.

Period 1:
l: 2016 and 2017

Period 2: 2018 and 2019

Period 3: 2020 and first half of 2021

Period 4: second half of 2021 and 2022

Period 5: 2016 to 2022

The periods 1-4 are relatively equal in terms of observations and the timespan are at most 2
years. Period 5 constitutes the whole period, meaning all the years from 2016 to 2022.

Furthermore, Table 3 is added as an additional benchmark for the empirical analysis. As

portfolios themselves might provide some insights, our goal is to achieve meaningful insights
8
in how Western Bulk might optimize its capital allocation
allocation8. .

7
The choice of periods is discussed further in the section "Robustness
“Robustness of data and results”.
results".
8
8
Capital allocation stems from the section "Background
“Background for portfolio optimization"
optimization” where contracts are
considered investments. Consequently, the overall investments can be considered capital allocation.

Page 23 of 41
 .
Spot Period Atlantic Indian Pacific
1 I •. ' •.
2 58,61 % 10,41 % 30,97 %
3 52,83 % 24,57 % 22,60 %
4 38,33 % 34,66 % 27,01 %
5 45,93 % 25,89 % 28,18 %
Short 1 48,19 % 22,31 % 29,50 %
2 40,07 % 24,38 % 35,55 %
3 26,52 % 28,60 % 44,88 %
4 25,20 % 29,26 % 45,54 %
5 37,19 % 25,46 % 37,35 %
Index 1 99,21 % 0,00% 0,79 %
2 93,37 % 6,59 % 0,04%
3 77,69 % 22,17 % 0,14%
4 52,89 % 46,55 % 0,56%
5 85,57 % 14,27 % 0,16%
FC 1 62,00 % 1,66% 36,34 %
2 50,28 % 6,88 % 42,83 %
3 47,69 % 13,99 % 38,31 %
4 47,77 % 16,11 % 36,11 %
5 48,79 % 13,88 % 37,33 %
Overall allocation " 39,69 %" 24,09 %" 36,22 %.

Table 3 - Summary of capital allocation by region and contract.

The table depicts the relative capital allocation in each strategy in its respective region and
period. An interesting observation is that Western Bulk allocates a large proportion of its
capital to the Atlantic region across all years and strategies. This observation could indicate
the well-known Atlantic premium in dry bulk shipping. It could also be a reflection of other
factors such as the trading performance of the "Atlantic
“Atlantic trading desk”,
desk", or that WBs
headquarters is in the Atlantic basin.

Empirical results
In this section we put forward our empirical results, as well as discussing the implications and
limitations of them. First, we look at the portfolio optimization between contract types in the
five mentioned periods between 2016 and 2022. Here we look at both minimum variance
portfolios (MVPs) and optimal portfolios. Secondly, we interpret these results and provide a
discussion of the implications of these results for Western Bulk. Finally, we will discuss the
robustness of the results and provide a discussion of limitations of the same results.

Page 24 of 41
Portfolio Optimization

Following the operational returns from Western Bulk in the period 2016-2022, we have
obtained various results regarding portfolio optimization of contracts. From the data used in
this thesis, a possible number of 12 contracts were available. As mentioned previously this
included four contract types over 3 available geographical regions. After applying the
framework of Markowitz, we obtained both MVPs and optimal portfolios for the five time
periods. These results are summarized in Table 10 and Table 11
JOand JJ below.

Pivotal to the portfolio optimization of Markowitz is the covariance among available assets in
the investment set. Due to differences in the number of observations between the contracts to
Western Bulk, this causes some contracts to be left out of the calculation. The reason behind
this is described in further detail in the subchapter "Robustness
“Robustness of data and results”.
results". As a
result, the available investment set contains 6-7 contracts instead of the possible maximum of
12. Addressing this issue is Period 5, which includes all 9 possible contracts. As mentioned,
this period considers the whole period 2016-2022. The designated set for each period is given
in the table below.

Atlantic- Indian- Pacific Atlantic- Indian- Pacific- Atlantic- Indian- Pacific-

Spot Spot Spot Period Period Period Forward Forward Forward
Return{%) I ,, a I I

2016-2017
SD 9,98 1,82 0,75 0,90 5,30

Return{%) 370,51 % 262,12 % -42,09 % 13,80 % -5,42 % 276,67 %

2018-2019
SD 7,58 6,19 1,97 0,92 1,77 10,27

Return{%) 234,20 % 196,25 % 99,94 % 64,88 % 58,91 % -123,30 %

2020-2021
SD 13,92 15,97 6,24 1,99 1,27 5,58 4,36

Return{%) I 371,83 % 187,11 % 2,79 % 427,54 % 327,93 %

2021-2022
SD 4,81 11,09 8,31 1,23 7,91 5,91

Return{%) 269,86 % 349,62 % 239,74 % 12,68 % 191,69 % 20,69 % 207,99 % 265,47 % 235,26 %
2016-2022
SD 9,39 12,58 6,70 1,84 1,92 1,29 7,67 7,48 6,99

Table 4 -Returns(%)
– Returns (%) and standard deviation (SD) for contracts in the investment set in all
periods.

Given this reduced investment set, the subsequent portfolios offer less diversification
possibilities. However, this is still preferable as the data set contains more observations on the
contracts.99 With the chosen investment set we calculated the covariance between
remaining contracts.

9
9
The data set and following calculations/results are discussed later in the results chapter, under "Robustness
“Robustness of
data and results”.
results".

Page 25 of 41
each contract, before creating a covariance matrix for each period. These matrices are all
summarized in tables 5 to 9.

Atlantic- Pacific- Atlantic- Indian- Pacific- Atlantic-

Spot Spot Period Period Period Forward
Atlantic-Spot 2,29215
Pacific-Spot -0,13790 1,52554
Atlantic-Period -0,03633 -0,21865 0,21856
Indian-Period 0,03433 0,03440 -0,00839 0,01722
Pacific-Period -0,02211 0,04523 0,00472 0,00626 0,02399
Atlantic-Forward -0,06597 -0,03572 -0,02780 0,03228 -0,00417 1,02976

Table 5 - Covariance matrix

matrix/or
for period 1:
l: 2016 and 2017.

Atlantic- Pacific- Atlantic- Indian- Pacific- Atlantic-

Spot Spot Period Period Period Forward
Atlantic-Spot 2,12334
Pacific Spot
PacificSpot -0,16395 1,36438
Atlantic-Period -0,07642 -0,04666 0,12140
Indian-Period -0,03459 -0,00782 0,00214 0,01955
Pacific-Period 0,00084 0,00403 0,00598 -0,00014 0,00394
Atlantic-Forward -0,29718 0,18933 0,19678 -0,02511
-0,02511 0,04297 2,15897

Table 6 - Covariance matrix for period 2: 2018 and 2019.

Atlantic- Indian- Pacific- Indian- Pacific- Atlantic- Pacific-

Spot Spot Spot Period Period Forward Forward
Atlantic-Spot 10,18841
I : : '.4
Indian-Spot -2,33216 9,51801
Pacific Spot
PacificSpot -1,06746 0,73202 1,35951
Indian-Period 0,04072 -0,20068 0,08495 0,16405
Pacific-Period -0,27913 0,15868 0,00635 -0,00978 0,06376
Atlantic-Forward 0,22658 -0,66235 -0,05352 0,00116 -0,01964 0,98538
Pacific-Forward -0,47072 -0,43274 0,25293 0,01194 0,03342 0,02600 0,76028

Table 7 - Covariance matrix for period 3: 2020 and first half of 2020.

Atlantic- Indian- Pacific- Pacific- Atlantic- Pacific-

' .. .
Spot Spot Spot Period Forward Forward
Atlantic-Spot 0,69341
Indian-Spot -0,05291 2,27063
Pacific Spot
PacificSpot -0,10202 0,04546 2,36426
Pacific-Period 0,04908 0,09823 -0,03998 0,05163
Atlantic-Forward -0,16626 -0,56694 0,14225 -0,00901 1,12701
Pacific-Forward -0,05801 -0,38784 -0,02541 -0,04006 0,29350 0,93249

Table 88-- Covariance matrix/or

matrix for period 4: second half of
o/2021
2021 and 2022.

Page 26 of 41
 Atlantic- Indian- Pacific- Atlantic- Indian- Pacific- Atlantic- Indian- Pacific-
Spot Spot Spot Period Period Period Forward Forward Forward
Atlantic-Spot
Indian-Spot 0,46318 5,32592
Pacific-Spot 0,18592 -0,31273 0,59159
Atlantic-Period 0,05069 0,09488 -0,04957 0,10307
Indian-Period -0,06246 -0,00599 -0,00079 -0,00019 0,01512
Pacific-Period -0,02317 0,00198 0,00384 -0,00307 0,00317 0,00963
Atlantic-Forward 0,01480 0,19984 -0,09719 -0,05324 -0,01203 -0,00451 0,64363
Indian-Forward 0,24963 -0,29234 0,22946 -0,03427 -0,02016 0,00157 0,00999 0,70750
Pacific-Forward 0,02592 -0,54746 0,04341 -0,13497 -0,00478 0,03665 0,04787 0,14734 1,62059

Table 99-- Covariance matrix for period 5: 2016 to 2022

Minimum Variance Portfolio

Using the abovementioned covariance tables, we are able to construct the Minimum Variance
Portfolios. Applying this with the mentioned methodology in a statistical computational
software as Excel, the MVPs are summarized in the table below.

Atlantic- Indian- Pacific- Atlantic- Indian- Pacific- Atlantic- Indian- Pacific-

Perie d E[r] (%)
Spot Spot Spot Period Period Period Forward Forward Forward

l I I ', 0,00 % 7,07 % 60,57 % 32,33 % 0,00 % 0,11 26,88 % 2,38

2 I • I ', 0,30 % 0,02 % 26,52 % 72,56 % 0,00 % 0,06 2,7 4 % 0,07

0,60 % 2,03 % 23,21 % 65,06 % 4,58 % 2,16 % 0,18 76,43 % 4,17

0,00 % 3,18 % 85,80 % 2,12 % 7,53 % 0,20 45,41 % 2,15

0,00 % 0,99 % 8,77 % 33,52 % 51,75 % 2,67 % 1,35 % 0,00 % 0,07 90,16 % 11,96

Table 10
l 0 - Minimum Variance Portfolios (MVPs) for all periods (2016-2022).

The table depicts all five assigned periods in the timespan of 2016 to 2022. Note that “-”
that"-"
refers to the contract not being part of the investment set in the corresponding period.
Looking at the table it is evident that period contracts dominate the %-share in the portfolio
for each period. In a risk management perspective, this can indicate that period contracts are
preferable in terms of minimizing risk.

Efficient frontier and optimal portfolio

Moving on from the Minimum Variance Portfolio, we use its implications in determining the
efficient frontier and the subsequent optimal portfolio. Using the MVP as a starting point we

Page 27 of 41
have calculated the efficient frontier for each of the five periods. These are all depicted in
Figures 4 to 8 below. Also included in the figures are the capital allocation line (CAL) and
the optimal (tangent) portfolio.

Assessing the optimal portfolio requires utilization of the risk-free interest rate. In our case
we are faced with an investment set with assets/contracts of different lengths. This differs
from traditional equity portfolios comprised of e.g., stocks, where this is not an issue. The
risk-free security should therefore optimally, be equal in duration with these contracts
(Emilson, 2020). During the period of 2016-2022 the spot-, period- and forward contracts
had an average duration of 33, 132 and 36 days respectively. Given these numbers a contract
has an average duration of 67 days, or 2,2 months. If adjusted for the percentage invested in
each contract (majority in period contracts) the average duration is 78 days, or 2,6 months.
Hence, a risk-free security with a duration of 3 months will be appropriate to match the
duration of the available contracts.

For the duration of

o f 3 months, the US 3-month Treasury bill (T-bill) is commonly used as the
risk-free asset in financial practice (Samo
(Sarno & Thornton, 2003). This asset is therefore chosen
as the determinant of the risk-free rate in our case of portfolio optimization (MarketWatch,
2023). With the approach of matching this risk-free rate with the average contract duration,
we assume that the yield curve is upward sloping. This entails that the interest rate of the bill
increases as the time to maturity increases. Investors are more exposed to fluctuations in the
interest rate with longer durations and will therefore require compensation accordingly
(Kloster, 2000).

Applying the risk-free rate in the calculation of a portfolio with maximum Sharpe Ratio
yields the optimal portfolio (Figures 4-8). The %-wise composition of contracts in each
optimal portfolio has been summarized in the table below. In the same table the standard
deviation (SD), expected return and Sharpe Ratio for each portfolio is also given.

Page 28 of 41
Period Atlantic- Indian- Pacific Atlantic- Indian- Pacific- Atlantic- Indian- Pacific-
E[r] (%)
Spot Spot Spot Period Period Period Forward Forward Forward

12,16 % 25,23 % 37,84 % 0,20 77,54 % 3,93

2 • ', 10,81 % 0,00 % 68,61 % 0,00 % 8,28 % 0,23 106,28 % 4,44

1,24 % 6,28 % 26,47 % 58,64 % 4,03 % 0,00 % 0,19 90,02 % 4,63

19,68 % 3,48 % 0,00 % 28,56 % 19,55 % 0,42 335,17 % 7,93

5 •. t 0,48 % 1,58 % 4,69 % 83,42 % 0,00 % 3,61 % 3,45 % 1,09 % 0,10 189,74 % 18,61

Table 11 - Optimal (tangent) portfolios for all periods (J

(1 to 5).

Compared to the MVP portfolios one can observe that a large share of the portfolio still
consists of period contracts. However, this share has been somewhat smaller, as a larger
share of spot- and forward contracts are now included in the portfolios. This coincides with
what was shown in the investment set, where the spot contracts generally offer superior
returns compared to the other contract types. Keep in mind that the spot contracts are riskier
and have significant volatility, thus limiting the share of it which can be placed in the optimal
portfolios.

Efficient Frontier and Optimal Portfolio

300,00%

250,00%

200,00%
RETURN (%)

z
150,00%
t:i
0:::

100,00%

50,00%

0,00%
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
STANDARD DEVIATION (SD)
STANDARD (SD)

Efficient Frontier - CAL

CAL e Tangent Portfolio e MVP
MVP

Figure 4 - Efficient frontier and optimal portfolio in Period 1:

J: 2016
20J6 and 2017.
20J 7.

Page 29 of 41
 Efficient Frontier and Optimal Portfolio
400,00%

350,00%

300,00%

250,00%
RETURN (%)

z 200,00%
0:::
:J
I-
LU
0:::
150,00%

100,00%

50,00%

0,00%
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1
STANDARD DEVIATION (SD)
STANDARD

Efficient Frontier -- CAL e Tangent Portfolio e MVP

MVP

Figure 5 --Efficient
Efficient frontier and optimal portfolio in Period 2: 2018 and 2019.

Efficient Frontier and Optimal Portfolio

250,00%

200,00%

g:150,00%
RETURN (%)

z
0:::
:J
t:i
o:::100,00%

50,00%

0,00%
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1

STANDARD DEVIATION (SD)

STANDARD

Efficient Frontier - CAL

CAL e Tangent Portfolio e MVP

Figure 6 --Efficient
Efficient frontier and optimal portfolio in Period 3: 2020 and first half of 2021.

Page 30 of 41
 Efficient Frontier and Optimal Portfolio
500,00%

450,00%

400,00%

350,00%

~ 300,00%
e
RETURN (%)

z
250,00%
I-
LU
o:::200,00%

150,00%

100,00%

50,00%

0,00%
0 0,1 0,2 0,3 0,4 0,5 0,6 0,7
STANDARD DEVIATION (SD)
STANDARD (SD)

Efficient Frontier -- CAL

CAL e Tangent Portfolio • MVP

Figure 77-Efficientfrontier
- Efficient frontier and optimal portfolio in Period 4: second half of
o/2021
2021 and 2022.

Efficient Frontier and Optimal Portfolio

300,00%

250,00%

200,00%
RETURN (%)

z
150,00%
t:i
0:::

100,00%

50,00%

0,00%
0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5
STANDARD DEVIATION (SD)
STANDARD

Efficient Frontier - CAL

CAL • Tangent Portfolio
Tangent e MVP

Figure 8 --Efficient
Efficient frontier and optimal portfolio in Period 5: 2016 -2022.
– 2022.

Page 31 of 41
Implications of results

From the optimal portfolios we can map out what is the ideal distribution of contracts for
Western Bulk. From all five periods we have looked at we can see a general pattern between
the contract types. The majority will be held in period contracts, with two significantly
smaller positions in spot and forward contracts. Keep in mind that this pattern is subject to
deviations in each period, but generally keeps its form among these three contract types.
Previously mentioned constraints on segment concentration are possible and could be
implemented in the optimal portfolios. Given the large share of period contracts in all
periods, it could be reasonable to apply such a constraint in practice.

Additionally, the results from these portfolios relate to what would have been optimal
portfolios ex-post. This would imply that Western Bulk have perfect knowledge of what
would be the outcome of each 2-year period. Having the foresight and understanding of such
allocation changes between these periods can be considered difficult and unlikely in a real-
world setting. Thus, we have included results over a longer period (2016 -– 2022), referred to
as Period 5. This will provide a more realistic assessment of what could be considered
common conditions for Western Bulk. Moreover, looking at a longer period will make it
more likely to coincide with shipping cycles and patterns.

Derived from the same optimal portfolios we obtain info on the geographical placement of
these contracts. This insight would benefit the desks of Western Bulk, as their operations are
structured in separate units. Each of their desks covers a respective ocean basin, and with our
results they are provided with a benchmark of where to place their contracts. Using the
optimal portfolios in each period, the distribution among the operational areas is the
following.

Page 32 of 41
 Geographical distribution of contracts
contracts
90,00%

80,00%

70,00%

60,00%

--··
50,00%
% share

(11
.c.
; 40,00%

30,00%

20,00%

10,00%

0,00%
1l 2 3 4 5
Period
A t l aAtlantic
n t i c (optimal)
{optimal) Indian (optimal)
{optimal) Pacific (optimal)
{optimal)

- • - Atlantic (actual)
{actual) - • - Indian (actual)
{actual) - • - Pacific (actual)
{actual)

Figure 9 - Actual and optimal geographical distribution of contracts per period.

Looking at the distribution obtained from optimal portfolios (solid lines), the figure depicts
significant variations among the distribution in each region. In all periods except period 1J one
region predominantly outweighs the others, with a share of over 50%. This pattern is
reinforced if we look at the full period/all years (Period 5), where the share in Indian
contracts is equal to 87%. If we compare this with the historical figures for Western Bulk, we
Ifwe
will be able to obtain more insight into how their contracts can be distributed among regions.

In the same figure we have included the historical figures for invested capital among regions
(dashed lines). The historical distribution among the regions deviates to what is implied by
the optimal portfolios. Most noticeable is that the optimal portfolio gives a larger variation in
regions than historical distributions. For example, the share of capital allocated in Indian is
significantly smaller than what has shown to be optimal in certain periods. This is also the
case for the full period when comparing the actual and optimal share.

Based on what the optimal portfolio suggests, it would be preferable to increase the share in
the Indian region. However, it would be problematic from a risk management perspective to
use these results without adjustments. Applying the insights from the optimal portfolio would
entail having 87% of contracts in the Indian region. This constitutes a portfolio with too great
a concentration in one geographical segment. Having this overexposure increases the risk of
Page 33 of 41
possible drawdowns, if wrong about the region. As mentioned in the theory section about
potential constraints on portfolio composition, it would be possible to include constraints on
geographical segment exposure.

Implementing these constraints would be determined by Western Bulk and their

preferences/practices. Not knowing these metrics, we have not included these constraints in
our preceding calculations and results. Consequently, the share in the Indian region will be
below the abovementioned optimal share, conditional on what the chosen constraints are.
This depicts a limitation regarding our results, as the potential realism and impact is
somewhat unadjusted. Thus, our results lean towards a higher degree of theoretical relevance,
compared to practical uses. By including these and other constraints, it would improve the
practical application and realism of our results.

Given the cyclicality of the bulk freight market, a portfolio approach may indicate certain
portfolio patterns (Figure 9) which coincide with various parts of the cycles. We know that a
complete short shipping cycle consists of four stages: trough, recovery, peak and collapse
(Scarsi, 2007). Note that the duration this paper has examined is 7 full years (2016-2022),
which coincides with the industry rule of thumb for duration of a cycle (Chistè
(Chiste & van
Vuuren, 2013). Thus, meaning that by understanding how optimal portfolios of contracts
have been historically, it can be used in setting a basis for distribution of future contracts.
However, it is not given that these periods coincide perfectly with each other. It would
therefore be preferable to look at this over a longer period and investigate if these cycles
indicate a pattern in terms of portfolio allocation.

Robustness of data and results

Contrary to stock prices (and returns), the returns to Western Bulk are not equal in terms of
daily quotes. The observations of movements in price between two or more stocks are
commonly quoted in equal observations, making comparison an easier task than with the case
of Western Bulk. Since the returns Western Bulk achieves are from contracts with various
durations, the observations among the contracts also vary. This poses a problem in
determining the covariance between contracts, as covariance calculation requires data sets of
equal length. To cope with this, we have opted for random sampling in obtaining an equal
number of observations between contracts.

Page 34 of 41
The discrepancy between contracts is obvious in all intra year periods from 2016 to 2022.
Thus, we have joined various periods together to obtain more observations in each period.
Doing this allows for a greater number of assets/contracts in the investment set. As
mentioned, the following periods have been used in the preceding calculations and portfolio
optimization.

Period 1:
l: 2016 and 2017

Period 2: 2018 and 2019

Period 3: 2020 and first half of 2021

Period 4: second half of 2021 and 2022

Period 5: from 2016 to 2022

To reduce the discrepancy in the number of observations between contracts in each period, a
minimum number of 30 observations for each contract have been used. Hence, the available
investment set in each period only consists of contracts with 30 or more return observations.
This allows for a larger random sample to be drawn and making the covariance calculations
more robust. A drawback with this approach is that certain contracts are not included in the
investment set. Thus, a trade-off is made between the number of contracts and the difference
in the number of observations between contracts.

Performing the random sampling approach requires insight into the number of observations to
each contract in the five designated periods. As they differ, it puts clear constraints on how
the sample size is determined. Since sample size can’t
can't be larger than population size, the
contract with the smallest population points out how large the sample size can be. In each of
the five periods the smallest population (observations) of a contract is respectively 45, 37, 31
3l
,37 and 51. Consequently, the sample size can’t
can't be greater than these sizes in their respective
periods. Using these population sizes, it is possible to calculate the optimal sample size
through the following equation:

N
𝑁𝑁 (21.))
(21.
n==
𝑛𝑛
+ 𝑁𝑁(𝑒𝑒)
11 + N(e)2 2

The sample size is given by n, N is the size of the population and e is the level of
𝑛𝑛, where 𝑁𝑁
precision (Yamane, 1967). With a 95% confidence level and corresponding P-value
P-value== 0,05
the sample sizes calculated are respectively 40, 34, 29, 34 and 45 for periods 1l to 5. In our

Page 35 of 41
simulation a sample of the abovementioned sizes is drawn from each contract (in each
period). Doing so generates a table of equal observation/rows for the contracts in the period
in question. From this table the covariance between contracts is calculated and stored in a
covariance matrix.

Simulation of random sampling

Having determined the sample sizes for the random sampling in each period, we have
extended this process to be repeated through simulation. The goal is to make the covariance
calculation between contracts statistically robust. Hence it is convenient to remove any bias
that might occur from the random sampling. Repeating the random sampling procedure
contributes in that manner. For this reason, we have created a simulation of the random
sampling process which is used in each of the five time periods. This simulation was created
with the statistical programming and computing language R.

In our simulation a sample of the abovementioned sizes is drawn from each contract type (in
each period). Doing so generates a table of equal observation/rows for the contracts, in the
period in question. From this table the covariance between contracts is calculated and stored
in a covariance matrix. This process of drawing a random sample is then repeated and again
stored in the covariance matrix, where the average of all constitutes the final covariance
matrix. For all the sampling iterations each one is made with replacement. The main reason
for choosing replacement is to ensure that the sample values are independent (Triola, 2011).

The replacement of samples is further needed in the simulation of drawing random samples.
As mentioned in the previous paragraph, we repeat the sampling to obtain an average for the
covariance matrix. In accordance with standard practices in statistical simulations, the
sampling process is repeated 10 000 times (Heijungs, 2020). Thus, replacement is necessary
to repeat the iterations 10
l O 000 times. Repeating this process ensures that the final covariance
matrix is robust (at the 5% level) to be used in the portfolio optimization calculations.

Discussion of limitations with the results

Our thesis has certain limitations which are likely to have implications for the realism and
credibility of our results. Firstly, as we have previously discussed there are elements that
affect an operator that aren’t
aren't captured in our analysis. Factors such as ballast and fuel risk are
part of the consideration an operator has to consider. These risks are not included in our

Page 36 of 41
analysis where the only risk included is the standard deviation calculated on the variance of
earnings. Therefore, our results are only to be considered assuming that all else is equal.

Another objection to our analysis is how Western Bulk conduct business compared to our
theoretical model. Our analysis has a numeric relation to risk and the fictitious business could
be run by these risk measures if wanted. Such an approach is not directly comparable to how
Western Bulk structure its risk measures. Western Bulk handles risk through its experience
and capital allocation to different regions (Husby, 2023). As an operator they do not care
about the relative volatility in different types of contracts based on historical data. Rather
they rely on their skill and knowledge to place ships in different regions. Also, by having a
large number of ships active throughout the dry bulk spectrum they have great intel on the
market minimizing unforeseen events disregarding market fluctuations.

The most adjacent theoretical concept in use is the concept of diversification. It is obvious
that an operator who allocates its resources to multiple geographies will more favorable
positioned compared to an operator located in few geographies. Although our calculations
show that most observations have a positive co-variance, they still have not the same
responds to volatility. Meaning that there are diversification effects to collect from operating
on more than a few locations. The operator’s
operator's mobility is another way of diversifying its
portfolio.

As mentioned, the size of the data set could optimally have been larger. With more
observations we would likely have had more assets in the possible investment set.
Consequently, this would likely have impacted on our results in terms of diversification
among more contract types, especially index contracts. Thus, we could gain more insight into
how an optimal portfolio could be constructed. Additionally, more observations would
provide a more robust risk/return perspective to the case we are looking at.

Finally, it has been mentioned that the optimization methodology doesn't

doesn’t include all possible
constraints. These constraints fathom aspects as limitations to contractual- and geographical
concentration, working capital and structure of trades. It will therefore be beneficial to
include the mentioned constraints in further research, as well as the other above-mentioned
limitations. Put together this will contribute to increasing the realism of the portfolio results
and the implications they can have in practice for Western Bulk.

Page 37 of 41
Concluding remarks and recommendation
Having looked at the returns from Western Bulk's
Bulk’s chartering contracts in the period from
2016 to 2022, we have obtained valuable insights through a portfolio approach. In a
geographical perspective Western Bulk have held a smaller position in the Indian region, than
what we have found to be optimal. Complementary to this is how the chartering contracts
should be distributed, where we have found it optimal to hold a majority in period contracts
and smaller positions in spot and forward contracts. These recommendations are more
relevant for capital allocation in a longer time perspective, as depicted with the whole period
of 2016-2022 (period 5). Furthermore, we recommend extending our findings by including
the mentioned constraints in any further portfolio optimization. Doing so will improve the
practical use of the corresponding results.

Even though the portfolio approach provides valuable insight into the operations of Western
Bulk, the approach can be a bit too static. If you follow this approach rigorously you will
likely miss opportunities in the market which are deviations from the portfolio weights. The
operations of Western Bulk are centered around exploiting these market imperfections. Thus,
we recommend that they continue with this strategy, but include portfolio insights as a
benchmark and useful reference point for capital allocation over a time horizon of equal
length as 2016-2022.

Page 38 of 41
References
Adland, R., & Jia, H. (2017). Simulation physical basis risks in the Capesize freight market.
Maritime Economic Logistics, 196-210.

Adland, R., & Prochazka, V. (2021). The value of timecharter optionality in the drybulk
market. Transportation Research Part E: Logistics and Transportation Review.

Adland, R., Benth, F., & Koekebakker, S. (2018). Multivariate modeling and analysis of
regional ocean freight rates. Transportation Research Part E, 194-221.

Adland, R., Bjerknes, F., & Herje, C. (2017). Spatial efficiency in the bulk freight market.
Maritime policy & management, 413-425.

Algarvio, H., Lopez, F., Sousa, J., & Lagarto, J. (2017). Multi-Agent Electricity Markets:
Retailer portfolio optimization using Markowitz theory. Electric Power Systems
Research, 282-294.

Atmaca, E. M., & Gökgöz, F. (2012). Financial optimization in the Turkish electricity
market: Markowitz's mean-variance approac. Renewable and Sustainable Energy
Review, 357-368.

Berg-Andreassen, J. A. (2011). A portfolio approach to strategic chartering decisions.

Maritime Policy & Management, 375-389.

Berk, J., & DeMarzo, P. (2017). Corporate Finance

Finance.. Harlow: Pearson Education.
Education .

Bodie, Z., Kane, A., & Marcus, A. J. (2014). Investments. New York: McGraw Hill.

Carisa, Y. K., Tsz, L. Y., & Siu, K. C. (2019). Optimal portfolio choice for ship leasing
L.Y.,
investments. Maritime Policy & Management, 884-900.

Chistè,
Chiste, C., & van Vuuren, G. (2013, April 29). Investigating the cyclical behaviour of the dry
bulk market. Maritime Policy & Management, pp. 1-19.

Clarksons Research. (2023, March). World Fleet Register. Retrieved from Clarksons
Research.

Cullinane, K. (1995). A portfolio analysis of market investments in dry bulk shipping. Centre
for International Shipping & Transport, 181-200.

Page 39 of 41
Cullinane, K. P. (1989, October). The Application of Modern
Modem Portfolio Theory to Hedging in
the Dry Bulk Shipping Markets. Plymouth, England.

Elton, E. J., Gruber, M. J., Brown, S. J., & Goetzman, W. N. (2014). Modern Portfolio
Theory and Investment Analysis. Hoboken: Wiley.

Emilson, M. (2020). Valuing using risk-free rate based on cash-flow duration now more
relevant. BDO Insights. Retrieved from BDO Australia.

George Alexandridis, S. S.-W. (2018). Shipping risk management practice revisited: A new
portfolio approach. Transportation Research Part A, 274-290.

Heijungs, R. (2020). On the number of Monte Carlo runs in comparative probabilistic LCA.
The International Journal of Life Cycle Assessment, 394-402.

Husby, E. (2023, February 10). Guest lecture NHH, ENE430 Commodity Trading and
Transport. Freight Trading. Bergen, Norway: Western Bulk.

Jacobs, B. I., & Levy, K. N. (2005). Market Neutral Strategies. New Jersey: John Wiley &
Inc. .
Sons, Inc..

Kasimati, E., & Veraros, N. (2018). Accuracy of forward freight agreements in forecasting
future freight rates. Applied Economics, 743-756.

Kloster, A. (2000, January). Beregning og tolkning av renteforventinger. Penger og Kreditt,

pp. 29-36.

Lorange, P. (2009). Shipping Strategy. New York: Cambridge University Press.

MarketWatch. (2023, May 09). MarketWatch: US.

U.S. 3 Year Treasury Bill. Retrieved from
MarketWatch:
https://www.marketwatch.com/investing/bond/tmubmusd03m/charts?countrycode=bx
&mod=mw_quote_tab

Markowitz, H. (1952). Portfolio Selection.

Selection . The Journal of Finance, 77-91.

Odean, T. (1999, December). Do Investors Trade Too Much? The American Economic
Review, pp. 1279-1298.

Sarno, L., & Thornton, D. L. (2003). The dynamic relationship between the federal funds rate
Samo,
and the Treasury bill rate; An empirical investigation. Journal of Banking & Finance,
1079-1110.

Page 40 of 41
Scarsi, R. (2007, December 14). The bulk shipping business: market cycles and shipowners'
biases. Maritime Policy & Management, pp. 577-590.

Sokri, A. (2012). Defence portfolio selection under uncertainty. Defence Research and
Development Canada.

Stopford, M. (1997). Maritime Economics. London: Routledge.

Sydsæter, K., Hammond, P., & Strøm, A. (2012). Essential Mathematics for Economic
Analysis. Harlow: Pearson Education.

Triola, M. F. (2011). Essentials of Statistics. Boston: Pearson Education.

Yamane, T. (1967). Statistics, An Introductory Analysis. New York: Harper and Row.

Page 41 of 41