This action might not be possible to undo. Are you sure you want to continue?

# Royal Institute of Technology

Master’s Thesis

Portfolio management using structured

products

The capital guarantee puzzle

Markus Hveem

hveem@kth.se

Stockholm, January 2011

Abstract

The thesis evaluates the concept of a structured products fund and investigates how a fund based

on structured products should be constructed to be as competitive as possible. The focus lies

on minimizing the risk of the fund and on capital guarantee. The diﬃculty with this type of

allocation problem is that the available products mature before the investment horizon, thus

the problem of how the capital should be reinvested arises. The thesis covers everything from

naive fund constructions to more sophisticated portfolio optimization frameworks and results in

recommendations regarding how a portfolio manager should allocate its portfolio given diﬀerent

settings. The study compares diﬀerent fund alternatives and evaluates them against, competing,

benchmark funds.

The thesis proposes a framework called the modiﬁed Korn and Zeytun framework which

allocates a portfolio based on structured products, which have maturity prior to the end of the

investment horizon, in optimal CVaR sense (i.e. appropriate for funds).

The study indicates that the most important concept of a structured products fund is trans-

action costs. A structured products fund cannot compete against e.g. mixed funds on the market

if it cannot limit its transaction costs at approximately the same level as competing funds. The

results indicate that it is possible to construct a fund based on structured products that is com-

petitive and attractive given low commission and transaction costs.

Keywords: Structured products fund, structured products, portfolio theory, portfolio optimiza-

tion, portfolio management, Conditional Value-at-Risk, transaction costs

Acknowledgements

I would like to thank my supervisor at KTH Mathematical Statistics, Filip Lindskog, for great

feedback and guidance as well as my supervisors at Nordea Markets, Karl Lindqvist and William

Sjöberg, for essential knowledge in the ﬁeld of structured products, feedback and ideas.

Stockholm, January 2011

Markus Hveem

v

Table of Contents

1 Introduction 1

1.1 Market risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Capital guarantee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.3 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Theoretical Background 5

2.1 Risk measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.1.1 Absolute lower bound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.1.2 Value-at-Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.1.3 Conditional Value-at-Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.2 Principal component analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.3 Scenario based optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.3.1 Conditional Value-at-Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.3.2 Minimum regret . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.4 Transaction costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

3 Financial Assets 15

3.1 Deﬁnition of a structured product . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3.1.1 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

3.2 Zero-coupon bond . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.3 Plain vanilla call option . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3.3.1 A note regarding the critique against B & S . . . . . . . . . . . . . . . . . 18

4 The Capital Protection Property 19

4.1 Capital guarantee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.2 Naive fund constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

4.2.1 Naive fund construction number 1 . . . . . . . . . . . . . . . . . . . . . . 20

4.2.2 Naive fund construction number 2 . . . . . . . . . . . . . . . . . . . . . . 25

4.2.3 Analysis and summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.2.4 Prevalent risk factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

4.3 Investigation of the option portfolio . . . . . . . . . . . . . . . . . . . . . . . . . 31

4.3.1 Portfolio construction and modeling . . . . . . . . . . . . . . . . . . . . . 31

4.3.2 Portfolios and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

4.3.3 Analysis and implications . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.4 Comparison with competition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.4.1 Client base . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.4.2 Benchmark fund . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.4.3 Structured products fund 1 . . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.4.4 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4.4.5 Results, structured products fund vs benchmark . . . . . . . . . . . . . . 39

4.4.6 Analysis and summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

4.4.7 Additional stress test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4.4.8 Backtesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

vii

4.4.9 Structured products fund 2 . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4.5 Alternative deﬁnitions of capital protection . . . . . . . . . . . . . . . . . . . . . 51

4.5.1 Deﬁnition 1 - Absolute lower bound . . . . . . . . . . . . . . . . . . . . . 51

4.5.2 Deﬁnition 2 - CVaR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5 Modeling Financial Assets 53

5.1 Volatility and option pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.2 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.3 Yield curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.4 A PCA model for the yield curve and index . . . . . . . . . . . . . . . . . . . . . 54

5.4.1 Data set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

5.4.2 PCA results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

5.4.3 Historical simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5.4.4 Monte Carlo simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

6 Portfolio Optimization 65

6.1 Fixed portfolio weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

6.2 Rolling portfolio weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

6.3 Benchmark fund . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

6.4 Dynamic portfolio weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

6.4.1 Modifying the Korn and Zeytun framework . . . . . . . . . . . . . . . . . 73

6.4.2 Dependence of the initial yield curve . . . . . . . . . . . . . . . . . . . . . 75

6.4.3 Observed paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

6.5 Transaction costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

6.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

6.7 Backtesting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

7 Conclusions 83

Bibliography 85

Appendices 91

A Naive fund constructions - Additional scenarios 91

B Option portfolios - Additional portfolios 93

C SPF vs Benchmark - Additional scenarios 95

D Correlations between the market index and the yield curve 99

E Optimization problem - Arbitrary start yield curve 101

M. Hveem viii (103)

CHAPTER 1

Introduction

Structured products are a structured form of investment vehicles that consists of a bond and

a market exposed ﬁnancial instrument. The popularity of investing in Equity-Linked Notes,

and other structured products, has been signiﬁcant exceeding 55.1 billion SEK in Sweden alone

during 2009 (My News Desk, 2010, [32]). One reason for their popularity is that they possess

the property of capital guarantee. Thus many investors have a belief that these products are safe

and do not realize the impact of credit risk. An investor can by investing in structured products

participate in the market besides the capital guaranteed part (the bond). Thus investors are

protected in bear markets and participate in bull markets.

The other major risk for an investor is market risk; the key to avoid market risk is by

diversifying. Market risk imposes a demand for applying a diversiﬁcation framework, similar

to the one of Markowitz’s (1952, [30]) seminal work on basic portfolio management theory,

applicable to structured products. Most studies in the area have been conducted by using a

structured product, a bond, the underlying and an option on the underlying as the investment

universe in a one period model (as in Martellini et al. (2005, [31])), which is inadequate to describe

the investment universe for e.g. a hedge fund. The ﬁrst study conducted on the subject, using a

longer investment horizon than the time to maturity for the structured product, was performed

by Korn and Zeytun (2009, [27]). This setup is more relevant for an investor that must rebalance

its portfolio at diﬀerent intermediate time points, and especially when the investor is allowed to

invest in more than just one structured product.

1.1 Market risk

It has been shown in studies of the Swedish market for structured products that an investor using

structured products in a diversiﬁed portfolio can achieve a fair return during bull markets and

take advantage of the capital protection property (to some extent) of the structured products in

the bear markets (Hansen and Lärfars, 2010, [19]; Johansson and Lingnardz, 2010, [25]).

These studies indicate that there are advantages of diversiﬁcation when investing in structured

products against market risk. Many investors on the Swedish market of structured products are

small private investors and thus not willing to commit most of their capital in one investment

category (Shefrin, 2002, [35]). Hence to oﬀer the advantages of diversiﬁcation in structured

products to small investors the concept of a fund based on structured products emerges. By

creating a fund based on structured products it is possible to decrease the required amount

invested for each individual investor and practically gain the same advantages as if the investor

had the possession itself.

1.2 Capital guarantee

The concept of market risk and diversifying using a set of numerous assets is very appealing to an

investor, especially if the investment is capital guaranteed. The main issue is that the portfolio

based on several structured products with diﬀerent maturities will not be capital guaranteed.

1

Chapter 1. Introduction

When, in the setting of structured products, a product is capital guaranteed it means that

the investor will at least receive the invested amount, at maturity (when disregarding the impact

of defaults). In the setting of a fund the investment horizons will instead be overlapping and

of approximately one to ﬁve years, thus the capital guarantee should be measured during each

of these investment periods. Problem arises when the fund holds positions in products with

maturity beyond the investment horizon, since all structured products can have a value of zero

before maturity due to changes in interest, movement of the underlying etc. Also if products have

maturity prior to the investment horizon’s end the issue of how the payoﬀ should be reinvested

arises. Consider that the fund should not only be guaranteed during one investment horizon,

but several, each new investor wants to have the property of capital guarantee over its own

investment horizon. Thus if the fund should be capital guaranteed for each investor, with an

investment horizon of τ

i

years, the return of the portfolio of every single τ

i

year long time period

should be positive. It is obvious that it is diﬃcult to gain the property of capital guarantee in

combination with the possibility of a high return.

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 92.84

Quarter

$

Price Underlying

NAV Fund

Start of investment

Figure 1.1: The underlying has 1% in return each quarter until quarter 12, when the investor buys

a share of the fund, after quarter 12 the underlying has −5% in return quarterly, σ =

0.15, r = 0.015 p.q., hence the fund is not capital guaranteed.

The example in Figure 1.1 illustrates the value of a fund that consists of rolling capital

guaranteed Equity-Linked Notes (notional amount of the bond equals to the price of the product)

each quarter with a maturity of three years. Thus each product is capital guaranteed but the

fund is not capital guaranteed over the three-year time horizon. The investor in this example is

an investor with an investment horizon of three years and holds only capital guaranteed products

(a total of 23 products, 12 at a time) with a maximum time to maturity of three years. During

this time period the fund value decreases with 7.16%, hence the simple naive fund construction

is not capital protected.

As the price of the underlying goes up the value of the fund increases since the options gain

more in value. This implies that after twelve quarters, when the underlying has gone up a total

of 12.68%, the fund’s NAV consists of options in a high extent, which makes the investment more

risky and erasing the capital guarantee.

1.3 Purpose

The purpose of this thesis is to investigate if it is possible to construct a portfolio based on

structured products that possesses the property of capital protection. The focus lies on developing

an alternative deﬁnition of capital protection, since there obviously does not exist any structured

products fund (SPF) fulﬁlling the requirements of absolute capital guarantee. The objective is

M. Hveem 2 (103)

1.4. Outline

also to investigate how a fund with as low downside risk as possible, with a decent expected

return (should be competitive in relation to competing funds), should be constructed. The

investigation is performed numerically based on simulating trajectories of future price patterns

for available structured products and by solving an optimization problem allocating the portfolio

based on CVaR constraints as in Uryasev (2000, [38]). Since the thesis will only consider investing

in products with the same underlying the importance of using a scheme based on Black and

Litterman’s (1992) studies in [8] is reduced, thus exploring the impact of active subjective views

on the investment choice is left for future research, as well as the possibility of multiple underlying.

The thesis focuses on market associated risks and will not evaluate the impact of credit risk

on the portfolio choice. Thus when investigating if the products are capital protected the impact

of defaults is neglected, this is usually how the terminology capital protection is used in the

concept of structured products.

1.4 Outline

Chapter 2 covers the most fundamental basics which the thesis has its foundation in, such as the

theory of risk measures, scenario based optimization and principal component analysis. It is not

necessary for the reader to go through this chapter, since it is not imperative to understand these

theories to understand the result. In Chapter 3 the deﬁnition of a structured product is covered

and how the individual components are priced. Chapter 4 covers the issue of capital guarantee,

the deﬁnition of capital guarantee, how the portfolio should be allocated initially to attain as

high degree of capital protection as possible. The chapter covers everything from naive fund

constructions to benchmarking with potential competition. A scheme to minimize the downside

risk is also described in the chapter.

Chapter 5 covers how the ﬁnancial assets are modeled and discloses the details regarding the

PCA. Chapter 6 covers three proposed allocation schemes, where a modiﬁed Korn and Zeytun

framework is recommended. In the last chapter conclusions are drawn regarding the results and

recommendations for future studies are given.

M. Hveem 3 (103)

CHAPTER 2

Theoretical Background

This chapter covers the theoretical background that the reader should be familiar with to under-

stand how the study is conducted. By understanding the concepts in this chapter it is also easier

to understand how to replicate the results and also the assumptions aﬀecting the results. The

chapter will cover everything from risk measures and principal component analysis (which is an

integral part of the modeling) to scenario based optimization and transaction costs. Notable, it

is not necessary to understand the concepts discussed in this chapter to understand the results

of the study.

To ensure full clarity for the reader matrices are written as bold upper case characters, vectors

as bold lower case characters and transposed with superscript T, not only in this chapter but

also throughout the whole paper.

2.1 Risk measures

The concept of risk has been around ﬁnance for more than ﬁfty years. Markowitz introduced

his seminal work within portfolio theory during 1952 [30] where he used standard deviation

(volatility) as risk measure to ﬁnd the optimal tradeoﬀ between risk and return. Since then

it has been shown that assets’ log returns are not multivariate normal distributed and that the

distribution of stock returns often exhibit negative skewed kurtosis (Fisher, 1999, [14]), especially

around extreme events such as the 1987 stock market crash, the Black Monday 19th October

when the Dow Jones Industrial Average dropped by 508 points to 1738.74, a 22.61% decrease

(Browning, 2007, [11]). Such extreme events prove that measuring risk with the variance is not

adequate, not even for stock returns.

A portfolio containing derivatives is not, in general, symmetric and variance (even if stock

returns are relative symmetric) is thus not an adequate risk measure. Some risk measures have

been introduced to capture these heavy tailed, skewed distributions such as Value-at-Risk (VaR),

Conditional-Value-at-Risk (CVaR, closely related to Expected Shortfall) (Acerbi, 2002, [3]). An-

other risk measure, which is not as commonly used as VaR and CVaR is the absolute lower

bound, which measures the worst-case outcome.

Further the net worth of the fund portfolio at time T is denoted as X = X

T

−X

t

, where X

t

is the value of the portfolio at time t.

2.1.1 Absolute lower bound

An acceptable portfolio is a portfolio which ﬁnal net worths are those that are guaranteed to

exceed a certain ﬁxed number (e.g. a percentage of the initial investment) c, i.e. the set of

acceptable portfolios must fulﬁll these requirements,

A = {X ∈ X : X ≥ c} ,

which gives the risk measure,

ρ (X) = min {m : m(1 +r

f

) +X ∈ A} = min {m : m(1 +r

f

) +X ≥ c} .

5

Chapter 2. Theoretical Background

x

0

= x

0

(X) is denoted as the smallest value that X can take, notice that,

ρ (X) = min {m : m(1 +r

f

) +X ≥ c} =

c −x

0

1 +r

f

.

Thus the deﬁnition is regarding the worst possible outcome of the position at time T. When

regarding absolute capital guarantee the portfolio satisﬁes ρ (X) ≤ 0 (Hult and Lindskog, 2009,

[23]).

2.1.2 Value-at-Risk

Value-at-Risk (VaR) is one of the most important concepts within risk management and is also

regulated by the FSA in many countries and by the guidelines from the second Basel Accord

regarding minimum capital requirements (BIS, 2006, [5]). The idea of VaR is that VaR

α

(L

w

) is

the value that the portfolio’s loss will be less than or equal to with a probability of α. Thus VaR

is given as (a similar notation is given in [27]),

• Let L

w

denote the loss of the portfolio with the portfolio weights w, and the probability of

L

w

not exceeding a threshold m as,

ψ (w, m) = P (L

w

≤ m) .

• Then Value-at-Risk VaR

α

(L

w

) is deﬁned as the loss with a conﬁdence level of α ∈ [0, 1]

by,

VaR

α

(L

w

) = min {m ∈ R : ψ (w, m) ≥ α} ,

or,

VaR

α

(L

w

) = min {m ∈ R : P (L

w

≤ m) ≥ α} .

• Where L

w

= −R

w

, where R

w

is the return associated with the portfolio vector w, and the

return is deﬁned as,

R

w

=

ﬁnal wealth

initial wealth

−1.

There exists several areas of critique against VaR; one of the most common critiques is that it

is not a coherent risk measure (Hult and Lindskog, 2009, [23]). Also it is not very useful for

distributions that have heavy tails, since it only measures the outcomes at the quantile α. Many

blame Value-at-Risk of being an integral part why a ﬁnancial system may fail, since the model

underestimates the risk/implication of market crashes (Brooks and Persand, 2000, [10]) and

does not take into account the size of the losses exceeding VaR, thus traders have the possibility

to hide risk in the tail. Also, VaR is a non-convex and non-smooth function which in many

cases has multiple local maximums and minimums, thus it is very hard to construct a portfolio

optimization scheme which is eﬀective enough to still be robust and give valid results (Uryasev,

2000, [38]).

Instead of VaR many authors propose Conditional Value-at-Risk (CVaR), which is much

easier to implement in portfolio optimization as proposed by Uryasev et al. in [28, 33, 38]. Also

Alexander et al. (2006, [4]) have recently developed an optimization scheme which is very eﬃcient

for minimizing CVaR and VaR for portfolios of derivatives, but it is much more complex than

the one proposed by Uryasev et al., and Alexander et al.’s theory will not be covered in this

paper.

M. Hveem 6 (103)

2.2. Principal component analysis

2.1.3 Conditional Value-at-Risk

CVaR is based on the deﬁnition of VaR since it is the expected loss under the condition that the

loss exceeds or equals the VaR, i.e. it is deﬁned as,

CVaR

α

(L

w

) = E[L

w

| L

w

≥ VaR

α

(L

w

)] .

CVaR is a coherent risk measure and more adequate than VaR since it discloses the expectation

of the loss if the loss exceeds the VaR, which is an important concept in risk management.

The goal of portfolio optimization is often to maximize the return subject to a demand on the

maximum risk acceptable. The optimization problem reduces to a linear optimization problem

with linear constraints, in the CVaR case. The advantage with a linear optimization problem is

that it can be solved using the simplex method.

Krokhmal et al. (2002, [28]) have shown that the solutions to the optimization problems,

min

x∈X

−R(x) , s.t. CVaR

α

(L

w

) ≤ C, (2.1)

and,

min

x∈X

−R(x) , s.t. F

α

(w, β) ≤ C, (2.2)

give the same minimum value where,

F

α

(w, β) = β +

1

1 −α

_

y∈R

m

[L(w, y) −β]

+

p (y) dy,

CVaR

α

(L

w

) = min

β∈R

F

α

(w, β) ,

and L(w, y) is the loss function associated with the portfolio vector w, y ∈ R

m

is the set of

uncertainties which determine the loss function. If the CVaR constraint in 2.1 is active, then

(w

∗

, β

∗

) minimizes 2.2 if and only if w

∗

minimizes 2.1 and β

∗

∈ arg min

β∈R

F

α

(w, β). When

β

∗

∈ arg min

β∈R

F

α

(w, β) reduces to a single point, then β

∗

gives the corresponding VaR with

conﬁdence level of α (Korn and Zeytun, 2009, [27]). Thus it is possible to solve the optimization

problem 2.2 instead, which is easily transformed to a linear optimization problem (disclosed

in Section 2.3), since it generates the same solution as 2.1, for the proof and more detailed

information please consult [28, 33, 38].

2.2 Principal component analysis

Principal component analysis (PCA) is a common tool to generate scenarios for changes in

the yield curve and returns for other types of assets. The idea of PCA is that data sets of

intercorrelated quantities can be separated into orthogonal variables, which explains the variance

and dependence in the data in a simpler way. Thus the number of factors to simulate can be

reduced drastically by reducing the number of risk factors to model. A whole yield curve can often

be reduced into only three factors, or so-called principal components, thus PCA is a useful tool to

generate scenarios that are parsimonious. Studies have shown that by using principal component

analysis only two or three principal components are often enough to describe more than 95% of

the variation in a yield curve (Barber and Copper, 2010, [6]; Litterman and Scheinkman, 1991,

[29]). The main idea is to transform the data to a new orthogonal coordinate system such that

the greatest variance by any projection of data comes to lie on the ﬁrst coordinate (i.e. ﬁrst

principal component), the second greatest variance by any projection on the second coordinate

and so on.

M. Hveem 7 (103)

Chapter 2. Theoretical Background

Consider a setting with m assets, n number of observations and that the asset returns are

given as a n×m matrix denoted

˜

R. It is important to center the returns/data by subtracting its

mean to perform the PCA and also normalize the data by dividing with

_

(n −1) (or by

√

n).

Let denote R =

˜

R−µ

√

n−1

, this type of PCA is referred to as covariance PCA since the matrix R

T

R

is a covariance matrix (the covariance matrix of the returns). It is also common with correlation

PCA, in which each variable is divided by its norm, making R

T

R a correlation matrix (most

common in statistical packages such as MATLAB and R), correlation PCA will be used in this

thesis (Abdi, 2010, [2]).

By using singular value decomposition R can be written on the following form,

R = P∆Q

T

,

where P is a n×n orthonormal matrix, Q is a m×m orthonormal matrix (called loading matrix

and each column corresponds to one PC) and ∆ is a n × m matrix with non-negative values

on the diagonal. The PCA creates new variables called principal components, which are linear

combinations of the original variables and are deﬁned in a way such that the amount of variation

associated with them are in decreasing order and orthogonal to each other. Let denote the factor

scores F (observations of the principal components) as,

F = P∆,

thus,

F = P∆= P∆Q

T

Q = RQ,

which implies that the ith observation of the jth original variable is expressed as follows,

r

i,j

= Q

T

1,j

F

i,1

+... +Q

T

m,j

F

i,m

.

Next R

T

R, is investigated,

R

T

R =

_

P∆Q

T

_

T

P∆Q

T

= Q∆

T

P

T

P∆Q

T

= Q∆

T

∆Q

T

= Q∆

2

Q

T

.

∆is a diagonal matrix and ∆

2

equals the diagonal matrix Λ containing the (positive) eigenvalues

λ

1

, ..., λ

n

to R

T

R (since R

T

R is a positive semi-deﬁnite matrix), Q is an orthogonal matrix.

An orthogonal matrix has the property Q

T

= Q

−1

, thus Q

T

Q = QQ

T

= I. The columns in

Q, q

1

, ...q

m

are the corresponding eigenvectors of R

T

R, which are orthonormal. It is possible

to assume, without loss of generality, that the columns of Λ and Q are ordered such that the

diagonal elements in Λ appear in descending order. Note that,

Cov

_

F

T

_

= E

_

Q

T

R

T

RQ

¸

= Q

T

Cov (R) Q = Q

T

R

T

RQ = Λ,

thus the components of F are uncorrelated and have variances λ

1

≥ ... ≥ λ

m

, in that order. It

can be shown that the returns R are uncorrelated expressed on the orthogonal basis, which is

shown by Hult et al. (2010) in [24].

The idea of PCA is as mentioned to reduce the number of variables needed to describe

the data. Only the variables that add important information to the sample are interesting

and withdrawn, thus the PCs that add the most variability. To investigate the contribution of

variability of each PC the ratio

λ

i

m

j=1

λ

j

is studied. It is most often possible to describe the whole

dependence structure for a data set with just the ﬁrst K PCs, as when modeling the yield curve.

Thus the returns are given by,

r

i,j

=

K

k=1

Q

T

k,j

F

i,k

+ε

i,j

.

Hence to simulate new returns; simulate factor scores (PCs) and use the factor loadings to

calculate the return. The sample must also be rescaled with its standard deviation and mean.

For more information regarding PCA and yield curve modeling please advise [2, 29, 36].

M. Hveem 8 (103)

2.3. Scenario based optimization

2.3 Scenario based optimization

There are in general two approaches to portfolio optimization problems: mean-variance and

scenario optimization. One of the frontrunners within mean-variance optimization was Markowitz

(1952, [30]), also many others have been well awarded for their contributions within this ﬁeld. As

the concept of skewness and kurtosis has become more prevalent the importance of scenario based

optimization has increased. Thus one of the most important uses of scenario based optimization

is that it actually allows derivatives/options to be part of the product mix, which is not in general

the case with a mean-variance optimization (Grinold, 1999, [17]).

This thesis considers portfolios of structured products, thus it is imperative to use scenario

optimization when allocating amongst the assets. Important to mention is that the optimization

is totally dependent on the scenarios, thus with the wrong assumptions or scenarios the result

will most likely be sub-optimal.

The idea of scenario based optimization is to turn a stochastic problem into a deterministic

problem by simulating future scenarios for all available assets. The problem stops being stochastic

when the scenarios are generated and the problem is (most often) transformed to linear form,

which can be solved by mathematical linear techniques.

To gain a qualitative solution adequate scenarios must be generated, in particular Scherer

(2004, [34]) states that the scenarios must be:

• Parsimonious - as few scenarios as possible to save computation power, or time.

• Representative - the scenarios must be representative and give a realistic representation of

the relevant problems and not induce estimation error.

• Arbitrage-free - scenarios should not allow arbitrage to exist.

There exist several diﬀerent methods of simulating data, two of these are bootstrapping

historical empirical data and Monte Carlo simulation, i.e. drawing samples from a parametric

distribution.

When bootstrapping historical empirical data, also known as historical simulation, the user

draws random samples from the empirical distribution, e.g. if the user is trying to simulate annual

returns the user may draw 12 monthly returns from the empirical distribution, thus generating

an annual return. The draws are performed with replacement, thus if the user is simulating 1,000

yearly returns the user draws for example 12,000 samples of monthly return from the empirical

distribution. Notable is that bootstrapping leaves correlation amongst the samples unchanged,

but destroys autocorrelation. Since bootstrapping is done by repeated independent draws, with

replacement, the data will look increasingly normal as the number of samples increases.

Monte Carlo simulation is similar to historical simulation where the sample is, instead of

drawn from the empirical distribution, drawn from a parametric distribution. The parametric

distribution may be attained by ﬁtting a statistical distribution to a historical sample of data.

If the assets are independent of each other the user can ﬁt an individual distribution to each

asset. A popular way of simulating from a parametric distribution is by using e.g. copulas or

autoregressive (AR) models.

In a scenario based optimization N available assets to allocate in are considered with the added

feature of S possible return scenarios. To solve a scenario based optimization problem there are

usually two steps to consider:

Step 1. Simulate S paths of returns for the assets i = 1, ..., N.

Step 2. Deﬁne a linear optimization problem on those simulated paths, which can be solved

using the simplex method.

M. Hveem 9 (103)

Chapter 2. Theoretical Background

2.3.1 Conditional Value-at-Risk

The CVaR problem can be (as mentioned earlier) converted to a linear optimization problem,

for more details regarding how the problem is transformed please advise [28, 33, 38]. The linear

CVaR problem based on scenario simulation is deﬁned as follows,

max

w,z,β

1

S

S

k=1

R

w

T,k

,

such that:

R

w

T,k

= w

1

R

1

T,k

+... +w

N

R

N

T,k

, k = 1, ..., S

R

w

T,k

+β +z

k

≥ 0, k = 1, ..., S

β +

1

S (1 −α)

S

k=1

z

k

≤ C,

z

k

≥ 0, k = 1, ..., S

w

1

+...w

N

= 1,

w

i

≥ 0, i = 1, ..., N

where, α is the conﬁdence level of CVaR, β is a free parameter which gives VaR in the optimum

solution of the CVaR problem, R

i

T,k

is the return for asset i for scenario k until time T and

w

i

the weight held in asset i (Krokhmal et al., 2002, [28]). The index k corresponds to which

scenario, the index i corresponds to which asset, S is the number of simulated paths and N the

number of assets. The problem can be solved using the simplex method, which is to prefer due to

its eﬃciency. Thus the simulation based CVaR problem is a problem that is relative well deﬁned

since the increasing power of today’s personal computers enables the possibility to solve these

problems to a reasonable cost.

It is also possible to write the portfolio choice problem with the CVaR as minimization

objective, resulting into the following linear approximation based on scenarios,

min

w,z,β

β +

1

S (1 −α)

S

k=1

z

k

,

such that:

R

w

T,k

= w

1

R

1

T,k

+... +w

N

R

N

T,k

, k = 1, ..., S

R

w

T,k

+β +z

k

≥ 0, k = 1, ..., S

1

S

S

k=1

R

w

T,k

≥ R

target

,

z

k

≥ 0, k = 1, ..., S

w

1

+...w

N

= 1,

w

i

≥ 0, i = 1, ..., N

Notable is that the two diﬀerent problems generate the same eﬃcient frontier.

2.3.2 Minimum regret

Minimum regret is an optimization scheme based on scenario optimization, which maximizes

the least favorable outcome of S scenarios given a certain demand on return. It is possible to

M. Hveem 10 (103)

2.4. Transaction costs

formulate the optimization problem as a linear problem based on scenarios and it is deﬁned as

follows (Scherer, 2004, [34]).

Let R

min

, be the worst possible outcome and R

T,k

be the expected return vector over all

scenarios thus,

max

w∈R

N

R

min

such that:

w

1

R

1

T,k

+... +w

N

R

N

T,k

≥ R

min

, k = 1, ..., S

w

T

R

T,k

≥ R

target

,

w

1

+...w

N

= 1,

w

i

≥ 0, i = 1, ..., N.

By using this scheme the downside is restricted, this type of allocation is preferable for really

risk-averse investors since they know the extent of their worst outcome. It is possible to modify

the setup to an equivalent optimization scheme that is deﬁned as follows,

max

w∈R

N

w

T

R

T,k

such that:

w

1

R

1

T,k

+... +w

N

R

N

T,k

≥ R

min

, k = 1, ..., S

w

1

+...w

N

= 1

w

i

≥ 0, i = 1, ..., N

2.4 Transaction costs

Transaction costs are important to take into consideration since they can change the proﬁtability

of an investment. An investor will have to pay transaction costs every time it rebalances its

portfolio due to several factors. The most common transaction costs are such as brokerage

commission, bid-ask spread, market impact (volume etc.) Scherer (2004, [34]) suggests that

transaction costs, tc, are of the following functional form,

tc = Commission +

Bid

Ask

-Spread +θ

¸

Trade volume

Daily volume

.

The bid-ask spread is expressed as a percentage and θ is a constant that needs to be estimated

from the market. The problem with this model is that data is needed on daily trading volumes,

which is not always available. Instead an appropriate way to estimate transaction costs is by

separating the costs into one ﬁxed and one linear part. There are many types of optimization

problems incorporating transaction costs and only those relevant to the study will be discussed.

There are in general two diﬀerent approaches to transactions costs; the direct approach restricts

the actual cost from happening by introducing actual transaction costs, which are deducted from

the return, thus having an impact on the result. The second is to put up restrictions upon actions

that have transaction costs linked to them, thus preventing the reactive transaction costs, e.g.

restricting turnover and/or trading constraints.

Let w

i

be the weight invested in asset i, w

initial

i

the weight invested in asset i prior to the

reallocation, w

+

i

as a positive weight change and w

−

i

as a negative weight change (asset sold).

Thus the weight invested in asset i after the reallocation is given as: w

i

= w

initial

i

+w

+

i

−w

−

i

.

M. Hveem 11 (103)

Chapter 2. Theoretical Background

Proportional Transaction Costs

One of the most common types of transaction costs is proportional transaction costs, which

means that the transaction costs are proportional to the amount bought or sold of the asset. Let

TC

+

i

be the proportional transaction cost associated with buying asset i and TC

−

i

be the cost

associated with selling asset i. The budget constraint in the traditional portfolio optimization

problem is

i

w

i

= 1 which now must be modiﬁed since the transactions have to be paid out of

the existing budget, thus instead the budget constraint is as follows,

n

i=1

w

i

+

n

i=1

_

TC

+

i

w

+

i

+TC

−

i

w

−

i

_

= 1.

By introducing transaction costs the reward function is changed from

n

i=1

w

i

µ

i

to

n

i=1

w

i

(1 +µ

i

)

since

n

i=1

w

i

≤ 1.

The linear CVaR problem based on scenario based optimization and proportional transaction

costs can be formulated as follows,

max

w,w

+

,w

−

,z,β

1

S

S

k=1

_

1 +R

w

T,k

_

,

such that:

β +

1

S (1 −α)

S

k=1

z

k

≤ C,

1 +R

w

T,k

=

_

1 +R

1

T,k

_

w

1

+... +

_

1 +R

N

T,k

_

w

N

, k = 1, ..., S

R

w

T,k

+β +z

k

≥ 0, k = 1, ..., S

z

k

≥ 0, k = 1, ..., S

N

i=1

w

i

+

N

i=1

_

TC

+

i

w

+

i

+TC

−

i

w

−

i

_

= 1,

w

i

= w

initial

i

+w

+

i

−w

−

i

, i = 1, ..., N

w

i

≥ 0, i = 1, ..., N

w

+

i

≥ 0, i = 1, ..., N

w

−

i

≥ 0, i = 1, ..., N

Proportional transaction costs are popular to model with since they do not add a lot of complexity

to the optimization program. Also most of the transaction costs in the market are proportional,

thus making it a good and adequate choice (Krokhmal et al., 2002, [28]).

Fixed Transaction Costs

Fixed transaction costs arise as soon as a particular asset is traded. The ﬁxed transaction costs

are not dependent on the trade size, thus a trade of $1 and $1,000,000 will generate the same

cost. Besides proportional transaction costs, ﬁxed transaction costs is one of the most common

type of transaction costs in the market. To be able to use ﬁxed transaction costs integer variables

δ

±

i

must be introduced, which takes the value one if trading takes place in asset i (w

±

i

is positive)

and zero otherwise. Including proportional transaction costs, the new budget constraint is given

as,

n

i=1

w

i

. ¸¸ .

Holdings

+

n

i=1

_

FC

+

i

δ

+

i

+FC

−

i

δ

−

i

_

. ¸¸ .

Fixed TC

+

n

i=1

_

TC

+

i

w

+

i

+TC

−

i

w

−

i

_

. ¸¸ .

Proportional TC

= 1,

M. Hveem 12 (103)

2.4. Transaction costs

where,

w

+

i

≤ δ

+

i

w

max

,

w

−

i

≤ δ

−

i

w

max

,

δ

±

i

∈ {0, 1} ,

and w

max

is a large number.

The problem of using ﬁxed transaction costs is associated with the integer variables δ

±

i

, which

transforms the linear optimization problem, which can be solved using the simplex method, to

a mixed integer linear program. A mixed integer linear program has a higher complexity and

takes more computation power, or time, to solve thus making it inadequate for many scenarios.

Hence it is preferable to consider an optimization program that does not contain binary or integer

variables. The mixed integer linear program for CVaR as constraint and the return as target

function is deﬁned as follows,

max

w,w

+

,w

−

,δ

+

,δ

−

,z,β

1

S

S

k=1

_

1 +R

w

T,k

_

,

such that:

β +

1

S (1 −α)

S

k=1

z

k

≤ C,

1 +R

w

T,k

=

_

1 +R

1

T,k

_

w

1

+... +

_

1 +R

N

T,k

_

w

N

, k = 1, ..., S

R

w

T,k

+β +z

k

≥ 0, k = 1, ..., S

z

k

≥ 0, k = 1, ..., S

N

i=1

w

i

+

n

i=1

_

FC

+

i

δ

+

i

+FC

−

i

δ

−

i

_

+

+

N

i=1

_

TC

+

i

w

+

i

+TC

−

i

w

−

i

_

= 1,

w

i

= w

initial

i

+w

+

i

−w

−

i

, i = 1, ..., N

w

i

≥ 0, i = 1, ..., N

0 ≤ w

+

i

≤ δ

+

i

w

max

, i = 1, ..., N

0 ≤ w

−

i

≤ δ

−

i

w

max

, i = 1, ..., N

δ

±

i

∈ {0, 1} , i = 1, ..., N

M. Hveem 13 (103)

CHAPTER 3

Financial Assets

This chapter covers the basics behind the ﬁnancial assets, e.g. what types of instruments that

are available and how they are priced. It is not necessary to read the chapter for someone who

is well familiar with structured products. In this thesis three types of assets are available for the

investor, these are: a risk-less asset (bond), a risky index (an Exchange Traded Fund (ETF))

and structured products (Equity-Linked Notes) with the risky index as underlying. The bond

is a theoretical asset where the default probability of the issuer is zero. The ETF is assumed to

follow the index perfectly, thus this market is relative theoretical. The structured products are

combinations of bonds and derivatives with the index as underlying.

3.1 Deﬁnition of a structured product

Structured products are in general synthetic investment instruments that are created to meet

special needs for customers that cannot be met by the current market. These needs are often

focused on low downside risk on one hand and the possibility of growth on the other. This is

usually achieved by constructing a portfolio consisting of securities and derivatives. Hence there

are endless combinations of possible structured products, thus there exists no standard product.

According to Martellini et al. (2005, [31]) was the ﬁrst structured form of asset management

the introduction of portfolio insurance such as the constant proportion or option based portfolio

insurance strategy. Later on this has spurred the development of more exotic structures and

creative constructions. The main reason for such constructions is to ﬁt the investors’ risk and

aspiration preferences.

Today the most common setup for a structured product is the combination of a zero-coupon

bond and an option. The type of option varies widely and most often a plain vanilla option with

a certain equity index, equity, currency, etc., as underlying is used; these products are usually

called Equity-Linked Notes (ELN). Since only the option part is exposed to the market, the

products are often also called capital guaranteed investment vehicles. Investment structures that

possess the property of capital guarantee are usually called notes, thus structured products that

posses a minimum of 100% capital guarantee are often called Principal protected notes (PPN).

Examples of other types of options used in structured products are such as basket options,

rainbows, look-back options, path dependent options and barrier options (HSBC, 2010, [21]).

These products will not be used in this thesis due to their computational heavy pricing models

involving, in many cases, Monte Carlo based option pricing methods. In the future when referring

to a structure product this thesis will refer to an Equity-Linked Note.

The example in Figure 3.1 is of a structured product that consists of a zero-coupon bond

and an option. The notional amount (amount paid at maturity T = 1) of the zero-coupon

bond equals the price of the structured product at time t = 0. Thus the investment is capital

guaranteed since the minimum amount that the investor (conditional on that the issuer of the

zero-coupon bond has not defaulted) receives is the price at time t = 0. Note that the bond’s

face value is deterministic and the payoﬀ of the option is stochastic.

The idea of the structured product is usually that most of its value is contributed by the

15

Chapter 3. Financial Assets

0 1

0

50

100

150

Bond Value

Option Value

Bond Value

(Deterministic)

Option Payoff

(Stochastic)

Structured product

Time

P

r

o

p

o

r

t

i

o

n

[

%

]

Figure 3.1: Structure of a structured product, time t = 0 is time of issuance and T = 1 is the time of

maturity, the bond payoﬀ is deterministic and the option payoﬀ is stochastic.

zero-coupon bond. The relation between how much is invested in the zero-coupon bond and the

option depends on the price of the zero-coupon bond (in theory, in practice it also depends on

the fees taken by the issuer). Thus diﬀerent constructions are available depending on the current

market climate, e.g. in a regime with low interest rates the zero-coupon bonds are expensive

and thus the amount left for buying options is relative small. The leveraged exposure to the

underlying is called participation rate (HSBC, 2010, [21]). The participation rate is measured in

percentage (or you might call it number of contracts).

3.1.1 Dynamics

Equity-Linked Notes in this thesis have notional amount of the bond and strike price equal to the

price of the underlying when issued, i.e. the options are plain vanilla at-the-money call options.

Let denote,

T as the time of maturity

S

0

as the value of the underlying at the time of issuance

S

T

as the value of the underlying at maturity

k as the participation rate

C as the price of the option

B as the price of the zero-coupon bond with notional amount S

0

r as the risk-free interest rate

Thus the payoﬀ at maturity of the structured product can be written as,

S

0

.¸¸.

Bond part

+k max (S

T

−S

0

, 0)

. ¸¸ .

Option part

,

where the participation rate, k, is given by,

k =

S

0

−B

0

C

.

Thus the amount paid out to the investor at maturity, T, in dollars per dollar invested is given

by,

1 +k max

_

S

T

−S

0

S

0

, 0

_

,

M. Hveem 16 (103)

3.2. Zero-coupon bond

i.e. k controls how much the investor will participate in the market. Notable is that k is increasing

in r and decreasing in σ (volatility) as seen in Figure 3.2. The option prices are calculated with

Black-Scholes formula as disclosed in Section 3.3. As the volatility increases the price of the

option increases, thus decreasing the participation rate. When the interest rate increases the

price of the zero-coupon bond decreases, thus increasing the participation rate.

0

2

4

6

8

10

5

10

15

20

25

30

35

40

45

50

0

10

20

30

40

50

60

70

80

90

100

Interest rate r

t

[%]

k=k(σ,r)

Volatility σ [%]

P

a

r

t

i

c

i

p

a

t

i

o

n

r

a

t

e

k

[

%

]

Figure 3.2: The surface plot shows the relation between the interest rate r

t

, the volatility σ and the

participation rate k (r

t

, σ) for T = 1.

3.2 Zero-coupon bond

A zero-coupon bond is a bond that pays no coupons, the only cash ﬂow generated from a zero-

coupon bond is the notional amount (face value) paid at maturity. Thus using a continuous

compounding for the interest rate the price of a risk-less zero-coupon bond at time t, with

maturity T and notional amount $1 is given by,

B

t

= e

−r

t

(T−t)

,

where r

t

is the risk-free rate at time t.

3.3 Plain vanilla call option

A plain vanilla call option, also called European call option is a product that gives the holder the

right, not the obligation, at maturity, time T, to buy the underlying asset S for the predeﬁned

strike price K. The price of the asset S at time T is denoted as S

T

. Since the holder does not

have the obligation to exercise the option, the option will have the following payoﬀ X at time T,

X = max (S

T

−K, 0) .

The price at time t of the European call option is denoted C (S

t

, K, t, T, r, σ, δ) where S

t

is the

price of the underlying at time t, K the strike price, T the time of maturity, r the risk-free rate,

M. Hveem 17 (103)

Chapter 3. Financial Assets

σ the volatility and δ the continuous dividend yield. It is common to price these options with

Black-Scholes formula, which is given below (Black and Scholes, 1973, [9]; Hull, 2002, [22]),

C (S

t

, K, t, T, r, σ, δ) = Π(t; X) ,

where,

Π(t; X) = S

t

e

−δ(T−t)

N [d

1

(t, S

t

)] −e

−r(T−t)

KN [d

2

(t, S

t

)] , (3.1)

and,

d

1

(t, s) =

1

σ

√

T −t

_

ln

_

s

K

_

+

_

r −δ +

1

2

σ

2

_

(T −t)

_

,

d

2

(t, s) = d

1

(t, s) −σ

√

T −t.

N is the cumulative distribution function for the standard normal distribution. It is important

to understand the underlying assumptions of Black-Scholes formula. These are such as that

the stock returns are lognormal distributed, constant volatility, constant risk-free rate, etc. The

Black-Scholes market model under the probability measure P is given as (Björk, 2009, [7]),

dS (t) = µS (t) dt +σS (t) dW (t) , S (0) = s

dB(t) = rB(t) dt, B(0) = 1

where W is a Brownian Motion (i.e. Wiener process), µ the drift and σ the volatility. Notable

is that the transform under this market model to the risk-neutral probability measure Q only

transforms the drift µ to r by using Girsanov’s theorem, which implies that the volatility σ is

the realized volatility observed at the market (Björk, 2009, [7]).

Notable is that the assumption of a continuous dividend yield is adequate in the setting of

modeling an index, since it is reasonable to assume that the dividends are spread all over the

year for diﬀerent components of the index.

3.3.1 A note regarding the critique against B & S

It is commonly known that the market model above has sustained a lot of critique since Black and

Scholes proposed it in 1973 [9]. The main area of criticism is the assumption of lognormal returns

and thus the dependence of the normal distribution (Björk, 2009, [7]). Instead many authors such

as Heston (1993, [20]) and Hagan et al. (2007, [18]) propose models that incorporate stochastic

volatility and the volatility-smile. Empirical evidence has shown that asset returns are skewed

and exhibit kurtosis. Since the Black-Scholes formula seems inadequate to price plain vanilla

options using the original deﬁnition with realized (historical) volatility, practitioners usually use

the so-called implied volatility instead. Implied volatility is the volatility so that Black-Scholes

formula gives the correct market price.

Since the thesis is regarding structured products the pricing of these options is only relevant in

the context of structured products. Wasserfallen and Schenk (1996, [39]) found that the prices of

structured products are not aﬀected systematically by using either realized volatility or implied

volatility. The study compared the theoretical value of the structured products (using both

realized and implied volatility) with the observed price at the primary and secondary market.

Thus the eﬀect of using realized volatility in comparison to implied volatility in the pricing

of the structured products can be disregarded in this thesis since the error in comparison with

the primary and secondary market is negligible (Wasserfallen and Schenk, 1996, [39]). Hence

Black-Scholes formula is an adequate choice for pricing the ELNs.

M. Hveem 18 (103)

CHAPTER 4

The Capital Protection Property

One of the most appealing properties of Equity-Linked Notes is the capital guarantee. Most

Equity-Linked Notes are constructed such that the notional amount of the bond equals the issue

price. Thus the investor is guaranteed to not loose any capital. Investors suﬀer from a syndrome

called loss-aversion, which means that they often make irrational decisions just to avoid loosing

any capital (Shefrin, 2002, [35]). This is one main factor why people invest in structured products,

the neat construction of market participation and capital guarantee.

The concept of a structured products fund is very appealing since the investor gains diver-

siﬁcation amongst the assets, with diﬀerent strikes and maturities (and by several underlying,

which is excluded in this paper). As mentioned in the introduction, several studies recent years

have shown that structured products actually are a good investment choice for rational investors,

as long as the investor uses a diversiﬁcation framework. The problem with structured products

is that they usually require that the investor invests a minimum amount in each structured

product. The market for structured products is quite large and during 2009 the total market

for ELNs was 55.1 billion SEK in Sweden alone, a large proportion of these investors are small

private investors. If an investor should diversify its portfolio amongst structured products the

invested amount increases to relative high levels, since each product has a required minimum

invested amount.

Investors do not, according to Shefrin (2002, [35]), prefer to invest a huge percentage of their

total capital in one single type of investment. Thus the only reasonable way for small investors to

diversify in structured products, without requiring a huge amount of capital, would be to invest

in a fund that is based on structured products. Problem arises when trying to construct a fund

in a way such that the property of capital guarantee is retained.

This chapter covers a relative broad spectrum of topics. First of all the deﬁnition of capital

guarantee is discussed and how it relates to a fund. A large part of the chapter is used to

discuss so called naive fund constructions. These constructions are simple forms of funds based

on structured products that are used to evaluate how the fund should be constructed so that

they carry as little downside risk as possible. The chapter also covers a similar study conducted

for option portfolios, where it is investigated how an investor should allocate amongst ATM call

options, at issuance, to limit the downside risk as much as possible. The results from these two

sections are then combined to investigate certain limits for the possession allowed in structured

products, to limit the downside risk. In the end alternative deﬁnitions of capital protection are

discussed to decide under which risk measure the structured products fund should be allocated.

4.1 Capital guarantee

When describing a fund based on capital guaranteed products it is quite easy to believe that

the fund itself also would be capital guaranteed, it is not as easy as that. A capital guaranteed

product is considered to be capital guaranteed over a certain investment period, e.g. three years.

Thus the investment horizon is ﬁnite and ﬁxed for every investor. When considering a fund,

the investor base is widely varying and most investors have diﬀerent investment horizons and

19

Chapter 4. The Capital Protection Property

investment times. This means that for a fund to be capital guaranteed it needs to be capital

guaranteed for all maturities and all overlapping time periods. Most ﬁnancial products can have

a value of zero prior to their maturity based on diﬀerent market risk factors, thus the value of

the fund at intermediate time points may converge to zero with a positive probability. It is now

quite obvious that a fund that holds these properties is a fund practically only investing in the

risk-free rate, which is not a desirable result.

Thus it is now quite clear that a fund based on structured products will not hold the property

of capital guarantee for all investors, maybe not for even anyone. This implies that the termi-

nology must be changed from capital guarantee to capital protection. Since capital protection

is a more diﬀuse deﬁnition that only states that the capital is protected, not guaranteed. Thus

a fund that uses the terminology capital protection in its marketing campaign only needs to

show that some of the capital base is protected. Thus it is important to notice the diﬀerence

between capital guarantee and capital protection, while smaller private investors may not notice

the diﬀerence, the diﬀerence is signiﬁcant.

In the following sections it will be covered how a fund should be constructed to attain as

much capital protection as possible. This includes naive fund constructions and investigations

of option portfolios etc.

4.2 Naive fund constructions

This section covers naive fund constructions; these constructions are created to understand more

of the dynamics of funds constructed of only structured products. The naive fund constructions

are probably the simplest possible funds based on structured products, thus called naive fund

constructions. By understanding more of the factors aﬀecting the return of the naive fund, it is

possible to gain knowledge of how a portfolio should be allocated amongst diﬀerent structured

products, to have as much capital protection as possible.

This section covers two diﬀerent naive fund constructions and how changes in the diﬀerent

risk factors aﬀect the return of the fund over an investment period. These diﬀerent changes are

stressed through using predeﬁned scenarios, not simulation. The reason why simulation is not

used is that the outcomes should be independent on the market model, as far as possible. Many

more scenarios, than the ones disclosed in this thesis, have been tested but only some the most

unfavorable scenarios are disclosed, since these scenarios are the only interesting scenarios when

regarding capital guarantee.

All structured products, in this chapter, have a time to maturity of three years at issuance;

a new structured product is issued each quarter. Thus it takes twelve quarters until the ﬁrst

product has matured, thus it is for simplicity assumed that the fund is launched to the public

during quarter twelve. It is assumed that investors have an investment horizon of three years,

thus the scenarios will be for six years, three prior to the investment (since the price of structured

products are path dependent) and three years after the investment.

The ﬁrst fund construction is a fund that is rolling capital guaranteed structured products.

Hence the notional amount of the bond equals the price of the structured product at issuance.

Thus the fund is just rolling capital guaranteed products with twelve diﬀerent maturities. A new

product is bought every quarter with the payoﬀ of the product that matures the same quarter.

The second fund construction is a fund that buys a standardized ELN at the start of the

fund and buys a new customized structured product each following quarter. The new customized

structured product has the same ratio of option:bond value as the fund prior to the rebalancing,

thus maintaining the fund’s ratio options:bonds relatively intact.

4.2.1 Naive fund construction number 1

A new structured product, that is 100% capital guaranteed, is issued every quarter where its

price equals the notional amount of the bond as well as the price of the underlying. Thus the

M. Hveem 20 (103)

4.2. Naive fund constructions

ratio options:bonds value is determined by the interest rate and the volatility, each product has

maturity three years after they are issued. The structured products follow the dynamics given

in Section 3.1.1.

Naive fund construction number one is a fund that is rolling the available structured products.

Every structured product matures after three years, which means that the fund buys the newly

issued structured product every quarter with the payoﬀ of the matured product (it is assumed

that the fund can hold inﬁnitesimal fractions of the structured products). Hence the fund will

hold a maximum of twelve products each quarter.

The fund has to start somewhere and since the product prices are path dependent it is

necessary to start the fund three years prior that the investor invests in it (in this setting).

Below follows a more detailed description of the fund. Denote,

i as the quarter the product was issued

β

i

t

as the value of the bond issued at quarter i at time t

O

i

t

as the value of the ATM call option issued at quarter i at time t

S

t

as the price of the underlying at time t

V

t

as the value of the fund at time t

c (S

t

, K, t, T, r, σ) as the value of a call option at time t with strike K, maturity time T

k

i

as the participation rate in the option issued at quarter i

r

f

t

as the risk-free interest rate at time t

r

i

t

as the return of product i between t −1 and t

r

t

as the return vector between t −1 and t

σ

t

as the volatility at time t

w

i

t

as the weight allocated in the product issued at quarter i at time t

w

t

as the weight vector at time t

As mentioned above, the structured products follow the dynamics given in Section 3.1.1 thus

the following formulas describe the prices of the structured products, the bonds are priced as,

β

i

t

=

_

_

_

S

i

e

−r

t

(i+12−t)

, if i ≤ t ≤ i + 12

0, otherwise

the participation rate of the option issued at quarter i is given as,

k

i

=

_

S

i

−β

i

i

_

c (S

i

, S

i

, i, i + 12, r

i

, σ

i

)

=

S

i

_

1 −e

−12r

i

_

c (S

i

, S

i

, i, i + 12, r

i

, σ

i

)

,

the value of the call option issued at quarter i at time t is given as,

O

i

t

=

_

¸

¸

¸

_

¸

¸

¸

_

k

i

c (S

t

, S

i

, t, i + 12, r

t

, σ

t

) , if i ≤ t ≤ i + 11

k

i

max (S

t

−S

i

, 0) , if t = i + 12

0, otherwise

Hence the return between t −1 and t for the structured product issued at quarter i is given as,

r

i

t

=

β

i

t

+O

i

t

β

i

t−1

+O

i

t−1

−1.

M. Hveem 21 (103)

Chapter 4. The Capital Protection Property

Now that the return for each structured product between each time period is known the attention

can again be turned towards the fund (the return of each product is the only necessary information

to calculate the return of the fund). The portfolio weights will be given diﬀerently between

quarters 0-11 and 12-24, since there is only one product available at quarter 0, only two products

available at quarter 1 and so on. Thus the portfolio sells of some of its capital each quarter to

allocate this in the newly issued product, until there are twelve products. At quarter 0 the fund

buys the newly issued product, at quarter 1 the fund sells of 50% of its possession to allocate

in the newly issued product, at quarter 2 the fund sells of 33.33% of its possession to allocate

in the newly issued product, at quarter 3 the fund sells of 25% of its possession to allocate in

the newly issued product and so on until the ﬁrst product matures, thus the portfolio weight for

asset i at time t between quarters 0-11 is given as,

w

i

t

| 0 ≤ t ≤ 11 =

_

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

_

_

1 +r

i

t

_

w

i

t−1

(1 +r

t

)

T

w

t−1

t

t + 1

, otherwise

1/(i + 1), if i = t

0, if i > t

The ﬁrst product matures during quarter twelve, thus there is a diﬀerent allocation scheme to

consider from quarter twelve and onwards. The payoﬀ of the matured product is as from quarter

twelve reinvested in the newly issued product (it is only rolling over the product). Thus the

portfolio weight for asset i at time t from quarter twelve and onwards is given as,

w

i

t

| t ≥ 12 =

_

¸

¸

¸

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

¸

¸

¸

_

_

1 +r

i

t

_

w

i

t−1

(1 +r

t

)

T

w

t−1

, otherwise

_

1 +r

i−12

t

_

w

i−12

t−1

(1 +r

t

)

T

w

t−1

, if i = t

0, if i ≤ t −12 or i > t

The fund’s return is the weighted return of all the assets’ returns, thus the value of the fund is

given as,

V

t

=

_

S

t

, if t = 0

V

t−1

(1 +r

t

)

T

w

t−1

, otherwise

Scenarios, underlying

The next step is to investigate some of the most important scenarios for the fund construction.

These scenarios should stress the fund construction such that weaknesses are disclosed. By

disclosing the worst-case scenarios it is possible to counter these characteristics by changing the

fund construction.

Note that only some of the most unfavorable scenarios are disclosed in this thesis, less unfa-

vorable scenarios are not disclosed, since they are not of importance (see Appendix A for more

scenario examples).

As mentioned earlier, it is assumed that the investors have an investment horizon of three

years and invest after three years (when the fund is announced on the market). It is important

that the investment horizon coincides with the time to maturity of the structured products, since

the individual structured product issued at the start of the investment provides absolute capital

guarantee for the investors and is their alternative investment. The scenarios are constructed to

stress the negative outcomes and as shown in Section 4.2.4 the underlying is the most prevalent

M. Hveem 22 (103)

4.2. Naive fund constructions

risk, thus the scenarios are mainly conducted for changes in the underlying. Therefor the other

parameters are set to be constant, σ = 0.10 and r

f

= 0.015 per quarter, for all quarters with a

ﬂat yield curve.

Scenario 1

The ﬁrst scenario is a scenario where the underlying has a continuous return of 10% during the

ﬁrst twelve quarters, then a continuous return of −15% the last twelve quarters; the result is

shown in Figure 4.1.

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 72.34

Quarter

$

Price Underlying

NAV Fund

Start of investment

Figure 4.1: Scenario 1, fund construction number 1: The value of the fund deteriorates as the price of

the underlying decreases, as the options are decreasing in value. The fund reallocates to

more options and cannot utilize the eﬀect of the growing value of the bonds in the same

extent as a regular structured product, since the fund buys new structured products each

quarter. Fund return quarters 12-24: −27.66%.

Figure 4.1 indicates that the price of the fund increases steadily during the quarters 1-12,

since all of the options end up in the money and increase in value. This implies that the ratio

between the value contained in options and bonds increases. Thus an investor who starts its

investment at quarter 12 will buy a high proportion of options, much higher than an investor

who bought the fund at quarter 0, hence the investor has no capital guarantee. On the other

hand an investor who invested at quarter 7 will actually receive a slight capital protection on

its previous positive return. The fund invests each quarter in a new structured product, thus

decreasing the ratio of options:bonds prior quarter 12, thus increasing the capital protection for

existing investors.

Figure 4.1 discloses that the value of the options decreases, towards zero, as the price of

the underlying continues to decrease. The problem with this allocation scheme is that the ratio

options:bonds increases as new products are introduced to the market, thus allocated in the fund.

The result indicates that the fund always strives to attain the original ratio by rebalancing from

bonds to options (or from options to bonds in the increase case), resulting in a further loss on the

option part. Thus there exists no capital guarantee since the fund is always buying new options.

The portfolio has lost most of its value in options, so how does the portfolio respond to an

increase in the price of the underlying after quarter 24? Figure 4.2 shows that there are still

some options left, but their value was really small prior to the increase, the portfolio still has the

same level of participation rate for the structured products.

An interesting feature with this portfolio construction is that; after an extreme decline a new

investor actually gains a degree of capital protection, as disclosed in Figure 4.3. In this setting

each product that matures is invested in the newly issued structured products, which means

that the amount invested in the new structured product is the notional amount of the previous

product, which implies that both products have the same notional amount. Thus during a large

M. Hveem 23 (103)

Chapter 4. The Capital Protection Property

0 5 10 15 20 25 30 35 40

0

20

40

60

80

100

120

140

X: 36

Y: 137.8

Quarter

$

Price Underlying

NAV Fund

Start of investment

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 108.1

Quarter

$

Price Underlying

NAV Fund

Start of investment

Figure 4.2: Extensions of scenario 1: The fund still has a lot of options held in the portfolio at quarter

24, thus an appreciation in the underlying generates a high return even though the previous

decrease in the price.

0 10 20 30 40 50 60

0

20

40

60

80

100

120

Quarter

$

Price Underlying

NAV Fund

Start of investment

Figure 4.3: The fund has a ﬂoor which it does not cross (given a ﬂat and constant yield curve) since

the portfolio is in practice just rolling bonds.

decline the fund’s value has a period of 12 quarters, where the value repeats itself. This means

that an investor at quarter 25 or 37 will in practice hold an identical position in bonds and

options as seen in Figure 4.3 (note that this is a special case with the ﬂat and constant yield

curve), thus the portfolio is in some sense capital guaranteed over the twelve month period in

this case assuming a ﬂat and constant yield curve, given the previous decline in the underlying.

On the other hand this happens since the investor buys almost only bonds and a small fraction

of options, thus the upwards potential is limited during the ﬁrst quarters of a bull market.

Scenario 2

The second scenario is a scenario where the underlying has zero in return during the ﬁrst twelve

quarters, then a continuous return of −15% the last twelve quarters; the result is shown in Figure

4.4. It can be expected that the outcome of this scenario should be very similar as the previous

one, since the scenarios do not diﬀer a lot from each other.

The fund value increases slightly during quarters 0-12 due to the interest of the bonds, while

the option value declines due to the theta value of the options. As the underlying crashes during

quarters 12-24 so does the option value, thus most of the value is contributed by the bonds at

M. Hveem 24 (103)

4.2. Naive fund constructions

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 91.89

Quarter

$

Price Underlying

NAV Fund

Start of investment

Figure 4.4: Scenario 2, fund construction number 1: 0% in return of the underlying for twelve quarters,

then continuous return of −15% each quarter. Fund return quarters 12-24: −8.11%.

quarter 15. This implies that, as the underlying continues to drop in price, there is not much

value left in options to aﬀect, thus after a few quarters the fund is almost rolling bonds (since

the decrease in option value each quarter reﬂects the increase of the bond value proportion of

the total value).

Thus the portfolio actually has, in this scenario, a restricted downside over the time horizon.

Hence the key to capital protection is to avoid holding products that are far in the money, since

they have a large percentage of value contained in the options. Therefor the portfolio should be

rebalanced such that these assets are underweighted.

4.2.2 Naive fund construction number 2

The naive fund construction number 2 diﬀers slightly from the previous one. The ﬁrst product

follows the same dynamics as in the previous subsection (100% capital guaranteed) but all the

other products are not standardized. Instead the structured products issued after quarter 0 are

customized in a way such that the ratio between the value contained in options and in bonds is

maintained relative stable for the fund. Thus every new structured product that is issued has

a participation rate such that the ratio between its option value and bond value, at issuance,

equals the fund’s ratio between option value and bond value at this quarter. The price of the

structured product equals the price of the underlying at issuance, which is also the option’s strike

price.

Every structured product matures three years after issuance, which means that the fund buys

the newly issued structured product every quarter with the payoﬀ of the matured product (it is

assumed that the fund can hold inﬁnitesimal fractions of the structured products). Hence the

fund will hold a maximum of twelve products each quarter.

The fund has to start somewhere and since the product prices are path dependent it is

necessary to start the fund three years prior that the investor invests in it (in this setting).

Below follows a more detailed description of the fund (the same notations as in the previous

subsection will be used).

The bonds’ notional amounts depend on both the price of the underlying and the ratio

options:bonds in the fund. The fund’s percentage of capital held in bonds at quarter i is given

as,

i−1

j=0

β

j

i

_

β

j

i

+O

j

i

_w

j

i

, and the percentage held in options as,

i−1

j=0

O

j

i

_

β

j

i

+O

j

i

_w

j

i

M. Hveem 25 (103)

Chapter 4. The Capital Protection Property

thus the bond prices are given as,

β

i

t

=

_

¸

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

¸

_

S

i

e

−r

t

(i+12−t)

, if i = 0

i−1

j=0

β

j

i

_

β

j

i

+O

j

i

_w

j

i

S

i

e

12r

i

−r

t

(i+12−t)

, if i ≤ t ≤ i + 12

0, otherwise

the participation rate of the option issued at quarter i is given as,

k

i

=

_

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

_

_

S

i

−β

i

i

_

c (S

i

, S

i

, i, i + 12, r

i

, σ

i

)

=

S

i

_

1 −e

−12r

i

_

c (S

i

, S

i

, i, i + 12, r

i

, σ

i

)

, if i = 0

i−1

j=0

O

j

i

_

β

j

i

+O

j

i

_w

j

i

S

i

c (S

i

, S

i

, i, i + 12, r

i

, σ

i

)

, otherwise

the value of the call option issued at quarter i at time t is given as,

O

i

t

=

_

¸

¸

¸

_

¸

¸

¸

_

k

i

C (S

t

, S

i

, t, i + 12, r

t

, σ

t

) , if i ≤ t ≤ i + 11

k

i

max (S

t

−S

i

, 0) , if t = i + 12

0, otherwise

Hence the return between t −1 and t for the structured product issued at quarter i is given as,

r

i

t

=

β

i

t

+O

i

t

β

i

t−1

+O

i

t−1

−1.

Now that it is known how the return for each structured product between each time period is

calculated the attention can again be turned towards the fund and how the portfolio weights are

calculated (since the return of the products each quarter depends on the weights the previous

quarter). The portfolio weights will be given diﬀerently between quarters 0-11 and 12-24, since

there is only one product available at quarter 0, only two products available at quarter 1 and so

on. Thus the portfolio sells of some of its capital each quarter to allocate this in the newly issued

product, until there are twelve products. At quarter 0 the fund buys the newly issued product,

at quarter 1 the fund sells of 50% of its possession to allocate in the newly issued product, at

quarter 2 the fund sells of 33.33% of its possession to allocate in the newly issued product, at

quarter 3 the fund sells of 25% of its possession to allocate in the newly issued product and so on

until the ﬁrst product matures, thus the portfolio weight for asset i at time t between quarters

0-11 is given as,

w

i

t

| 0 ≤ t ≤ 11 =

_

¸

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

¸

_

_

1 +r

i

t

_

w

i

t−1

(1 +r

t

)

T

w

t−1

t

t + 1

, otherwise

w

i

t

= 1/(i + 1), if i = t

w

i

t

= 0, if i > t

M. Hveem 26 (103)

4.2. Naive fund constructions

The ﬁrst product matures during quarter twelve, thus there is a diﬀerent allocation scheme to

consider from quarter twelve and onwards. The payoﬀ of the matured product is as from quarter

twelve reinvested in the newly issued product. Thus the portfolio weight for asset i at time t

from quarter twelve and onwards is given as,

w

i

t

| t ≥ 12 =

_

¸

¸

¸

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

¸

¸

¸

_

_

1 +r

i

t

_

w

i

t−1

(1 +r

t

)

T

w

t−1

, otherwise

_

1 +r

i−12

t

_

w

i−12

t−1

(1 +r

t

)

T

w

t−1

, if i = t

0, if i ≤ t −12 or i > t

The fund’s return is the weighted return of all the assets’ returns, thus the value of the fund is

given as,

V

t

=

_

S

t

, if t = 0

V

t−1

(1 +r

t

)

T

w

t−1

, otherwise

Scenarios, underlying

The next step is to investigate some of the most important scenarios for the fund construction.

These scenarios should stress the fund construction such that weaknesses are disclosed. By

disclosing the worst-case scenarios it is possible to counter these characteristics by changing the

fund construction.

Note that only some of the most unfavorable scenarios are disclosed in this thesis, less unfa-

vorable scenarios are not disclosed, since they are not of importance.

As mentioned earlier, it is assumed that the investors have an investment horizon of three

years and invest after three years (when the fund is announced on the market). It is important

that the investment horizon coincides with the time to maturity of the structured products, since

the individual structured product issued at the start of the investment provides absolute capital

guarantee for the investors and is their alternative investment. The scenarios are constructed to

stress the negative outcomes and as shown in Section 4.2.4 the underlying is the most prevalent

risk, thus the scenarios are mainly conducted for changes in the underlying. Therefor the other

parameters are set to be constant, σ = 0.10 and r

f

= 0.015 per quarter, for all quarters with a

ﬂat yield curve.

Scenario 1

The ﬁrst scenario is a scenario where the underlying has a continuous return of 10% during the

ﬁrst twelve quarters, then a continuous return of −15% the last twelve quarters; the result is

shown in Figure 4.5.

Figure 4.5 shows that the value of the fund increases a lot as the underlying increases, due to

the increase of the options and their convexity. Thus an investor that buys the fund at quarter

twelve buys a high proportion of options instead of bonds, since the fund maintains its ratio of

options:bonds relative stable when rebalancing. The fund’s proportion held in options converges

towards zero after seventeen quarters, thus bonds constitute almost all of the value. Figure 4.5

discloses that the fund actually has a positive return after quarter 17, due to the positive return

of the bonds, even though the market continues to decrease, as the value of the option part is

already zero.

If the market goes down during twelve consecutive months an increase in the market would

not have any aﬀect on the value of the fund, since it would not contain any options anymore.

This fund construction creates a certain level of capital protection, but not in the desired way

M. Hveem 27 (103)

Chapter 4. The Capital Protection Property

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 61.68

Quarter

$

Price Underlying

NAV Fund

Start of investment

Figure 4.5: Scenario 1, fund construction number 2: Similar result as in the previous fund construc-

tion. As the price of the underlying increases the ratio options:bonds increases heavily

to approximately 1:1 from 16.47:83.53, thus a new investor at quarter twelve can loose

approximately half of its investment. Fund return quarters 12-24: −38.32%.

since it is useless after a huge decline. It is also too risky since gains from the options are not

reallocated to bonds, creating a huge downside after a bull market. Also, when the market turns

bust during consecutive quarters the portfolio reduces its participation heavily with the market,

which can imply that the fund is only allocating in bonds and that it is no longer a structured

products fund.

Scenario 2

The second scenario is a scenario where the underlying has zero in return during the ﬁrst twelve

quarters, then a continuous return of −15% the last twelve quarters; the result is shown in Figure

4.6.

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 116.4

Quarter

$

Price Underlying

NAV Fund

Start of investment

Figure 4.6: Scenario 2, fund construction number 2: The ratio options:bonds is very low since the

price of the underlying does not move until quarter 12. Thus the fall of the price of the

underlying does not really aﬀect the value of the fund a lot. Fund return quarters 12-24:

16.36%.

All options end out of the money, thus the option part’s value quickly converges towards zero

M. Hveem 28 (103)

4.2. Naive fund constructions

and the portfolio gains a positive return from the remaining bonds.

4.2.3 Analysis and summary

The most preferable fund construction is construction number one since it has less extreme

outcomes. Also fund construction number one will continue to participate in the market even

though the market has decreased a lot. The fund also provides capital protection, when the

market has gone up, for investors who experienced the bull-market.

Based on these results it is easy to conclude that there is a big advantage to rebalance the

portfolio at intermediate time points so that the ratio options:bonds does not exceed a certain

limit. This implies that it is possible to protect new investors from for example scenario 1. It is

also desirable to increase the ratio options:bonds in the same way as it is done in fund construction

number 1 after a market crash, if the fund wants to increase the possibility of appreciation, but

this should not be conducted when trying to minimize the risk. Using the allocation scheme in

fund construction number 1 implies that the fund slowly rebalances towards its original ratio

options:bonds. On the other hand rebalancing up the ratio options:bonds each quarter when it

has decreased under a certain limit would imply that the portfolio’s risk increases i.e. there are

pros and cons with every decision.

Fund construction number 2 is not desirable at all since it is a very risky investment strategy

and does not hold the desired characteristics.

Concluded, it is important that the allocation scheme has a focus on rebalancing the rate

options:bonds, where there exists a maximum acceptable limit. This can be done by for example

solving a linear optimization problem.

The question now arises regarding how the option portfolio should be constructed, i.e. how

the weights in the diﬀerent options (and thus also the structured products) should be constructed

to retain as much capital protection as possible. The next section investigates how an investor

should allocate the option portfolio to have as little downside risk as possible. The reason why

it is possible to generalize the result of options to the structured products is that a change in

the underlying is the most prevalent risk of the structured products, which will be discussed in

the next subsection.

4.2.4 Prevalent risk factors

This short subsection discusses the most prevalent risk factors for an ELN. It is important to

determine the most prevalent risk factors to be able to construct a portfolio of structured products

such that the risk is minimized.

The most volatile part of a structured product is the option, the bond is less sensitive to

movements aﬀecting its price, i.e. the yield curve. Consider an Equity-Linked-Note with a price

of $100, price of the underlying of $100, y

tm

= 0.06, σ = 0.2 p.a. and a time to maturity of three

years. The ratio options:bonds would be 16.47:83.53, a shift of the underlying downwards of 50%

would induce a decrease of the product’s value of 15.9743. The yield to maturity would have

to increase to 13.08% to cause a price drop that big. There is not a huge risk with changes in

the yield curve that are not extreme since if the yield curve goes up bonds will be less expensive

and the bonds bought will generate a higher return, as seen in Figure 4.7 which depicts a huge

change in the interest rate. Also the interest rate risk is very small for buy and hold strategies,

since the investor is guaranteed a ﬁxed amount in the future, thus a ﬁxed yield.

Figure 4.8 depicts the sensitivity of an ELN w.r.t. the underlying and the interest rate, the

ELN is much more sensitive to changes in the underlying than in the yield curve. Thus it is

more important to focus on changes in the underlying than the interest rate. When considering

a buy and hold portfolio (which most funds practice, it is reasonable to assume that the whole

portfolio will not rebalanced each quarter due to transaction costs) it is a much larger risk that

the option is out of the money close to maturity than that the bond has a low price close to

M. Hveem 29 (103)

Chapter 4. The Capital Protection Property

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 100.3

Quarter

[

%

]

Interest Rate

NAV Fund

Start of investment

Figure 4.7: The interest rate goes from 2% p.a. to 14% p.a. during just a few quarters. The investor

does not have a negative return over the whole period. The big risk is if a change occurs

just prior to the end of the investment horizon, but as it can be seen, the impact is still

limited.

maturity. Thus the interest rate risk of holding an existing bond is much smaller than holding

the option.

0

50

100

150

200

0

5

10

15

60

80

100

120

140

160

S

t

T = 3

Interest Rate r [%]

E

L

N

P

r

i

c

e

[

%

]

70

80

90

100

110

120

130

140

150

0

50

100

150

200

0

5

10

15

90

100

110

120

130

S

t

T = 0.5

Interest Rate r [%]

E

L

N

P

r

i

c

e

[

%

]

95

100

105

110

115

120

125

Figure 4.8: The surface plots describe how changes in the yield and price of the underlying aﬀect the

price of an ELN. σ = 0.2, K = 100, as time decreases so does the sensitivity to changes in

the yield curve.

There may also be a risk with a decrease in volatility. But, as a consequence of the deﬁnition

of volatility, the volatility increases as the price goes down (since the drift is in general positive),

thus the two risk factors counter each other’s impact on the price. Thus a decrease in volatility

is not a big concern since it happens when the price of the underlying goes up, which aﬀects the

option price more than the volatility as seen in Figure 4.9.

Hence it is possible to generalize the optimal allocation of options, in min risk sense, to a

portfolio of structured products since the options in the ELNs are signiﬁcantly more risky than

the bonds. The next section will therefor investigate how the option portfolio should be allocated

to minimize the negative outcomes.

M. Hveem 30 (103)

4.3. Investigation of the option portfolio

0

50

100

150

200

0

10

20

30

40

50

0

20

40

60

80

100

S

t

[%] Volatility σ [%]

O

p

t

i

o

n

P

r

i

c

e

0

10

20

30

40

50

60

70

80

90

Figure 4.9: The surface plot describes the relationship between changes in volatility and the price,

price is the dominant factor. Strike K = 100, r = 0.06, T = 3.

4.3 Investigation of the option portfolio

The purpose of this section is to investigate how the option portfolio should be allocated to min-

imize the negative outcomes in sense of the underlying, since the minimum-risk option portfolio

should in most cases yield the minimum-risk ELN portfolio.

The options are priced with Black-Scholes formula as given in Equation 3.1. The option

prices are assumed to not be smaller than 0.5% of the price of the underlying, before maturity,

to prevent unrealistic results. Without setting such boundaries some options may have returns

in the size of 10

30

between two quarters.

The study covers diﬀerent portfolio constructions (only those relevant are shown) and how

the scenarios are modeled to stress the diﬀerent portfolio allocations. The section also generates

a recommendation regarding how an option portfolio should be allocated such that the risk is

minimized.

4.3.1 Portfolio construction and modeling

The portfolios that are investigated consist of only plain vanilla at-the-money call options at

issuance on the underlying index. To test which portfolio setups that are best in min risk sense

the study will conduct Monte Carlo simulation. It is not important to replicate a certain index,

since the study is looking for patterns, which should be independent as much as possible of the

assumptions of the underlying. The index OMXS30 at Nasdaq OMX serves as a base for the

simulation, thus daily log returns are gathered between 1st of October 2009 to 30th of September

2010. The log returns are ﬁtted to a student’s t-distribution to attain slightly heavier tails than

for the normal distribution, as well as a better ﬁt. Since the investigation is investigating patterns

and worst-case outcomes it does not really matter what the drift µ is set to be, thus µ is for

simplicity set to zero, the other ﬁtted parameters for the t-distribution are: σ = 0.0105, ν = 5.69

(on a days basis, it is not important what the actual parameters are since the focus lies on

patterns, which should not be sensitive to the distribution of the underlying).

The following parameters are also used: r

f

= 6% p.a. and the volatility used when pricing

the options is the actual realized volatility during the period of 20.5% p.a.

M. Hveem 31 (103)

Chapter 4. The Capital Protection Property

A new ATM call option is issued each quarter with maturity in three years, e.g. the option

issued at quarter t has strike K

t

= S

t

, where S

t

is the price of the underlying at time t. Thus

there are at each intermediate time point twelve options available to invest in, with diﬀerent

strikes and maturity times.

Monte Carlo simulation is used to create outcomes of the underlying, a total of 60,000 three-

year periods are simulated (excluding the three-year time period used to create the necessary

price paths for the options).

The weight in option j held by the portfolio i is denoted as w

i

j

where j is the option that

has a maximum of j quarters left to maturity, j = 1, 2, 3, ..., 12. The portfolio weight vector of

portfolio i is denoted as w

i

.

The best result is measured in the sense of worst outcome (since intrinsic capital protection

is investigated in this chapter, another option would be CVaR), it is also desirable that the

distribution of the outcomes is positive skewed, thus imposing less risk to an investor. The

reason for measuring the risk of holding the option portfolio as the worst-case outcome and not

by just using a measure such as the volatility is that the return distributions of derivatives (e.g.

options) often are skewed and exhibit kurtosis.

The next subsection describes the diﬀerent relevant fund constructions and discloses the result

of them as well. It turns out that the products with the longest time to maturity exhibit the

lowest risk.

4.3.2 Portfolios and results

Eight portfolios are investigated in this section using the simulation described above, these

portfolio are some of the most relevant constructions (for more construction please see Appendix

B). The option portfolios are as follows,

The ﬁrst portfolio has portfolio weights that are equal to each other, thus the portfolio is totally

diversiﬁed over all available options.

w

1

=

_

1

12

1

12

1

12

1

12

1

12

1

12

1

12

1

12

1

12

1

12

1

12

1

12

_

T

The second portfolio is a portfolio that has the ﬁrst eleven portfolio weights equal to each other,

the last one equal to zero.

w

2

=

_

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

0

_

T

The third portfolio is a portfolio that has the last eleven portfolio weights equal to each other,

the ﬁrst one equal to zero.

w

3

=

_

0

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

_

T

The fourth portfolio is a portfolio that has the ﬁrst eleven portfolio weights equal to zero, and

the last one equal to one.

w

4

=

_

0 0 0 0 0 0 0 0 0 0 0 1

_

T

The ﬁfth portfolio is a portfolio that has decreasing portfolio weights, with w

5

j

= 13 −j.

w

5

=

_

12

78

11

78

10

78

9

78

8

78

7

78

6

78

5

78

4

78

3

78

2

78

1

78

_

T

The sixth portfolio is a portfolio that has increasing portfolio weights, with w

6

j

= j.

w

6

=

_

1

78

2

78

3

78

4

78

5

78

6

78

7

78

8

78

9

78

10

78

11

78

12

78

_

T

M. Hveem 32 (103)

4.3. Investigation of the option portfolio

The seventh portfolio is a portfolio that has equal portfolio weights in assets 2-10, zero otherwise.

w

7

=

_

0

1

9

1

9

1

9

1

9

1

9

1

9

1

9

1

9

1

9

0 0

_

T

The eighth portfolio is a portfolio that has equal portfolio weights in assets 2-11, zero otherwise.

w

8

=

_

0

1

10

1

10

1

10

1

10

1

10

1

10

1

10

1

10

1

10

1

10

0

_

T

The results for the diﬀerent option portfolios are disclosed in Figure 4.10, there are also

additional option portfolios disclosed in Appendix B.

Figure 4.10 and Appendix B show that the portfolios where the options closest to maturity

are underweighted have “better” distributions since they exhibit less negative skewness. Also the

worst-case outcome is much better for portfolios that have underweighted the options closest to

maturity. Fewer number of assets in the portfolio results in, in most cases, a more risky portfolio,

both in worst-case and in distribution. Notable is that the portfolio that has the best worst-case

outcome is the portfolio that holds 100% of its holdings in the option with the longest time to

maturity (Portfolio 4).

Figure 4.12a shows that options with a longer time to maturity are less sensitive to the

changes in time (θ =

∂P

∂t

). Thus the value reduction coming in form of the reduction in time,

ceteris paribus, is less if the investor holds options with a long time to maturity. Also the delta

(∆ =

∂P

∂s

) for options that are in the money is higher for options close to maturity than options

with a long time to maturity as depicted in Figure 4.12b.

As a fund manager it would be quite unreasonable to allocate all the capital in the same

asset, thus it is adequate to demand that the fund diversiﬁes itself, to not overweight. Usually

a portfolio manager is not allowed to invest more than a certain percentage (say 30%) in a

particular asset, it is in this case beneﬁcial to diversify along almost all options, except the one

closest to maturity (which has the highest risk, since it might be “all or nothing”). The results

in Figure 4.10, B.1 and B.2 indicate that it does not really matter how the portfolio diversiﬁes

amongst the assets since the results are quite similar as long as the portfolio holds a relative

large proportion of the holdings in products with a longer time to maturity. Thus it might be

reasonable to diversify by holding only four options with a long time to maturity as in Portfolio

16 in Appendix B or for example diversify over the whole spectrum and attain approximately

the same results.

It would be desirable if the return distribution for the portfolio of options over 36 months

would look somewhat similar to the return distribution of a single option from issue to maturity

as seen in Figure 4.11. The results shown in the ﬁgures indicate that it is not possible to attain

the same type of distribution as for a single option when investing in several options over 36

months.

−5 0 5 10 15 20

0

1000

2000

3000

4000

5000

6000

7000

Plain vanilla option return

Return [%]

O

u

t

c

o

m

e

s

Figure 4.11: The histogram shows the distribution of a plain vanilla option’s one-year return, issued

ATM (T = 1).

M. Hveem 33 (103)

Chapter 4. The Capital Protection Property

−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

180

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(a) Portfolio 1, −22.37%.

−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(b) Portfolio 2, −21.88%.

−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

180

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(c) Portfolio 3, −17.64%.

−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

50

100

150

200

250

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(d) Portfolio 4, −16.15%.

−0.3 −0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

180

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(e) Portfolio 5, −27.69%.

−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

180

200

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(f ) Portfolio 6, −20.71%.

−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

180

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(g) Portfolio 7, −19.80%.

−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

180

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(h) Portfolio 8, −18.91%.

Figure 4.10: Histograms over the outcomes for option portfolios 1-8, the worst-case average 36 month

return is given in the caption.

M. Hveem 34 (103)

4.3. Investigation of the option portfolio

An optimal portfolio would be positive skewed, with a limited downside. The most beneﬁ-

cial construction fulﬁlling these criteria is Portfolio 3, as indicated in Figure 4.10c (as well as

considering the worst-case outcome). Hence it is important to diversify over a large set of assets

to reduce the risk, when constructing a portfolio only consisting of options. On the other hand

as the time to maturity decreases, the risk increases (the theta and delta (given that the option

is in the money) in particular, as disclosed in Figure 4.12). Thus when constructing a portfolio

with structured products it is important to not overweight in the shorter tenors.

The main result is that the fund should diversify over a number of assets and also avoid

investing in the products with a short time to maturity. Next follows an analysis of how the

results found in Section 4.2 and in this section should be combined to construct a desirable

structured products fund.

0

50

100

150

200

0

10

20

30

40

−4.5

−4

−3.5

−3

−2.5

−2

−1.5

−1

−0.5

0

S

t

Months to maturity

O

p

t

i

o

n

T

h

e

t

a

(

Θ

)

(a) The plot describes a plain vanilla call option’s theta

(time value), the theta is heavily decreasing in time,

implying that the holder looses a lot of capital from

a time change close to maturity, ceteris paribus.

0

50

100

150

200

0

10

20

30

40

0

0.2

0.4

0.6

0.8

1

Months to maturity

S

t

O

p

t

i

o

n

D

e

l

t

a

(

∆

)

(b) The plot describes a plain vanilla call option’s delta,

options closer to maturity are more sensitive when

they are in the money than options with a longer

time to maturity.

Figure 4.12: The Figure a) illustrates an option’s theta value, the Figure b) illustrates an option’s

delta value for diﬀerent levels of in the money and time to maturity.

4.3.3 Analysis and implications

In this subsection follows a combination of a summary, an analysis and a description of the

implications from the studies in this chapter so far.

It is known from Section 4.2 that it is desirable to rebalance the portfolio when the ratio

options:bonds exceeds a predetermined level (which depends on the risk proﬁle of the fund) to

limit the downside risk. The reason why the ratio options:bonds suddenly exceeds a certain level

is (most often) due to either an increase in the price of the underlying (increasing the value

of the options) or in the yield curve (reducing the value of the bonds). Thus by rebalancing

the portfolio the downside risk is limited since the proportion of options is reduced. This is

an important concept since it protects new investors, the fund must be able to guarantee new

investors that they do not buy more than a certain percentage of options. Without this guarantee

the purpose of the fund is lost, by appealing to investors’ loss aversion (Shefrin, 2002, [35]).

By rebalancing the portfolio after a period of growth the value held in bonds increases. Thus

the existing investors will receive a certain level of capital protection on their gains. The tradeoﬀ

is that the portfolio at the same time limits the upside of the investment (the tradeoﬀ between risk

and return). These characteristics appeal to psychological concepts within Behavioral Finance,

where investors tend to sell assets too early during growth, since they are conservative and

people do not expect the market to continue appreciating (disposition eﬀect) (Frazzini, 2006,

[15]; Kahneman and Tversky, 1979, [26]). Thus investors want to rebalance their portfolio after

M. Hveem 35 (103)

Chapter 4. The Capital Protection Property

a period of appreciation.

During bear markets, when the portfolio decreases in value, it may be desirable to rebalance

the portfolio’s ratio of options:bonds towards a desired level. The rebalancing should not be

conducted in the same manner as when the ratio exceeds a certain level, instead it may be

desirable to slowly re-establish the portfolio’s ratio options:bonds to the original level. Without

this reallocation the fund will not have as much upside for new investors. Thus it is a balance

act for the fund how it should change the ratio options:bonds in a way that both new investors

and current investors appreciate. The next section will investigate how this ratio/level should

be chosen and its implications on downside risk.

Since Equity-Linked Notes are considered in this paper the options are far more risky than the

bonds, thus by allocating the portfolio according to the minimum risk choice w.r.t. the options

should also yield the minimum risk choice for the structured products.

As stated previously, there is no need to investigate changes of volatility, since the volatility

and the price of the underlying usually counter each other (Glot, 2005, [16]). Thus the impact

of volatility can be neglected when choosing portfolio weights. The inﬂation goes up when

the market goes up, thus also interest rates. When markets go down, investors ﬂee to safer

instruments such as bonds, thus pushing up the price and decreasing the yield. This implies that

the price changes of bonds and call options are negatively correlated.

As seen in Section 4.3 the best choice to minimize the downside risk for a portfolio of options

is by diversiﬁcation in the longer (and middle) tenors. Thus it is not desirable to hold positions

in options that have a short time to maturity.

When considering a portfolio of structured products it might be non-desirable to overweight

the newly issued products during bear markets since they will have a substantial proportion of

option value in them compared to the other products. Thus if the portfolio overweights/allocates

to the newly issued ELNs it will always reallocate capital to newly issued options, which is not

desirable in bear markets. Hence instead of allocating according to Portfolio 3, as when only

considering an option portfolio, a higher degree of capital protection during bear markets will be

attained by investing using either Portfolio 7 or 8. Thus it is reasonable to invest using Portfolio

3 in general and after drops in the index allocate the portfolio according to Portfolio 7.

It is desirable to rebalance the portfolio such that the ratio options:bonds are held under

control, but how should this be conducted, for which levels etc? These issues will be investigated

in the next section where a comparison with existing mixed funds results in optimal choices of

the ratio options:bonds.

4.4 Comparison with competition

A lot of assumptions have to be made in able to construct the fund. One way to make adequate

assumptions is to relate to the fund’s competition. Investors will not invest in the fund if it is not

capable of beating its competition, at least at some aspects. The comparison will be based on

the results found in the previous sections. Thus the purpose of this section is to investigate how

a potential structured products fund (SPF) performs in relation to its competition (mixed funds)

during extreme scenarios. The analysis results in appropriate boundaries for the percentage of

value in the SPF contributed by options given a certain competing/benchmark fund.

A competing fund is a fund that has the same client base as the SPF. Thus it is imperative

to establish the SPF’s client base, so that a competing benchmark fund can be identiﬁed. After

that the competing fund has been identiﬁed the study investigates how diﬀerent SPFs can be

constructed, such that the SPF provides a higher degree of capital protection than the benchmark

fund.

The diﬀerent constructions are both stress tested with predeﬁned extreme scenarios as well

as simulated scenarios. The diﬀerent SPF strategies are also backtested against historical data.

The section covers two types of SPFs, one where the weights are based on the possession in

the structured products and one SPF where the weights in the structured products are based on

M. Hveem 36 (103)

4.4. Comparison with competition

the possession in the options. It turns out that the second one is more theoretical than the ﬁrst

one. The result in this section indicates that a structured products fund should try to have a

possession of 5-25% option value attained in the portfolio, with a target level of 15% to be more

attractive for investors in min risk sense than a competing mixed fund.

4.4.1 Client base

The fund based on structured products has a high focus on capital protection. Thus the investors

that it is suitable for are those who are looking for investments with a limited downside, or

a small risk, as well as a possibility to have a return connected to the market. These are

risk-aversive investors who usually are looking for a combination of bonds and stocks. Today

many of these investors invest in mixed funds, which diversify their portfolios in both debt

and equity instruments to lower the risk level, hence it is this alternative which is an adequate

benchmark. Thus mixed funds are identiﬁed to be the competition to the SPF, hence the

structured products fund must provide a lower downside risk than competing mixed funds to be

an attractive investment alternative.

Next follows a more detailed description of the benchmark fund.

4.4.2 Benchmark fund

It is assumed that a reasonable benchmark fund is a mixed fund that aims at holding 30% (ε

bench

)

of its capital in stocks and 70% in bonds with up to three years until maturity (to match the

SPF). The benchmark fund rebalances its position every three months, if the level of stocks held

is not within 30% ± 10% (δ

bench

). The investment within stocks is considered in this study to

be conducted in the index (or an ETF), which is also the underlying for the SPF. An example

of a price development of this type of benchmark fund is disclosed in Figure 4.13. The ﬁgure

shows that the benchmark fund gains value as the market is appreciating and does not decrease

signiﬁcantly as the market takes a turn for the worse. Thus it has some of the desired properties

for a SPF, imposing an adequate benchmark. The focus is now turned to the construction of the

structured products fund.

0 50 100 150

50

100

150

200

250

300

350

400

450

Month

$

Price Underlying

NAV Benchmark

Figure 4.13: The co-movements of the underlying and the benchmark fund, for

bench

= 0.3, δ

bench

=

0.1 and r = 0.06 p.a, random scenario.

4.4.3 Structured products fund 1

The fund constructions in this section are based on the results gained in Section 4.2 and 4.3, thus

the portfolio diversiﬁes broadly amongst the available products. Portfolio 7 will be used, since

M. Hveem 37 (103)

Chapter 4. The Capital Protection Property

the test is conducted for bear markets (in Appendix C results for Portfolio 3 are disclosed), thus

the portfolio holds equal weights in products i = 2 − 10 where i denotes the maximum number

of quarters until maturity for the product as previously.

A new capital guaranteed structured product is issued each quarter with maturity in three

years, price equal to the price of the underlying and also the notional amount of the bond, as

in Section 4.2.1. The fund also has the possibility of investing in short-term debt, in this case

bonds that have maturity the next quarter. Thus the fund can by investing in the short-term

debt reduce the percentage of value placed in options. A target level of options held in the fund’s

portfolio is also introduced, this target level is the desired percentage of the fund’s value that

should be constituted by options, let denote this target level as ε

SPF

. Notable is that the fund

is not allowed to short the short-term debt.

The fund rebalances its portfolio of structured products each quarter as previously, according

to the weights and maturity amongst the structured products, and also the proportion placed in

additional bonds.

The fund adjusts the proportion placed in the short-term debt subject to its level of option

value as a percentage of the total fund. The portfolio is allowed to deviate a percentage δ

SPF

from

its target level of options. Thus the portfolio is rebalanced if the percentage of value contributed

by the options exceeds ε

SPF

+δ

SPF

(or is less than ε

SPF

−δ

SPF

) by allocating in short-term debt

rather than the structured products, thus attaining the target level.

˜ w denotes the weight vector for the structured products excluding the short-term debt (the

weights amongst the structured products), the weight in the short-term debt at time t is denoted

as w

b

t

. Thus the actual weights (except w

b

t

) for each time-period is given by:

w

t

=

_

1 −w

b

t

_

˜ w,

and since the focus lies on Portfolio 7 from Section 4.3 the following weights are used,

˜ w =

_

0

1

9

1

9

1

9

1

9

1

9

1

9

1

9

1

9

1

9

0 0

_

T

.

Note that w

b

0

= 0 and w

0

= ˜ w, the proportion of value of the fund contributed by bonds is

denoted α

b

, and α

o

is the proportion of value of the fund contributed by options. Which gives

us the following deﬁnitions,

α

b

t

= w

b

t

+

_

1 −w

b

t

_

n

i=1

˜ w

i

β

i

t

β

i

t

+O

i

t

,

α

o

t

=

_

1 −w

b

t

_

n

i=1

˜ w

i

O

i

t

β

i

t

+O

i

t

,

where n is the number of available products (12 in this case), thus the portfolio is rebalanced

each quarter if,

α

o

t

≤ ε

SPF

−δ

SPF

or α

o

t

≥ ε

SPF

+δ

SPF

.

The portfolio is rebalanced with the aim that α

o

should equal

SPF

, this is not always possible

since it is not allowed to short the short-term debt, which implies that the new portfolio weights

are given as,

w

new

t

=

_

¸

¸

¸

¸

¸

_

¸

¸

¸

¸

¸

_

w

t

ε

SPF

α

o

t

, if

i

w

i

t

ε

SPF

α

o

t

≤ 1

˜ w, if

i

w

i

t

ε

SPF

α

o

t

> 1

M. Hveem 38 (103)

4.4. Comparison with competition

these weights are the weights excluding the short-term debt, thus the new allocation in the

short-term debt is as follows,

w

b

t

= 1 −1

T

w

new

t

.

The fund is not allowed to short the short-term debt (to minimize the risk), thus it cannot hold

negative positions in w

b

t

. This implies that the short-term debt is used as a safe haven to avoid

a too high exposure to the market.

The purpose with the following subsections is to determine the adequate levels for ε

SPF

and

δ

SPF

such that the fund has a greater protection (towards extreme downside risk) than the

benchmark portfolio(s), and still has a good upside potential. The levels will be determined such

that the SPF provides a greater capital protection for market crashes of −20% p.a. and more.

4.4.4 Scenarios

This subsection describes the scenarios that the SPF and benchmark fund are stressed with. The

scenarios are not simulated since it is desirable that the SPF provides capital protection for some

speciﬁed market crashes.

The scenarios should stress the SPF and the benchmark fund towards negative outcomes, thus

expose the funds in an extent where they disclose weakness (which is a market crash immediately

after a strong increase in the market, as seen in the previous sections). The focus lies on scenarios

with an up-going market followed by a market crash (as the worst-case scenarios in Section 4.2),

ﬁve diﬀerent scales of market crashes are investigated, −80%, −50%, −30%, −20% and −10%

p.a. during four years. Two diﬀerent settings prior to the crash will be investigated, the ﬁrst

setting with a bull market with a return of 30% in return p.a. between months 0-48 (Scenario A).

The second scenario is quite similar, 30% return p.a. during months 0-48 and a return of 50%

each month for months 48-50 (Scenario B). The idea is that since the funds can only rebalance

to short-term debt every three months that the funds can be stressed even more if an extreme

increase occurs during a period where the funds cannot rebalance. Thus these two types of

scenarios should cover the worst types of extreme outcomes for the funds (for more extreme

scenarios please consult Appendix C).

Hence the market crash in Scenario A is between months 48-96 and in Scenario B between

months 50-98.

After the crash the market experiences an increase of 30% p.a. until month 144 (12 years).

An annual interest rate of 6% is assumed for the bonds as well as the short-term debt (ﬂat yield

curve is assumed for simplicity), the scenarios can be seen in Figure 4.14.

The reason why predeﬁned stressed scenarios are studied instead of simulation in this sub-

section is that the study is trying to stress test the funds in particular scenarios and see if the

SPF has the desired properties. Also it is imperative that the results gained from this study are

independent of the models selected for the underlying and that they can be generalized to any

distribution or market model, thus it is appropriate to select scenarios without simulation.

The next subsection covers the results over how the funds perform in the diﬀerent scenarios

and how the results should be interpreted.

4.4.5 Results, structured products fund vs benchmark

The focus lies on the worst-case outcomes of the SPF (based on Portfolio 7) and the benchmark

fund given diﬀerent scenarios and values for ε and δ. The objective is to construct a SPF such

that it has more favorable worst-case outcomes than the benchmark portfolio.

Various results are disclosed in Figure 4.14, Table 4.1 and 4.2. Table 4.1 discloses the result

regarding the SPF, its worst-case outcome given the diﬀerent scenarios, ε

SPF

and δ

SPF

. Table 4.2

discloses in the same manner the result regarding the benchmark fund, its worst-case outcome

given the diﬀerent scenarios, ε

bench

and δ

bench

.

M. Hveem 39 (103)

Chapter 4. The Capital Protection Property

Return Index

Months 48-96 /

50-98 p.a.

ε

SPF

δ

SPF

Scenario A Worst-

Case Return

Scenario B Worst-

Case Return

-80% 1 1 -18.53% -56.07%

-80% 0.3 0.1 -18.53% -37.62%

-80% 0.25 0.1 -11.95% -32.18%

-80% 0.20 0.1 -11.35% -22.62%

-80% 0.15 0.1 -2.85% -15.20%

-80% 0.15 0.05 -2.54% -19.22%

-80% 0.10 0.05 -1.06% -9.59%

-50% 1 1 -17.46% -48.50%

-50% 0.3 0.1 -17.46% -33.04%

-50% 0.25 0.1 -11.57% -25.42%

-50% 0.20 0.1 -11.03% -18.39%

-50% 0.15 0.1 -3.42% -11.48%

-50% 0.15 0.05 -6.20% -16.71%

-50% 0.10 0.05 -1.82% -7.59%

-30% 1 1 -16.64% -37.40%

-30% 0.3 0.1 -16.64% -22.49%

-30% 0.25 0.1 -13.24% -19.26%

-30% 0.20 0.1 -11.31% -13.63%

-30% 0.15 0.1 -5.00% -5.41%

-30% 0.15 0.05 -7.32% -10.23%

-30% 0.10 0.05 -3.45% -3.54%

-20% 1 1 -14.60% -26.84%

-20% 0.3 0.1 -14.60% 14.64%

-20% 0.25 0.1 -11.13% -10.59%

-20% 0.20 0.1 -9.72% -6.15%

-20% 0.15 0.1 -3.95% 0.34%

-20% 0.15 0.05 -6.09% -3.38%

-20% 0.10 0.05 -2.55% 2.90%

-10% 1 1 -9.60% -14.08%

-10% 0.3 0.1 -9.60% -4.75%

-10% 0.25 0.1 -6.78% -1.81%

-10% 0.20 0.1 -5.81% 0.43%

-10% 0.15 0.1 -1.65% 3.82%

-10% 0.15 0.05 -3.11% 1.23%

-10% 0.10 0.05 0.22% 4.79%

Table 4.1: SPF 1: Worst-case outcomes for SPF 1 based on Portfolio 7 given scenarios A and B, ε

SPF

and δ

SPF

, the worst-case period is always the 36 month following the initiation of the crash.

M. Hveem 40 (103)

4.4. Comparison with competition

Return Index

Months 48-96 /

50-98 p.a.

ε

bench

δ

bench

Scenario A Worst-

Case Return

Scenario B Worst-

Case Return

-80% 0.7 0.1 -94.49% -94.48%

-80% 0.5 0.1 -87.35% -87.80%

-80% 0.3 0.1 -62.94% -63.04%

-80% 0.20 0.1 -43.27% -44.37%

-80% 0.10 0.05 -17.36% -18.17%

-50% 0.7 0.1 -72.62% -72.16%

-50% 0.5 0.1 -54.86% -55.73%

-50% 0.3 0.1 -34.13% -33.07%

-50% 0.20 0.1 -15.61% -15.26%

-50% 0.10 0.05 -0.37% 0%

-30% 0.7 0.1 -47.71% -47.22%

-30% 0.5 0.1 -31.84% -32.5%

-30% 0.3 0.1 -15.68% -12.84%

-30% 0.20 0.1 -2.59% -2.39%

-30% 0.10 0.05 7.45% 7.94%

-20% 0.7 0.1 -32.85% -31.70%

-20% 0.5 0.1 -18.36% -18.91%

-20% 0.3 0.1 -6.66% -4.61%

-20% 0.20 0.1 2.58% 4.78%

-20% 0.10 0.05 10.24% 11.56%

-10% 0.7 0.1 -15.57% -13.80%

-10% 0.5 0.1 -4.76% -5.03%

-10% 0.3 0.1 1.58% 4.75%

-10% 0.20 0.1 7.87% 9.48%

-10% 0.10 0.05 13.09% 14.24%

Table 4.2: Benchmark fund: Worst-case outcomes given scenarios A and B, ε

bench

and δ

bench

.

M. Hveem 41 (103)

Chapter 4. The Capital Protection Property

The structured products fund provides, as the tables indicate, (in general for the same levels

of ε and δ as the benchmark fund) a much higher safety in extreme downturn markets (−80%

drop), since it does not short bonds to reallocate its portfolio. It is not a huge diﬀerence between

the worst-case outcomes for diﬀerent extreme scenarios (−30 to −80% drops) when regarding

the SPF (Table 4.1), since the value of the options does not decrease a lot more in the −80%

case than in the −30% case. Table 4.2 shows that the benchmark fund on the other hand suﬀers

to a high extent when the index has huge negative returns since it reallocates every three-month

to the index, thus not providing a limit where it cannot lose more capital. It is important to

notice that the SPF appreciates quite a lot when the market starts to go up again, which is

very desirable as seen in Figure 4.14. The SPF has a greater upside for huge increases than

the benchmark fund (Figure 4.14). Notable is that it takes longer time for the SPF than the

benchmark fund to appreciate after the market turns bust, it is this eﬀect which also creates the

capital protection property, since it takes a few quarters to increase the participation rate.

Table 4.1 shows that the SPF is in general more sensitive to a large drop in the stock market

after an extreme increase without having the possibility to rebalance the portfolio (Scenario B)

than to Scenario A. The SPF is less sensitive to Scenario B if it is only followed by a correction

in the market (return of −10% to −20% p.a.). This is due to that the SPF does not suﬀer a lot

of the decline between months 50-51 and the portfolio is rebalanced at month 51 to lower levels,

thus making it less sensitive to a market crash. On the other hand the portfolio does not beneﬁt

so much of the rebalancing possibility during month 51 in the case of a huge decline between

months 50-51, such as in the −80% drop scenario (Table 4.1). Notable is that the results are

quite similar if there is a drop of −80% per month between months 50-51, the results for this

scenario are disclosed in Appendix C

1

.

Since the fund rebalances its portfolio at month 48 a sense of upper loss limit, which the

investor cannot lose more than if it starts its investment horizon at quarter 48, is created. This

implies that if the fund experiences a total market crash (−100%) the investor still holds its

position in bonds, while the market linked assets would by worth zero. Thus the investor loses

a “maximum” (in sense of the underlying) of ε + δ which is applicable to both the SPF and

benchmark fund. But the value invested in options is convex and more volatile than investing

in the underlying. Thus a decline with 50% may erase almost all value attained in the options.

Hence the criteria ε

SPF

+ δ

SPF

< ε

bench

+ δ

bench

must hold if the SPF should have a higher

capital protection for declines of the scale 30-50%.

One interesting property that is not disclosed in the tables is that the fund can limit its

downside signiﬁcantly using two diﬀerent δ

SPF

, one for the upside (δ

SPF

+

) and one for the downside

(δ

SPF

−

). By increasing the downside limit δ

SPF

−

the fund has a higher tolerance for holding short-

term debt than when having a symmetrical δ

SPF

. Important to notice is that as δ

SPF

−

increases

it will be harder for the SPF to appreciate after a bear market since it accepts higher levels of

short-term debt and does not reallocate in a high extent to structured products.

Next follows an analysis regarding the results and an interpretation regarding their implica-

tions for a portfolio manager.

4.4.6 Analysis and summary

As mentioned earlier an adequate benchmark fund is a fund which has ε

bench

= 0.3 and δ

bench

=

0.1. Thus the desired weights of ε

SPF

and δ

SPF

should provide better capital protection against

downside risk than the benchmark fund. It is reasonable to demand that the SPF should be less

sensitive to market drops of −20% p.a. or more (it is diﬃcult to ﬁnd portfolio weights for the

fund that would be less sensitive at a −10% market drop

2

, as seen in Table 4.1 and 4.2). By

inspecting Table 4.1 and 4.2 it is easy to notice that the adequate levels for ε

SPF

and δ

SPF

are

0.15 respectively 0.1 (notable is that these boundaries also provides a better worst-case outcome

1

The scenarios considered in the appendix are worse, but if those scenarios would occur the investor would

have, most certainly, bigger problems to consider than its position in the SPF.

2

Since the price of the options is more volatile than for stocks and indices.

M. Hveem 42 (103)

4.4. Comparison with competition

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(a) Scenario A, −80% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(b) Scenario B, −80% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(c) Scenario A, −30% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(d) Scenario B, −30% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(e) Scenario A, −15% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(f ) Scenario B, −15% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(g) Scenario A, −10% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(h) Scenario B, −10% market crash.

Figure 4.14: Illustrative example of the scenarios and movements of the SPF 1 (Portfolio 7) and the

benchmark fund for ε

SPF

= 0.15, ε

bench

= 0.3, δ

SPF

= 0.1, δ

bench

= 0.1.

M. Hveem 43 (103)

Chapter 4. The Capital Protection Property

for a −15% market crash as depicted in Figure 4.14). Examples of outcomes for Portfolio 7 using

diﬀerent

SPF

and δ

SPF

with diﬀerent scenarios are disclosed in Figure 4.14.

Thus a SPF should have the parameters ε

SPF

= 0.15, δ

SPF

= 0.1 to provide better capital

protection than a benchmark fund that has ε

bench

= 0.3 and δ

bench

= 0.1.

These are just some of the possible boundaries; by increasing ε

SPF

to 0.2 or 0.25 the fund is

more aggressive and can take advantage of the momentum gained during bull markets (as well

as increasing the possibility of a higher return after bear markets, Figure 4.14). Thus the level of

ε

SPF

needs to be decided in comparison with the competition and how the SPF is managed, the

fund’s goals etc. It is important to contemplate regarding what the fund is trying to accomplish

and what tradeoﬀs diﬀerent portfolio choices imply.

To summarize the previous sections: it is important to diversify the SPF over a numerous

set of assets (Section 4.3). The fund should avoid investing in structured products that have

maturity in the nearest future, since these are very volatile and in practice almost only a bet

with a high theta (cost of making the bet). Also the fund should during a bear market avoid

overweighting newly issued products (Section 4.3). In a downturn market investing in newly

issued products increases the value held in options in the portfolio, thus increasing the risk.

Thus the fund manager should, when trying to minimize the downside risk, allocate broadly over

the spectrum of available products, except in the products that are about to mature (especially

during bull markets) and newly issued products during bear markets.

It is important that the fund manager reallocates some of the return to either cash or short-

term debt, as the SPF value increases, hence maintaining the ratio between the capital allocated

in options and bonds at a reasonable level (Section 4.2 and 4.4). This implies that the fund

locks in some of its previous return as it appreciates, thus protecting the previous proﬁts. It

also protects new investors from investing in a (too) high proportion of options in relation to

the holdings in bonds. These eﬀects are really desirable and in line with human behavior to

rebalance due to loss-aversion (Kahneman and Tversky, 1979, [26]). When the manager does not

rebalance the portfolio investors may invest in a fund with a high risk, due to the volatility of

the option payoﬀs. It is, as shown in this study, imperative that the fund manager understands

the risk of investing in structured products in a sophisticated portfolio, with diﬀerent maturities,

and how the capital protection is aﬀected by the market returns and the reallocations.

It is not only important to rebalance the portfolio after a series of appreciations to decrease

the risk of the fund. The fund manager should after periods of deprecation rebalance (if any

existing) holdings from short-term debt to the structured products to increase the optionality (in

this case the proportion of options) in the portfolio to avoid that new investors will suﬀer from

the fund’s previous decline (Section 4.4). It is also important to rebalance the portfolio after a

decline to give the previous investors the possibility to “win the losses back” as people prefer this

re-investment strategy due to get-evenitis (Shefrin, 2002, [35]).

Thus when formulating an optimization algorithm, that allocates amongst diﬀerent structured

products, there are several constraints that should be made such that the risk of the portfolio

is held under control. The proportion held in options should be restricted as done in this

section, the fund manager decides the appropriate limits depending on what the SPF is trying

to attain in capital protection sense and in relation to competition. It is also wise to limit the

proportion invested in each structured product (to diversify adequately), and when minimizing

the absolute downside it is recommended to avoid overweighting the newly issued products during

bear markets and products maturing the upcoming period.

The SPF 1 has been stressed for the predeﬁned scenarios, and the appropriate levels of ε

SPF

and δ

SPF

have been attained. Next follows an additional stress test to assure that these levels

also provide better outcomes than the benchmark fund under simulated scenarios.

4.4.7 Additional stress test

The stress test is conducted to ensure that the SPF 1 yields a higher degree of capital protection

than a benchmark fund given the attained levels of ε

SPF

and δ

SPF

also for other scenarios than

M. Hveem 44 (103)

4.4. Comparison with competition

−0.4 −0.2 0 0.2 0.4 0.6 0.8 1 1.2 1.4

0

500

1000

1500

2000

2500

3000

3500

Histogram over the SPF:s 36 month (overlapping) returns

Return

O

u

tc

o

m

e

s

(a) SPF 1: min: -23.05%, mean:

16.19%, max: 127.19%.

−0.4 −0.2 0 0.2 0.4 0.6 0.8 1 1.2

0

500

1000

1500

2000

2500

3000

Histogram over the Benchmark funds 36 month (overlapping) returns

Return

O

u

tc

o

m

e

s

(b) Benchmark: min: -24.01%,

mean: 15.35%, max: 105.45%.

−1 0 1 2 3 4 5

0

500

1000

1500

2000

2500

3000

3500

4000

Histogram over the Underlying 36 month (overlapping) returns

Return

O

u

tc

o

m

e

s

(c) Underlying: min: -80.59%,

mean: 6.3%, max: 455.26%.

Figure 4.15: Histograms over 36 months returns for the competing funds and the stress test.

the ones speciﬁed above. Thus SPF 1 has ε

SPF

= 0.15 and δ

SPF

= 0.1, which should yield better

worst-case outcomes than the benchmark fund according to the study conducted above.

The same type of Monte Carlo simulation as performed in Section 4.3 is used, thus the daily

log returns for the index are modeled with a student’s t-distribution with ν = 5.69, σ = 0.0105

and µ = 0 on a days basis. The following parameters are also used: r

f

= 6% p.a. and the

volatility used when pricing the options is 20.5% p.a.

The drift equals to zero, since the model should be stressed towards negative outcomes,

notable is that it is not important exactly what the parameters are since the study searches

for patterns. The study is conducted through investigating 500,000 36-month periods and the

36 months return. The result is disclosed in Figure 4.15, which indicates that SPF 1 has the

most desirable distribution of the three and the most limited downside. Thus the result (Figure

4.15 and the speciﬁed outcomes) shows that SPF 1 is a better investment alternative than the

benchmark fund, and certainly the index when considering downside risk. Notable is also that

SPF 1 has a higher upside than the benchmark fund (comparing Figure 4.15a and 4.15b), thus

it may aspire to investors who are looking for a decent upside and a limited downside. Figure

4.15a indicates that SPF 1 has a positive skewed distribution, which is optimal for risk-averse

investors, kurtosis is also displayed on the positive side and not the negative side. The results

for a stress test where µ equals −30% p.a. is discloses in Appendix C (Figure C.1), notable is

that SPF 1 is superior in extreme worst-case scenarios as well.

Now that it is concluded how a SPF should be constructed w.r.t. ε

SPF

and δ

SPF

it is interesting

to investigate how the fund would have performed in the past through a backtest. Thus the next

subsection focuses on a backtest for the SPF and how it would have performed historically.

4.4.8 Backtesting

A backtest means that a modeled is tested on historical data and evaluated on that basis. Thus

the strategy for the SPF described above (using Portfolio 7) is applied with the Nasdaq OMXS30

Index as underlying between 1996-01-02 −2010-09-30. The volatility used is the realized volatility

over every last three-month period. The interest rates for the zero-coupon bonds are given by

the STIBOR (Stockholm Interbank Oﬀered Rate). This is a good proxy for the interest rates

for ELNs in the Nordic market where the bonds are issued by local investment banks, thus the

STIBOR rate reﬂects the rate of the zero-coupon bonds for ELNs issued in Sweden in SEK. The

risk-free rate is assumed to equal the STIBOR, since this is a quite conservative choice and a

common rate to use as the risk-free rate (a higher r

f

will yield a lower return for the SPF). For

tenors over one year quoted SEK swap rates are used, available on Bloomberg. The backtest-

results for diﬀerent portfolios are disclosed below, for diﬀerent ε

SPF

and δ

SPF

. Higher ε

SPF

and

δ

SPF

implies a higher downside risk and the historical worst-case scenario is best for the most

restrictive choice of ε

SPF

and δ

SPF

. Thus as mentioned earlier, the choice of these parameters

should depend on the goals of the fund and thus determine the risk-level.

M. Hveem 45 (103)

Chapter 4. The Capital Protection Property

The time horizon considering capital protection is of three years as mentioned earlier. The

worst-case outcome over all overlapping three years periods for the benchmark is −14.17% and

for the index −67.79%, SPF 4.16a: −2.13%; SPF 4.16b: −8.15%; SPF 4.16c: −9.95%; SPF

4.16d: −16.30%. Thus most of the portfolios had a better worst-case return than the benchmark

fund for the OMXS30 index between 1996 to the third quarter of 2010.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(a) ε

SPF

= 0.10, δ

SPF

= 0.05.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(b) ε

SPF

= 0.15, δ

SPF

= 0.1.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(c) ε

SPF

= 0.2, δ

SPF

= 0.1.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(d) ε

SPF

= 0.25, δ

SPF

= 0.1.

Figure 4.16: Backtest results: OMXS30 vs SPF 1 (Portfolio 7) and Benchmark.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(a) ε

SPF

= 0.10, δ

SPF

= 0.05.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(b) ε

SPF

= 0.15, δ

SPF

= 0.1.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(c) ε

SPF

= 0.2, δ

SPF

= 0.1.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(d) ε

SPF

= 0.25, δ

SPF

= 0.1.

Figure 4.17: Backtest results: OMXS30 vs SPF 1 (Portfolio 3) and Benchmark.

The results indicate that the SPFs have the desired properties. Notable is that the comparison

is done with the OMXS30, which is not including dividends and a benchmark fund, based on

this index. Thus it is not a perfect benchmark and a more adequate benchmark is disclosed in

Chapter 6. This standard index version is used in this chapter to be able to cover two market

crashes; the total return version only exists since 2002.

The next subsection is regarding an alternative to the SPF 1 described above. The big

diﬀerence lies in their allocation schemes for the structured products, but it will not be covered

as extensively as SPF 1 since SPF 2 is mostly of theoretical interest.

M. Hveem 46 (103)

4.4. Comparison with competition

4.4.9 Structured products fund 2

The structured products in SPF 1 were allocated using a ﬁxed weights scheme, which indicated

the weights allocated in the certain structured products. Thus it is impossible to guarantee that

the fund will hold the desired amount of options from the diﬀerent tenors in comparison to each

other using SPF 1. This means that the portfolio may overweight certain options signiﬁcantly,

even though the total value held in options are under control, through the limits of ε

SPF

and

δ

SPF

. The optimal portfolio weights that minimized the risk of movements in the underlying

were discovered in Section 4.3. Thus to minimize the risk of movements in the underlying the

fund should allocate the portfolio based rather on the options itself than based on the structured

products. This means that the allocation algorithm calculates the weights for the structured

products based on the proportion of options:bonds in the products and the desired relation of

concentration amongst the options.

Now consider the same setting as for the SPF 1, the only thing that is diﬀerent is that the

weights are attained through a diﬀerent scheme. The downside risk should be minimized for

movements in the underlying. Thus option Portfolio 3 will be used as a base to calculate the

adequate portfolio weights for the structured products. Important to notice is that this section is

very theoretical and has less practical applications than SPF 1, which is quite realistic. SPF 2 is

quite theoretical since the ratio options:bonds may decline signiﬁcantly for some products if the

market turns bust. Thus inﬁnitesimal positions in some products and huge positions in others

may be necessary to attain the desired level of options held in each tenor. This implies that

the scheme may be adequate in times of normal market movements, but is disrupted in stressed

scenarios. This scheme would also generate massive transaction costs, since the portfolio weights

between two time periods often would change signiﬁcantly, since the value of options is more

volatile than the whole structured product.

The same products are available as for SPF 1. The fund reallocates its portfolio in the same

manner as for SPF 1 (using the short-term debt etc.), the only diﬀerence is that ˜ w diﬀers. The

weight vector amongst the options is denoted w

options

and is in this case given as follows,

w

options

=

_

0

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

1

11

_

T

,

let ξ be the ﬁrst non-zero element of w

options

, j its index and o

i

t

=

O

i

t

β

i

t

+O

i

t

. A temporary weight

vector w

temp

(t) is created, given as,

w

temp

i

(t) =

_

¸

¸

¸

_

¸

¸

¸

_

0, if i < j

ξ, if i = j

w

options

i

o

j

t

o

i

t

, if i > j

thus ˜ w is given as,

˜ w

t

=

w

temp

(t)

1

T

w

temp

(t)

.

The rest of the allocation scheme is exactly the same as in Section 4.4.3 for the SPF 1.

The same analysis is conducted for this fund construction as for SPF 1. The results for

Scenario A and B are given in Table 4.3. By comparing the results in Table 4.1 and 4.3 it is

easy to see that the SPF 2 has much better worst-case outcomes including every scenario and

every selection of ε

SPF

and δ

SPF

than SPF 1. The Portfolio 3 was the best choice to minimize

downside risk with in Section 4.3, thus it is reasonable that SPF 2 should beat SPF 1 in terms

of worst-case outcome w.r.t. the underlying.

Figure 4.14 and 4.19 indicate that SPF 2 does not only provide a better protection during

the market crash but also a better future potential when the market starts to appreciate and

turns into a bull market.

M. Hveem 47 (103)

Chapter 4. The Capital Protection Property

In the SPF 2 case it is possible to choose ε

SPF

2

= 0.20 and still have a lower downside risk

than the benchmark fund.

The same analysis, implications etc. are applicable to SPF 2 in the exact same way as SPF

1. Hence only the results are disclosed for the interested reader.

The following section discusses the terminology capital protection and how it should be used

further in this thesis.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(a) ε

SPF

= 0.10, δ

SPF

= 0.05

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(b) ε

SPF

= 0.15, δ

SPF

= 0.1

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(c) ε

SPF

= 0.2, δ

SPF

= 0.1

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(d) ε

SPF

= 0.25, δ

SPF

= 0.1

Figure 4.18: Backtest results: OMXS30 vs SPF 2 (Portfolio 3) and Benchmark.

M. Hveem 48 (103)

4.4. Comparison with competition

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(a) Scenario A, −80% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(b) Scenario B, −80% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(c) Scenario A, −30% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(d) Scenario B, −30% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(e) Scenario A, −15% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(f ) Scenario B, −15% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(g) Scenario A, −10% market crash.

0 50 100 150

0

20

40

60

80

100

120

140

Month

$

Price Underlying

NAV Fund

NAV Benchmark

(h) Scenario B, −10% market crash.

Figure 4.19: Illustrative example of scenarios and movements of the SPF 2 (Portfolio 3) and benchmark

fund for ε

SPF

= 0.15, ε

bench

= 0.3, δ

SPF

= 0.1, δ

bench

= 0.1.

M. Hveem 49 (103)

Chapter 4. The Capital Protection Property

Return Index

Months 48-96 /

50-98 p.a.

ε

SPF

δ

SPF

Scenario A Worst-

Case Return

Scenario B Worst-

Case Return

-80% 1 1 -11.79% -30.67%

-80% 0.3 0.1 -11.79% -20.76%

-80% 0.25 0.1 -11.79% -18.51%

-80% 0.20 0.1 -4.74% -14.24%

-80% 0.15 0.1 -4.58% -10.11%

-80% 0.15 0.05 0.00% -8.22%

-80% 0.10 0.05 1.20% -2.61%

-50% 1 1 -10.89% -21.42%

-50% 0.3 0.1 -10.89% -14.66%

-50% 0.25 0.1 -10.89% -12.07%

-50% 0.20 0.1 -4.30% -6.10%

-50% 0.15 0.1 -2.47% -2.51%

-50% 0.15 0.05 -3.26% -3.90%

-50% 0.10 0.05 1.24% 2.25%

-30% 1 1 -9.48% -15.39%

-30% 0.3 0.1 -9.48% -9.08%

-30% 0.25 0.1 -9.48% -6.76%

-30% 0.20 0.1 -3.70% -3.11%

-30% 0.15 0.1 -2.24% 0.93%

-30% 0.15 0.05 -2.49% -0.74%

-30% 0.10 0.05 -1.16% 4.08%

-20% 1 1 -8.17% -11.80%

-20% 0.3 0.1 -8.17% -6.90%

-20% 0.25 0.1 -8.17% -4.69%

-20% 0.20 0.1 -3.98% -1.20%

-20% 0.15 0.1 -1.61% 2.82%

-20% 0.15 0.05 -1.72% 0.88%

-20% 0.10 0.05 1.76% 5.54%

-10% 1 1 -5.88% -7.25%

-10% 0.3 0.1 -5.88% -3.77%

-10% 0.25 0.1 -5.88% -1.60%

-10% 0.20 0.1 -1.67% 0.80%

-10% 0.15 0.1 0.00% 4.69%

-10% 0.15 0.05 -0.60% 2.36%

-10% 0.10 0.05 3.29% 7.16%

Table 4.3: SPF 2: Worst-case outcomes for the SPF 2 based on Portfolio 3 option weights given

scenarios A and B, ε

SPF

and δ

SPF

.

M. Hveem 50 (103)

4.5. Alternative deﬁnitions of capital protection

4.5 Alternative deﬁnitions of capital protection

Since it is not possible to construct a capital guaranteed SPF it is imperative to discuss the usage

of the terminology capital protection. The terminology capital guarantee is something absolute,

which indicates that the portfolio will not have a negative return. It is absolute in the sense that

it cannot be distorted and it is not possible to argue that a placement is capital guaranteed if it

may have a negative return (still not taking the credit risk into consideration). The terminology

capital protection on the other hand is more diﬀuse and ﬂexible.

Still, it is quite easy to have the same associations to the word protection as the word guar-

antee, and many investors mix up the terminologies (Chuan, 2008, [12]). Thus while it is not

possible to call the fund capital guaranteed it is essential that the fund uses the terminology

capital protected in its communication with the investors to appeal to their loss-aversion (Kah-

neman and Tversky, 1979, [26]). What does it mean that an investment is capital protected?

Capital protection means, according to UBS (2010, [37]), that the investor is guaranteed a certain

minimum repayment of the invested amount in the end of its investment term (time horizon).

Thus when claiming that a fund is capital protected it should have a limited downside risk, i.e.

a certain percentage of the investment should be guaranteed.

This leads us to the purpose of this section, alternative deﬁnitions of capital protection. By

using the original deﬁnition and modify it slightly it may be possible to construct a fund which

fulﬁlls the requirements for an alternative deﬁnition of capital protection.

The focus lies on two alternative deﬁnitions of capital protection. These deﬁnitions are

evaluated regarding their pros and cons and which deﬁnition is best to use. Note that these

alternative deﬁnitions are for capital protection, not guarantee. The capital protection property

is regarding a certain time horizon, i.e. if the investors hold their position during this time horizon

they will receive a certain level of capital protection.

4.5.1 Deﬁnition 1 - Absolute lower bound

This deﬁnition is a worst-case scenario based deﬁnition, which means that the portfolio’s perfor-

mance should be Monte Carlo simulated with a certain number of scenarios, N, which generates

a worst-case outcome for the portfolio. The aim of the portfolio allocation is that the worst-case

scenario should exceed the invested amount X

0

times the degree of capital protection κ (de-

termines the degree of capital protection, can be any positive number). Hence the portfolio is

considered as capital protected if,

min

i

X

i

T

> κX

0

i = 1, ..., N,

where X

i

T

is the portfolio value at time T (end of the investment horizon) for the ith scenario.

Pros

This is probably the most concrete deﬁnition of capital protection since the worst outcome

simulated actually would exceed the minimum acceptable level of return. The investor can be

quite conﬁdent that it will not lose more capital than the given level over the time horizon. It is

also possible to use a sophisticated algorithm to solve the linear optimization problem as given

in Section 2.3.2, ensuring a well-deﬁned solution to the problem.

Cons

The point of using this deﬁnition is to quantify the worst-case outcome, but the worst-case

outcome will always be based on the scenarios, the assumptions regarding the market model and

M. Hveem 51 (103)

Chapter 4. The Capital Protection Property

how the underlying is modeled, the options, the yield curve etc. Also, how many scenarios are

needed? Choosing between N = 10,000, 50,000 or 100,000 scenarios, which one is adequate?

The problem is that when increasing the number of scenarios, the number of extreme outliers

are also increased, so how is it possible to determine the adequate number of scenarios to use,

since the worst-case scenario would for most portfolios converge to −100% in return given that

N → ∞? Another problem with this deﬁnition is that it can be extremely hard to ﬁnd a

portfolio that actually fulﬁlls this requirement (depending on the choice of κ, e.g. if κ = 1 the

portfolio may end up in only investing in bonds and almost nothing in options, thus eliminating

the possibility to gain when the market goes up). A negative side of this deﬁnition is also that

the risk measure is only saying something about the worst-case outcome, not anything about the

remaining unfavorable outcomes.

4.5.2 Deﬁnition 2 - CVaR

This deﬁnition is based on CVaR, which means that the CVaR for the portfolio at some quantile α

must be less than (1 −κ) X

0

, for a determined level of capital protection κ. For more information

regarding the deﬁnition of CVaR please consult Section 2.1.3. Thus the portfolio is considered

to be capital protected if,

CVaR

α

(L

w

) = E[L

w

| L

w

≥ VaR

α

(L

w

)] ≤ (1 −κ) X

0

Pros

A positive side of this deﬁnition is that it is more realistic (in sense that a portfolio may actually

fulﬁll the requirement even for high values of κ) than the worst-case deﬁnition. It is more robust,

it is a coherent risk measure and the result converges as the number of scenarios grows. Also it

may be a more adequate deﬁnition than the worst-case deﬁnition, since it discloses, “if it goes

bad, how bad does it get?” and not only for the worst-case scenario. For example by using the

quantile α = 0.95 CVaR is the expected value of the loss for the 5% worst outcomes. It is also a

risk measure that can capture the risk of heavy tails and skewness. Thus it is possible to capture

a lot of the risk using CVaR, much more than using a worst-case outcome deﬁnition. Another

advantage is that the linear optimization problem for CVaR is well deﬁned and easy to solve

using the simplex method.

Cons

The negative aspect is that it does not provide an absolute deﬁnition of capital protection, if the

CVaR is positive it does not guarantee that there are no negative outcomes.

CVaR is the risk measure that will be used further on in this thesis, due to its major beneﬁts.

Note that it is possible to use the alternative deﬁnition 1 in the same manner as with alternative

deﬁnition 2 in the upcoming scenario optimization, by solving the minimum regret optimization

problem instead. The portfolio weights will in general diﬀer for the two diﬀerent risk measures,

since they measure diﬀerent things. The focus on minimizing the worst-case outcome may impose

a high risk for the investor, leading to a suboptimal portfolio.

M. Hveem 52 (103)

CHAPTER 5

Modeling Financial Assets

Modeling the ﬁnancial assets is one of the most important issues in portfolio optimization and

allocation problems. It is of great importance that the risk factors are modeled in a way such

that they reﬂect the underlying risk factors. This thesis studies Equity-Linked Notes, thus the

components that must be modeled are: the yield curve, the underlying and the volatility. These

risk factors determine the fair price of the fund. The reason why the previous study has been so

extensive is to avoid, as far as possible, dependence on the chosen market model in this chapter.

5.1 Volatility and option pricing

It is important that the options in this study can be priced through a closed-form expression, since

the study involves the usage of Monte Carlo simulation and a simulation based pricing model

would thus be too computer intensive. The Black-Scholes formula for option pricing will therefor

be used for pricing the plain vanilla options, since it provides a fast and adequate pricing model.

The problem with Black-Scholes formula is that it does not always price the options correctly

due to the misspeciﬁcation that the price of an asset follows a geometric brownian motion. The

implied volatility is the volatility that should be used in Black-Scholes formula so that it gives

the fair price of the option. Thus by using Black-Scholes formula and implied volatility the

correct price is attained. The thesis is regarding ELNs and as discussed earlier, in Section 3.3.1,

there are no systematic diﬀerences for the prices of ELNs when using either implied volatility or

realized volatility (Wasserfallen and Schenk, 1996, [39]). Thus the volatility will be modeled as

the realized volatility during one year, thus it is based on 52 weekly observations of the index

log returns.

It is important to notice that the liquidity of options on OMXS30 is quite poor, which implies

that it is hard to ﬁnd appropriate historical data such that the price of the options can be modeled

through implied volatility (surfaces). Thus it may reduce the misspeciﬁcation and error in pricing

on the Nordic market to use realized volatility instead of implied volatility.

The dividend rate of the index must be estimated, since the OMXS30 index is not including

dividends. There exists a total return version of OMXS30, where dividends are reinvested in

the index, which is available on Bloomberg. When comparing the total return index with the

standardized version it is easy to see that the total return index yields a higher return of approx-

imately 3.0% p.a. using continuous compounding during the observed time period (since 2002

when it ﬁrst was started to 15th of November 2010), which implies that the expected dividend

yield is 3.0% p.a. in the Black-Scholes model.

5.2 Index

There exists countless of models describing the movements of a stock or an index. Probably the

most popular model is to describe the movements of the underlying as a geometric brownian

motion as done by Black and Scholes (1973, [9]). It is also common to model the returns

by using for example normal variance mixtures (Aas, 2006, [1]; Hult and Lindskog, 2009, [23]).

53

Chapter 5. Modeling Financial Assets

Another popular model is proposed by Heston (1993) in [20] where he proposes a model including

stochastic volatility that results in a closed-form expression for option pricing.

Since the aim of this chapter is to model scenarios, for the risk factors so that the scenarios

can be evaluated in an optimization problem, it is desirable to have a model that is easy to sample

from, also it is important to maintain the correlation in the data. Not only the underlying should

be modeled but also the yield curve, thus it is necessary to consider the dependence structure

between the index and the yield curve. An adequate method to use is principal component

analysis (PCA) as described in Section 2.2. There exists signiﬁcant dependence between a yield

curve and its market’s corresponding index, e.g. in the Swedish market the STIBOR is not

independent of the stock index OMXS30. The correlation between OMXS30 and the Swedish

yield curve is negative for shorter tenors and positive for tenors above one year (also the absolute

value of the correlation is much higher for longer tenors). The correlation is in general higher for

weekly returns than for daily returns during the observed time period. The correlation matrices

for daily, weekly and yearly log returns between OMXS30 and the SEK rates between 1M to 3Y

are disclosed in Appendix D. Thus it is better to use weekly log returns for the simulation since

the dependence structure amongst the data is not contaminated as much as the daily log returns

by noise.

Since one of the most appropriate ways to model the yield curve is by PCA and since PCA

also is a good method to describe stock/index returns it is a natural choice to use PCA for the

two combined.

1998 2000 2002 2004 2006 2008 2010

400

600

800

1000

1200

1400

Year

I

n

d

e

x

Nasdaq OMXS30

Figure 5.1: The closing index levels of OMXS30 between January 1996 and November 2010.

5.3 Yield curve

It is as mentioned earlier a good choice to use PCA to model the yield curve changes since

a yield curve can, in general, be described by three dominant factors in a three-factor model,

these are the level, steepness and curvature (Litterman and Scheinkman, 1991, [29]). Therefor

the yield curve will be modeled as described in Section 2.2 together with the index to maintain

the dependence structure. Thus the PCA should result in four signiﬁcant factors, i.e. principal

components, which will be enough to explain most of the dependence structure in the data. This

choice of model guarantees that the sampling of scenarios will be parsimonious.

In the described market model the STIBOR will be used as the risk-free rate, since this is a

common choice as the risk-free rate when pricing in the Swedish market. STIBOR is the daily

interbank oﬀered rate in Stockholm, thus it also reﬂects the yield on the zero-coupon bonds that

are issued by the local investment banks. Hence the STIBOR is an appropriate rate both for

the risk-free rate and the zero-coupon rate. The Swedish Riksbank publishes daily quotes of the

STIBOR on their homepage http://www.riksbank.se.

5.4 A PCA model for the yield curve and index

A four-factor model is, as mentioned earlier, enough to describe the market movements, since

the yield curve is described by three factors and the index (mainly) by the remaining factor.

M. Hveem 54 (103)

5.4. A PCA model for the yield curve and index

1998 2000 2002 2004 2006 2008 2010

1

2

3

4

5

6

7

8

Year

[

%

]

1M

2M

3M

6M

9M

12M

2Y

3Y

Figure 5.2: SEK STIBOR and Swap rates between January 1996 and November 2010.

The index and the yield curve must be rewritten on a stationary form since they are not

stationary. Thus the index and the yield curve nodes are modeled with log returns, which are

weakly stationary. Even though it may seem strange to use log returns for the yield curve, since

these are in some sense already log returns, it is possible to prevent negative interest rates by this

procedure, which is very important. Also it seems more adequate for the observed data sample

to model with log returns than absolute changes.

5.4.1 Data set

The OMXS30 from the beginning of 1996 until the middle of November 2010 is used for the

simulation. The STIBOR is used as the risk-free rate up to one year and above one year the

SEK swap rates are used, the data set ranges from the beginning of 1996 until the middle of

November 2010. The volatility is modeled by using the one year realized volatility. STIBOR has

the tenors 1M, 2M, 3M, 6M, 9M, 12M and the swap rates the tenors 2Y and 3Y, thus there is a

total of nine PCs. A linear interpolation is performed for products with a time to maturity that

does not coincide with these tenors.

5.4.2 PCA results

A PCA is performed on the whole data set; the ﬁrst four PCs explain 97.94% of the variation in

the data. This is in line with the results found by Litterman and Scheinkman (1991, [29]) as well

as by Barber and Copper (2010, [6]) that most PCAs for yield curves can with only three PCs

describe more than 95% of the variation (notable is that by removing the index the ﬁrst three

factors explains 97.68% of the variation). Notable is that the four PCs describe more than 95%

of the variation, which is more than enough.

PC Explained variation [%] Cumulative explained variation [%]

1 68.00 68.00

2 16.81 84.82

3 9.62 94.44

4 3.50 97.94

5 0.72 98.65

6 0.57 99.22

7 0.37 99.59

8 0.30 99.89

9 0.11 100.00

Table 5.1: PCA results: Variation explained by each principal component and total explained variation

(cumulative explained variation).

By inspecting Figure 5.3 it is easy to see that the ﬁrst PC’s factor loadings are all positive and

can be interpreted as the level change, also the factor loading for the index is almost zero. The

M. Hveem 55 (103)

Chapter 5. Modeling Financial Assets

OMXS30 1M 2M 3M 6M 9M 1Y 2Y 3Y

−1

−0.8

−0.6

−0.4

−0.2

0

0.2

0.4

0.6

0.8

1

Risk Factor

F

a

c

t

o

r

L

o

a

d

i

n

g

s

Principle Components

PC 1 (Level)

PC 2 (Steepness)

PC 3 (Index)

PC 4 (Curvature)

Figure 5.3: Factor loadings for the ﬁrst four principle components, explaining changes in OMXS30 and

the yield curve up to three years.

second PC corresponds to steepness since its factor loadings change sign moving to longer tenors.

The third PC corresponds to the index since the factor loading is close to one for OMXS30 and

small for the rest of the factors. The fourth PC corresponds to curvature since products in the

middle tenors have positive factor loadings when the others have negative.

The factor loadings have been estimated during a period of approximately ﬁfteen years. It is

important that the factor loadings are stable through time if the distribution is used for sampling

(since the estimation is used as a proxy for the future). To investigate if the model is adequate

the factor loadings and explanatory power of overlapping PCAs must be studied through time

and investigated for stability.

To conduct the study the data between January 1996 and June 2010 is divided into over-

lapping three years intervals with a distance of six months. Thus this gives approximately 150

observations, which is enough to retrieve a good estimation of the factor loadings.

The results are disclosed in Table 5.2, the ﬁrst PC explains between 59% and 74% of the

variation and the other PCs exhibit similar patterns. On the other hand the total explanatory

power of the four PCs is always above 93.48%. Thus the four PCs explain most of the variation

in the data sample regardless of which sample period selected. Notable is that the explanatory

power for the whole period was 97.94% and that the explanatory power in the model is higher

during the latter part of the time period. The PCA is computed on weekly returns in this section,

performing the PCA on daily returns generates less stable results, with lower explanatory power.

The PCs’ factor loadings for the overlapping periods are disclosed in Figure 5.4. PC 1 has the

smallest variation, while the other three have higher levels of variation; the factor loadings have

the same structure for all samples. Concluded: the factor loadings are relative stable through

time and stable enough to serve for simulation, even though the factor loadings exhibit some

sample dependence.

5.4.3 Historical simulation

Historical simulation is a non-parametric model that draws samples from the empirical distri-

bution with replacement. Thus to simulate new PCs; samples of factor scores should be drawn

from the empirical distribution. This means sampling random dates (with equal probabilities)

that are used to extract the corresponding factor scores, and replace the sample before the next

draw. The advantage of using historical simulation is that the scenarios have happened in the

M. Hveem 56 (103)

5.4. A PCA model for the yield curve and index

Start End PC 1 PC 2 PC3 PC 4 Cumulative [%]

Jan 1996 Dec 1998 73.51 11.18 9.74 3.09 97.52

Jul 1996 Jun 1999 74.35 11.10 9.54 2.88 97.87

Jan 1997 Dec 1999 67.34 11.71 10.47 3.99 93.51

Jul 1997 Jun 2000 65.98 12.96 10.64 3.90 93.48

Jan 1998 Dec 2000 65.80 12.96 10.74 4.08 93.59

Jul 1998 Jun 2001 66.19 12.84 10.91 4.01 93.95

Jan 1999 Dec 2001 64.59 14.88 11.30 3.69 94.46

Jul 1999 Jun 2002 62.84 16.05 11.21 3.85 93.94

Jan 2000 Dec 2002 67.95 17.38 9.75 2.91 98.00

Jul 2000 Jun 2003 66.26 20.42 8.81 2.89 98.38

Jan 2001 Dec 2003 65.79 21.01 8.64 2.99 98.43

Jul 2001 Jun 2004 61.88 23.59 9.15 3.68 98.30

Jan 2002 Dec 2004 58.80 24.77 9.00 4.53 97.10

Jul 2002 Jun 2005 60.27 23.34 8.85 4.66 97.12

Jan 2003 Dec 2005 61.38 21.96 9.17 4.59 97.09

Jul 2003 Jun 2006 61.77 20.58 9.93 4.62 96.90

Jan 2004 Dec 2006 62.62 19.84 10.16 4.44 97.07

Jul 2004 Jun 2007 64.24 17.53 10.55 4.42 96.73

Jan 2005 Dec 2007 63.68 18.51 10.66 4.52 97.37

Jul 2005 Jun 2008 59.18 20.21 11.24 5.48 96.11

Jan 2006 Dec 2008 69.61 20.96 7.58 1.20 99.35

Jul 2006 Jun 2009 72.57 16.85 7.28 2.52 99.22

Jan 2007 Dec 2009 71.42 17.58 7.09 2.70 98.79

Jul 2007 Jun 2010 69.89 18.76 7.17 2.92 98.74

Jan 1996 October 2010 68.00 16.81 9.62 3.50 97.94

Table 5.2: Variation explained by each principal component and total explained variation (cumulative

explained variation) for 24 overlapping time periods between January 1996 and June 2010.

OMXS30 1M 2M 3M 6M 9M 1Y 2Y 3Y

−1

−0.5

0

0.5

1

Risk Factor

Principle Component 1

F

a

c

t

o

r

L

o

a

d

i

n

g

s

OMXS30 1M 2M 3M 6M 9M 1Y 2Y 3Y

−1

−0.5

0

0.5

1

Risk Factor

Principle Component 2

F

a

c

t

o

r

L

o

a

d

i

n

g

s

OMXS30 1M 2M 3M 6M 9M 1Y 2Y 3Y

−1

−0.5

0

0.5

1

Risk Factor

Principle Component 3

F

a

c

t

o

r

L

o

a

d

i

n

g

s

OMXS30 1M 2M 3M 6M 9M 1Y 2Y 3Y

−1

−0.5

0

0.5

1

Risk Factor

Principle Component 4

F

a

c

t

o

r

L

o

a

d

i

n

g

s

Figure 5.4: Factor loadings for the ﬁrst four PCs for 24 overlapping time periods between January

1996 and June 2010, each line corresponds to one sample.

M. Hveem 57 (103)

Chapter 5. Modeling Financial Assets

past, they are realistic and reﬂect the empirical distribution. Also it is really simple to implement

and use historical simulation. On the other hand the problem is that there is no possibility of

implementing new scenarios, thus the model is restricted to the outcomes of the past. This means

that if no extreme events have happened in the past the probability of generating an extreme

scenario is zero. But the model is not restricted in this setup since the return over a three-year

period is generated from weekly returns. The model would be limited when studying only weekly

returns, not yearly or monthly returns, thus simulating three-years returns using weekly data

does not put any serious restrictions on the simulation.

The ith sample of the jth variable’s return using the ﬁrst four PCs in a historical simulation

is given as,

r

i,j

=

4

k=1

Q

T

k,j

F

i,k

+ε

i,j

,

where,

ε

i,j

∼ N

_

0, Var

_

9

k=5

Q

T

k,j

F

•,k

__

.

The sample should also be rescaled with its historical mean and standard deviation to attain the

ﬁnal sampled data.

5.4.4 Monte Carlo simulation

The idea of a Monte Carlo simulation is to ﬁt a parametric distribution to the observed data;

random samples are drawn from the parametric distribution to generate new data. Thus in the

case of the PCA a parametric distribution should be ﬁtted to the observed factor scores.

By using QQ-plots, or other graphical illustrations such as histogram-ﬁts and scatter plots,

it is possible to illustrate and identify appropriate parametric distributions. It is easy to see

that the PCs can be described by t-distributions with diﬀerent levels of freedom, location and

scale, as shown in Figure 5.5. Scatter plots indicate that the four PCs come from a spherical

distribution and that there exists dependence between the samples (scatter plots have elliptical

shapes) as seen in Figure 5.6.

t-location scale

PC ˆ µ ˆ σ ˆ ν

1 0.00 1.0296 2.4189

2 0.00 0.8880 4.1763

3 0.00 0.8189 8.8734

4 0.00 0.2976 2.7824

Table 5.3: The ﬁt for a student’s t-distribution to each marginal distribution (factor scores) for the

ﬁrst four PCs attained through a maximum likelihood estimation.

Multivariate t-distribution

If ν (degrees of freedom) for the marginal distributions (in this case the individual PCs) are close

to each other it is possible to use a multivariate t-distribution, instead of using e.g. copulas. A

relative good ﬁt is actually attained for this sample with a multivariate t-distribution, but it is

not as good as the t-copula model as disclosed further down. The following parameters for the

M. Hveem 58 (103)

5.4. A PCA model for the yield curve and index

−30 −25 −20 −15 −10 −5 0 5

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

Data

D

ensity

Principal Component 1

Factor Scores PC 1

Student’s t−distribution

(a) PC 1

−10 −8 −6 −4 −2 0 2 4

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

Data

D

ensity

Principal Component 2

Factor Scores PC 2

Student’s t−distribution

(b) PC 2

−4 −3 −2 −1 0 1 2 3

0

0.1

0.2

0.3

0.4

0.5

0.6

Data

D

ensity

Principal Component 3

Factor Scores PC 3

Student’s t−distribution

(c) PC 3

−3 −2 −1 0 1 2 3 4 5 6

0

0.2

0.4

0.6

0.8

1

1.2

Data

D

ensity

Principal Component 4

Factor Scores PC 4

Student’s t−distribution

(d) PC 4

Figure 5.5: Histograms for the distribution of factor scores (PC 1-4), the red line illustrates the pdf of

a t-distribution, the t-distribution seems to be an adequate parametric distribution for the

PCs’ marginal distributions.

−35 −30 −25 −20 −15 −10 −5 0 5 10 15 20

−15

−10

−5

0

5

10

15

PC 1

P

C

2

−35 −30 −25 −20 −15 −10 −5 0 5 10 15 20

−15

−10

−5

0

5

10

15

PC 1

P

C

3

−35 −30 −25 −20 −15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 1

P

C

4

−15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 2

P

C

3

−15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 2

P

C

4

−10 −5 0 5 10

−6

−4

−2

0

2

4

6

PC 3

P

C

4

Figure 5.6: Scatter plots of pairwise PCs and corresponding factor scores for the empirical (observed)

samples.

M. Hveem 59 (103)

Chapter 5. Modeling Financial Assets

multivariate t-distribution are attained using a maximum likelihood estimation,

ˆ µ =

_

_

_

_

0

0

0

0

_

_

_

_

,

ˆ

Σ =

_

_

_

_

2.337 0 0 0

0 0.5770 0 0

0 0 0.3301 0

0 0 0 0.1201

_

_

_

_

, ˆ ν = 3.2326.

It is common to assume that the residual term is multivariate normal distributed. This is not a

necessary condition and the residual term ε can be rewritten as a matrix (factor loadings) times

a stochastic vector (vector of the remaining PCs). For the observed data set the stochastic vector

is approximately multivariate t-distributed. Thus,

r

i,j

=

4

k=1

Q

T

k,j

F

i,k

+

9

k=5

Q

T

k,j

F

ε

i,k

, (5.1)

where,

F ∼ t

ν

(0, Σ) , F

ε

∼ t

ν

ε

(0, Σ

ε

) ,

and,

ˆ

Σ

ε

=

_

_

_

_

_

_

0.0195 0 0 0 0

0 0.0154 0 0 0

0 0 0.0099 0 0

0 0 0 0.0082 0

0 0 0 0 0.0030

_

_

_

_

_

_

, ˆ ν

ε

= 2.8622.

Hence to perform a Monte Carlo simulation, using a multivariate t-distribution (in this setting),

draw samples of F and F

ε

and plug in the sampled values into Equation 5.1 above. The multi-

variate t-distribution is a normal variance mixture and hence easy to sample from e.g. MATLAB

has an inbuilt function to perform the operation.

Copula

The marginal distributions for the factor scores are approximately t-distributed as seen above.

Thus by using a Copula it is possible to maintain the marginal distributions intact and use the

dependence structure from another distribution. Thus an adequate choice is to use t-marginals

and a t-copula since the dependence structure is spherical and similar to the one sampled from

a t-copula. If all the marginals had the same levels of freedom it would be the exact same model

as with a multivariate t-distribution, but as seen in Table 5.3 this is not the case. The marginals

described in Table 5.3 will be used for the ﬁrst four PCs.

The best ﬁt (using MATLAB’s copulaﬁt) is a t-copula with ˆ ν = 3.0617, which has the

following dependence structure,

ˆ ρ =

_

_

_

_

1.0000 −0.4985 −0.2170 0.1680

−0.4985 1.0000 −0.0841 0.0733

−0.2170 −0.0841 1.0000 0.1202

0.1680 0.0733 0.1202 1.0000

_

_

_

_

.

To sample from the copula with t-marginals sample from a multivariate t-distribution with

ˆ ν = 3.0617 as above, use the cumulative distribution for a univariate t

ˆ ν

variable on the sample,

the samples are now uniformed samples from the t

ˆ ν

copula. Take the inverse of each cumulative

marginal distribution on the samples to attain the new centered and rescaled samples, multiply

these samples with each dispersion coeﬃcient and add the mean to attain the ﬁnal factor score

sample.

M. Hveem 60 (103)

5.4. A PCA model for the yield curve and index

−35 −30 −25 −20 −15 −10 −5 0 5 10 15 20

−15

−10

−5

0

5

10

15

PC 1

P

C

2

−35 −30 −25 −20 −15 −10 −5 0 5 10 15 20

−15

−10

−5

0

5

10

15

PC 1

P

C

3

−35 −30 −25 −20 −15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 1

P

C

4

−15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 2

P

C

3

−15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 2

P

C

4

−10 −5 0 5 10

−6

−4

−2

0

2

4

6

PC 3

P

C

4

Figure 5.7: Scatter plots of pairwise PCs and corresponding factor scores sampled from the multivariate

t-distribution.

The error term can be modeled either through a multivariate t-distribution or any other

parametric distribution such as a copula model. It does not matter a lot which distribution is

chosen since the error explains the remaining noise in the data (which is small, less than 2.5%

of the total variance). By ﬁtting a t-copula with t-marginals the following result is retrieved,

t-location scale

PC ˆ µ ˆ σ ˆ ν

5 0.00 0.1405 2.8771

6 0.00 0.1080 2.5897

7 0.00 0.0808 2.4963

8 0.00 0.0832 2.6872

9 0.00 0.0692 3.8576

Table 5.4: Parameters for the residual factor scores attained through a maximum likelihood estimation

of t-distributions.

ˆ ρ

ε

=

_

_

_

_

_

_

1.0000 0.0658 0.1179 0.1837 0.1133

0.0658 1.0000 0.1380 0.1260 −0.0183

0.1179 0.1380 1.0000 0.2901 0.1661

0.1837 0.1260 0.2901 1.0000 0.1338

0.1133 −0.0183 0.1661 0.1338 1.0000

_

_

_

_

_

_

, ˆ ν

ε

= 3.4124.

Figure 5.9 shows that the copula provides a much better ﬁt for the marginals than the multi-

variate t-distribution. The copula provides also a better dependence structure, when comparing

Figure 5.7 and 5.8 with Figure 5.6. Thus it is easy to conclude from the plots that the copula

provides a better ﬁt, notable is that the multivariate t-distribution does not provide a bad ﬁt at

all, just not as good as the copula model.

M. Hveem 61 (103)

Chapter 5. Modeling Financial Assets

−35 −30 −25 −20 −15 −10 −5 0 5 10 15 20

−15

−10

−5

0

5

10

15

PC 1

P

C

2

−35 −30 −25 −20 −15 −10 −5 0 5 10 15 20

−15

−10

−5

0

5

10

15

PC 1

P

C

3

−35 −30 −25 −20 −15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 1

P

C

4

−15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 2

P

C

3

−15 −10 −5 0 5 10 15

−10

−8

−6

−4

−2

0

2

4

6

8

10

PC 2

P

C

4

−10 −5 0 5 10

−6

−4

−2

0

2

4

6

PC 3

P

C

4

Figure 5.8: Scatter plots of pairwise PCs and corresponding factor scores sampled from the t-copula

with t-marginals.

Hence an appropriate choice is to use either the copula model with Monte Carlo simulation or

historical simulation, the result does not diﬀer a lot in practice. This is particularly the case since

weekly return data is used to simulate three years trajectories. The two models would diﬀer much

more if the return for just one week should be simulated. When sampling one week’s data only

one draw is required, thus limiting the outcomes to the empirical distribution. When sampling

three-years returns from weekly returns it is still possible to generate almost an inﬁnite number

of combinations using the empirical distribution, 156 return samples with replacement out of 781

observed weeks provides 781

156

diﬀerent combinations, which converges towards inﬁnity.

The advantage of using historical simulation instead of the parametric simulation is that it

is much faster to use, reducing the sample time in many cases from several hours to minutes.

Thus a historical simulation is used further in this thesis since a sensitivity analysis shows that

the results are in practice identical using both.

M. Hveem 62 (103)

5.4. A PCA model for the yield curve and index

−35 −30 −25 −20 −15 −10 −5 0 5 10

−50

−40

−30

−20

−10

0

10

20

Empirical Quantiles

M

u

l

t

i

v

a

r

i

a

t

e

t

−

d

i

s

t

r

i

b

u

t

i

o

n

M

a

r

g

i

n

Q

u

a

n

t

i

l

e

s

Principal Component 1

−35 −30 −25 −20 −15 −10 −5 0 5 10

−35

−30

−25

−20

−15

−10

−5

0

5

10

15

Empirical Quantiles

C

o

p

u

l

a

M

a

r

g

i

n

Q

u

a

n

t

i

l

e

s

Principal Component 1

−10 −5 0 5

−10

−8

−6

−4

−2

0

2

4

6

8

Empirical Quantiles

M

u

l

t

i

v

a

r

i

a

t

e

t

−

d

i

s

t

r

i

b

u

t

i

o

n

M

a

r

g

i

n

Q

u

a

n

t

i

l

e

s

Principal Component 2

−10 −5 0 5

−12

−10

−8

−6

−4

−2

0

2

4

6

Empirical Quantiles

C

o

p

u

l

a

M

a

r

g

i

n

Q

u

a

n

t

i

l

e

s

Principal Component 2

−5 −4 −3 −2 −1 0 1 2 3 4

−5

−4

−3

−2

−1

0

1

2

3

4

5

Empirical Quantiles

M

u

l

t

i

v

a

r

i

a

t

e

t

−

d

i

s

t

r

i

b

u

t

i

o

n

M

a

r

g

i

n

Q

u

a

n

t

i

l

e

s

Principal Component 3

−5 −4 −3 −2 −1 0 1 2 3 4

−5

−4

−3

−2

−1

0

1

2

3

4

Empirical Quantiles

C

o

p

u

l

a

M

a

r

g

i

n

Q

u

a

n

t

i

l

e

s

Principal Component 3

−4 −2 0 2 4 6 8

−4

−2

0

2

4

6

8

Empirical Quantiles

M

u

l

t

i

v

a

r

i

a

t

e

t

−

d

i

s

t

r

i

b

u

t

i

o

n

M

a

r

g

i

n

Q

u

a

n

t

i

l

e

s

Principal Component 4

−4 −2 0 2 4 6 8

−4

−2

0

2

4

6

8

Empirical Quantiles

C

o

p

u

l

a

M

a

r

g

i

n

Q

u

a

n

t

i

l

e

s

Principal Component 4

Figure 5.9: QQ-plots of the empirical marginal distributions against the multivariate t-distribution’s

marginals (to the left) and the copula’s marginals (to the right).

M. Hveem 63 (103)

CHAPTER 6

Portfolio Optimization

The fund should yield a high return as well as limit the risk under CVaR. CVaR is used as risk

measure since the distribution of the structured products is skewed and exhibits kurtosis, thus

risk measures such as standard deviation are inappropriate (as described earlier). In this chapter,

the fund is allowed to rebalance the portfolio every quarter, the investment horizon is three years,

the fund is allowed to allocate a maximum of 30% in each asset, CVaR will be considered for

α = 0.95. All structured products in this chapter are 100% capital guaranteed and follows the

dynamics given in Chapter 3, thus the structured products have a time to maturity of three years

at issuance, the bond’s notional amount equals the price of the underlying, the strike as well as

the price of the product itself at issuance.

Only proportional transaction costs are considered since ﬁxed transaction costs transforms

the linear CVaR optimization problem into a mixed integer linear problem, which increases the

complexity of the problem a lot.

The chapter covers three diﬀerent types of portfolio choices: ﬁxed portfolio weights, rolling

portfolio weights and dynamic portfolio weights. The diﬀerent alternatives are evaluated in

relation to a benchmark fund and the most beneﬁcial fund construction is investigated. The

results indicate that the most beneﬁcial setup is the dynamic portfolio weights scheme, where a

modiﬁed Korn and Zeytun framework is proposed. The study indicates that the most imperative

factor that aﬀects the return is in fact not the portfolio selection itself but instead the transaction

costs.

The ﬁrst portfolio allocation strategy/scheme that is described is the ﬁxed portfolio weight

strategy.

6.1 Fixed portfolio weights

Fixed portfolio weights is the setup where product one always is the product closest to maturity

and product twelve the product with the longest time to maturity (since it exists only twelve

products), thus it is similar to the portfolios used in Section 4.3, where there is always twelve

capital guaranteed structured products available. The portfolio will be allocated using the CVaR

algorithm described in Chapter 2 with the modiﬁcation that no binary variables will be used (no

ﬁxed transaction costs), since they turn the problem into a mixed integer linear problem, which

is not desirable. The return is used as target function since this problem is better deﬁned than

having the return as a constraint (needs less iterations via the simplex method). Since the ﬁxed

weights concept is used, which is a relative theoretical concept, it is interesting to investigate

which products are more optimal than others to allocate in, when considering tenors. Thus the

study disregards transaction costs initially to investigate the optimal choice in a perfect market

setting. The result should be seen as a guideline for optimal portfolio choice, i.e. which types of

products that are better to invest in than others, regarding the tenor.

The results presented in Section 4.3 are in line with the results found in this section. The

products closest to maturity are associated with the highest risk, i.e. CVaR, where it was previ-

ously found that the products closest to maturity have the worst outcomes of equivalent option

65

Chapter 6. Portfolio Optimization

portfolios. Investing in the products closest to maturity imposes an “all or nothing situation”

for the options, while retrieving a more smooth return distribution for newly issued products as

proven earlier (even though the interest rate was ignored in the previous chapters). Since similar

results are retrieved when including the eﬀect of interest rate changes (as seen in Table 6.1) the

assumption that the risk of movements in the underlying is the most prevalent risk factor can

be conﬁrmed. Notable is that the expected return is higher for products with higher risk, thus it

seems to be adequate to use a CVaR optimization scheme to attain a smooth eﬃcient frontier.

The analysis is performed via 100,000 six-year trajectory samples based on the model described

in Chapter 5. The return over the last three years of each sample trajectory is considered, since

the ﬁrst three years are needed to generate relevant price data as the price process is path depen-

dent. The last three years observed price data on the market is not used as start data since the

analysis should be conducted independent of the current market. Instead diﬀerent start levels

for the yield curve will be considered, i.e. the yield curve that is in the beginning of the six-year

period. In this section, the yield curve as of 15th November 2010, representing a low yield case

will be used. In the appendix the case with an arbitrary yield curve is disclosed. The results are

in general the same, mostly imposing just higher expected return and risk.

It is important to understand that the result may not look exactly the same for all markets.

It is also important to notice that a more general concept is investigated in this section since the

previous three-year trajectories are simulated. Thus the eﬃcient frontier would look diﬀerent

if an investor would want to invest in structured products today since the current yield curve,

index level, strikes and participation rates are known. Hence the results in this chapter are

mainly general results, i.e. in the long run the eﬃcient frontiers etc. would look like this, not

necessary all the time.

Thus the hypothesis that was presented in Section 4.3 was correct, the highest risk is associ-

ated with products that are closest to maturity and the newly issued products have a lower risk

as disclosed in Table 6.1. Thus it is reasonable to assume that the investment strategy proposed

earlier in Chapter 4 is still adequate.

2 4 6 8 10 12 14 16

16.5

17

17.5

18

18.5

19

19.5

20

Efficient Frontier Fixed Portfolio Weights

Portfolio CVaR

α = 0.95

[%]

E

x

p

e

c

t

e

d

P

o

r

t

f

o

l

i

o

R

e

t

u

r

n

[

%

]

Figure 6.1: Eﬃcient frontier for the optimization choice using the ﬁxed portfolio weights setting, α =

0.95.

The CVaR algorithm described in Chapter 2 is applied to produce a robust eﬃcient frontier;

the result is disclosed in Figure 6.1. Transaction costs have been disregarded and the restriction

that it is only allowed to invest a maximum of 30% in each asset is in place. Since the optimization

is computer intensive, it is not possible to use a lot more than approximately 1,000 scenarios

for the eﬃcient frontier (in a timely manner on a regular PC). Thus to attain robust results

100 eﬃcient frontiers are simulated for diﬀerent risk levels (CVaR) as constraint and the average

of these weights are calculated to attain robust portfolio weights. Thus yielding a more robust

eﬃcient frontier, which reﬂects the intrinsic frontier in a better manner.

Figure 6.1 indicates that the expected return increases as the risk increases, and the frontier

M. Hveem 66 (103)

6.1. Fixed portfolio weights

is quite similar the frontier for a classic Markowitz optimization problem with standard deviation

as risk measure.

Product Number, CVaR

0.95

[%]

1 2 3 4 5 6 7 8 9 10 11 12

18.46 14.88 12.41 10.52 9.33 8.04 7.08 6.40 4.68 4.23 3.92 3.74

Product Number, Expected return [%]

1 2 3 4 5 6 7 8 9 10 11 12

20.14 19.71 19.63 19.07 18.12 17.71 17.30 16.93 16.48 16.14 15.81 15.51

Table 6.1: The table describes the CVaR at the 95% conﬁdence level and expected return for the

diﬀerent products over the three-year period using the ﬁxed portfolio weights setting.

Next the sensitivity of each scenario’s portfolio weights is analyzed. It is important to analyze

the standard deviation of the portfolio weights, since it discloses how robust the weights are, and

if it is necessary to use multiple samples to attain a representative solution to the optimization

problem. 100 samples of 1,000 scenarios for three diﬀerent levels of CVaR are used, one low risk,

one medium risk and one high-risk level. The portfolio weights are approximately equally robust

for the low risk scenarios and for high-risk scenarios, with slightly more robust portfolio weights

for the high-risk level, as Table 6.2 indicates.

Weights, standard deviation [%]

CVaR

0.95

1 2 3 4 5 6 7 8 9 10 11 12

4% 0.80 0.95 8.69 9.7 0.04 5.25 8.82 10.67 9.07 11.03 11.94 10.06

8% 6.42 7.37 5.50 6.91 3.73 8.01 9.00 11.05 8.01 7.69 6.37 4.95

12% 9.75 9.94 1.80 5.73 2.62 3.72 2.47 3.92 1.57 1.11 0.39 0.50

Table 6.2: The table describes the robustness (standard deviation) of the weights for a portfolio al-

location using 1,000 scenarios and for three diﬀerent CVaR levels using the ﬁxed portfolio

weights setting.

The sensitivity of the portfolio weights is higher with less sample scenarios. Also it is more

adequate to use CVaR as a constraint and maximizing the return as the target, than the other

way around, since it does not only require less iterations but also generates more stable portfolio

weights. Table 6.2 shows that there exists signiﬁcant standard deviation for the portfolio weights.

So what does this imply? Since each sample has diﬀerent paths up till the start of investment a

high standard deviation implies that it is important to not use a generalized allocation scheme,

since the weights shift much depending on the scenarios.

From the distribution and CVaR results for the individual products it is reasonable to assume

that lower levels of CVaR allocate most of the capital to the products with the longest time to

maturity (which is the case). But as the maximum CVaR allowed increases the portfolio will

allocate in shorter products to yield a higher expected return (since the expected return increases

almost linear over the twelve assets). Thus the frontier portfolios (weights), disclosed in Figure

6.2, are attained.

Hence depending on the desired maximum risk level the portfolio weights should be chosen

according to Figure 6.2 and the portfolio would receive the expected return according to Figure

6.1.

Thus following the path of capital guarantee; a fund should allocate mainly in the last four

- ﬁve products to minimize the CVaR for the portfolio. It should also, depending on its risk

tolerance (and transaction costs), sell of assets after a number of quarters elapsed. The risk

level and expected return will be very dependent on the transaction costs, which is important

to notice. Consider transaction costs of 2% (2% in proportional transaction costs each time

the fund sells or buys an asset) and a portfolio which holds equal weights in the longest tenors

w

9

= w

10

= w

11

= w

12

. The portfolio has without transaction costs a CVaR

0.95

of 3.62% and

M. Hveem 67 (103)

Chapter 6. Portfolio Optimization

4.2 5.2 6.2 7.2 8.2 9.2 10.2 11.2 12.2 13.2 14.2

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Portfolio CVaR

α = 0.95

[%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

Figure 6.2: Portfolio weights for the eﬃcient frontier using the ﬁxed portfolio weights setting, α = 0.95.

an expected return of 15.98% over three years. Including the transaction costs it will have a

CVaR

0.95

of 9.29% and an expected return of 9.16%. On the other hand a portfolio holding

equal amounts invested in products twelve throughout seven has a CVaR

0.95

of 7.52% and an

expected return of 11.75% including the transaction cost. Thus the problem must be modiﬁed

to ﬁnd a solution including transaction costs.

To solve how to allocate in an optimal way, including transaction costs, an alternative CVaR

algorithm with proportional transaction costs is introduced, which can approximate the induced

transaction costs of a portfolio with ﬁxed weights. The weight held in asset 12 one quarter is

approximately held in asset 11 the next quarter and so on. Thus the fund can, by solving the opti-

mization problem below, attain portfolio weights that are desirable when taking in consideration

transaction costs.

max

w,w

+

,w

−

,z,β

1

S

S

k=1

_

1 +R

w

T,k

_

,

such that:

β +

1

S (1 −α)

S

k=1

z

k

≤ C,

1 +R

w

T,k

=

_

1 +R

1

T,k

_

w

1

+... +

_

1 +R

N

T,k

_

w

N

, k = 1, ..., S

R

w

T,k

+β +z

k

≥ 0, k = 1, ..., S

z

k

≥ 0, k = 1, ..., S

N

i=1

w

i

+

N−1

i=1

_

TC

+

i

w

+

i

+TC

−

i

w

−

i

_

+TC

+

N

w

N

= 1,

w

i

= w

i+1

+w

+

i

−w

−

i

, i = 1, ..., N −1

w

N

= w

1

+w

+

N

−w

−

N

w

i

≥ 0, w

±

i

≥ 0, i = 1, ..., N

The optimal portfolio weights are investigated for the cases of no transaction costs and three

diﬀerent levels of transaction costs, when minimizing the CVaR, given a positive return. The

optimal weights are investigated in a setting of a low initial yield curve and the setting of an

arbitrary, random start yield curve, the results are disclosed in Figure 6.3. It is in general most

M. Hveem 68 (103)

6.2. Rolling portfolio weights

optimal to allocate in the ﬁrst four products, and as the transaction costs rises it is less optimal

to rebalance the portfolio often, thus leading to that the fund allocates to more products.

0 1 2 3

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Proportional Transaction Costs [%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

(a) Low yield

0 1 2 3

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Proportional Transaction Costs [%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

(b) Generalized yield

Figure 6.3: Optimal portfolio weights for the ﬁxed portfolio weight scheme for diﬀerent proportional

transaction costs, minimizing CVaR

0.95

given the requirement of a positive return, accord-

ing to the algorithm described above.

6.2 Rolling portfolio weights

Rolling portfolio weights is the setup where product one is a product that has one quarter left to

maturity and is rolled over at maturity to a new equivalent capital guaranteed structured product

with three years to maturity. Product two has two quarters to maturity before it is rolled over to

a new capital guaranteed structured product, and so on. Thus this rolling product is a perpetual

product (similar to the bond roll) where transaction costs occur when the investor rolls over to

the new product (thus each product will over a three-year period have equal transaction costs

in percentage). A CVaR optimization algorithm, the same as in the beginning of the previous

section, is performed for the ﬁxed portfolio weights. The transaction costs will not impact the

portfolio choice itself since all products are associated with the same degree of transaction costs

(when the roll is performed), thus transaction costs are not considered in this section.

The CVaR algorithm with CVaR as constraint and maximizing the expected return as the

target function is performed 100 times (for robustness) to produce a robust eﬃcient frontier; the

result is disclosed in Figure 6.4.

Figure 6.4 shows that the expected return increases as the risk increases, but only in a very

small extent. The expected return is almost similar for all risk levels. It is preferable to invest

in product twelve since its time to maturity for the ﬁrst roll equals the investment horizon, thus

providing intrinsic capital guarantee. The risk associated with the products is less for products

that are close to roll over than products in the middle spectrum. The reason why the return of

the eﬃcient frontier is almost equal for all risk levels is that almost all products have the same

expected return, since they are over a three-year horizon almost equivalent products in return

sense.

Table 6.3 indicates that the products that are closer to their ﬁrst roll over are more risky

than products that have a long time to their ﬁrst roll. The diﬀerent risk levels for the products

are derived from the participation rate. A higher participation rate (due to often a higher yield)

imposes a higher risk. The probability of attaining a higher participation rate is higher for

products that roll over later. On the other hand the risk associated with options is higher when

the time to maturity is smaller. Thus the products that have the longest time until their roll will

M. Hveem 69 (103)

Chapter 6. Portfolio Optimization

3 4 5 6 7 8 9

17.8

17.81

17.82

17.83

17.84

17.85

17.86

17.87

17.88

Efficient Frontier Rolling Portfolio Weights

Portfolio CVaR

α = 0.95

[%]

E

x

p

e

c

t

e

d

P

o

r

t

f

o

l

i

o

R

e

t

u

r

n

[

%

]

Figure 6.4: Eﬃcient frontier for the optimization choice using the rolling portfolio weights setting,

α = 0.95.

risk receiving higher participation rates in the end, but they will only go through a few quarters

(the ones with the least risk) until the investment horizon ends with this high participation rate.

Also these products will beneﬁt of the previous roll before the high participation rate that with

a high probability had a low participation rate. Thus the average participation rate over the

investment horizon is higher for products with 1-9 quarters to the next roll, imposing a higher

risk. The risk levels and expected returns for the twelve diﬀerent rolls are given in Table 6.3.

Product (months until next roll), CVaR

0.95

[%]

1 2 3 4 5 6 7 8 9 10 11 12

9.28 10.57 11.43 10.63 10.68 10.31 9.73 9.12 8.15 6.86 5.06 0.00

Product (months until next roll), Expected return [%]

1 2 3 4 5 6 7 8 9 10 11 12

18.04 18.01 17.98 17.57 17.57 17.54 17.58 17.60 17.64 17.69 17.72 17.81

Table 6.3: The table describes the CVaR at the 95% conﬁdence level and expected return for the

diﬀerent products over the three-year period for the diﬀerent rolls.

The robustness of the weights is also studied through investigating the standard deviations

of the weights as previously, the result is presented in Table 6.4.

Weights, standard deviation [%]

CVaR

0.95

1 2 3 4 5 6 7 8 9 10 11 12

4% 9.35 11.65 7.86 5.89 4.32 4.83 2.98 2.78 6.59 8.34 10.44 4.72

6% 12.76 12.19 10.43 7.94 7.89 7.22 5.86 5.05 12.66 13.40 11.89 12.30

8% 12.25 11.83 9.89 7.95 8.71 7.77 8.69 5.62 13.87 13.73 12.03 12.34

Table 6.4: The table describes the robustness (standard deviation) of the weights for a portfolio allo-

cation using 1,000 scenarios and for three diﬀerent CVaR levels using the rolling portfolio

weights setting.

The standard deviation result is similar to the result in the previous section, which indicates

that it is important as a portfolio manager to not use a generalized scheme for the allocation of

the portfolio since the portfolio weights depends a lot on the previous path.

Investigating the portfolio weights for the robust eﬃcient frontier shows that the portfolio

should, to attain lower levels of CVaR, allocate to the products that have a lower individual

CVaR, which is reasonable as disclosed in Figure 6.5.

The expected return will in the long run be very similar for all rolls, while the risk proﬁle of

M. Hveem 70 (103)

6.3. Benchmark fund

4.6 5.6 6.6 7.6 8.6

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Portfolio CVaR

α = 0.95

[%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

Figure 6.5: Portfolio weights for the eﬃcient frontier using the rolling portfolio weights setting, α =

0.95.

the products diﬀers. Thus it is better to invest in products with a long time to the next roll. The

high standard deviation of the portfolio weights implies that it is very important to evaluate the

portfolio depending on the observed paths (yield curve, participation rate, index level etc) since

they will impact both the expected return and the risk proﬁle.

Thus when investing with the rolling weight scheme, it is better to allocate to the newly issued

products (as stated in the previous chapters and sections). It is better to invest with rolling

portfolio weights than the ﬁxed weight scheme if the expected return it yields is satisfactory.

Thus by requiring a maximum of 3.80% in CVaR

0.95

the portfolio will actually have an expected

return of 17.82% over three years (excluding transaction costs, which would increase the CVaR

and decrease the expected return), compared to the ﬁxed weights which only yields an expected

return of 16.88% for the same risk level, which also induces much higher transaction costs.

As described earlier in this paper, a portfolio’s risk and return is only relevant in a certain

context, thus it is imperative that the diﬀerent portfolio constructions are compared with a

benchmark fund.

6.3 Benchmark fund

The analysis would be less relevant without a comparison with a relevant benchmark. Observed

market data for the Swedish market has been used throughout the analysis in this chapter,

thus the SPF must be benchmarked against alternatives at the Swedish market. Consider the

benchmark fund described earlier, which places 30% of its capital in the index, plus minus 10%,

and the rest of the capital in a basket of bonds with time to maturity between three months to

three years (approximately equally weighted, rolling). This yields an expected return, over the

three years investment horizon, of 19.81% including the expectation of 3% annual dividends in

the index and a CVaR

0.95

of 9.49%. The distribution of the fund returns is given in Figure 6.6.

When disregarding the expectation of the dividends the portfolio instead receives an expected

return of 16.12% and a CVaR

0.95

of 10.75%.

Thus the distributions of the two, with or without dividend, are of course almost identical,

the only thing that diﬀers is approximately a parallel shift. The results indicate that it is hard

for the SPF to compete with the benchmark fund including dividends (in terms of return) since

the benchmark has a much higher expected return given the same risk level. Thus the SPF

must compete with the benchmark portfolio given low risk. In Table 6.5 results for diﬀerent

benchmark funds are disclosed. Notable is that transaction costs have been disregarded, which

would decrease the expected return of these products.

M. Hveem 71 (103)

Chapter 6. Portfolio Optimization

−0.5 0 0.5 1 1.5 2 2.5

0

100

200

300

400

500

600

700

Benchmark Fund

CVaR

0.95

= 9.49%

(a) Incl. dividend

−0.5 0 0.5 1 1.5 2 2.5

0

50

100

150

200

250

300

350

400

450

Benchmark Fund

CVaR

0.95

= 10.75%

(b) Excl. dividend

Figure 6.6: Histograms over the sampled three-year returns for the benchmark fund, ε

bench

= 0.3.

Incl. Dividends. Excl. Dividends.

Index allocation Expected Return CVaR

0.95

Expected Return CVaR

0.95

10% 11.40 -1.66 10.16 -1.20

20% 15.43 3.25 13.05 4.38

30% 19.50 8.72 16.12 10.75

40% 23.88 14.35 19.16 17.13

Table 6.5: The table discloses the expected return and CVaR for diﬀerent levels of index weights over

the three-year period for a benchmark fund.

Table 6.1, 6.3 and 6.5 indicate that the SPF can compete initially (without transaction costs)

in the low risk spectrum, where the rolling weights portfolio generates higher expected returns

relative to the risk, which is also the case for the ﬁxed portfolio weights. Thus these types of

funds would be able to compete with the benchmark fund in the low risk spectrum. This is also

in line with the idea of the SPF, which is that it should be a low risk fund, and the results show

that it is only possible for the SPF to compete against other low risk funds to yield a higher

expected return.

The next section will investigate the last allocation scheme, the dynamic portfolio weights

allocation scheme.

6.4 Dynamic portfolio weights

The allocation problem is quite complex for the SPF. It is assumed that product one is the

product with one quarter until maturity, product two is the product with two quarters until

maturity and so on as previously, hence there exists twelve products (the products are also

standard). The fund manager has the possibility today to invest in twelve structured products,

where only one of the products does not mature before the end of the investment horizon. Thus

when considering to allocate the portfolio over a three-year period the fund manager needs to

take in count how the payoﬀ of the products that mature before the end of the investment

horizon should be reinvested. It is necessary to consider the full three-year time horizon since

a portfolio allocated on a three-month basis would be suboptimal and induce high transaction

costs. Thus it is desirable to investigate the portfolio choice for the whole three-year time period

to avoid excessive transaction costs (since this is the time to maturity of the longest product that

exists). Korn and Zeytun (2009, [27]) propose a method to solve optimal investment problems

with structured products under CVaR constraints, where the investor’s investment horizon is

longer than the time to maturity. Korn and Zeytun proposes the following modiﬁed multiple

M. Hveem 72 (103)

6.4. Dynamic portfolio weights

period CVaR allocation algorithm,

max

w,ν,z,β

1

S

S

k=1

R

w,ν

T,k

,

such that:

R

w,ν

T,k

= w

1

R

1,ν

T,k

+... +w

N−1

R

N−1,ν

T,k

+w

N

R

N

T,k

, k = 1, ..., S

R

j,ν

T,k

=

_

1 + Π

j

k

__

ν

j

j+1

_

1 +r

j+1

T−j,k

_

+... +ν

j

N

_

1 +r

N

T−j,k

_

_

−1, j = 1, ..., N −1

R

w,ν

T,k

+β +z

k

≥ 0, k = 1, ..., S

β +

1

S (1 −α)

S

k=1

z

k

≤ C,

z

k

≥ 0, k = 1, ..., S

w

1

+...w

N

= 1,

w

i

≥ 0, i = 1, ..., N

ν

j

≥ 0, j = 1, ..., N −1

where ν

j

l

is the percentage of the product j’s payoﬀ that is allocated to asset l after maturity,

r

l

T−j,k

is the return of asset l for the remaining time period (time until the end of the investment

horizon from time j), Π

j

k

is the return of asset j from the start of the investment period until

its maturity and k corresponds to the scenario. Thus each product that matures is reinvested

using the weights ν

j

l

. Note that the last product does not mature prior to the investment horizon

thus R

N

T,k

= Π

N

k

. The optimization problem above is without transaction costs, the proportional

transaction costs constraints and budget constraints can be added in the exact same way as for

the regular CVaR optimization problem.

The problem with this setting is that it is no longer a linear optimization problem. Thus

it cannot be solved with only a simplex procedure. It must instead be solved by creating op-

timization loops, where each intermediate rebalancing point creates a new loop. Eleven outer

optimization loops, with twelve possible products to allocate within, are needed since the portfo-

lio has eleven intermediate rebalancing points. It is not feasible to allocate with this sophisticated

optimization algorithm in this setting since the time to attain a solution would be months, not

minutes or hours, and for robust weights it would take years to solve the optimization problem.

Also, the optimization problem is not always convex, thus resulting in diﬀerent solutions, given

diﬀerent start values in the optimization.

6.4.1 Modifying the Korn and Zeytun framework

The algorithm described by Korn and Zeytun is adequate for a limited number of intermediate

time periods, thus relative few assets, since the iterations and as a consequence the comput-

ing time grows exponentially. Concluded: it is not possible to use this algorithm to solve the

allocation problem in this setting.

The optimization problem is modiﬁed by setting constraints on ν

j

l

, such that the problem is

easier to solve. Let ν

j

l

be the weights that minimize the CVaR for the particular product j includ-

ing the reinvestment, thus reducing the total problem to only solving twelve linear optimization

problems.

Thus the modiﬁed algorithm is as follows,

M. Hveem 73 (103)

Chapter 6. Portfolio Optimization

i. Sample N trajectories from the model described in Chapter 5 to attain returns for

the structured products.

ii. Solve the CVaR optimization problem with minimizing CVaR as target and the con-

straint of a positive return for the period between N − 1 to N to attain ν

N−1

,

r

N−1

T−(N−2),k

and complete the full return trajectory for product N −1.

iii. Repeat ii until the full price trajectories for all products are known, thus between

N −2 to N, N −3 to N and so on (in this case eleven optimization problems).

iv. Solve the well-deﬁned CVaR optimization problem between 1 and N to attain initial

weights w (since all scenario trajectories are known).

v. Repeat steps i - iv 100 times to attain robust portfolio weights.

The results attained through this allocation scheme are much more beneﬁcial than the pre-

vious two allocation schemes. Thus it is this allocation scheme that should be used in relation

to the benchmark portfolio, since it can generate higher expected returns to the same risk level.

Without taking in count transaction costs the robust eﬃcient frontier in Figure 6.7 is attained.

3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8

17.45

17.46

17.47

17.48

17.49

17.5

17.51

17.52

17.53

Efficient Frontier Dynamic Portfolio Weights

Portfolio CVaR

α = 0.95

[%]

E

x

p

e

c

t

e

d

P

o

r

t

f

o

l

i

o

R

e

t

u

r

n

[

%

]

Figure 6.7: Eﬃcient frontier for the optimization choice using the dynamic portfolio weights and the

modiﬁed Korn and Zeytun framework, α = 0.95.

The expected return does not increase a lot as the portfolio increases its risk tolerance, most

structured products in this setting have similar expected returns. Most of the reinvested capital

will be placed in the newly issued product, since the products with the longest time to maturity

have the lowest risk. Thus the solution is in practice very similar to the one of rolling portfolio

weights. Notable is that this is for the low yield setting and the solution to the problem is not

always the same, thus the solution will in many cases be similar to the one of rolling portfolio

weights, but since the returns are path dependent, the risk proﬁle will be diﬀerent, depending on

the previous price data. Thus the big diﬀerence between these two models is that the modiﬁed

Korn and Zeytun framework is dynamic and minimizes the risk of the reinvested capital and

allocates based on that knowledge. Hence even though the solution in this case is extremely

similar to the rolling weights problem, it is not the same (it generates lower risk, but also lower

return). The modiﬁed Korn and Zeytun algorithm allocates even more in the newly issued

products, especially for the lower risk segments. The expected return and CVaR for this setup

are disclosed in Table 6.6.

The standard deviation of the weights are analyzed in the same way as in previous sections

and disclosed in Table 6.7.

The standard deviation result indicates that it is important to use the scheme actively, and

not just use a static allocation scheme, to choose the portfolio weights depending on the previous

M. Hveem 74 (103)

6.4. Dynamic portfolio weights

Product, months until maturity, CVaR

0.95

[%]

1 2 3 4 5 6 7 8 9 10 11 12

8.80 10.04 10.96 10.26 10.24 9.87 9.48 8.74 7.74 6.38 4.68 0.00

Product, months until maturity, Expected return [%]

1 2 3 4 5 6 7 8 9 10 11 12

17.64 17.62 17.66 17.27 17.24 17.24 17.20 17.25 17.31 17.36 17.39 17.45

Table 6.6: The table describes the CVaR at the 95% conﬁdence level and expected return for the

diﬀerent products over the three-year period using the dynamic portfolio weights and the

modiﬁed Korn and Zeytun framework .

Weights, standard deviation [%]

CVaR

0.95

1 2 3 4 5 6 7 8 9 10 11 12

4% 11.24 9.20 9.34 1.33 3.28 3.98 0.68 4.60 8.46 12.36 10.05 4.31

6% 12.81 13.06 12.97 2.37 6.22 7.44 4.06 6.66 9.67 12.49 13.74 12.52

8% 12.23 13.54 13.87 6.42 7.31 8.22 4.71 7.28 9.88 12.42 12.61 12.85

Table 6.7: The table describes the robustness (standard deviation) of the weights for a portfolio allo-

cation using 1,000 scenarios and for three diﬀerent CVaR levels using the dynamic portfolio

weights and the modiﬁed Korn and Zeytun framework.

paths.

When analyzing the portfolio weights for the diﬀerent risk levels it is easy to notice that the

products with a long time to maturity are very attractive, since they both have high expected

return and low risk. Also the products that are close to maturity have a higher expected return

and a lower risk than the products in the middle spectrum. Therefor the algorithm allocates

mainly in products 1-3 and 10-12, as disclosed in Figure 6.8. Notable is that the algorithm is

best for low risk strategies, since it minimizes the risk of the reinvested capital, which is the goal

of this thesis.

3.8 4.8 5.8 6.8 7.8

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Portfolio CVaR

α = 0.95

[%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

Figure 6.8: Portfolio weights for the eﬃcient frontier using the dynamic portfolio weights and the

modiﬁed Korn and Zeytun framework, α = 0.95.

6.4.2 Dependence of the initial yield curve

It is important to note that the risk level and level of expected return are signiﬁcantly dependent

on the initial yield curve. In this chapter the yield curve from middle of November 2010 on

the Swedish market is used as an initial yield curve of the six-year period. Thus the results are

M. Hveem 75 (103)

Chapter 6. Portfolio Optimization

dependent on this initial yield curve. In Appendix E the corresponding ﬁgures and tables are

disclosed where an arbitrary start yield curve is used instead to attain unbiased results. Notable

is that the results are very similar to each other, the thing that diﬀers the most is the level of

expected return and CVaR. The optimal portfolio weights are very similar to each other for both

low-risk and high-risk strategies.

The ﬁxed portfolio weights scheme gains a much higher expected return given the same risk

levels for the arbitrary yield curve case. The same patterns of weights for diﬀerent risk levels are

shown for the arbitrary yield setting, thus a low risk strategy in one setting can be expected to

also be a low risk strategy in the other setting. Notable is also that by inspecting the eﬃcient

frontier for the ﬁxed portfolio weights, it is possible to see that the low risk strategy of ﬁxed

portfolio weights cannot compete with the low risk strategy of the rolling portfolio weights, but

on the other hand there are portfolios which impose higher expected returns than for the rolling

and dynamic weights in the ﬁxed portfolio weight scheme. Thus using a strategy where the fund

always buys structured products close to maturity is the only way to trigger a higher expected

return in the SPF, which on the other hand increases the risk dramatically. Notable is also that

the ﬁxed portfolio weights strategy will in most cases impose a lot more transaction costs than

the other strategies, thus reducing its attractiveness.

Almost the exact same risk levels are attained in the rolling portfolio weights setting to a

much higher expected return in the arbitrary yield case (compared to the low yield case). This

can be derived from the fact that increasing the limit for CVaR allowed does not necessary

increase the expected return. Since most products have approximately equal expected return

the only diﬀerence is that the arbitrary yield curve case has a higher expected return and a

slightly diﬀerent risk level.

For the benchmark fund the expectations on the dividends in the future are really important,

and it is important for the portfolio manager to understand the impact of having diﬀerent expec-

tations for the dividends. The results are a little bit diﬀerent when comparing the benchmark

fund with a low yield curve against an arbitrary yield curve, than for the SPF. The expected

return increases (due to the higher yield) but the risk of the index does not increase, thus actually

reducing the risk and increasing the expected return.

Concluded: the SPF is more beneﬁcial for investors in a market state of a low yield, since

it imposes less risk for the investor in the SPF (lower participation rate) at the same time as

competing mixed funds actually have increased risk.

By comparing the alternatives of rolling and dynamic portfolio weights against investing in

a competing benchmark portfolio it is possible to notice that there are levels where it is more

beneﬁcial to invest in the SPF. This is what the thesis is looking for, if the SPF can generate a

lower risk for the same level of expected return as existing funds, it is deﬁnitely a good choice

to invest in the SPF.

6.4.3 Observed paths

There exists in general an optimal choice for the portfolio weights given that the portfolio is

resampled. On the other hand the standard deviation of the portfolio weights was quite high

which indicates that it is important to choose the portfolio weights actively. A high standard

deviation for the portfolio weights indicates that the optimal portfolio weights are depending on

the previous paths. Therefor the optimal portfolio choice with an investment horizon that starts

the 16th of November 2010 in the Swedish market will now be studied. Figure 6.9 discloses the

closing prices of OMXS30 during the three-year time period prior to the day of investment and

the robust eﬃcient frontier for the investment choice according to the modiﬁed Korn and Zeytun

algorithm.

The expected return is signiﬁcantly higher for this scenario than for the generalized case

described above. Also this scenario has a much lower minimum risk level for the CVaR.

It is interesting to investigate the standard deviation for the portfolio weights given the

previous price paths. By analyzing the standard deviation of the portfolio weights it is possible

M. Hveem 76 (103)

6.4. Dynamic portfolio weights

2008 2009 2010

600

700

800

900

1000

1100

Year

I

n

d

e

x

Nasdaq OMXS30

1 2 3 4 5 6 7 8

26.4

26.6

26.8

27

27.2

27.4

27.6

Efficient Frontier Dynamic Portfolio Weights

Portfolio CVaR

α = 0.95

[%]

E

x

p

e

c

te

d

P

o

r

tf

o

lio

R

e

tu

r

n

[

%

]

Figure 6.9: Observed price paths of OMXS30 during the three years leading up to 16th of November

2010 and the eﬃcient frontier using the dynamic portfolio weights and the modiﬁed Korn

and Zeytun framework for 16th of November 2010.

to assess if 1,000 scenarios is enough to determine the optimal portfolio weights, the result is

disclosed in Table 6.8.

Weights, standard deviation [%]

CVaR

0.95

1 2 3 4 5 6 7 8 9 10 11 12

1% 1.82 1.25 0.10 0.00 0.00 0.00 0.00 0.52 1.60 0.34 0.00 0.00

4% 0.00 8.96 5.70 0.00 0.00 0.00 0.00 0.00 0.00 3.94 0.00 0.00

7% 0.00 4.57 2.85 0.00 0.00 0.00 0.00 0.00 0.00 0.00 5.82 0.00

Table 6.8: The table describes the robustness (standard deviation) of the weights for the portfolio allo-

cation using 1,000 scenarios and for three diﬀerent CVaR levels using the dynamic portfolio

weights and the modiﬁed Korn and Zeytun framework for 16th of November 2010.

The result indicates that the speciﬁed setting (known paths) has a much lower standard devi-

ation for the portfolio weights than the generalized investment scheme. The standard deviation

is higher for the scenarios with higher risk, but the products that have standard deviation are

the products that the portfolio allocates in. Thus it is the same assets that are more or less

most optimal in each optimization, but the amount allocated in each asset diﬀers slightly. Thus

the result indicates that it is reasonable to use 1,000 simulations to attain relative stable port-

folio weights. On the other hand the quality of the result is still enhanced when resampling the

portfolio weights (as most often).

1.8 2.8 3.8 4.8 5.8 6.8 7.8

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Portfolio CVaR

α = 0.95

[%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

Figure 6.10: Portfolio weights for the eﬃcient frontier using the dynamic portfolio weights and the

modiﬁed Korn and Zeytun framework for 16th of November 2010, α = 0.95.

The portfolio weights for the eﬃcient frontier for the optimal investment choice as of the 16th

November 2010 are disclosed in Figure 6.10. The weights are quite similar as in the generalized

M. Hveem 77 (103)

Chapter 6. Portfolio Optimization

case, but they diﬀer slightly (adjusted for the particular scenario). Thus it is imperative to adjust

the portfolio optimization depending on the previous observed paths.

6.5 Transaction costs

This section focuses on the dynamic portfolio weights setting and how transaction costs impact

the risk and return levels in comparison to the benchmark fund. The SPF is compared with

a benchmark fund, which includes dividends from the index (the most realistic benchmark),

for diﬀerent participation levels as well as for diﬀerent levels of proportional transaction costs.

The transaction costs occur when an asset is bought or sold, all assets have the same level of

transaction costs, for buying as well as selling. Thus rebalancing a portfolio induces transaction

costs, both when the asset is sold and when a new asset is bought. As seen in the previous section;

the most desirable setting for the dynamic portfolio weights is to minimize the portfolio’s risk,

thus this will be conducted given the constraint of a positive return. Both the setting with a

low start yield curve (three years prior to the initial investment horizon) and the setting with

an arbitrary start yield curve are investigated. Fixed transaction costs are disregarded to avoid

transforming the problem into a mixed integer linear problem. In Table 6.9 the expected return

and CVaR

0.95

for diﬀerent benchmark funds and SPFs are disclosed.

Table 6.9 indicates that the SPF can compete against diﬀerent benchmark funds during

diﬀerent market settings. In the low yield case the SPF provides, in general, lower risk to a

higher expected return than the benchmark fund with an aim of 20% index investments (when

assuming the same transaction costs level). Also the SPF is a lot less risky than the benchmark

funds at the 30% and 40% levels.

In the arbitrary yield setting: the SPF has higher risk associated with it than the benchmark

fund at 20% level (assuming equal transaction costs for the two funds). On the other hand the

SPF has a higher expected return and lower risk than the 30% benchmark fund.

The risk of the SPF increases as the yield increases, since the participation rate increases.

Notable is that the risk of the bond part decreases (a bond’s delta is increasing in yield (higher

yield, less sensitive to yield changes)), which implies that the benchmark portfolio’s risk decreases

as the yield increases. Thus a fund manager must be aware of these characteristics and how they

impact the portfolio’s expected return and risk. With a higher yield the bond portfolio gains

a higher return, which explains the higher expected return for both the SPF and benchmark

portfolio in the arbitrary yield setting.

The results indicate that transaction costs is one of the most important concepts for the

fund manager to consider in its portfolio. Assume that the proportional transaction costs for

structured products are approximately 3% and for the benchmark portfolio 1%, when studying

the arbitrary yield case. Thus the SPF’s expected return is 21.83%, has a CVaR

0.95

of 7.36%

and the benchmark fund which allocates 30% of its assets in the index has an expected return

of 23.08%, a CVaR

0.95

of 6.32% (as shown in Table 6.9). Thus the benchmark fund has both a

higher expected return and a lower risk than the SPF, thus it would be an irrational decision to

invest in the SPF.

It is not unreasonable that big funds (or investment banks) have the possibility to invest to

as low transaction costs as 0.5% or even 0.25%. Thus the SPF must in the arbitrary yield case

have as low as 1.5% in proportional transaction costs to be more attractive than the benchmark

fund investing 30% of its capital in the index. It is important that the reader takes care and

understands that these levels are dependent on the data that is used in the modeling, as well

as the underlying assumptions. The same patterns are exhibited for a huge variety of data sets,

but the expected returns, CVaR at diﬀerent levels varies of course. The low yield setting reﬂects

the setting of having a yield of approximately 1.5% p.a. in the three-month tenor and 2.3% p.a.

for the three-year tenor.

Concluded: the whole idea of a structured products fund is very dependent on the transaction

costs and it is only possible for the fund to compete with other funds, such as mixed funds, if

M. Hveem 78 (103)

6.5. Transaction costs

SPF Dynamic Weights Low yield SPF Dynamic Weights Arbitrary yield

TC [%] Expected Return [%] CVaR

0.95

[%] Expected Return [%] CVaR

0.95

[%]

0.00 17.69 2.79 27.19 3.99

0.25 17.26 3.10 26.73 4.29

0.50 16.82 3.41 26.25 4.59

1.00 16.00 3.98 25.32 5.16

1.50 15.21 4.54 24.41 5.73

2.00 14.44 5.08 23.53 6.28

2.50 13.70 5.61 22.67 6.82

3.00 12.97 6.13 21.83 7.36

Benchmark 20% Low yield Benchmark 20% Arbitrary yield

TC Expected Return [%] CVaR

0.95

[%] Expected Return [%] CVaR

0.95

[%]

0.00 15.43 3.25 20.81 0.13

0.25 15.11 3.39 20.51 0.23

0.50 14.78 3.70 20.18 0.58

1.00 14.22 4.18 19.58 1.14

1.50 13.65 4.70 18.99 1.69

2.00 13.05 5.09 18.36 2.09

2.50 12.56 5.56 17.82 2.62

3.00 11.94 6.10 17.26 3.07

Benchmark 30% Low yield Benchmark 30% Arbitrary yield

TC Expected Return [%] CVaR

0.95

[%] Expected Return [%] CVaR

0.95

[%]

0.00 19.50 8.72 24.37 5.43

0.25 19.26 8.97 24.04 5.79

0.50 18.93 9.09 23.75 6.04

1.00 18.30 9.66 23.08 6.32

1.50 17.69 10.13 22.59 6.90

2.00 17.08 10.51 21.97 7.30

2.50 16.48 10.93 21.28 7.89

3.00 15.84 11.44 20.68 8.20

Benchmark 40% Low yield Benchmark 40% Arbitrary yield

TC Expected Return [%] CVaR

0.95

[%] Expected Return [%] CVaR

0.95

[%]

0.00 23.88 14.35 28.18 11.46

0.25 23.56 14.41 27.85 11.64

0.50 23.20 14.57 27.40 12.00

1.00 22.58 15.25 26.78 12.41

1.50 21.87 15.59 26.16 12.75

2.00 21.32 16.02 25.56 13.17

2.50 20.64 16.44 24.91 13.61

3.00 20.07 16.78 24.19 14.15

Table 6.9: The table discloses the expected return and CVaR

0.95

for diﬀerent levels of proportional

transaction costs for the SPF and benchmark funds over a three-year investment horizon

for both the low yield setting and the arbitrary yield setting.

M. Hveem 79 (103)

Chapter 6. Portfolio Optimization

the fund keeps down the transaction costs. If the SPF can have the same level of transaction

costs as the benchmark funds it will be able to generate a higher expected return to a lower risk.

Notable is that as the interest increases the ability for the SPF to compete with the benchmark

portfolios decreases (in terms of low risk). Thus it is very important that the fund manager

understands how its choices aﬀect the return and risk of the fund.

The most beneﬁcial investment setup to use is the modiﬁed Korn and Zeytun framework. The

scheme usually implies that the fund allocates most of its capital in the three newest products

(the products with the longest time to maturity) and almost all of the remaining capital in the

product closest to maturity, since these can soon be reallocated in the newly issued products.

6.6 Summary

How can the results found in Chapter 4 and 6 be combined into an investment strategy? It has

been established that the products with the longest time to maturity have, in general, the lowest

CVaR (and expected return). Chapter 4 showed that it is important (if the fund manager wants

to minimize the risk) to limit the level of options held in the portfolio. Thus when competing

with a benchmark fund the fund manager could by allowing between 5-25 percent of the portfolio

value to be allocated in options provide a better protection against downside risk. The investors

are more or less guaranteed, by limiting the option portfolio in this manner, that a market crash

will not erase more than 25% of the fund’s portfolio value. Rebalancing in this manner implies

that the fund gains capital protection on its previous gains, since it locks them in by investing

in bonds. Obviously by minimizing the risk in this manner the potential return is also reduced.

Section 4.4 indicated that the fund should not overweight products that have been issued

the last quarter during downturn markets and also avoid products close to maturity to attain

protection to the worst-case outcomes. Minimizing the worst-case outcome imposes a slightly

diﬀerent portfolio than minimizing CVaR. Hence it is very important for the investor/portfolio

manager to understand how the diﬀerent decisions aﬀect the risk proﬁle. Investing only in

products close to maturity would provide a higher expected return and impose a higher risk.

This paper is not stating that any option is better than the other; it is only presenting the

impact of diﬀerent portfolio options and goals.

The other main investment alternative is to minimize CVaR with the whole investment uni-

verse available as done in this chapter, thus allocating most of the capital in the newly issued

products since they have the best risk proﬁle in CVaR sense. When comparing the results of the

three constructions (ﬁxed, rolling and dynamic) it is concluded that dynamic portfolio weights is

the best choice in risk adjusted return sense (generates lower risk to the same return levels). The

modiﬁed Korn and Zeytun framework will not overweight newly issued assets during downturn

markets since it is minimizing the risk, thus rather investing in the products with a low ratio

options:bonds. The downside with this strategy is that it may induce relative high transaction

costs.

The constraint of holding a maximum of 25% value in options can be added, thus forcing

the portfolio manager to reduce the worst-case outcome risk at the same time as the CVaR (but

notable minimizing the CVaR will in most cases avoid allocating too much capital in the options).

Diﬀerent portfolio managers will make diﬀerent choices regarding which alternative is best,

none is better than the other since they are constructed to restrict diﬀerent events.

Transaction costs will be a central question no matter which investment strategy is chosen.

The results are only relevant in relation to the investment environment, thus it is necessary that

the SPF has the possibility of being more attractive than the benchmark fund, either at risk or

return. Notable is that transaction costs are more important for the expected return and risk

level than the portfolio choice itself. Thus given high transaction costs it is impossible for a SPF

to compete with a benchmark fund with low transaction costs. Hence if the reader is considering

to start a SPF it should focus on the transaction costs, it is impossible to outrun the mixed funds

available at the market in the long run without limiting the transaction costs.

M. Hveem 80 (103)

6.7. Backtesting

6.7 Backtesting

This section investigates how the modiﬁed Korn and Zeytun algorithm would have performed

historically. The model is backtested from the beginning of 2002 to the middle of 2010 (data

for OMXS30 total return is only available since 2002 from Bloomberg). In each time step the

simulation is based on the last four year’s data (log returns), to capture approximately a market

cycle. Proportional transaction costs at the levels 0%, 1%, 2% and 3% are considered. The

CVaR is minimized given the constraint of a positive return. Notable is that the scheme is only

rebalancing matured capital, to avoid as much unnecessary transaction costs as possible. The

results are compared with the benchmark fund described earlier (with 30% market participation),

which is based on the total return version of OMXS30, and with the actual OMXS30 total return

version. Figure 6.11 shows that the SPF had the desired characteristics during the ﬁnancial

turmoil during 2002 and during 2008-2009, imposing a high level of capital protection. Also the

SPF did not actually yield any return during the latter part of 2009 and 2010, rather negative

return. This is due to that the portfolio is rebalanced quite signiﬁcantly during the last quarters

(imposing negative returns due to transaction costs) and the volatility drops signiﬁcantly, which

implies that the value of the options decreases. The interest rate movements also impact the

return during these quarters negative, the increase in the yield results in lower bond prices.

2003 2004 2005 2006 2007 2008 2009 2010

60

80

100

120

140

160

Year

[

%

]

Nasdaq OMXS30 Total Return

NAV Fund

SPF Benchmark

(a) TC = 0%, SPF total return 51.73%.

2003 2004 2005 2006 2007 2008 2009 2010

60

80

100

120

140

160

Year

[

%

]

Nasdaq OMXS30 Total Return

NAV Fund

SPF Benchmark

(b) TC = 1%, SPF total return 43.35%.

2003 2004 2005 2006 2007 2008 2009 2010

60

80

100

120

140

160

Year

[

%

]

Nasdaq OMXS30 Total Return

NAV Fund

SPF Benchmark

(c) TC = 2%, SPF total return 36.73%.

2003 2004 2005 2006 2007 2008 2009 2010

60

80

100

120

140

160

Year

[

%

]

Nasdaq OMXS30 Total Return

NAV Fund

SPF Benchmark

(d) TC = 3%, SPF total return 29.25%.

Figure 6.11: Backtesting results for the dynamic portfolio weights SPF against a benchmark mixed

fund with 30% participation rate and OMXS30.

The transaction costs impact the return signiﬁcantly (notable is that the benchmark fund

suﬀers in approximately the same extent). The problem with investing using a structured prod-

ucts fund is that the structured products mature and the capital must be reinvested in new

products, imposing new transaction costs. Thus if the fund instead invests in products that have

a longer time to maturity it is possible to both reduce the risk (since products with a longer time

to maturity have lower risk) and the transaction costs.

A backtest is also conducted for the ﬁxed portfolio weights strategy using the algorithm from

Section 6.1 and in Figure 6.3, which takes in count the transaction costs, thus allocating mostly

in the newly issued products, the result is disclosed in Figure 6.12. This strategy generates a

higher return than the dynamic portfolio weights strategy during the time period, which is more

due to a coincidence than an optimal portfolio choice, since this is just one observation. The ﬁxed

portfolio weights setting generates higher transaction costs than the dynamic portfolio weights

setting.

It is really important to not rebalance the whole portfolio every quarter, instead the fund

should mainly rebalance capital that has matured, in a setting with high transaction costs.

Rebalancing the whole portfolio every quarter would in the dynamic portfolio weights setting,

M. Hveem 81 (103)

Chapter 6. Portfolio Optimization

2003 2004 2005 2006 2007 2008 2009 2010

60

80

100

120

140

160

Year

[

%

]

Nasdaq OMXS30 Total Return

NAV Fund

SPF Benchmark

(a) TC = 0%, SPF total return 63.64%.

2003 2004 2005 2006 2007 2008 2009 2010

60

80

100

120

140

160

Year

[

%

]

Nasdaq OMXS30 Total Return

NAV Fund

SPF Benchmark

(b) TC = 1%, SPF total return 45.86%.

2003 2004 2005 2006 2007 2008 2009 2010

60

80

100

120

140

160

Year

[

%

]

Nasdaq OMXS30 Total Return

NAV Fund

SPF Benchmark

(c) TC = 2%, SPF total return 33.81%.

2003 2004 2005 2006 2007 2008 2009 2010

60

80

100

120

140

160

Year

[

%

]

Nasdaq OMXS30 Total Return

NAV Fund

SPF Benchmark

(d) TC = 3%, SPF total return 30.60%.

Figure 6.12: Backtesting results for the ﬁxed portfolio weights SPF including proportional transaction

costs against a benchmark mixed fund with 30% participation rate and OMXS30.

with 3% in transaction costs, yield 15% in return instead of approximately 29%. Thus the

portfolio manager should assume a type of buy and hold strategy and mainly reallocate matured

capital. If the fund assumes a buy and hold strategy the portfolio manager should be aware

of that it still needs to rebalance the whole portfolio sometimes, e.g. when the option portfolio

value exceeds its allowed boundary.

Thus it is recommended that, in a setting of high transaction costs, the portfolio manager

uses the modiﬁed Korn and Zeytun framework to initially allocate its capital, and to reallocate

the matured capital. The importance of using a more active portfolio strategy is higher when

the possibility of diversifying in several structured products with diﬀerent underlying exists.

M. Hveem 82 (103)

CHAPTER 7

Conclusions

The thesis has studied the concept of a structured products fund (SPF) and how it should be

constructed to be competitive in risk sense. The study shows that the concept should not be

disregarded and that it is possible to construct a SPF that is competitive.

Transaction costs aﬀect the expected return and risk level for the SPF in a high extent. Thus

the study indicates that the most important factor for a portfolio manager to consider is the

transaction costs. A SPF with high transaction costs will not have the possibility of beating the

competing mixed funds at either risk or return. Thus the fund needs low transaction costs (in

level with the competing funds) in order to not increase the risk and decrease the return in a too

high extent.

The study focuses a lot on the issue of capital guarantee since it is desirable that the SPF

has the same characteristics as a capital guaranteed product. The study indicates that it is not

possible to attain a portfolio based on structured products that is capital guaranteed. The thesis

instead studies the concept of alternative capital protection and ﬁnds that a fund can be capital

protected at a certain level for a given risk measure, the thesis uses the risk measure CVaR.

Thus a fund that has a CVaR

0.90

less than for example 0% can claim that the portfolio is capital

protected on a 100% level with a conﬁdence level of 90%. CVaR is used as a risk measure since

it captures both skewness and kurtosis and is adequate to use in scenario based optimization. It

is also common to describe a fund’s risk characteristics via CVaR, thus making it an adequate

risk measure to benchmark through.

The thesis ﬁnds that a portfolio manager has mainly two investment schemes to allocate

the portfolio with, to minimize the downside risk, thus attaining as much capital protection as

possible. The ﬁrst investment scheme is a type of rolling investment scheme where the portfolio

manager invests broadly in almost all the available products (preferably equal weights). The

fund should sell of the products when they are close to maturity (a quarter or two left), since

these assets exhibit a lot of risk especially if they are in the money. The capital that is gained

when assets are sold should be reinvested in the newly issued products since they have in general

the lowest risk associated with them. On the other hand as the index starts to decrease (during

bear markets) the portfolio should underweight newly issued products (products with the longest

time to maturity) since they are risky products in downturn markets. This scheme is used to

both restrict the risk and the induced transaction costs. The beneﬁt with this scheme is that

it is easy to implement and that it serves as a reasonable investment choice during most time

periods. On the other hand the scheme can induce a high-risk portfolio during unwanted market

states, since it does not adjust to changes in the market. Thus it is impossible to control the

expected return of the investment as well as the risk level. Hence the thesis instead proposes a

more sophisticated allocation algorithm, which is referred to as the modiﬁed Korn and Zeytun

framework.

The modiﬁed Korn and Zeytun framework is used to minimize the risk (CVaR) given the

constraint of the expected return. This framework is very beneﬁcial and creates low risk portfolios

with high expected return. The framework that is presented in detail in Section 6.4 shows that

it is possible to generate portfolios that have a higher expected return than competing funds to

a lower risk, given that the transaction costs are low. The study is conducted with both low

83

Chapter 7. Conclusions

yield and arbitrary yield start trajectories. The results show that a SPF is most beneﬁcial during

times of low yield since the risk of the SPF increases a lot as the yield goes up.

The results indicate that a SPF can be a superb investment vehicle for investors searching

for low risk alternatives with a limited downside. They also show that it is imperative that a

portfolio manager understands how the portfolio allocation aﬀects the characteristics of the fund,

since it is a rather complex issue.

It is strongly recommended to use the modiﬁed Korn and Zeytun framework to allocate the

portfolio with, for a SPF, since it generates well-balanced portfolios in risk and return sense.

Allocating according to the modiﬁed Korn and Zeytun framework will in most cases lead to

allocating in the best available portfolio (given the constraints) in every time step since it can

be used for speciﬁc paths (adjust for the current market data).

The thesis ﬁnds that it is important, no matter which scheme is used, that the portfolio is

not allowed to hold more than a certain degree of value attained in options (a limit), to avoid

downside risk for its investors. A recommended level is that the SPF is allowed to allocate a

maximum of 25% of its capital in options (within the structured products) with a target level of

15%. The portfolio should be rebalanced if this limit is exceeded, hence reducing the downside

risk.

The thesis studies ELNs with a time to maturity of three years and the results can be

generalized to products with a longer time to maturity at issuance (newly issued products, less

risk). The thing that diﬀers for a capital guaranteed ELN with a time to maturity of three years

at issuance and one with a longer time to maturity (or shorter) is the ratio options:bonds. Thus

a newly issued ELN with ﬁve years to maturity has a higher risk than a newly issued ELN with

three years to maturity since it has a higher proportion options. Thus it is very important to use

the boundaries of allowed proportion options:bonds especially when combining diﬀerent types of

ELNs. The big advantage by using the modiﬁed Korn and Zeytun framework is that it adjusts

for these discrepancies and that the portfolio manager does not need to modify the allocation

model if it modiﬁes its investment universe.

Thus the main result is: yes it is possible to construct a fund based on structured products

that is competitive and exhibits a type of capital protection (i.e. in CVaR sense). Only one

underlying for the ELNs has been used and the results can be generalized without serious loss

of generality to the multi-dimensional case. A Black-Litterman approach to the optimization

problem is left for future studies where it, in combination with the multi-underlying case, can

provide superior return due to subjective views on the market.

M. Hveem 84 (103)

Bibliography

[1] Aas, K. (2006). The Generalised Hyperbolic Skew Student’s t-distribution. Journal of Fi-

nancial Econometrics, 4 (2), 275-309.

[2] Abdi, H., & Williams, L. J. (2010). Principal Component Analysis. Wiley Interdisciplinary

Reviews: Computational Statistics, 2, 1-47.

[3] Acerbi, C. (2002). Spectral measures of risk: A coherent representation of subjective risk

aversion. Journal of Banking & Finance, 26 (7), 1505-1518.

[4] Alexander, S., Coleman, T.F., & Li, Y. (2006). Minimizing CVaR and VaR for a portfolio

of derivatives. Journal of Banking & Finance, 30 (2), 583-605.

[5] Bank For International Settlements. (2006). Basel II: International Convergence of Capital

Measurement and Capital Standards: A Revised Framework. Basel: BIS

[6] Barber, J. R., & Copper, M. L. (2010). Principal component analysis of yield curve move-

ments. Journal of Economics and Finance, 1-16, doi:10.1007/s12197-010-9142-y.

[7] Björk, T. (2009). Arbitrage Theory in Continuous Time (3rd ed.). Oxford: Oxford Finance.

[8] Black, F., & Litterman, R. (1992). Global Portfolio Optimization. Financial Analysts Jour-

nal, 48 (5), 28-43.

[9] Black, F., & Scholes, M. (1973). The Pricing of Options and Corporate Liabilities. Journal

of Political Economy, 81 (3), 637-654.

[10] Brooks, C., & Persand, G. (2000). Value at Risk and Market Crashes. Discussion Paper in

Finance 2000-01. University of Reading.

[11] Browning, E., S. (2007, October 15). Exorcising Ghosts of Octobers Past, Wall Street Jour-

nal, p. C1.

[12] Chuan, T. (2008, August, 5). Capital Protected is not Capital Guaranteed. Financial Plan-

ning Central:

http://www.ifa-sg.com/capital-protected-is-not-capital-guaranteed/

[13] Deng, G., Li, G., Liu, S., Mallett, J., & McCann, C. (2009). Structured Products In the

Aftermath of Lehman Brothers. Working Papers Securities Litigation & Consulting Group.

[14] Fisher, A. (1999). Multivariate Stock Returns Around Extreme Events: A Reassessment of

Economic Fundamentals and the 1987 Market Crash. NYU Working Paper, No. FIN-99-071.

[15] Frazzini, A. (2006). The Disposition Eﬀect and Underreaction to News. The Journal of

Finance, 61 (4), 2017-2046.

[16] Glot, P. (2005). Relationship Between Implied Volatility Indexes and Stock Index Returns.

The Journal of Portfolio Management, 31 (3), 92-100.

[17] Grinold, R., C. (1999). Mean-Variance and Scenario-Based Approaches to Portfolio Selec-

tion. Journal of Portfolio Management, 25 (2), 10-22.

85

Bibliography

[18] Hagan, P. S., Kumar, D., Lesniewski, A. S., & Woodward, D. E. (2002). Managing Smile

Risk. Wilmott Magazine September 2002, 84-108.

[19] Hansen, P., & Lärfars, M. (2010). Evaluating Long-Term Performance of Structured Prod-

ucts. Stockholm: Kungliga Tekniska Högskolan.

[20] Heston, S. L. (1993). A Closed-Form Solution for Options with Stochastic Volatility with

Applications to Bond and Currency Options. The Review of Financial Studies, 6 (2), 327-

343.

[21] HSBC. (2010, September 14). Structured Product Glossary. HSBC:

http://www.hsbcnet.com/treasury/structured-products/wms-

global/struc_prod_glossary.html

[22] Hull, J.C. (2002). Options, Futures, and Other, Derivatives (5th ed.). New Jersey: Prentice

Hall.

[23] Hult, H., & Lindskog, F. (2009). Lecture notes in Risk management. Stockholm: Royal

Institute of Technology.

[24] Hult, H., Lindskog, F., Hammarlid, O., & Rehn, C.-J. (2010). Risk and portfolio analysis:

principles and methods, Part I. Stockholm: Royal Institute of Technology.

[25] Johansson, T., & Lingnardz, H. (2010). Evaluation of Equity-Linked Notes - an Empirical

Study of SEB Equity-Linked Notes. Lund: Lund University.

[26] Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk.

Econometrica, 47 (2), 263-291.

[27] Korn, R., & Zeytun, S. (2009). Solving optimal investment problems with structured prod-

ucts under CVaR constraints. Optimization, 58 (3), 291-304.

[28] Krokhmal, P., Palmquist, J., & Uryasev, S. (2002). Portfolio optimization with conditional

value-at-risk objective and constraints. Journal of Risk, 4 (2), 11-27.

[29] Litterman, R., & Scheinkman, J. (1991). Common Factors Aﬀecting Bond Returns. The

Journal of Fixed Income, 1 (1), 54-61.

[30] Markowitz, H. (1952). Portfolio selection. Journal of Finance, 7 (1), 77-91.

[31] Martellini, L., Simsek, K., & Goltz, K. (2005). Structured forms of investment strategies

in institutional investors portfolios. Technical report, EDHEC Risk and Asset Management

Research Centre.

[32] My News Desk. (2010, January 29). Swedbank störst på indexobligationer. My News Desk:

http://www.mynewsdesk.com/se/pressroom/swedbank/pressrelease/

view/swedbank-stoerst-paa-indexobligationer-366818

[33] Rockafellar, R., T., & Uryasev, P. (2001). Conditional Value-at-Risk for General Loss Dis-

tributions. ISE Dept. Working Paper No. 2001-5.

[34] Scherer, B. (2004) Portfolio Construction and Risk Budgeting (2nd ed.). London: Risk

Books.

[35] Shefrin, H. (2002). Beyond Greed and Fear (2nd ed.). Oxford: Oxford University Press.

[36] Soto, G. (2004) Using Principal Component Analysis to Explain Term Structure Movements:

Performance and Stability. Available online at: http://ssrn.com/abstract=985404.

M. Hveem 86 (103)

Bibliography

[37] UBG AG. (2010, October, 12). Banking terms. UBS AG:

http://www.ubs.com/1/e/gcc/bankingterms.html.

[38] Uryasev, P. (2000). Conditional Value-at-Risk: Optimization Algorithms and Applications.

Financial Engineering News, 14, 1-5.

[39] Wasserfallen, W., & Schenk, C. (1996). Portfolio Insurance for the Small Investor in Switzer-

land, The Journal of Derivatives, 3 (3), 37-43.

M. Hveem 87 (103)

Appendices

89

APPENDIX A

Naive fund constructions - Additional scenarios

Naive fund construction 1

0 5 10 15 20 25

0

50

100

150

200

250

X: 24

Y: 92.13

Quarter

$

Price Underlying

NAV Fund

Start of investment

(a) Scenario 3, −7.87%.

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 100

Quarter

$

Price Underlying

NAV Fund

Start of investment

(b) Scenario 4, 0%.

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 92.28

Quarter

[

%

]

Volatility

NAV Fund

Start of investment

(c) Scenario 5, −7.72%.

0 5 10 15 20 25

0

20

40

60

80

100

120

X: 24

Y: 89.76

Quarter

[

%

]

Interest Rate

NAV Fund

Start of investment

(d) Scenario 6, −10.24%.

Figure A.1: Scenarios 3-6, return between Q12-Q24.

91

APPENDIX B

Option portfolios - Additional portfolios

w

9

=

_

1 0 0 0 0 0 0 0 0 0 0 0

_

T

w

10

=

_

0 1 0 0 0 0 0 0 0 0 0 0

_

T

w

11

=

_

0 0 0 0 0 0 0 0 0 0 1 0

_

T

w

12

=

_

0 0

1

10

1

10

1

10

1

10

1

10

1

10

1

10

1

10

1

10

1

10

_

T

w

13

=

_

0 0 0 0

1

8

1

8

1

8

1

8

1

8

1

8

1

8

1

8

_

T

w

14

=

_

1

4

1

4

1

4

1

4

0 0 0 0 0 0 0 0

_

T

w

15

=

_

0 0 0 0

1

4

1

4

1

4

1

4

0 0 0 0

_

T

w

16

=

_

0 0 0 0 0 0 0 0

1

4

1

4

1

4

1

4

_

T

w

17

=

_

0 0 0 0 0 0 0 0

1

10

3

10

3

10

3

10

_

T

w

18

=

_

1

8

1

8

1

8

1

8

1

8

1

8

1

8

1

8

0 0 0 0

_

T

w

19

=

_

1

8

1

8

1

8

1

8

0 0 0 0

1

8

1

8

1

8

1

8

_

T

−1 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4

0

200

400

600

800

1000

1200

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

tc

o

m

e

s

(a) Portfolio 9, −100%.

−0.5 −0.4 −0.3 −0.2 −0.1 0 0.1 0.2

0

500

1000

1500

2000

2500

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

tc

o

m

e

s

(b) Portfolio 10, −47.37%.

−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

50

100

150

200

250

300

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

tc

o

m

e

s

(c) Portfolio 11, −19.02%.

Figure B.1: Histograms over the outcomes for option portfolios 9-11, the worst-case average 36 month

return is given in the caption.

93

Appendix B. Option portfolios - Additional portfolios

−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

180

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(a) Portfolio 12, −20.55%.

−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

50

100

150

200

250

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(b) Portfolio 13, −20.93%.

−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3

0

100

200

300

400

500

600

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(c) Portfolio 14, −39.94%.

−0.3 −0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

50

100

150

200

250

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(d) Portfolio 15, −26.46%.

−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

50

100

150

200

250

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(e) Portfolio 16, −20.44%.

−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

50

100

150

200

250

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(f ) Portfolio 17, −20.74%.

−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(g) Portfolio 18, −24.04%.

−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

0

20

40

60

80

100

120

140

160

180

Option portfolio average monthly returns (36 months periods)

Average return per month

O

u

t

c

o

m

e

s

(h) Portfolio 19, −24.47%.

Figure B.2: Histograms over the outcomes for option portfolios 12-19, the worst-case average 36 month

return is given in the caption.

M. Hveem 94 (103)

APPENDIX C

SPF vs Benchmark - Additional scenarios

Scenario C & D

Return Index

Months 49-96 /

51-98 p.a.

ε

bench

δ

bench

Scenario C Worst-

Case Return

Scenario D Worst-

Case Return

-80% 0.7 0.1 -97.36% -98.12%

-80% 0.5 0.1 -91.23% -94.01%

-80% 0.3 0.1 -71.41% -78.26%

-80% 0.20 0.1 -51.70% -60.40%

-80% 0.10 0.05 -23.56% -32.10%

-50% 0.7 0.1 -87.92% -91.10%

-50% 0.5 0.1 -71.47% -79.34%

-50% 0.3 0.1 -51.05% -62.53%

-50% 0.20 0.1 -29.10% -40.95%

-50% 0.10 0.05 -8.77% -18.58%

-30% 0.7 0.1 -77.85% -83.52%

-30% 0.5 0.1 -57.83% -69.09%

-30% 0.3 0.1 -37.75% -52.21%

-30% 0.20 0.1 -19.82% -32.96%

-30% 0.10 0.05 2.26% -12.32%

-20% 0.7 0.1 -71.86% -78.88%

-20% 0.5 0.1 -50.02% -63.15%

-20% 0.3 0.1 -32.27% -47.57%

-20% 0.20 0.1 -14.70% -29.60%

-20% 0.10 0.05 0.76% -10.23%

-10% 0.7 0.1 -64.91% -73.57%

-10% 0.5 0.1 -42.02% -57.13%

-10% 0.3 0.1 -26.33% -43.11%

-10% 0.20 0.1 -11.22% -25.90%

-10% 0.10 0.05 2.69% -8.10%

Table C.1: Benchmark fund: Worst-case outcome given scenarios C and D, ε

bench

and δ

bench

.

95

Appendix C. SPF vs Benchmark - Additional scenarios

Scenario C Scenario D

Months 0-48, return 30% p.a. Months 0-48, return 30% p.a.

Months 48-49, return -80% p.m. Months 48-50, return 50% p.a.

Months 49-96, return -X% p.a. Months 50-51, return -80% p.m.

Months 96-144, return 30% p.a. Months 51-98, return -X% p.a.

Months 98-144, return 30% p.a.

Return Index

Months 49-96 /

51-98 p.a.

ε

SPF

δ

SPF

Scenario C Worst-

Case Return

Scenario D Worst-

Case Return

-80% 1 1 -20.35% -61.49%

-80% 0.3 0.1 -20.35% -61.49%

-80% 0.25 0.1 -11.25% -51.12%

-80% 0.20 0.1 -10.42% -50.03%

-80% 0.15 0.1 1.32% -30.87%

-80% 0.10 0.05 3.79% -25.69%

-50% 1 1 -20.65% -62.31%

-50% 0.3 0.1 -20.65% -62.31%

-50% 0.25 0.1 -11.58% -52.17%

-50% 0.20 0.1 -10.76% -51.10%

-50% 0.15 0.1 0.94% -32.25%

-50% 0.10 0.05 3.41% -27.27%

-30% 1 1 -24.26% -63.95%

-30% 0.3 0.1 -24.26% -63.95%

-30% 0.25 0.1 -15.61% -54.24%

-30% 0.20 0.1 -14.82% -53.22%

-30% 0.15 0.1 -3.65% -35.28%

-30% 0.10 0.05 -1.30% -30.43%

-20% 1 1 -25.671% -64.62%

-20% 0.3 0.1 -25.67% -64.62%

-20% 0.25 0.1 -17.18% -55.09%

-20% 0.20 0.1 -16.41% -54.09%

-20% 0.15 0.1 -5.45% -36.48%

-20% 0.10 0.05 -3.14% -31.72%

-10% 1 1 -26.26% -64.91%

-10% 0.3 0.1 -26.26% -64.91%

-10% 0.25 0.1 -17.84% -55.46%

-10% 0.20 0.1 -17.07% -54.47%

-10% 0.15 0.1 -6.20% -37.00%

-10% 0.10 0.05 -3.91% -32.28%

Table C.2: SPF 1: Worst-case outcomes for the SPF 1 Portfolio 7 given the scenarios C and D, ε

SPF

and δ

SPF

.

M. Hveem 96 (103)

Portfolio 3, Scenario A & B

Return Index

Months 48-96 /

50-98 p.a.

ε

SPF

δ

SPF

Scenario A Worst-

Case Return

Scenario B Worst-

Case Return

-80% 1 1 -26.11% -57.88%

-80% 0.25 0.1 -26.11% -40.90%

-80% 0.20 0.1 -19.68% -35.51%

-80% 0.15 0.1 -15.98% -26.94%

-80% 0.15 0.05 -14.77% -29.43%

-80% 0.10 0.05 -13.02% -20.77%

-50% 1 1 -24.87% -49.59%

-50% 0.25 0.1 -24.87% -31.78%

-50% 0.20 0.1 -18.96% -27.15%

-50% 0.15 0.1 -16.01% -19.28%

-50% 0.15 0.05 -16.69% -23.33%

-50% 0.10 0.05 -12.71% -14.78%

-30% 1 1 -20.46% -37.45%

-30% 0.25 0.1 -20.46% -22.40%

-30% 0.20 0.1 -15.21% -16.62%

-30% 0.15 0.1 -12.92% -9.19%

-30% 0.15 0.05 -13.19% -13.63%

-30% 0.10 0.05 -8.82% -12.84%

-20% 1 1 -16.53% -26.71%

-20% 0.25 0.1 -16.53% -12.91%

-20% 0.20 0.1 -11.56% -8.31%

-20% 0.15 0.1 -8.71% -1.35%

-20% 0.15 0.05 -9.36% -5.38%

-20% 0.10 0.05 -5.12% -0.79%

-10% 1 1 -9.92% -13.72%

-10% 0.25 0.1 -9.92% -3.5%

-10% 0.20 0.1 -5.94% -1.28%

-10% 0.15 0.1 -0.53% 5.44%

-10% 0.15 0.05 -4.49% 0.00%

-10% 0.10 0.05 0.00% 3.35%

Table C.3: SPF 1: Worst-case outcomes for the SPF 1 Portfolio 3 given scenarios A and B, ε

SPF

and

δ

SPF

.

M. Hveem 97 (103)

Appendix C. SPF vs Benchmark - Additional scenarios

Stress testing

µ = −30% p.a. for underlying instead of 0% p.a. as in Section 4.4.7

−0.4 −0.2 0 0.2 0.4 0.6 0.8 1

0

500

1000

1500

2000

2500

3000

3500

Histogram over the SPF:s 36 month (overlapping) returns

Return

O

u

tc

o

m

e

s

(a) SPF 1: ε

SPF

= 0.15, δ

SPF

=

0.1, min: −20.59%, mean:

8.01%, max: 63.24%.

−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 0.4 0.5

0

500

1000

1500

2000

2500

Histogram over the Benchmark funds 36 month (overlapping) returns

Return

O

u

tc

o

m

e

s

(b) Benchmark: ε

bench

= 0.3,

δ

bench

= 0.1, min: −37.83%,

mean: −7.70%, max: 45.03%.

−1 −0.5 0 0.5 1 1.5

0

500

1000

1500

2000

2500

3000

3500

4000

Histogram over the Underlying 36 month (overlapping) returns

Return

O

u

tc

o

m

e

s

(c) Underlying: min: −92.60%,

mean: −56.87%, max: 129.82%.

Figure C.1: Histograms over 36 months returns for the competing funds and the modiﬁed stress test

Backtesting

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(a) ε

SPF

= 0.10, δ

SPF

= 0.05.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(b) ε

SPF

= 0.15, δ

SPF

= 0.1.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(c) ε

SPF

= 0.2, δ

SPF

= 0.1.

1996 1998 2000 2002 2004 2006 2008 2010 2012

100

150

200

250

300

350

400

Year

$

Nasdaq OMXS30

NAV Fund

NAV Benchmark

(d) ε

SPF

= 0.25, δ

SPF

= 0.1.

Figure C.2: Backtest results: OMXS30 vs SPF 1 (Portfolio 4) and Benchmark..

M. Hveem 98 (103)

APPENDIX D

Correlations between the market index and the yield curve

Correlation matrix between OMXS30 and SEK yield curve, daily log returns (Jan 1996 - Nov 2010).

Index 1M 2M 3M 6M 9M 1Y 2Y 3Y

Index 1.0000 -0.0080 -0.0023 -0.0117 0.0012 0.0070 0.0193 0.1900 0.1891

1M -0.0080 1.0000 0.8970 0.8559 0.7678 0.7081 0.6855 0.3166 0.2382

2M -0.0023 0.8970 1.0000 0.9149 0.8422 0.7850 0.7628 0.3610 0.2802

3M -0.0117 0.8559 0.9149 1.0000 0.8947 0.8429 0.8229 0.3914 0.3070

6M 0.0012 0.7678 0.8422 0.8947 1.0000 0.9362 0.9214 0.4487 0.3657

9M 0.0070 0.7081 0.7850 0.8429 0.9362 1.0000 0.9610 0.4783 0.4100

1Y 0.0193 0.6855 0.7628 0.8229 0.9214 0.9610 1.0000 0.4981 0.4317

2Y 0.1900 0.3166 0.3610 0.3914 0.4487 0.4783 0.4981 1.0000 0.8916

3Y 0.1891 0.2382 0.2802 0.3070 0.3657 0.4100 0.4317 0.8916 1.0000

Correlation matrix between OMXS30 and SEK yield curve, weekly log returns (Jan 1996 - Nov 2010).

Index 1M 2M 3M 6M 9M 1Y 2Y 3Y

Index 1.0000 -0.0504 -0.0484 -0.0458 -0.0211 -0.0120 -0.0000 0.1741 0.1899

1M -0.0504 1.0000 0.9431 0.8971 0.8016 0.7497 0.7226 0.3765 0.2666

2M -0.0484 0.9431 1.0000 0.9578 0.8897 0.8471 0.8191 0.4488 0.3378

3M -0.0458 0.8971 0.9578 1.0000 0.9345 0.8965 0.8693 0.4975 0.3915

6M -0.0211 0.8016 0.8897 0.9345 1.0000 0.9711 0.9513 0.5965 0.4993

9M -0.0120 0.7497 0.8471 0.8965 0.9711 1.0000 0.9864 0.6438 0.5646

1Y -0.0000 0.7226 0.8191 0.8693 0.9513 0.9864 1.0000 0.6684 0.5972

2Y 0.1741 0.3765 0.4488 0.4975 0.5965 0.6438 0.6684 1.0000 0.9412

3Y 0.1899 0.2666 0.3378 0.3915 0.4993 0.5646 0.5972 0.9412 1.0000

Correlation matrix between OMXS30 and SEK yield curve, monthly log returns (Jan 1996 - Nov 2010).

Index 1M 2M 3M 6M 9M 1Y 2Y 3Y

Index 1.0000 -0.1041 -0.1328 -0.1320 -0.1029 -0.0935 -0.0757 0.1121 0.1233

1M -0.1041 1.0000 0.9608 0.9258 0.8415 0.7791 0.7473 0.5317 0.3837

2M -0.1328 0.9608 1.0000 0.9777 0.9166 0.8634 0.8325 0.5900 0.4410

3M -0.1320 0.9258 0.9777 1.0000 0.9553 0.9134 0.8849 0.6419 0.4912

6M -0.1029 0.8415 0.9166 0.9553 1.0000 0.9800 0.9626 0.7399 0.5965

9M -0.0935 0.7791 0.8634 0.9134 0.9800 1.0000 0.9932 0.7940 0.6677

1Y -0.0757 0.7473 0.8325 0.8849 0.9626 0.9932 1.0000 0.8223 0.7048

2Y 0.1121 0.5317 0.5900 0.6419 0.7399 0.7940 0.8223 1.0000 0.9563

3Y 0.1233 0.3837 0.4410 0.4912 0.5965 0.6677 0.7048 0.9563 1.0000

99

APPENDIX E

Optimization problem - Arbitrary start yield curve

This appendix uses an arbitrary start yield curve, thus the initial yield curve is drawn randomly

from the observed sample. The initial yield curve is drawn arbitrary between January 1996 to

November 2010.

Fixed portfolio weights

Product Number, CVaR

0.95

[%]

1 2 3 4 5 6 7 8 9 10 11 12

22.58 18.78 16.27 14.19 12.53 11.20 10.21 9.63 7.17 6.75 6.58 6.68

Product Number, Expected return [%]

1 2 3 4 5 6 7 8 9 10 11 12

28.73 28.76 28.45 27.99 27.20 26.76 26.33 25.94 25.91 25.64 25.38 25.14

Table E.1: The table describes the CVaR at the 95% conﬁdence level and expected return for the

diﬀerent products over the three-year period using the ﬁxed portfolio weights setting.

6.2 7.4 8.4 9.6 10.6 11.8 12.8 14 15 16 17

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Portfolio CVaR

α = 0.95

[%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

Figure E.1: Portfolio weights for the eﬃcient frontier using the ﬁxed portfolio weights setting, α = 0.95.

101

Appendix E. Optimization problem - Arbitrary start yield curve

Rolling portfolio weights

Product (months until next roll), CVaR

0.95

[%]

1 2 3 4 5 6 7 8 9 10 11 12

7.68 10.27 11.95 13.18 13.96 14.41 14.72 14.49 13.89 12.41 9.99 0

Product (months until next roll), Expected return [%]

1 2 3 4 5 6 7 8 9 10 11 12

26.95 26.87 26.83 26.73 26.67 26.64 26.63 26.64 26.76 26.89 26.97 27.13

Table E.2: The table describes the CVaR at the 95% conﬁdence level and expected return for the

diﬀerent products over the three-year period using the rolling portfolio weights setting.

4.6 4.8 5.4 5.8 6.2 6.6 7 7.4 7.8

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Portfolio CVaR

α = 0.95

[%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

Figure E.2: Portfolio weights for the eﬃcient frontier using the rolling portfolio weights setting, α =

0.95.

Benchmark fund

Incl. Dividends. Excl. Dividends.

Index allocation Expected Return CVaR

0.95

Expected Return CVaR

0.95

10% 17.25 -3.76 16.09 -3.17

20% 20.81 0.13 18.46 1.31

30% 24.37 5.43 20.83 7.57

40% 28.18 11.46 23.33 14.41

Table E.3: The table discloses the expected return and CVaR for diﬀerent levels of index weights over

the three-year period for a benchmark fund.

M. Hveem 102 (103)

Dynamic portfolio weights

Product, months until maturity, CVaR

0.95

[%]

1 2 3 4 5 6 7 8 9 10 11 12

9.85 12.34 13.70 14.45 14.63 14.53 14.00 13.14 12.01 10.34 7.70 0.00

Product, months until maturity, Expected return [%]

1 2 3 4 5 6 7 8 9 10 11 12

26.69 26.58 26.52 26.41 26.35 26.31 26.35 26.46 26.52 26.61 26.74 26.88

Table E.4: The table describes the CVaR at the 95% conﬁdence level and expected return for the

diﬀerent products over the three-year period using the dynamic portfolio weights and the

modiﬁed Korn and Zeytun framework .

5.8 6.8 7.8 8.8

0

10

20

30

40

50

60

70

80

90

100

W

e

i

g

h

t

[

%

]

Portfolio CVaR

α = 0.95

[%]

w

1

w

2

w

3

w

4

w

5

w

6

w

7

w

8

w

9

w

10

w

11

w

12

Figure E.3: Portfolio weights for the eﬃcient frontier using the dynamic portfolio weights and the

modiﬁed Korn and Zeytun framework, α = 0.95.

M. Hveem 103 (103)

Abstract The thesis evaluates the concept of a structured products fund and investigates how a fund based on structured products should be constructed to be as competitive as possible. The focus lies on minimizing the risk of the fund and on capital guarantee. The diﬃculty with this type of allocation problem is that the available products mature before the investment horizon, thus the problem of how the capital should be reinvested arises. The thesis covers everything from naive fund constructions to more sophisticated portfolio optimization frameworks and results in recommendations regarding how a portfolio manager should allocate its portfolio given diﬀerent settings. The study compares diﬀerent fund alternatives and evaluates them against, competing, benchmark funds. The thesis proposes a framework called the modiﬁed Korn and Zeytun framework which allocates a portfolio based on structured products, which have maturity prior to the end of the investment horizon, in optimal CVaR sense (i.e. appropriate for funds). The study indicates that the most important concept of a structured products fund is transaction costs. A structured products fund cannot compete against e.g. mixed funds on the market if it cannot limit its transaction costs at approximately the same level as competing funds. The results indicate that it is possible to construct a fund based on structured products that is competitive and attractive given low commission and transaction costs. Keywords: Structured products fund, structured products, portfolio theory, portfolio optimization, portfolio management, Conditional Value-at-Risk, transaction costs

.

Acknowledgements I would like to thank my supervisor at KTH Mathematical Statistics. Karl Lindqvist and William Sjöberg. for essential knowledge in the ﬁeld of structured products. Stockholm. Filip Lindskog. feedback and ideas. for great feedback and guidance as well as my supervisors at Nordea Markets. January 2011 Markus Hveem v .

.

. .3 Investigation of the option portfolio . . . . . . . . . . . . . . . . . . . . . . . . .3. . .2 Capital guarantee 1. 4. . . . . . . . . . . . . 3 Financial Assets 3. 4. . . . 4. . . . . .3 Scenario based optimization . . . . . . . .4 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . 3. . . . . . . .2. . . . .4. . . . . . . . . . . . . . . . .1 Dynamics . . . . . . . . . 2. . . . . .4 Comparison with competition . . . . . . . . . . 4. . . .3. . . . . . . . . . . . . . . . . . . . . . .2 Zero-coupon bond . . . .3 Structured products fund 1 . . . . . . . . . . . structured products fund vs benchmark 4. . . .1 Conditional Value-at-Risk 2. . . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Plain vanilla call option . . . . . . . . . . . . . . . . . . . . .Table of Contents 1 Introduction 1. . . . . . . . vii . . . 4. . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . .8 Backtesting . . . . .3 Analysis and implications . . . . . . . . . . . . . . . . . . .4 Transaction costs . . . .4 Outline .4 Prevalent risk factors . . . . . . . . . 4. . 4. 4. . . .3 Conditional Value-at-Risk 2. . . . . . .4. . . . . . . . 4. . . . . . . . . . . . . . . . . . . .2. . . . . . . . .1. . . . . . . . . . . . . . . . . . . .4. . 4. . . . . . . . . . . . . . . . . . . . . . . . . . .2 Minimum regret . . . . . . . . . 2. . . . . . . . . . . . . 4 The Capital Protection Property 4. . . . . . . . . . . . . . . . . .2 Principal component analysis . . . . . . . . . . . . . . . . . . . . . . . .1 A note regarding the critique against . . . . 4. . . . . . . . . . . . . . . . . . .1 Naive fund construction number 1 . . . . . . . . . . .7 Additional stress test . . . . . .6 Analysis and summary . . . . . . . . . . . . . .1 Absolute lower bound . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . 3. . . . . . . . . . . . . . . . . . 2. . . . . . . . . . 4. . . . . . . . . . . .1. . . . . . . .2 Benchmark fund . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. . . . . . . . . . . .2 Value-at-Risk . .4. . . . . . . 1. . . . . . . .3. . . 3. . . . .1 Client base . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. . .3. . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . .5 Results. . . . . . 1 1 1 2 3 5 5 5 6 7 7 9 10 10 11 15 15 16 17 17 18 19 19 20 20 25 29 29 31 31 32 35 36 37 37 37 39 39 42 44 45 2 Theoretical Background 2. . . . . . . . . . . . . . . . 4. S .2 Naive fund constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1 Portfolio construction and modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 1. . . . . . . 4. . . . . . . . . . . . . . .1. . . . . . . . . 2. . . . . . . . .1 Risk measures . . . . . . . . . .1 Deﬁnition of a structured product . . . . . . . .2 Portfolios and results . . . . . . . . . . . . . . . . .4. . . . . . . 4. . . . . . . . . . .3 Analysis and summary . .1 Market risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2 Naive fund construction number 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . .3 Purpose . . . . . .1 Capital guarantee . . . . . . . . . . . . . . .2. . . . . . . . 2. . . . . . . . . .1. . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . B& . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . .2 PCA results . . . .1 Data set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .Absolute lower bound 4. . . . . . . . . . . . . . . . . . . . . 5. . . . .2 Deﬁnition 2 . . . .4. . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. .2 Rolling portfolio weights . . . . . . . . . . . .4 Monte Carlo simulation . . . . 6. . . . . . . . . . . . . . . . . . . . . 6. . . . . . . . . 6. . . . . . . . . . . . . . .4. . .4. . . . . . . . . . . . .Additional scenarios B Option portfolios . . . . . . . . . . . . . . . . . . . . . .3 Historical simulation . . . . . . . . . . . . . . . . . . . .6 Summary . . . .4. . . . . . . . . . . . . . . . . . . . . . . . .4. . . . .Arbitrary start yield curve M. . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . . . . . . . . . . . .Additional scenarios . . . . . . . . . . . . . .1 Fixed portfolio weights . . 6. . . 47 51 51 52 53 53 53 54 54 55 55 56 58 65 65 69 71 72 73 75 76 78 80 81 83 85 91 91 93 95 99 101 5 Modeling Financial Assets 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Backtesting . . 6. .9 Structured products fund 2 . . . . . . .5. . . . . . . . . . .3 Observed paths .5. . . Hveem viii (103) . . . . . . . . . . . . . . . . . . . . .5 Transaction costs . . . . . .3 Benchmark fund . . . . . . . . . . . . . .4 Dynamic portfolio weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 4. . Alternative deﬁnitions of capital protection 4. . . . . .4. . 6 Portfolio Optimization 6. . . . . .1 Deﬁnition 1 . . . . .2 Dependence of the initial yield curve . . . . . . . . . . . . . . . . . . . . .4 A PCA model for the yield curve and index 5. . . 6. . . . . 5. . . . 6. . . .Additional portfolios C SPF vs Benchmark . . . . . . . . . . . D Correlations between the market index and the yield curve E Optimization problem . .CVaR . . . . . . . . . . . . . . . . . 5. . . . . . . . . . . . . . .4. . . . .3 Yield curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. . . . . . . . . .2 Index . . . . . . . . . . . . . . . . . . . . . . . . . .1 Volatility and option pricing . . . .1 Modifying the Korn and Zeytun framework 6. . . . . . . . . 6. . . . . . . . . . . . . . 5. . . 7 Conclusions Bibliography Appendices A Naive fund constructions . . . . . . . . . . . . . . . . . . . . . . . . . . .4. .

a hedge fund. 2010. Many investors on the Swedish market of structured products are small private investors and thus not willing to commit most of their capital in one investment category (Shefrin.CHAPTER 1 Introduction Structured products are a structured form of investment vehicles that consists of a bond and a market exposed ﬁnancial instrument. similar to the one of Markowitz’s (1952. 1 . [30]) seminal work on basic portfolio management theory. The other major risk for an investor is market risk. 2010.2 Capital guarantee The concept of market risk and diversifying using a set of numerous assets is very appealing to an investor. [31])). One reason for their popularity is that they possess the property of capital guarantee. the underlying and an option on the underlying as the investment universe in a one period model (as in Martellini et al. This setup is more relevant for an investor that must rebalance its portfolio at diﬀerent intermediate time points. 1. 2002. [35]). which is inadequate to describe the investment universe for e. Most studies in the area have been conducted by using a structured product. (2005. and other structured products. Market risk imposes a demand for applying a diversiﬁcation framework. using a longer investment horizon than the time to maturity for the structured product. Johansson and Lingnardz. An investor can by investing in structured products participate in the market besides the capital guaranteed part (the bond). especially if the investment is capital guaranteed. [32]). 1. was performed by Korn and Zeytun (2009. By creating a fund based on structured products it is possible to decrease the required amount invested for each individual investor and practically gain the same advantages as if the investor had the possession itself. [19]. [25]). a bond. and especially when the investor is allowed to invest in more than just one structured product. Hence to oﬀer the advantages of diversiﬁcation in structured products to small investors the concept of a fund based on structured products emerges. Thus many investors have a belief that these products are safe and do not realize the impact of credit risk. The main issue is that the portfolio based on several structured products with diﬀerent maturities will not be capital guaranteed.1 billion SEK in Sweden alone during 2009 (My News Desk.1 Market risk It has been shown in studies of the Swedish market for structured products that an investor using structured products in a diversiﬁed portfolio can achieve a fair return during bull markets and take advantage of the capital protection property (to some extent) of the structured products in the bear markets (Hansen and Lärfars. [27]).g. These studies indicate that there are advantages of diversiﬁcation when investing in structured products against market risk. applicable to structured products. 2010. The popularity of investing in Equity-Linked Notes. has been signiﬁcant exceeding 55. Thus investors are protected in bear markets and participate in bull markets. the key to avoid market risk is by diversifying. The ﬁrst study conducted on the subject.

The focus lies on developing an alternative deﬁnition of capital protection.3 Purpose The purpose of this thesis is to investigate if it is possible to construct a portfolio based on structured products that possesses the property of capital protection. movement of the underlying etc. with an investment horizon of τi years. when the investor buys a share of the fund. Consider that the fund should not only be guaranteed during one investment horizon. Introduction When. The objective is M.16%. the fund’s NAV consists of options in a high extent. 1.q. hence the simple naive fund construction is not capital protected. hence the fund is not capital guaranteed. in the setting of structured products.. 120 Price Underlying NAV Fund 100 Start of investment X: 24 Y: 92. Hveem 2 (103) . It is obvious that it is diﬃcult to gain the property of capital guarantee in combination with the possibility of a high return. each new investor wants to have the property of capital guarantee over its own investment horizon. Thus if the fund should be capital guaranteed for each investor.015 p. In the setting of a fund the investment horizons will instead be overlapping and of approximately one to ﬁve years. after quarter 12 the underlying has −5% in return quarterly. since all structured products can have a value of zero before maturity due to changes in interest. σ = 0. As the price of the underlying goes up the value of the fund increases since the options gain more in value. 12 at a time) with a maximum time to maturity of three years. the return of the portfolio of every single τi year long time period should be positive. Also if products have maturity prior to the investment horizon’s end the issue of how the payoﬀ should be reinvested arises. r = 0. During this time period the fund value decreases with 7. thus the capital guarantee should be measured during each of these investment periods. when the underlying has gone up a total of 12. since there obviously does not exist any structured products fund (SPF) fulﬁlling the requirements of absolute capital guarantee.84 80 60 $ 40 20 0 0 5 10 15 20 25 Quarter Figure 1. The investor in this example is an investor with an investment horizon of three years and holds only capital guaranteed products (a total of 23 products.1: The underlying has 1% in return each quarter until quarter 12. Problem arises when the fund holds positions in products with maturity beyond the investment horizon. This implies that after twelve quarters. at maturity (when disregarding the impact of defaults).68%. Thus each product is capital guaranteed but the fund is not capital guaranteed over the three-year time horizon. which makes the investment more risky and erasing the capital guarantee. but several.Chapter 1.15. The example in Figure 1.1 illustrates the value of a fund that consists of rolling capital guaranteed Equity-Linked Notes (notional amount of the bond equals to the price of the product) each quarter with a maturity of three years. a product is capital guaranteed it means that the investor will at least receive the invested amount.

the deﬁnition of capital guarantee. The thesis focuses on market associated risks and will not evaluate the impact of credit risk on the portfolio choice. Since the thesis will only consider investing in products with the same underlying the importance of using a scheme based on Black and Litterman’s (1992) studies in [8] is reduced. 1. with a decent expected return (should be competitive in relation to competing funds). how the portfolio should be allocated initially to attain as high degree of capital protection as possible. Chapter 4 covers the issue of capital guarantee. In the last chapter conclusions are drawn regarding the results and recommendations for future studies are given. The chapter covers everything from naive fund constructions to benchmarking with potential competition.1. where a modiﬁed Korn and Zeytun framework is recommended.4. thus exploring the impact of active subjective views on the investment choice is left for future research. Chapter 5 covers how the ﬁnancial assets are modeled and discloses the details regarding the PCA. In Chapter 3 the deﬁnition of a structured product is covered and how the individual components are priced. [38]). Hveem 3 (103) . this is usually how the terminology capital protection is used in the concept of structured products. such as the theory of risk measures.4 Outline Chapter 2 covers the most fundamental basics which the thesis has its foundation in. Chapter 6 covers three proposed allocation schemes. scenario based optimization and principal component analysis. A scheme to minimize the downside risk is also described in the chapter. should be constructed. It is not necessary for the reader to go through this chapter. Thus when investigating if the products are capital protected the impact of defaults is neglected. The investigation is performed numerically based on simulating trajectories of future price patterns for available structured products and by solving an optimization problem allocating the portfolio based on CVaR constraints as in Uryasev (2000. Outline also to investigate how a fund with as low downside risk as possible. M. as well as the possibility of multiple underlying. since it is not imperative to understand these theories to understand the result.

.

Since then it has been shown that assets’ log returns are not multivariate normal distributed and that the distribution of stock returns often exhibit negative skewed kurtosis (Fisher. the set of acceptable portfolios must fulﬁll these requirements. [3]). where Xt is the value of the portfolio at time t. i.74. Such extreme events prove that measuring risk with the variance is not adequate. Conditional-Value-at-Risk (CVaR.1 Risk measures The concept of risk has been around ﬁnance for more than ﬁfty years. 5 . a 22. a percentage of the initial investment) c. Markowitz introduced his seminal work within portfolio theory during 1952 [30] where he used standard deviation (volatility) as risk measure to ﬁnd the optimal tradeoﬀ between risk and return. skewed distributions such as Value-at-Risk (VaR). [14]). To ensure full clarity for the reader matrices are written as bold upper case characters. 1999.1 Absolute lower bound An acceptable portfolio is a portfolio which ﬁnal net worths are those that are guaranteed to exceed a certain ﬁxed number (e. Notable. The chapter will cover everything from risk measures and principal component analysis (which is an integral part of the modeling) to scenario based optimization and transaction costs. By understanding the concepts in this chapter it is also easier to understand how to replicate the results and also the assumptions aﬀecting the results. not only in this chapter but also throughout the whole paper. which is not as commonly used as VaR and CVaR is the absolute lower bound.e. in general.CHAPTER 2 Theoretical Background This chapter covers the theoretical background that the reader should be familiar with to understand how the study is conducted.61% decrease (Browning. closely related to Expected Shortfall) (Acerbi. not even for stock returns.g. Further the net worth of the fund portfolio at time T is denoted as X = XT − Xt . the Black Monday 19th October when the Dow Jones Industrial Average dropped by 508 points to 1738. which gives the risk measure. Some risk measures have been introduced to capture these heavy tailed. vectors as bold lower case characters and transposed with superscript T. symmetric and variance (even if stock returns are relative symmetric) is thus not an adequate risk measure. it is not necessary to understand the concepts discussed in this chapter to understand the results of the study. especially around extreme events such as the 1987 stock market crash. 2. Another risk measure.1. [11]). ρ (X ) = min {m : m (1 + rf ) + X ∈ A} = min {m : m (1 + rf ) + X ≥ c} . which measures the worst-case outcome. 2002. 2. 2007. A portfolio containing derivatives is not. A = {X ∈ X : X ≥ c} .

Hveem 6 (103) .. [10]) and does not take into account the size of the losses exceeding VaR. m) = P (Lw ≤ m) . 2000. thus traders have the possibility to hide risk in the tail. [5]). but it is much more complex than the one proposed by Uryasev et al. • Where Lw = −Rw . notice that. VaR is a non-convex and non-smooth function which in many cases has multiple local maximums and minimums. 1] by. since the model underestimates the risk/implication of market crashes (Brooks and Persand. VaRα (Lw ) = min {m ∈ R : ψ (w. 2. When regarding absolute capital guarantee the portfolio satisﬁes ρ (X ) ≤ 0 (Hult and Lindskog. 2009. and the return is deﬁned as. or. since it only measures the outcomes at the quantile α. Also Alexander et al.Chapter 2.1. ﬁnal wealth − 1. which is much easier to implement in portfolio optimization as proposed by Uryasev et al. Thus VaR is given as (a similar notation is given in [27]). 1 + rf Thus the deﬁnition is regarding the worst possible outcome of the position at time T . [23]). m) ≥ α} . [4]) have recently developed an optimization scheme which is very eﬃcient for minimizing CVaR and VaR for portfolios of derivatives. ρ (X ) = min {m : m (1 + rf ) + X ≥ c} = c − x0 . thus it is very hard to construct a portfolio optimization scheme which is eﬀective enough to still be robust and give valid results (Uryasev. Rw = initial wealth There exists several areas of critique against VaR. 38]. in [28. Also it is not very useful for distributions that have heavy tails. 2006. 33. and Alexander et al. [23]). Many blame Value-at-Risk of being an integral part why a ﬁnancial system may fail. Also. 2009. where Rw is the return associated with the portfolio vector w. • Let Lw denote the loss of the portfolio with the portfolio weights w. M. • Then Value-at-Risk VaRα (Lw ) is deﬁned as the loss with a conﬁdence level of α ∈ [0. Theoretical Background x0 = x0 (X ) is denoted as the smallest value that X can take.’s theory will not be covered in this paper. (2006. one of the most common critiques is that it is not a coherent risk measure (Hult and Lindskog. ψ (w. The idea of VaR is that VaRα (Lw ) is the value that the portfolio’s loss will be less than or equal to with a probability of α.2 Value-at-Risk Value-at-Risk (VaR) is one of the most important concepts within risk management and is also regulated by the FSA in many countries and by the guidelines from the second Basel Accord regarding minimum capital requirements (BIS. Instead of VaR many authors propose Conditional Value-at-Risk (CVaR). 2000. and the probability of Lw not exceeding a threshold m as. VaRα (Lw ) = min {m ∈ R : P (Lw ≤ m) ≥ α} . [38]).

CVaRα (Lw ) ≤ C. y) − β] p (y) dy. 2010. x∈X min −R (x) . β) reduces to a single point.1) and. Fα (w. (2. s. for the proof and more detailed information please consult [28. it is deﬁned as. (2002.e. [6]. A whole yield curve can often be reduced into only three factors.2 instead. Thus the number of factors to simulate can be reduced drastically by reducing the number of risk factors to model. then (w∗ . Hveem 7 (103) . β).3). (2. x∈X min −R (x) . Principal component analysis 2.2 Principal component analysis Principal component analysis (PCA) is a common tool to generate scenarios for changes in the yield curve and returns for other types of assets. Thus it is possible to solve the optimization problem 2.1. in the CVaR case.1. s.e.1 is active. CVaRα (Lw ) = E [Lw | Lw ≥ VaRα (Lw )] . thus PCA is a useful tool to generate scenarios that are parsimonious.2.t. The optimization problem reduces to a linear optimization problem with linear constraints. CVaR is a coherent risk measure and more adequate than VaR since it discloses the expectation of the loss if the loss exceeds the VaR. The goal of portfolio optimization is often to maximize the return subject to a demand on the maximum risk acceptable.1 and β ∗ ∈ arg minβ∈R Fα (w. When β ∗ ∈ arg minβ∈R Fα (w. Litterman and Scheinkman. then β ∗ gives the corresponding VaR with conﬁdence level of α (Korn and Zeytun. 38]. which is an important concept in risk management. Studies have shown that by using principal component analysis only two or three principal components are often enough to describe more than 95% of the variation in a yield curve (Barber and Copper. M. If the CVaR constraint in 2. 1991. the second greatest variance by any projection on the second coordinate and so on. [29]). The idea of PCA is that data sets of intercorrelated quantities can be separated into orthogonal variables. β) = β + 1 1−α [L (w. The advantage with a linear optimization problem is that it can be solved using the simplex method. Fα (w. y ∈ Rm is the set of uncertainties which determine the loss function. β ∗ ) minimizes 2. Krokhmal et al. 2. 2009. ﬁrst principal component). since it generates the same solution as 2. The main idea is to transform the data to a new orthogonal coordinate system such that the greatest variance by any projection of data comes to lie on the ﬁrst coordinate (i.2) give the same minimum value where. [28]) have shown that the solutions to the optimization problems. or so-called principal components. y∈Rm + CVaRα (Lw ) = min Fα (w. 33.2. β) .2 if and only if w∗ minimizes 2. which is easily transformed to a linear optimization problem (disclosed in Section 2. β) ≤ C. β∈R and L (w. y) is the loss function associated with the portfolio vector w. i. [27]).3 Conditional Value-at-Risk CVaR is based on the deﬁnition of VaR since it is the expected loss under the condition that the loss exceeds or equals the VaR.t. which explains the variance and dependence in the data in a simpler way.

. M. k. 29. in which each variable is divided by its norm.m ..Chapter 2. making RT R a correlation matrix (most common in statistical packages such as MATLAB and R). is investigated. without loss of generality. Only the variables that add important information to the sample are interesting and withdrawn. thus.. correlation PCA will be used in this thesis (Abdi. The columns in Q. An orthogonal matrix has the property QT = Q−1 . Theoretical Background Consider a setting with m assets.. Cov FT = E QT RT RQ = QT Cov (R) Q = QT RT RQ = Λ. which are orthonormal. thus the components of F are uncorrelated and have variances λ1 ≥ . K ri. + QT Fi..1 + . thus QT Q = QQT = I. It can be shown that the returns R are uncorrelated expressed on the orthogonal basis. It is also common with correlation PCA. in that order. To investigate the contribution of λ variability of each PC the ratio m i λj is studied. (2010) in [24]. 2010..j = k=1 QT Fi. which is shown by Hult et al. For more information regarding PCA and yield curve modeling please advise [2. 36]. thus the PCs that add the most variability. n number of observations and that the asset returns are ˜ given as a n × m matrix denoted R. ∆ is a diagonal matrix and ∆2 equals the diagonal matrix Λ containing the (positive) eigenvalues λ1 . λn to RT R (since RT R is a positive semi-deﬁnite matrix).. The sample must also be rescaled with its standard deviation and mean.j m.. Q is a m × m orthonormal matrix (called loading matrix and each column corresponds to one PC) and ∆ is a n × m matrix with non-negative values on the diagonal. The idea of PCA is as mentioned to reduce the number of variables needed to describe the data. R = P∆QT .k + εi. this type of PCA is referred to as covariance PCA since the matrix RT R is a covariance matrix (the covariance matrix of the returns). [2]). Note that.. It is possible to assume.. Hveem 8 (103) . Q is an orthogonal matrix. . that the columns of Λ and Q are ordered such that the diagonal elements in Λ appear in descending order. which implies that the ith observation of the jth original variable is expressed as follows. which are linear combinations of the original variables and are deﬁned in a way such that the amount of variation associated with them are in decreasing order and orthogonal to each other. 1. where P is a n × n orthonormal matrix. Let denote the factor scores F (observations of the principal components) as. ˜ By using singular value decomposition R can be written on the following form. RT R = P∆QT T P∆QT = Q∆T PT P∆QT = Q∆T ∆QT = Q∆2 QT . as when modeling the yield curve. It is important to center the returns/data by subtracting its √ mean to perform the PCA and also normalize the data by dividing with (n − 1) (or by n).j = QT Fi. Thus the returns are given by. ≥ λm . It is most often possible to describe the whole j=1 dependence structure for a data set with just the ﬁrst K PCs.qm are the corresponding eigenvectors of RT R. The PCA creates new variables called principal components. R−µ Let denote R = √n−1 . F = P∆ = P∆QT Q = RQ.j Hence to simulate new returns. ri. q1 . simulate factor scores (PCs) and use the factor loadings to calculate the return.j . F = P∆.j Next RT R.

As the concept of skewness and kurtosis has become more prevalent the importance of scenario based optimization has increased. e. N .3 Scenario based optimization There are in general two approaches to portfolio optimization problems: mean-variance and scenario optimization. copulas or autoregressive (AR) models. • Representative . [34]) states that the scenarios must be: • Parsimonious . The idea of scenario based optimization is to turn a stochastic problem into a deterministic problem by simulating future scenarios for all available assets. Thus one of the most important uses of scenario based optimization is that it actually allows derivatives/options to be part of the product mix.as few scenarios as possible to save computation power. 1999. Step 2. which is not in general the case with a mean-variance optimization (Grinold. Since bootstrapping is done by repeated independent draws. The parametric distribution may be attained by ﬁtting a statistical distribution to a historical sample of data.e. [30]). [17]).the scenarios must be representative and give a realistic representation of the relevant problems and not induce estimation error.000 yearly returns the user draws for example 12. or time.g... When bootstrapping historical empirical data. Deﬁne a linear optimization problem on those simulated paths. which can be solved by mathematical linear techniques. Simulate S paths of returns for the assets i = 1. Important to mention is that the optimization is totally dependent on the scenarios. • Arbitrage-free .scenarios should not allow arbitrage to exist. . thus with the wrong assumptions or scenarios the result will most likely be sub-optimal. The problem stops being stochastic when the scenarios are generated and the problem is (most often) transformed to linear form. There exist several diﬀerent methods of simulating data. with replacement. also known as historical simulation. To gain a qualitative solution adequate scenarios must be generated. thus it is imperative to use scenario optimization when allocating amongst the assets. If the assets are independent of each other the user can ﬁt an individual distribution to each asset. but destroys autocorrelation. instead of drawn from the empirical distribution. in particular Scherer (2004.000 samples of monthly return from the empirical distribution.g. drawing samples from a parametric distribution. if the user is trying to simulate annual returns the user may draw 12 monthly returns from the empirical distribution. Scenario based optimization 2. Notable is that bootstrapping leaves correlation amongst the samples unchanged.. A popular way of simulating from a parametric distribution is by using e. thus if the user is simulating 1. i. The draws are performed with replacement. This thesis considers portfolios of structured products.3. Hveem 9 (103) . two of these are bootstrapping historical empirical data and Monte Carlo simulation. the user draws random samples from the empirical distribution. the data will look increasingly normal as the number of samples increases. To solve a scenario based optimization problem there are usually two steps to consider: Step 1. which can be solved using the simplex method.2. Monte Carlo simulation is similar to historical simulation where the sample is. also many others have been well awarded for their contributions within this ﬁeld. drawn from a parametric distribution. In a scenario based optimization N available assets to allocate in are considered with the added feature of S possible return scenarios. One of the frontrunners within mean-variance optimization was Markowitz (1952. thus generating an annual return. M.

.3. 2002. w RT.. w RT. max 1 S S w RT. 38].. Thus the simulation based CVaR problem is a problem that is relative well deﬁned since the increasing power of today’s personal computers enables the possibility to solve these problems to a reasonable cost. S + β + zk ≥ 0. ..k = w1 RT. w1 + .k . w1 + . The index k corresponds to which scenario...Chapter 2.z. wi ≥ 0.3.k . 1 S S k=1 zk ≥ 0. for more details regarding how the problem is transformed please advise [28. RT. k=1 zk ≥ 0. S i = 1. The problem can be solved using the simplex method.. S + β + zk ≥ 0.k + .. Hveem 10 (103) .wN = 1... It is possible to M..k . . k=1 such that: w 1 N RT... S k = 1.z. S i = 1...k = w1 RT...k ≥ Rtarget . α is the conﬁdence level of CVaR.wN = 1. + wN RT. min β + 1 S (1 − α) S w. which is to prefer due to its eﬃciency.k is the return for asset i for scenario k until time T and wi the weight held in asset i (Krokhmal et al.. k=1 w.k + ..β zk . 1 S (1 − α) S β+ zk ≤ C. S is the number of simulated paths and N the number of assets. . k = 1.. k = 1. wi ≥ 0... resulting into the following linear approximation based on scenarios.. . S k = 1. 33. .... N where.. + wN RT. the index i corresponds to which asset. The linear CVaR problem based on scenario simulation is deﬁned as follows. It is also possible to write the portfolio choice problem with the CVaR as minimization objective. .k k = 1.β such that: w 1 N RT.2 Minimum regret Minimum regret is an optimization scheme based on scenario optimization... β is a free parameter which gives VaR in the optimum i solution of the CVaR problem.. w RT.. Theoretical Background 2..k k = 1. 2.. [28]). . N Notable is that the two diﬀerent problems generate the same eﬃcient frontier..1 Conditional Value-at-Risk The CVaR problem can be (as mentioned earlier) converted to a linear optimization problem. which maximizes the least favorable outcome of S scenarios given a certain demand on return.

..) Scherer (2004..k + . T By using this scheme the downside is restricted. k = 1.k be the expected return vector over all scenarios thus. . market impact (volume etc. Transaction costs formulate the optimization problem as a linear problem based on scenarios and it is deﬁned as follows (Scherer. 2. wi ≥ 0. . [34]). restricting turnover and/or trading constraints. wi as a positive weight change and wi as a negative weight change (asset sold). thus having an impact on the result. which are deducted from the return.wN = 1 wi ≥ 0. i = 1.k such that: 1 N w1 RT.k ≥ Rmin . this type of allocation is preferable for really risk-averse investors since they know the extent of their worst outcome. There are many types of optimization problems incorporating transaction costs and only those relevant to the study will be discussed. w∈RN max wT RT.wN = 1.. [34]) suggests that transaction costs. There are in general two diﬀerent approaches to transactions costs. + − initial Thus the weight invested in asset i after the reallocation is given as: wi = wi + wi − wi ..g.. It is possible to modify the setup to an equivalent optimization scheme that is deﬁned as follows.k + . are of the following functional form.. The most common transaction costs are such as brokerage commission. S w RT. S i = 1. which is not always available.. initial Let wi be the weight invested in asset i.. .2. The problem with this model is that data is needed on daily trading volumes..4 Transaction costs Transaction costs are important to take into consideration since they can change the proﬁtability of an investment.. the direct approach restricts the actual cost from happening by introducing actual transaction costs..k ≥ Rtarget . + wN RT. k = 1. Let Rmin . Hveem 11 (103) . An investor will have to pay transaction costs every time it rebalances its portfolio due to several factors. thus preventing the reactive transaction costs.4. w∈RN max Rmin such that: 1 N w1 RT. e. tc. N w1 + . Instead an appropriate way to estimate transaction costs is by separating the costs into one ﬁxed and one linear part. ..k ≥ Rmin . be the worst possible outcome and RT. Daily volume The bid-ask spread is expressed as a percentage and θ is a constant that needs to be estimated from the market. bid-ask spread.... N. + wN RT.. tc = Commission + Bid -Spread + θ Ask Trade volume . The second is to put up restrictions upon actions that have transaction costs linked to them.. w1 + .. M.. 2004. wi the weight invested in asset i prior to the + − reallocation.

n n wi + i=1 i=1 + + − − T Ci wi + T Ci wi = 1.. n n By introducing transaction costs the reward function is changed from i=1 wi µi to i=1 wi (1 + µi ) n since i=1 wi ≤ 1... k=1 w...k k = 1.. N wi + i=1 i=1 + + − − T Ci wi + T Ci wi = 1. .. Hveem 12 (103) .. The budget constraint in the traditional portfolio optimization problem is i wi = 1 which now must be modiﬁed since the transactions have to be paid out of the existing budget. 2002. + − initial wi = wi + wi − wi .. + wi ≥ 0. N i = 1. . ﬁxed transaction costs is one of the most common type of transaction costs in the market. thus instead the budget constraint is as follows.. To be able to use ﬁxed transaction costs integer variables ± ± δi must be introduced.. Let + − T Ci be the proportional transaction cost associated with buying asset i and T Ci be the cost associated with selling asset i. Besides proportional transaction costs. N i = 1.. which takes the value one if trading takes place in asset i (wi is positive) and zero otherwise. max + − 1 S S w 1 + RT. The ﬁxed transaction costs are not dependent on the trade size. The linear CVaR problem based on scenario based optimization and proportional transaction costs can be formulated as follows. Fixed Transaction Costs Fixed transaction costs arise as soon as a particular asset is traded. . N i = 1.. − wi i = 1.Chapter 2. S + β + zk ≥ 0...000 will generate the same cost. . which means that the transaction costs are proportional to the amount bought or sold of the asset. S k = 1. N zk ≥ 0. [28]). thus a trade of $1 and $1. S k = 1. Also most of the transaction costs in the market are proportional. the new budget constraint is given as. thus making it a good and adequate choice (Krokhmal et al. .k .w . . N ≥ 0.... w RT.w .000. wi ≥ 0. Theoretical Background Proportional Transaction Costs One of the most common types of transaction costs is proportional transaction costs.. Including proportional transaction costs. k=1 w 1 N 1 + RT..β such that: β+ 1 S (1 − α) S zk ≤ C.k = 1 + RT. Proportional transaction costs are popular to model with since they do not add a lot of complexity to the optimization program.. n n n + + − − F Ci δi + F Ci δi + i=1 Fixed TC i=1 Proportional TC + + − − T Ci wi + T Ci wi = 1.z.k wN . wi + i=1 Holdings M... . + 1 + RT.k w1 + ..

k=1 w 1 N 1 + RT. to a mixed integer linear program. .... S k = 1. N i = 1.. and wmax is a large number.. ± δi ∈ {0.... k=1 such that: 1 β+ S (1 − α) w RT.. Transaction costs where.. Hveem 13 (103) .z.. ..2... to solve thus making it inadequate for many scenarios... w.. N 0≤ ± δi − wi ≤ − δi wmax . which can be solved using the simplex method. + + 0 ≤ wi ≤ δi wmax .w+ ... S + β + zk ≥ 0.k .w . .k S zk ≤ C. N wi + i=1 N i=1 − − + + F Ci δi + F Ci δi + + i=1 + + − − T Ci wi + T Ci wi = 1. which transforms the linear optimization problem. Hence it is preferable to consider an optimization program that does not contain binary or integer variables. 1} ... A mixed integer linear program has a higher complexity and takes more computation power.k w1 + .δ . − − wi ≤ δi wmax . + 1 + RT.. . N i = 1. i = 1. k = 1. .k wN .. ∈ {0. 1} . + − initial wi = wi + wi − wi .β max − + 1 S S w 1 + RT.k = 1 + RT. . ± The problem of using ﬁxed transaction costs is associated with the integer variables δi .4. + + wi ≤ δi wmax . . N i = 1...δ − . S k = 1.. . The mixed integer linear program for CVaR as constraint and the return as target function is deﬁned as follows. wi ≥ 0. n zk ≥ 0. or time. M. N i = 1.

.

thus this market is relative theoretical. etc. in many cases. The bond is a theoretical asset where the default probability of the issuer is zero. look-back options. According to Martellini et al. as underlying is used. Today the most common setup for a structured product is the combination of a zero-coupon bond and an option. Since only the option part is exposed to the market. The ETF is assumed to follow the index perfectly. In this thesis three types of assets are available for the investor. In the future when referring to a structure product this thesis will refer to an Equity-Linked Note. equity. what types of instruments that are available and how they are priced. thus structured products that posses a minimum of 100% capital guarantee are often called Principal protected notes (PPN). rainbows. Monte Carlo based option pricing methods. Thus the investment is capital guaranteed since the minimum amount that the investor (conditional on that the issuer of the zero-coupon bond has not defaulted) receives is the price at time t = 0. Later on this has spurred the development of more exotic structures and creative constructions.1 Deﬁnition of a structured product Structured products are in general synthetic investment instruments that are created to meet special needs for customers that cannot be met by the current market. Note that the bond’s face value is deterministic and the payoﬀ of the option is stochastic. This is usually achieved by constructing a portfolio consisting of securities and derivatives.1 is of a structured product that consists of a zero-coupon bond and an option. It is not necessary to read the chapter for someone who is well familiar with structured products. The example in Figure 3. [31]) was the ﬁrst structured form of asset management the introduction of portfolio insurance such as the constant proportion or option based portfolio insurance strategy. thus there exists no standard product. Hence there are endless combinations of possible structured products. 3. The main reason for such constructions is to ﬁt the investors’ risk and aspiration preferences. The idea of the structured product is usually that most of its value is contributed by the 15 .. These needs are often focused on low downside risk on one hand and the possibility of growth on the other. path dependent options and barrier options (HSBC. Investment structures that possess the property of capital guarantee are usually called notes. these products are usually called Equity-Linked Notes (ELN). (2005.CHAPTER 3 Financial Assets This chapter covers the basics behind the ﬁnancial assets. The structured products are combinations of bonds and derivatives with the index as underlying. the products are often also called capital guaranteed investment vehicles.g. a risky index (an Exchange Traded Fund (ETF)) and structured products (Equity-Linked Notes) with the risky index as underlying. e. The type of option varies widely and most often a plain vanilla option with a certain equity index. [21]). currency. these are: a risk-less asset (bond). These products will not be used in this thesis due to their computational heavy pricing models involving. Examples of other types of options used in structured products are such as basket options. 2010. The notional amount (amount paid at maturity T = 1) of the zero-coupon bond equals the price of the structured product at time t = 0.

in practice it also depends on the fees taken by the issuer).1: Structure of a structured product. 0). C Thus the amount paid out to the investor at maturity. in a regime with low interest rates the zero-coupon bonds are expensive and thus the amount left for buying options is relative small. k= S0 − B0 .1 Dynamics Equity-Linked Notes in this thesis have notional amount of the bond and strike price equal to the price of the underlying when issued. S0 Bond part + k max (ST − S0 .0 . T S0 ST k C B r as as as as as as as the the the the the the the time of maturity value of the underlying at the time of issuance value of the underlying at maturity participation rate price of the option price of the zero-coupon bond with notional amount S0 risk-free interest rate Thus the payoﬀ at maturity of the structured product can be written as. the options are plain vanilla at-the-money call options. is given by. Thus diﬀerent constructions are available depending on the current market climate. in dollars per dollar invested is given by. The leveraged exposure to the underlying is called participation rate (HSBC. 1 + k max ST − S0 . The relation between how much is invested in the zero-coupon bond and the option depends on the price of the zero-coupon bond (in theory.Chapter 3.e. 2010. time t = 0 is time of issuance and T = 1 is the time of maturity. the bond payoﬀ is deterministic and the option payoﬀ is stochastic. Let denote.g. e. [21]).1. Financial Assets Structured product 150 Option Payoff (Stochastic) 100 Proportion [%] Option Value Bond Value 50 Bond Value (Deterministic) 0 0 Time 1 Figure 3. S0 16 (103) M. zero-coupon bond. i. Option part where the participation rate. Hveem . The participation rate is measured in percentage (or you might call it number of contracts). k. 3. T .

the only cash ﬂow generated from a zerocoupon bond is the notional amount (face value) paid at maturity. the volatility σ and the participation rate k (rt .2. Bt = e−rt (T −t) . K. The option prices are calculated with Black-Scholes formula as disclosed in Section 3. X = max (ST − K.3. with maturity T and notional amount $1 is given by. σ. thus increasing the participation rate. As the volatility increases the price of the option increases.r) 100 90 80 Participation rate k [%] 70 60 50 40 30 20 10 10 6 0 50 4 45 40 35 30 25 2 20 15 10 5 0 8 Volatility σ [%] Interest rate rt [%] Figure 3.e.3 Plain vanilla call option A plain vanilla call option. k=k(σ. The price at time t of the European call option is denoted C (St .3.2: The surface plot shows the relation between the interest rate rt . r the risk-free rate. where rt is the risk-free rate at time t. Hveem 17 (103) . K the strike price. T. time T . 3. Notable is that k is increasing in r and decreasing in σ (volatility) as seen in Figure 3. σ) for T = 1. the option will have the following payoﬀ X at time T .2. to buy the underlying asset S for the predeﬁned strike price K. 0) . Since the holder does not have the obligation to exercise the option. Thus using a continuous compounding for the interest rate the price of a risk-less zero-coupon bond at time t. When the interest rate increases the price of the zero-coupon bond decreases. T the time of maturity. M. δ) where St is the price of the underlying at time t. t. k controls how much the investor will participate in the market. r. thus decreasing the participation rate. not the obligation. also called European call option is a product that gives the holder the right. Zero-coupon bond i. 3.2 Zero-coupon bond A zero-coupon bond is a bond that pays no coupons. at maturity. The price of the asset S at time T is denoted as ST .

The study compared the theoretical value of the structured products (using both realized and implied volatility) with the observed price at the primary and secondary market. Wasserfallen and Schenk (1996. 3. 2009. d1 (t. 2009. Hull. St )] . It is important to understand the underlying assumptions of Black-Scholes formula. 1973. [9]. σ. practitioners usually use the so-called implied volatility instead.3. Since the thesis is regarding structured products the pricing of these options is only relevant in the context of structured products.e. 2002. 2009. The Black-Scholes market model under the probability measure P is given as (Björk. [20]) and Hagan et al. s) − σ T − t. C (St . Empirical evidence has shown that asset returns are skewed and exhibit kurtosis. r. Instead many authors such as Heston (1993. dS (t) = µS (t) dt + σS (t) dW (t) . S (0) = s B (0) = 1 where W is a Brownian Motion (i. K 2 √ d2 (t. T. [22]). M. These are such as that the stock returns are lognormal distributed. where. K. [39]). dB (t) = rB (t) dt. X) = St e−δ(T −t) N [d1 (t. Π (t. Notable is that the transform under this market model to the risk-neutral probability measure Q only transforms the drift µ to r by using Girsanov’s theorem. µ the drift and σ the volatility. which is given below (Black and Scholes. constant volatility. Hence Black-Scholes formula is an adequate choice for pricing the ELNs. X) .Chapter 3. since it is reasonable to assume that the dividends are spread all over the year for diﬀerent components of the index. constant risk-free rate. Since the Black-Scholes formula seems inadequate to price plain vanilla options using the original deﬁnition with realized (historical) volatility. It is common to price these options with Black-Scholes formula. 1996. δ) = Π (t. Thus the eﬀect of using realized volatility in comparison to implied volatility in the pricing of the structured products can be disregarded in this thesis since the error in comparison with the primary and secondary market is negligible (Wasserfallen and Schenk. t. Wiener process). The main area of criticism is the assumption of lognormal returns and thus the dependence of the normal distribution (Björk. s) = d1 (t. Implied volatility is the volatility so that Black-Scholes formula gives the correct market price. Notable is that the assumption of a continuous dividend yield is adequate in the setting of modeling an index. [7]). [18]) propose models that incorporate stochastic volatility and the volatility-smile.1 A note regarding the critique against B & S It is commonly known that the market model above has sustained a lot of critique since Black and Scholes proposed it in 1973 [9]. (2007. which implies that the volatility σ is the realized volatility observed at the market (Björk. St )] − e−r(T −t) KN [d2 (t. [39]) found that the prices of structured products are not aﬀected systematically by using either realized volatility or implied volatility. [7]).1) N is the cumulative distribution function for the standard normal distribution. and. etc. s 1 ln + r − δ + σ 2 (T − t) . [7]). Hveem 18 (103) . Financial Assets σ the volatility and δ the continuous dividend yield. s) = 1 σ T −t √ (3.

the neat construction of market participation and capital guarantee. to limit the downside risk. This is one main factor why people invest in structured products. Most Equity-Linked Notes are constructed such that the notional amount of the bond equals the issue price. If an investor should diversify its portfolio amongst structured products the invested amount increases to relative high levels. The chapter also covers a similar study conducted for option portfolios. several studies recent years have shown that structured products actually are a good investment choice for rational investors. As mentioned in the introduction.CHAPTER 4 The Capital Protection Property One of the most appealing properties of Equity-Linked Notes is the capital guarantee. Investors suﬀer from a syndrome called loss-aversion. The problem with structured products is that they usually require that the investor invests a minimum amount in each structured product. to limit the downside risk as much as possible. Thus the only reasonable way for small investors to diversify in structured products. a large proportion of these investors are small private investors. without requiring a huge amount of capital. would be to invest in a fund that is based on structured products. at issuance. 2002. [35]).g. First of all the deﬁnition of capital guarantee is discussed and how it relates to a fund. it is not as easy as that. Thus the investor is guaranteed to not loose any capital. This chapter covers a relative broad spectrum of topics. Investors do not. In the end alternative deﬁnitions of capital protection are discussed to decide under which risk measure the structured products fund should be allocated. with diﬀerent strikes and maturities (and by several underlying. three years. according to Shefrin (2002. A capital guaranteed product is considered to be capital guaranteed over a certain investment period. These constructions are simple forms of funds based on structured products that are used to evaluate how the fund should be constructed so that they carry as little downside risk as possible. e. which is excluded in this paper).1 Capital guarantee When describing a fund based on capital guaranteed products it is quite easy to believe that the fund itself also would be capital guaranteed. Thus the investment horizon is ﬁnite and ﬁxed for every investor. Problem arises when trying to construct a fund in a way such that the property of capital guarantee is retained. where it is investigated how an investor should allocate amongst ATM call options. The concept of a structured products fund is very appealing since the investor gains diversiﬁcation amongst the assets. the investor base is widely varying and most investors have diﬀerent investment horizons and 19 . The results from these two sections are then combined to investigate certain limits for the possession allowed in structured products. prefer to invest a huge percentage of their total capital in one single type of investment. 4. [35]).1 billion SEK in Sweden alone. as long as the investor uses a diversiﬁcation framework. since each product has a required minimum invested amount. The market for structured products is quite large and during 2009 the total market for ELNs was 55. A large part of the chapter is used to discuss so called naive fund constructions. which means that they often make irrational decisions just to avoid loosing any capital (Shefrin. When considering a fund.

This section covers two diﬀerent naive fund constructions and how changes in the diﬀerent risk factors aﬀect the return of the fund over an investment period. have a time to maturity of three years at issuance. The ﬁrst fund construction is a fund that is rolling capital guaranteed structured products. In the following sections it will be covered how a fund should be constructed to attain as much capital protection as possible. not simulation.2 Naive fund constructions This section covers naive fund constructions. The Capital Protection Property investment times. By understanding more of the factors aﬀecting the return of the naive fund. Thus it is important to notice the diﬀerence between capital guarantee and capital protection. it is possible to gain knowledge of how a portfolio should be allocated amongst diﬀerent structured products. a new structured product is issued each quarter.2. Thus it is now quite clear that a fund based on structured products will not hold the property of capital guarantee for all investors. Thus the fund is just rolling capital guaranteed products with twelve diﬀerent maturities. maybe not for even anyone. since these scenarios are the only interesting scenarios when regarding capital guarantee. Hveem 20 (103) . Thus a fund that uses the terminology capital protection in its marketing campaign only needs to show that some of the capital base is protected. All structured products. This means that for a fund to be capital guaranteed it needs to be capital guaranteed for all maturities and all overlapping time periods. The reason why simulation is not used is that the outcomes should be independent on the market model. It is now quite obvious that a fund that holds these properties is a fund practically only investing in the risk-free rate. three prior to the investment (since the price of structured products are path dependent) and three years after the investment. This includes naive fund constructions and investigations of option portfolios etc. as far as possible. 4. These diﬀerent changes are stressed through using predeﬁned scenarios. The new customized structured product has the same ratio of option:bond value as the fund prior to the rebalancing. 4. thus called naive fund constructions. thus maintaining the fund’s ratio options:bonds relatively intact. This implies that the terminology must be changed from capital guarantee to capital protection. Thus it takes twelve quarters until the ﬁrst product has matured. that is 100% capital guaranteed. Most ﬁnancial products can have a value of zero prior to their maturity based on diﬀerent market risk factors. the diﬀerence is signiﬁcant. Thus the M. these constructions are created to understand more of the dynamics of funds constructed of only structured products. not guaranteed. while smaller private investors may not notice the diﬀerence. than the ones disclosed in this thesis. The second fund construction is a fund that buys a standardized ELN at the start of the fund and buys a new customized structured product each following quarter. thus it is for simplicity assumed that the fund is launched to the public during quarter twelve. have been tested but only some the most unfavorable scenarios are disclosed.1 Naive fund construction number 1 A new structured product. It is assumed that investors have an investment horizon of three years. thus the value of the fund at intermediate time points may converge to zero with a positive probability. in this chapter. is issued every quarter where its price equals the notional amount of the bond as well as the price of the underlying. thus the scenarios will be for six years. Many more scenarios. which is not a desirable result. The naive fund constructions are probably the simplest possible funds based on structured products. Since capital protection is a more diﬀuse deﬁnition that only states that the capital is protected. to have as much capital protection as possible.Chapter 4. Hence the notional amount of the bond equals the price of the structured product at issuance. A new product is bought every quarter with the payoﬀ of the product that matures the same quarter.

i k c (St . if i ≤ t ≤ i + 11 if t = i + 12 otherwise Hence the return between t − 1 and t for the structured product issued at quarter i is given as. ri . i + 12. T. i i βt i Ot St Vt c (St . rt . ri . Hence the fund will hold a maximum of twelve products each quarter. 0. σi ) the value of the call option issued at quarter i at time t is given as. r.1 thus the following formulas describe the prices of the structured products. Si e−rt (i+12−t) .1. Naive fund construction number one is a fund that is rolling the available structured products. t. the structured products follow the dynamics given in Section 3. Every structured product matures after three years. the bonds are priced as.4. which means that the fund buys the newly issued structured product every quarter with the payoﬀ of the matured product (it is assumed that the fund can hold inﬁnitesimal fractions of the structured products). Below follows a more detailed description of the fund. The structured products follow the dynamics given in Section 3.2. if i ≤ t ≤ i + 12 i βt = 0. i. 0) . i + 12. otherwise the participation rate of the option issued at quarter i is given as. σ) ki f rt i rt rt σt i wt wt as as as as as as as as as as as as as the quarter the product was issued the value of the bond issued at quarter i at time t the value of the ATM call option issued at quarter i at time t the price of the underlying at time t the value of the fund at time t the value of a call option at time t with strike K. i. i i βt−1 + Ot−1 M. each product has maturity three years after they are issued. i Ot = k i max (St − Si . Denote. maturity time T the participation rate in the option issued at quarter i the risk-free interest rate at time t the return of product i between t − 1 and t the return vector between t − 1 and t the volatility at time t the weight allocated in the product issued at quarter i at time t the weight vector at time t As mentioned above.1. Naive fund constructions ratio options:bonds value is determined by the interest rate and the volatility. K. t. Hveem 21 (103) .1. i + 12. ki = i Si 1 − e−12ri Si − βi = . σi ) c (Si . c (Si . σt ) . i rt = i i βt + Ot − 1. Si . Si . The fund has to start somewhere and since the product prices are path dependent it is necessary to start the fund three years prior that the investor invests in it (in this setting). Si .

less unfavorable scenarios are not disclosed. The Capital Protection Property Now that the return for each structured product between each time period is known the attention can again be turned towards the fund (the return of each product is the only necessary information to calculate the return of the fund). Hveem 22 (103) . since they are not of importance (see Appendix A for more scenario examples). thus there is a diﬀerent allocation scheme to consider from quarter twelve and onwards. i i 1 + rt wt−1 . These scenarios should stress the fund construction such that weaknesses are disclosed. at quarter 1 the fund sells of 50% of its possession to allocate in the newly issued product. The payoﬀ of the matured product is as from quarter twelve reinvested in the newly issued product (it is only rolling over the product). otherwise if i = t if i ≤ t − 12 or i > t The fund’s return is the weighted return of all the assets’ returns. only two products available at quarter 1 and so on. 0.4 the underlying is the most prevalent M.2. (1 + rt )T wt−1 0. it is assumed that the investors have an investment horizon of three years and invest after three years (when the fund is announced on the market). The portfolio weights will be given diﬀerently between quarters 0-11 and 12-24. thus the portfolio weight for asset i at time t between quarters 0-11 is given as.33% of its possession to allocate in the newly issued product. since there is only one product available at quarter 0. thus the value of the fund is given as. T if t = 0 otherwise Scenarios. At quarter 0 the fund buys the newly issued product. underlying The next step is to investigate some of the most important scenarios for the fund construction. As mentioned earlier. Vt = St .Chapter 4. at quarter 3 the fund sells of 25% of its possession to allocate in the newly issued product and so on until the ﬁrst product matures. otherwise if i = t if i > t i wt | 0 ≤ t ≤ 11 = The ﬁrst product matures during quarter twelve. Vt−1 (1 + rt ) wt−1 . By disclosing the worst-case scenarios it is possible to counter these characteristics by changing the fund construction. Thus the portfolio weight for asset i at time t from quarter twelve and onwards is given as. until there are twelve products. The scenarios are constructed to stress the negative outcomes and as shown in Section 4. Thus the portfolio sells of some of its capital each quarter to allocate this in the newly issued product. i i t 1 + rt wt−1 . It is important that the investment horizon coincides with the time to maturity of the structured products. Note that only some of the most unfavorable scenarios are disclosed in this thesis. at quarter 2 the fund sells of 33. T (1 + rt ) wt−1 i i−12 i−12 wt | t ≥ 12 = 1 + rt wt−1 . since the individual structured product issued at the start of the investment provides absolute capital guarantee for the investors and is their alternative investment. (1 + r )T w t+1 t t−1 1/(i + 1).

Scenario 1 The ﬁrst scenario is a scenario where the underlying has a continuous return of 10% during the ﬁrst twelve quarters. This implies that the ratio between the value contained in options and bonds increases.1 discloses that the value of the options decreases. but their value was really small prior to the increase. as the options are decreasing in value. which means that the amount invested in the new structured product is the notional amount of the previous product. so how does the portfolio respond to an increase in the price of the underlying after quarter 24? Figure 4. The result indicates that the fund always strives to attain the original ratio by rebalancing from bonds to options (or from options to bonds in the increase case).1. after an extreme decline a new investor actually gains a degree of capital protection.015 per quarter. Thus during a large M. thus decreasing the ratio of options:bonds prior quarter 12. Figure 4. Hveem 23 (103) .66%. thus allocated in the fund. much higher than an investor who bought the fund at quarter 0. thus the scenarios are mainly conducted for changes in the underlying. 120 Price Underlying NAV Fund 100 Start of investment 80 X: 24 Y: 72. The fund invests each quarter in a new structured product. Figure 4. The portfolio has lost most of its value in options.2 shows that there are still some options left. since the fund buys new structured products each quarter. as the price of the underlying continues to decrease. thus increasing the capital protection for existing investors. The problem with this allocation scheme is that the ratio options:bonds increases as new products are introduced to the market. On the other hand an investor who invested at quarter 7 will actually receive a slight capital protection on its previous positive return.1 indicates that the price of the fund increases steadily during the quarters 1-12. An interesting feature with this portfolio construction is that. as disclosed in Figure 4. the result is shown in Figure 4. The fund reallocates to more options and cannot utilize the eﬀect of the growing value of the bonds in the same extent as a regular structured product. Naive fund constructions risk. which implies that both products have the same notional amount. Thus an investor who starts its investment at quarter 12 will buy a high proportion of options.3. σ = 0. hence the investor has no capital guarantee. for all quarters with a ﬂat yield curve. resulting in a further loss on the option part. since all of the options end up in the money and increase in value. Therefor the other parameters are set to be constant. Fund return quarters 12-24: −27. towards zero. then a continuous return of −15% the last twelve quarters.10 and rf = 0.2. the portfolio still has the same level of participation rate for the structured products.34 60 $ 40 20 0 0 5 10 15 20 25 Quarter Figure 4. Thus there exists no capital guarantee since the fund is always buying new options.4.1: Scenario 1. In this setting each product that matures is invested in the newly issued structured products. fund construction number 1: The value of the fund deteriorates as the price of the underlying decreases.

The Capital Protection Property 140 Price Underlying NAV Fund 120 Start of investment 100 X: 36 Y: 137. thus most of the value is contributed by the bonds at M. the result is shown in Figure 4.8 120 Price Underlying NAV Fund 100 Start of investment X: 24 Y: 108.2: Extensions of scenario 1: The fund still has a lot of options held in the portfolio at quarter 24. thus the portfolio is in some sense capital guaranteed over the twelve month period in this case assuming a ﬂat and constant yield curve. where the value repeats itself. thus the upwards potential is limited during the ﬁrst quarters of a bull market. thus an appreciation in the underlying generates a high return even though the previous decrease in the price. As the underlying crashes during quarters 12-24 so does the option value. given the previous decline in the underlying.1 80 80 60 $ 60 40 40 20 20 0 $ 0 5 10 15 20 25 30 35 40 0 0 5 10 15 20 25 Quarter Quarter Figure 4. Scenario 2 The second scenario is a scenario where the underlying has zero in return during the ﬁrst twelve quarters. since the scenarios do not diﬀer a lot from each other. 120 Start of investment 100 Price Underlying NAV Fund 80 60 $ 40 20 0 0 10 20 30 40 50 60 Quarter Figure 4.Chapter 4. It can be expected that the outcome of this scenario should be very similar as the previous one. while the option value declines due to the theta value of the options. then a continuous return of −15% the last twelve quarters.3: The fund has a ﬂoor which it does not cross (given a ﬂat and constant yield curve) since the portfolio is in practice just rolling bonds. decline the fund’s value has a period of 12 quarters. Hveem 24 (103) .3 (note that this is a special case with the ﬂat and constant yield curve). This means that an investor at quarter 25 or 37 will in practice hold an identical position in bonds and options as seen in Figure 4. The fund value increases slightly during quarters 0-12 due to the interest of the bonds.4. On the other hand this happens since the investor buys almost only bonds and a small fraction of options.

equals the fund’s ratio between option value and bond value at this quarter. and the percentage held in options as. which means that the fund buys the newly issued structured product every quarter with the payoﬀ of the matured product (it is assumed that the fund can hold inﬁnitesimal fractions of the structured products). at issuance. Instead the structured products issued after quarter 0 are customized in a way such that the ratio between the value contained in options and in bonds is maintained relative stable for the fund. which is also the option’s strike price.2.2. Therefor the portfolio should be rebalanced such that these assets are underweighted. Hence the key to capital protection is to avoid holding products that are far in the money.11%. there is not much value left in options to aﬀect. Fund return quarters 12-24: −8. in this scenario. Naive fund constructions 120 Price Underlying NAV Fund 100 Start of investment X: 24 Y: 91. then continuous return of −15% each quarter. quarter 15. The fund’s percentage of capital held in bonds at quarter i is given as. Hence the fund will hold a maximum of twelve products each quarter. The bonds’ notional amounts depend on both the price of the underlying and the ratio options:bonds in the fund. a restricted downside over the time horizon.2 Naive fund construction number 2 The naive fund construction number 2 diﬀers slightly from the previous one. Every structured product matures three years after issuance. thus after a few quarters the fund is almost rolling bonds (since the decrease in option value each quarter reﬂects the increase of the bond value proportion of the total value). wi j j j j βi + O i βi + Oi j=0 j=0 M. Thus the portfolio actually has. Below follows a more detailed description of the fund (the same notations as in the previous subsection will be used). The ﬁrst product follows the same dynamics as in the previous subsection (100% capital guaranteed) but all the other products are not standardized. as the underlying continues to drop in price.4. Thus every new structured product that is issued has a participation rate such that the ratio between its option value and bond value. 4.89 80 60 $ 40 20 0 0 5 10 15 20 25 Quarter Figure 4. This implies that. since they have a large percentage of value contained in the options. fund construction number 1: 0% in return of the underlying for twelve quarters. The price of the structured product equals the price of the underlying at issuance. i−1 i−1 j j βi Oi j j wi .4: Scenario 2. Hveem 25 (103) . The fund has to start somewhere and since the product prices are path dependent it is necessary to start the fund three years prior that the investor invests in it (in this setting).

if i ≤ t ≤ i + 11 if t = i + 12 otherwise Hence the return between t − 1 and t for the structured product issued at quarter i is given as. only two products available at quarter 1 and so on. 0) . i + 12. rt . i Ot = k i max (St − Si . t. otherwise if i = t if i > t i wt | 0 ≤ t ≤ 11 = M. at quarter 1 the fund sells of 50% of its possession to allocate in the newly issued product. ri . T (1 + rt ) wt−1 t + 1 i wt = 1/(i + 1). i. until there are twelve products. σi ) i−1 j Oi j βi if i = 0 ki = j=0 + j Oi j wi Si . Si . Si . Si e t i−1 βj i if i = 0 j wi Si e12ri −rt (i+12−t) . at quarter 3 the fund sells of 25% of its possession to allocate in the newly issued product and so on until the ﬁrst product matures. i + 12. i. i wt = 0. −r (i+12−t) . i. σi ) otherwise the value of the call option issued at quarter i at time t is given as. i rt = i i βt + Ot − 1. The Capital Protection Property thus the bond prices are given as. i βt = j=0 0. i k C (St . since there is only one product available at quarter 0. σt ) . ri . i Si − βi Si 1 − e−12ri = . i i 1 + rt wt−1 t . Si . c (Si . i + 12. at quarter 2 the fund sells of 33. Si .33% of its possession to allocate in the newly issued product. The portfolio weights will be given diﬀerently between quarters 0-11 and 12-24. j βi + j Oi if i ≤ t ≤ i + 12 otherwise the participation rate of the option issued at quarter i is given as. ri . 0.Chapter 4. i i βt−1 + Ot−1 Now that it is known how the return for each structured product between each time period is calculated the attention can again be turned towards the fund and how the portfolio weights are calculated (since the return of the products each quarter depends on the weights the previous quarter). Hveem 26 (103) . c (Si . i + 12. At quarter 0 the fund buys the newly issued product. thus the portfolio weight for asset i at time t between quarters 0-11 is given as. σi ) c (Si . Thus the portfolio sells of some of its capital each quarter to allocate this in the newly issued product.

Naive fund constructions The ﬁrst product matures during quarter twelve. due to the positive return of the bonds. If the market goes down during twelve consecutive months an increase in the market would not have any aﬀect on the value of the fund. The scenarios are constructed to stress the negative outcomes and as shown in Section 4. Hveem 27 (103) .2. since they are not of importance. This fund construction creates a certain level of capital protection. T if t = 0 otherwise Scenarios. even though the market continues to decrease.4. As mentioned earlier. it is assumed that the investors have an investment horizon of three years and invest after three years (when the fund is announced on the market). Thus an investor that buys the fund at quarter twelve buys a high proportion of options instead of bonds. The payoﬀ of the matured product is as from quarter twelve reinvested in the newly issued product. since it would not contain any options anymore. Thus the portfolio weight for asset i at time t from quarter twelve and onwards is given as. Note that only some of the most unfavorable scenarios are disclosed in this thesis. since the individual structured product issued at the start of the investment provides absolute capital guarantee for the investors and is their alternative investment. Vt−1 (1 + rt ) wt−1 . Scenario 1 The ﬁrst scenario is a scenario where the underlying has a continuous return of 10% during the ﬁrst twelve quarters.10 and rf = 0.5 shows that the value of the fund increases a lot as the underlying increases.5. due to the increase of the options and their convexity. then a continuous return of −15% the last twelve quarters.5 discloses that the fund actually has a positive return after quarter 17. Vt = St . if i ≤ t − 12 or i > t The fund’s return is the weighted return of all the assets’ returns. thus bonds constitute almost all of the value. Figure 4. By disclosing the worst-case scenarios it is possible to counter these characteristics by changing the fund construction.4 the underlying is the most prevalent risk. less unfavorable scenarios are not disclosed. σ = 0. if i = t (1 + r )T w t t−1 0. thus the value of the fund is given as. Therefor the other parameters are set to be constant. as the value of the option part is already zero. since the fund maintains its ratio of options:bonds relative stable when rebalancing. otherwise (1 + rt )T wt−1 i i−12 i−12 wt | t ≥ 12 = 1 + rt wt−1 .2.015 per quarter. the result is shown in Figure 4. thus the scenarios are mainly conducted for changes in the underlying. It is important that the investment horizon coincides with the time to maturity of the structured products. for all quarters with a ﬂat yield curve. The fund’s proportion held in options converges towards zero after seventeen quarters. i i 1 + rt wt−1 . underlying The next step is to investigate some of the most important scenarios for the fund construction. Figure 4. but not in the desired way M. thus there is a diﬀerent allocation scheme to consider from quarter twelve and onwards. These scenarios should stress the fund construction such that weaknesses are disclosed.

since it is useless after a huge decline. which can imply that the fund is only allocating in bonds and that it is no longer a structured products fund. thus the option part’s value quickly converges towards zero M.47:83.53. thus a new investor at quarter twelve can loose approximately half of its investment. It is also too risky since gains from the options are not reallocated to bonds. fund construction number 2: Similar result as in the previous fund construction. Fund return quarters 12-24: −38. All options end out of the money. then a continuous return of −15% the last twelve quarters. 120 Price Underlying NAV Fund 100 Start of investment X: 24 Y: 116.5: Scenario 1. Also.32%.6.36%.4 80 60 $ 40 20 0 0 5 10 15 20 25 Quarter Figure 4. The Capital Protection Property 120 Price Underlying NAV Fund 100 Start of investment 80 X: 24 Y: 61. As the price of the underlying increases the ratio options:bonds increases heavily to approximately 1:1 from 16. when the market turns bust during consecutive quarters the portfolio reduces its participation heavily with the market. Scenario 2 The second scenario is a scenario where the underlying has zero in return during the ﬁrst twelve quarters.Chapter 4. the result is shown in Figure 4. creating a huge downside after a bull market. Hveem 28 (103) .6: Scenario 2. Thus the fall of the price of the underlying does not really aﬀect the value of the fund a lot.68 60 $ 40 20 0 0 5 10 15 20 25 Quarter Figure 4. Fund return quarters 12-24: 16. fund construction number 2: The ratio options:bonds is very low since the price of the underlying does not move until quarter 12.

as seen in Figure 4.e.47:83. and a time to maturity of three years.2 p. On the other hand rebalancing up the ratio options:bonds each quarter when it has decreased under a certain limit would imply that the portfolio’s risk increases i. Thus it is more important to focus on changes in the underlying than the interest rate. It is important to determine the most prevalent risk factors to be able to construct a portfolio of structured products such that the risk is minimized. Hveem 29 (103) . ytm = 0. but this should not be conducted when trying to minimize the risk. Also the interest rate risk is very small for buy and hold strategies. since the investor is guaranteed a ﬁxed amount in the future.08% to cause a price drop that big.3 Analysis and summary The most preferable fund construction is construction number one since it has less extreme outcomes. Concluded.06.r. It is also desirable to increase the ratio options:bonds in the same way as it is done in fund construction number 1 after a market crash.8 depicts the sensitivity of an ELN w.2.53. if the fund wants to increase the possibility of appreciation. 4.9743. The next section investigates how an investor should allocate the option portfolio to have as little downside risk as possible.e.e. when the market has gone up. The most volatile part of a structured product is the option. Consider an Equity-Linked-Note with a price of $100. σ = 0. the ELN is much more sensitive to changes in the underlying than in the yield curve. The fund also provides capital protection.4. the yield curve. for investors who experienced the bull-market. 4. Fund construction number 2 is not desirable at all since it is a very risky investment strategy and does not hold the desired characteristics. The reason why it is possible to generalize the result of options to the structured products is that a change in the underlying is the most prevalent risk of the structured products.2. there are pros and cons with every decision. The question now arises regarding how the option portfolio should be constructed. how the weights in the diﬀerent options (and thus also the structured products) should be constructed to retain as much capital protection as possible. the bond is less sensitive to movements aﬀecting its price. When considering a buy and hold portfolio (which most funds practice. thus a ﬁxed yield. The yield to maturity would have to increase to 13. There is not a huge risk with changes in the yield curve that are not extreme since if the yield curve goes up bonds will be less expensive and the bonds bought will generate a higher return. price of the underlying of $100.2. This implies that it is possible to protect new investors from for example scenario 1. Also fund construction number one will continue to participate in the market even though the market has decreased a lot.7 which depicts a huge change in the interest rate. i. the underlying and the interest rate.4 Prevalent risk factors This short subsection discusses the most prevalent risk factors for an ELN.t. Based on these results it is easy to conclude that there is a big advantage to rebalance the portfolio at intermediate time points so that the ratio options:bonds does not exceed a certain limit. which will be discussed in the next subsection.a. i. Using the allocation scheme in fund construction number 1 implies that the fund slowly rebalances towards its original ratio options:bonds. where there exists a maximum acceptable limit. Naive fund constructions and the portfolio gains a positive return from the remaining bonds. it is important that the allocation scheme has a focus on rebalancing the rate options:bonds. This can be done by for example solving a linear optimization problem. a shift of the underlying downwards of 50% would induce a decrease of the product’s value of 15. The ratio options:bonds would be 16. it is reasonable to assume that the whole portfolio will not rebalanced each quarter due to transaction costs) it is a much larger risk that the option is out of the money close to maturity than that the bond has a low price close to M. Figure 4.

but as it can be seen. to a portfolio of structured products since the options in the ELNs are signiﬁcantly more risky than the bonds. The big risk is if a change occurs just prior to the end of the investment horizon. which aﬀects the option price more than the volatility as seen in Figure 4. The Capital Protection Property 120 Interest Rate NAV Fund 100 X: 24 Y: 100. as time decreases so does the sensitivity to changes in the yield curve. the volatility increases as the price goes down (since the drift is in general positive).a.3 Start of investment 80 [%] 60 40 20 0 0 5 10 15 20 25 Quarter Figure 4. thus the two risk factors counter each other’s impact on the price.7: The interest rate goes from 2% p. the impact is still limited. maturity.8: The surface plots describe how changes in the yield and price of the underlying aﬀect the price of an ELN. σ = 0. M. Hveem 30 (103) . K = 100. Thus the interest rate risk of holding an existing bond is much smaller than holding the option.a. But. Thus a decrease in volatility is not a big concern since it happens when the price of the underlying goes up. Hence it is possible to generalize the optimal allocation of options. The investor does not have a negative return over the whole period. during just a few quarters. T=3 150 T = 0. as a consequence of the deﬁnition of volatility.9.5 125 140 160 130 140 130 120 120 ELN Price [%] 120 120 110 100 100 80 90 60 15 80 10 150 5 50 100 0 0 70 200 ELN Price [%] 115 110 110 100 105 90 15 10 150 5 50 100 0 0 95 200 100 Interest Rate r [%] St Interest Rate r [%] St Figure 4. The next section will therefor investigate how the option portfolio should be allocated to minimize the negative outcomes.Chapter 4. to 14% p. in min risk sense. There may also be a risk with a decrease in volatility.2.

Without setting such boundaries some options may have returns in the size of 1030 between two quarters.0105.69 (on a days basis. which should be independent as much as possible of the assumptions of the underlying.a. To test which portfolio setups that are best in min risk sense the study will conduct Monte Carlo simulation. Since the investigation is investigating patterns and worst-case outcomes it does not really matter what the drift µ is set to be. which should not be sensitive to the distribution of the underlying).3 Investigation of the option portfolio The purpose of this section is to investigate how the option portfolio should be allocated to minimize the negative outcomes in sense of the underlying. It is not important to replicate a certain index.a. r = 0.06. Investigation of the option portfolio 90 100 80 80 70 60 50 40 40 20 30 0 50 40 30 20 10 50 0 0 100 0 150 200 10 20 Option Price 60 Volatility σ [%] S [%] t Figure 4. and the volatility used when pricing the options is the actual realized volatility during the period of 20. Strike K = 100. Hveem 31 (103) . thus µ is for simplicity set to zero. The option prices are assumed to not be smaller than 0.1 Portfolio construction and modeling The portfolios that are investigated consist of only plain vanilla at-the-money call options at issuance on the underlying index.5% of the price of the underlying.9: The surface plot describes the relationship between changes in volatility and the price. to prevent unrealistic results. The index OMXS30 at Nasdaq OMX serves as a base for the simulation. as well as a better ﬁt. M.1. the other ﬁtted parameters for the t-distribution are: σ = 0. since the minimum-risk option portfolio should in most cases yield the minimum-risk ELN portfolio. it is not important what the actual parameters are since the focus lies on patterns.5% p. since the study is looking for patterns. before maturity. 4. The following parameters are also used: rf = 6% p. The options are priced with Black-Scholes formula as given in Equation 3. The study covers diﬀerent portfolio constructions (only those relevant are shown) and how the scenarios are modeled to stress the diﬀerent portfolio allocations. ν = 5.3.4. T = 3. The section also generates a recommendation regarding how an option portfolio should be allocated such that the risk is minimized. 4. The log returns are ﬁtted to a student’s t-distribution to attain slightly heavier tails than for the normal distribution. price is the dominant factor. thus daily log returns are gathered between 1st of October 2009 to 30th of September 2010.3.

Monte Carlo simulation is used to create outcomes of the underlying. another option would be CVaR). it is also desirable that the distribution of the outcomes is positive skewed. . 12. the last one equal to zero. It turns out that the products with the longest time to maturity exhibit the lowest risk. w5 = 12 78 11 78 10 78 9 78 8 78 7 78 6 78 5 78 4 78 3 78 2 78 1 78 T 6 The sixth portfolio is a portfolio that has increasing portfolio weights. w4 = 0 0 0 0 0 0 0 0 0 0 0 1 T 5 The ﬁfth portfolio is a portfolio that has decreasing portfolio weights. The option portfolios are as follows.g. 2.. thus imposing less risk to an investor. and the last one equal to one. the option issued at quarter t has strike K t = St . j = 1. the ﬁrst one equal to zero.2 Portfolios and results Eight portfolios are investigated in this section using the simulation described above. these portfolio are some of the most relevant constructions (for more construction please see Appendix B).g. e. The best result is measured in the sense of worst outcome (since intrinsic capital protection is investigated in this chapter. w6 = M. The next subsection describes the diﬀerent relevant fund constructions and discloses the result of them as well. The Capital Protection Property A new ATM call option is issued each quarter with maturity in three years. w2 = 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 0 T The third portfolio is a portfolio that has the last eleven portfolio weights equal to each other. Hveem 1 78 2 78 3 78 4 78 5 78 6 78 7 78 8 78 9 78 10 78 11 78 12 78 T 32 (103) . with wj = j. The ﬁrst portfolio has portfolio weights that are equal to each other.3.000 threeyear periods are simulated (excluding the three-year time period used to create the necessary price paths for the options).Chapter 4. with wj = 13 − j. w1 = 1 12 1 12 1 12 1 12 1 12 1 12 1 12 1 12 1 12 1 12 1 12 1 12 T The second portfolio is a portfolio that has the ﬁrst eleven portfolio weights equal to each other. i The weight in option j held by the portfolio i is denoted as wj where j is the option that has a maximum of j quarters left to maturity. The reason for measuring the risk of holding the option portfolio as the worst-case outcome and not by just using a measure such as the volatility is that the return distributions of derivatives (e. 4. 3.. w3 = 0 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 T The fourth portfolio is a portfolio that has the ﬁrst eleven portfolio weights equal to zero. options) often are skewed and exhibit kurtosis. Thus there are at each intermediate time point twelve options available to invest in. The portfolio weight vector of portfolio i is denoted as wi .. where St is the price of the underlying at time t. thus the portfolio is totally diversiﬁed over all available options. with diﬀerent strikes and maturity times. a total of 60.

issued ATM (T = 1). to not overweight. zero otherwise. Thus it might be reasonable to diversify by holding only four options with a long time to maturity as in Portfolio 16 in Appendix B or for example diversify over the whole spectrum and attain approximately the same results. except the one closest to maturity (which has the highest risk. Figure 4. M. w7 = 0 1 9 1 9 1 9 1 9 1 9 1 9 1 9 1 9 1 9 0 0 T The eighth portfolio is a portfolio that has equal portfolio weights in assets 2-11. it is in this case beneﬁcial to diversify along almost all options. w8 = 0 1 10 1 10 1 10 1 10 1 10 1 10 1 10 1 10 1 10 1 10 0 T The results for the diﬀerent option portfolios are disclosed in Figure 4. It would be desirable if the return distribution for the portfolio of options over 36 months would look somewhat similar to the return distribution of a single option from issue to maturity as seen in Figure 4. As a fund manager it would be quite unreasonable to allocate all the capital in the same asset. The results shown in the ﬁgures indicate that it is not possible to attain the same type of distribution as for a single option when investing in several options over 36 months.10.11. Investigation of the option portfolio The seventh portfolio is a portfolio that has equal portfolio weights in assets 2-10. both in worst-case and in distribution. Usually a portfolio manager is not allowed to invest more than a certain percentage (say 30%) in a particular asset. in most cases. a more risky portfolio. ∂t ceteris paribus. Hveem 33 (103) . The results in Figure 4.1 and B. there are also additional option portfolios disclosed in Appendix B. Figure 4.12b. since it might be “all or nothing”).4.10 and Appendix B show that the portfolios where the options closest to maturity are underweighted have “better” distributions since they exhibit less negative skewness.2 indicate that it does not really matter how the portfolio diversiﬁes amongst the assets since the results are quite similar as long as the portfolio holds a relative large proportion of the holdings in products with a longer time to maturity.12a shows that options with a longer time to maturity are less sensitive to the changes in time (θ = ∂P ).10. Notable is that the portfolio that has the best worst-case outcome is the portfolio that holds 100% of its holdings in the option with the longest time to maturity (Portfolio 4). Plain vanilla option return 7000 6000 5000 Outcomes 4000 3000 2000 1000 0 −5 0 5 10 15 20 Return [%] Figure 4. B. Fewer number of assets in the portfolio results in. Thus the value reduction coming in form of the reduction in time. thus it is adequate to demand that the fund diversiﬁes itself.11: The histogram shows the distribution of a plain vanilla option’s one-year return. is less if the investor holds options with a long time to maturity.3. Also the delta (∆ = ∂P ) for options that are in the money is higher for options close to maturity than options ∂s with a long time to maturity as depicted in Figure 4. Also the worst-case outcome is much better for portfolios that have underweighted the options closest to maturity. zero otherwise.

1 −0.91%. Option portfolio average monthly returns (36 months periods) Outcomes 100 80 60 −0.15 −0. −20.05 0 0.05 0.2 −0.1 −0.1 0.05 0.05 0 0.15 −0.05 0.2 −0.15 −0.1 −0.1 0. the worst-case average 36 month return is given in the caption.2 −0. −21.15 Average return per month Average return per month (g) Portfolio 7. −17.15 Average return per month Average return per month (c) Portfolio 3. The Capital Protection Property Option portfolio average monthly returns (36 months periods) 180 160 140 120 120 100 160 Option portfolio average monthly returns (36 months periods) 140 Outcomes 100 80 60 Outcomes −0.1 0.1 −0. −22.1 −0.15 80 60 40 40 20 0 −0.15 −0.15 Outcomes −0.1 0. Option portfolio average monthly returns (36 months periods) Outcomes 100 80 60 40 20 0 −0.15 −0. Figure 4.37%.05 0 0.15 0 −0.15 −0.3 40 20 0 −0.05 0 0.88%.15 −0.10: Histograms over the outcomes for option portfolios 1-8.2 Outcomes 150 100 50 −0.05 0 0.15 Average return per month Average return per month (a) Portfolio 1.1 0.05 0 0. −16.05 0 0.25 20 0 −0.25 −0.71%.15 Average return per month Average return per month (e) Portfolio 5. Option portfolio average monthly returns (36 months periods) 180 160 140 120 200 180 160 140 120 100 80 60 40 20 0 −0.64%. −18. −19.2 −0. M.05 0.25 −0.69%. Hveem 34 (103) .1 0.1 −0.15%.15 −0. Option portfolio average monthly returns (36 months periods) 180 160 140 120 200 250 (b) Portfolio 2.2 −0.1 −0.05 0. −27.Chapter 4.1 0.05 0 0.1 0.05 0.2 Outcomes −0.1 −0.05 0.80%.05 0.25 (d) Portfolio 4. Option portfolio average monthly returns (36 months periods) Outcomes 100 80 60 40 20 0 −0.2 −0.15 100 80 60 40 20 0 −0. (h) Portfolio 8. Option portfolio average monthly returns (36 months periods) 180 160 140 120 180 160 140 120 (f ) Portfolio 6.

Investigation of the option portfolio An optimal portfolio would be positive skewed.6 0. the theta is heavily decreasing in time. where investors tend to sell assets too early during growth.5 −3 −3.4. Kahneman and Tversky. The main result is that the fund should diversify over a number of assets and also avoid investing in the products with a short time to maturity. as indicated in Figure 4. Figure 4. [35]). time to maturity.5 −4 −4. the fund must be able to guarantee new investors that they do not buy more than a certain percentage of options.3.2 0 200 150 100 50 0 40 30 20 10 0 Months to maturity 0 St St Months to maturity (a) The plot describes a plain vanilla call option’s theta (b) The plot describes a plain vanilla call option’s delta.10c (as well as considering the worst-case outcome). 4. [15]. The tradeoﬀ is that the portfolio at the same time limits the upside of the investment (the tradeoﬀ between risk and return). an analysis and a description of the implications from the studies in this chapter so far.5 −2 −2. (time value). Next follows an analysis of how the results found in Section 4. 1979. Without this guarantee the purpose of the fund is lost. 2002. The most beneﬁcial construction fulﬁlling these criteria is Portfolio 3.2 that it is desirable to rebalance the portfolio when the ratio options:bonds exceeds a predetermined level (which depends on the risk proﬁle of the fund) to limit the downside risk. Hence it is important to diversify over a large set of assets to reduce the risk.5 40 30 20 10 0 50 100 150 200 0. options closer to maturity are more sensitive when implying that the holder looses a lot of capital from they are in the money than options with a longer a time change close to maturity. The reason why the ratio options:bonds suddenly exceeds a certain level is (most often) due to either an increase in the price of the underlying (increasing the value of the options) or in the yield curve (reducing the value of the bonds).12: The Figure a) illustrates an option’s theta value. Thus by rebalancing the portfolio the downside risk is limited since the proportion of options is reduced.2 and in this section should be combined to construct a desirable structured products fund. since they are conservative and people do not expect the market to continue appreciating (disposition eﬀect) (Frazzini. [26]). 0 −0. the Figure b) illustrates an option’s delta value for diﬀerent levels of in the money and time to maturity. with a limited downside.4 0.12). Hveem 35 (103) . Thus investors want to rebalance their portfolio after M. the risk increases (the theta and delta (given that the option is in the money) in particular. by appealing to investors’ loss aversion (Shefrin. These characteristics appeal to psychological concepts within Behavioral Finance. ceteris paribus. By rebalancing the portfolio after a period of growth the value held in bonds increases. This is an important concept since it protects new investors.3.8 Option Delta (∆) 0. Thus the existing investors will receive a certain level of capital protection on their gains.3 Analysis and implications In this subsection follows a combination of a summary. when constructing a portfolio only consisting of options. On the other hand as the time to maturity decreases. Thus when constructing a portfolio with structured products it is important to not overweight in the shorter tenors. It is known from Section 4. 2006. as disclosed in Figure 4.5 −1 1 Option Theta (Θ) −1.

Investors will not invest in the fund if it is not capable of beating its competition. It is desirable to rebalance the portfolio such that the ratio options:bonds are held under control. The Capital Protection Property a period of appreciation. Thus if the portfolio overweights/allocates to the newly issued ELNs it will always reallocate capital to newly issued options. As seen in Section 4. [16]). The section covers two types of SPFs. The diﬀerent constructions are both stress tested with predeﬁned extreme scenarios as well as simulated scenarios.Chapter 4. the options should also yield the minimum risk choice for the structured products. Thus it is a balance act for the fund how it should change the ratio options:bonds in a way that both new investors and current investors appreciate. thus pushing up the price and decreasing the yield. One way to make adequate assumptions is to relate to the fund’s competition. Thus it is reasonable to invest using Portfolio 3 in general and after drops in the index allocate the portfolio according to Portfolio 7. The inﬂation goes up when the market goes up. which is not desirable in bear markets. 4. instead it may be desirable to slowly re-establish the portfolio’s ratio options:bonds to the original level. at least at some aspects. Thus it is imperative to establish the SPF’s client base. Hveem 36 (103) . The rebalancing should not be conducted in the same manner as when the ratio exceeds a certain level. As stated previously. When markets go down. thus by allocating the portfolio according to the minimum risk choice w. This implies that the price changes of bonds and call options are negatively correlated. thus also interest rates. for which levels etc? These issues will be investigated in the next section where a comparison with existing mixed funds results in optimal choices of the ratio options:bonds. After that the competing fund has been identiﬁed the study investigates how diﬀerent SPFs can be constructed. The diﬀerent SPF strategies are also backtested against historical data. since the volatility and the price of the underlying usually counter each other (Glot. The next section will investigate how this ratio/level should be chosen and its implications on downside risk. Hence instead of allocating according to Portfolio 3.r. Thus it is not desirable to hold positions in options that have a short time to maturity. Thus the impact of volatility can be neglected when choosing portfolio weights. 2005. The comparison will be based on the results found in the previous sections. A competing fund is a fund that has the same client base as the SPF. but how should this be conducted. such that the SPF provides a higher degree of capital protection than the benchmark fund. it may be desirable to rebalance the portfolio’s ratio of options:bonds towards a desired level. When considering a portfolio of structured products it might be non-desirable to overweight the newly issued products during bear markets since they will have a substantial proportion of option value in them compared to the other products.3 the best choice to minimize the downside risk for a portfolio of options is by diversiﬁcation in the longer (and middle) tenors. Thus the purpose of this section is to investigate how a potential structured products fund (SPF) performs in relation to its competition (mixed funds) during extreme scenarios. a higher degree of capital protection during bear markets will be attained by investing using either Portfolio 7 or 8. one where the weights are based on the possession in the structured products and one SPF where the weights in the structured products are based on M. The analysis results in appropriate boundaries for the percentage of value in the SPF contributed by options given a certain competing/benchmark fund. investors ﬂee to safer instruments such as bonds.4 Comparison with competition A lot of assumptions have to be made in able to construct the fund. During bear markets. as when only considering an option portfolio. so that a competing benchmark fund can be identiﬁed. when the portfolio decreases in value.t. Without this reallocation the fund will not have as much upside for new investors. Since Equity-Linked Notes are considered in this paper the options are far more risky than the bonds. there is no need to investigate changes of volatility.

An example of a price development of this type of benchmark fund is disclosed in Figure 4.2 Benchmark fund It is assumed that a reasonable benchmark fund is a mixed fund that aims at holding 30% (εbench ) of its capital in stocks and 70% in bonds with up to three years until maturity (to match the SPF). It turns out that the second one is more theoretical than the ﬁrst one. 450 400 350 300 250 200 150 100 50 Price Underlying NAV Benchmark $ 0 50 100 150 Month Figure 4. These are risk-aversive investors who usually are looking for a combination of bonds and stocks. δ bench = 4. for 0. imposing an adequate benchmark. 4. hence the structured products fund must provide a lower downside risk than competing mixed funds to be an attractive investment alternative. The investment within stocks is considered in this study to be conducted in the index (or an ETF). The result in this section indicates that a structured products fund should try to have a possession of 5-25% option value attained in the portfolio. which diversify their portfolios in both debt and equity instruments to lower the risk level.3. which is also the underlying for the SPF.3 Structured products fund 1 The fund constructions in this section are based on the results gained in Section 4. thus the portfolio diversiﬁes broadly amongst the available products. Portfolio 7 will be used.2 and 4. since M. Comparison with competition the possession in the options.3. bench = 0. Hveem 37 (103) .4.4. Today many of these investors invest in mixed funds. if the level of stocks held is not within 30% ± 10% (δ bench ).a. The benchmark fund rebalances its position every three months. The ﬁgure shows that the benchmark fund gains value as the market is appreciating and does not decrease signiﬁcantly as the market takes a turn for the worse. with a target level of 15% to be more attractive for investors in min risk sense than a competing mixed fund. Thus the investors that it is suitable for are those who are looking for investments with a limited downside.4. Next follows a more detailed description of the benchmark fund.1 and r = 0. random scenario. hence it is this alternative which is an adequate benchmark. The focus is now turned to the construction of the structured products fund.1 Client base The fund based on structured products has a high focus on capital protection.13. or a small risk.13: The co-movements of the underlying and the benchmark fund.4.06 p. Thus mixed funds are identiﬁed to be the competition to the SPF.4. 4. Thus it has some of the desired properties for a SPF. as well as a possibility to have a return connected to the market.

thus the portfolio is rebalanced each quarter if. The portfolio is allowed to deviate a percentage δ SPF from its target level of options. Notable is that the fund is not allowed to short the short-term debt. as in Section 4. and also the proportion placed in additional bonds. The fund adjusts the proportion placed in the short-term debt subject to its level of option value as a percentage of the total fund. ˜ w= 0 1 9 1 9 1 9 1 9 1 9 1 9 1 9 1 9 1 9 0 0 T . o o αt ≤ εSPF − δ SPF or αt ≥ εSPF + δ SPF . this target level is the desired percentage of the fund’s value that should be constituted by options. the proportion of value of the fund contributed by bonds is b denoted α . this is not always possible since it is not allowed to short the short-term debt. εSPF wt o . Which gives us the following deﬁnitions.3 the following weights are used.1. The fund rebalances its portfolio of structured products each quarter as previously. let denote this target level as εSPF . the weight in the short-term debt at time t is denoted b b as wt . A new capital guaranteed structured product is issued each quarter with maturity in three years. thus the portfolio holds equal weights in products i = 2 − 10 where i denotes the maximum number of quarters until maturity for the product as previously. n b b b αt = wt + 1 − wt i=1 n o b αt = 1 − wt i=1 i βt . i i βt + O t where n is the number of available products (12 in this case). w denotes the weight vector for the structured products excluding the short-term debt (the ˜ weights amongst the structured products). if i i wt i i wt εSPF o ≤1 αt εSPF o >1 αt new wt = M. A target level of options held in the fund’s portfolio is also introduced. i + Ot wi ˜ i βt wi ˜ i Ot . price equal to the price of the underlying and also the notional amount of the bond. and αo is the proportion of value of the fund contributed by options. if αt ˜ w. in this case bonds that have maturity the next quarter.2. which implies that the new portfolio weights are given as. Thus the actual weights (except wt ) for each time-period is given by: b ˜ wt = 1 − wt w. Thus the fund can by investing in the short-term debt reduce the percentage of value placed in options. b ˜ Note that w0 = 0 and w0 = w. Thus the portfolio is rebalanced if the percentage of value contributed by the options exceeds εSPF + δ SPF (or is less than εSPF − δ SPF ) by allocating in short-term debt rather than the structured products.Chapter 4. The Capital Protection Property the test is conducted for bear markets (in Appendix C results for Portfolio 3 are disclosed). The portfolio is rebalanced with the aim that αo should equal SPF . thus attaining the target level. and since the focus lies on Portfolio 7 from Section 4. according to the weights and maturity amongst the structured products. Hveem 38 (103) . The fund also has the possibility of investing in short-term debt.

4.a.2 discloses in the same manner the result regarding the benchmark fund.1 and 4.a. The objective is to construct a SPF such that it has more favorable worst-case outcomes than the benchmark portfolio.4.4 Scenarios This subsection describes the scenarios that the SPF and benchmark fund are stressed with. the ﬁrst setting with a bull market with a return of 30% in return p. until month 144 (12 years). εbench and δ bench . εSPF and δ SPF . M. thus it is appropriate to select scenarios without simulation. thus the new allocation in the short-term debt is as follows. The next subsection covers the results over how the funds perform in the diﬀerent scenarios and how the results should be interpreted.2). This implies that the short-term debt is used as a safe haven to avoid a too high exposure to the market. 4. Table 4.14.a. The levels will be determined such that the SPF provides a greater capital protection for market crashes of −20% p. during months 0-48 and a return of 50% each month for months 48-50 (Scenario B). Table 4.4. and more. its worst-case outcome given the diﬀerent scenarios. The purpose with the following subsections is to determine the adequate levels for εSPF and SPF δ such that the fund has a greater protection (towards extreme downside risk) than the benchmark portfolio(s). −30%. Hence the market crash in Scenario A is between months 48-96 and in Scenario B between months 50-98. during four years. −80%. Various results are disclosed in Figure 4. The second scenario is quite similar. −50%. The reason why predeﬁned stressed scenarios are studied instead of simulation in this subsection is that the study is trying to stress test the funds in particular scenarios and see if the SPF has the desired properties. The fund is not allowed to short the short-term debt (to minimize the risk).a. Comparison with competition these weights are the weights excluding the short-term debt. 30% return p.2. between months 0-48 (Scenario A).14. 4. Hveem 39 (103) . b new wt = 1 − 1T wt . The idea is that since the funds can only rebalance to short-term debt every three months that the funds can be stressed even more if an extreme increase occurs during a period where the funds cannot rebalance. its worst-case outcome given the diﬀerent scenarios. Also it is imperative that the results gained from this study are independent of the models selected for the underlying and that they can be generalized to any distribution or market model. −20% and −10% p. as seen in the previous sections). An annual interest rate of 6% is assumed for the bonds as well as the short-term debt (ﬂat yield curve is assumed for simplicity). thus it cannot hold b negative positions in wt . Two diﬀerent settings prior to the crash will be investigated. The scenarios should stress the SPF and the benchmark fund towards negative outcomes.5 Results.a.1 discloses the result regarding the SPF. and still has a good upside potential. Table 4. The scenarios are not simulated since it is desirable that the SPF provides capital protection for some speciﬁed market crashes. thus expose the funds in an extent where they disclose weakness (which is a market crash immediately after a strong increase in the market. the scenarios can be seen in Figure 4. structured products fund vs benchmark The focus lies on the worst-case outcomes of the SPF (based on Portfolio 7) and the benchmark fund given diﬀerent scenarios and values for ε and δ. ﬁve diﬀerent scales of market crashes are investigated. Thus these two types of scenarios should cover the worst types of extreme outcomes for the funds (for more extreme scenarios please consult Appendix C).4. After the crash the market experiences an increase of 30% p. The focus lies on scenarios with an up-going market followed by a market crash (as the worst-case scenarios in Section 4.

10 1 0.05 1 0. Hveem 40 (103) .20 0.62% -15.64% -16.42% -18.82% -16.25 0.05 0.45% -14.1 0.15 0.82% 1.54% -26.65% -3.1 0.1 0.06% -17.10 1 0.1 0.1 0.79% 1 0.60% -9.18% -22.15 0.59% -48.15 0.05 0.20 0.1 0.20 0.95% -6.1: SPF 1: Worst-case outcomes for SPF 1 based on Portfolio 7 given scenarios A and B.1 0.13% -9.04% -25.3 0.05 1 0.25 0.64% -13.25 0.15 0.64% -10.25 0.11% 0.75% -1. -80% -80% -80% -80% -80% -80% -80% -50% -50% -50% -50% -50% -50% -50% -30% -30% -30% -30% -30% -30% -30% -20% -20% -20% -20% -20% -20% -20% -10% -10% -10% -10% -10% -10% -10% εSPF δ SPF Scenario A WorstCase Return -18.43% 3.48% -16.1 0.1 0. M.15 0.1 0.32% -3.15% 0.05 Table 4.a.35% -2.34% -3.62% -32.08% -4.60% -11.38% 2.15 0.05 0.1 0.59% -37.23% 4.59% -6.24% -11.00% -7.1 0.15 0.57% -11.09% -2.53% -18.Chapter 4.1 0.20% -19.1 0.42% -6. The Capital Protection Property Return Index Months 48-96 / 50-98 p.31% -5.05 1 0.3 0.71% -7.05 1 0.15 0.3 0.3 0.23% -3.10 1 0.20 0.10 1 0.81% 0.20 0.15 0.78% -5.90% -14.3 0.15 0.22% -9.49% -19.81% -1.10 1 0.05 0.1 0.22% Scenario B WorstCase Return -56.50% -33.84% 14.26% -13.60% -14.1 0.03% -3.95% -11.55% -9.46% -17.54% -1.25 0.1 0.72% -3.05 0.40% -22. εSPF and δ SPF .63% -5.1 0.1 0.1 0.41% -10. the worst-case period is always the 36 month following the initiation of the crash.53% -11.07% -37.39% -11.60% -6.20% -1.85% -2.46% -11.

1 0.10 0.5 0.85% -18.5% -12.5 0.24% -15.03% 4.10 0. M.80% -5.a.7 0.05 0.1 0.27% -17.1 0.39% 7.4.20 0.05 0.1 0.1 0.62% -54.61% 4.59% 7.10 0.20 0.7 0.61% -0.1 0.17% -72.20 0.1 0.1 0.68% -2.1 0.1 0.58% 7.71% -31.87% 13.84% -2.7 0.3 0.76% 1.20 0.20 0.70% -18.75% 9.4.86% -34.1 0.36% -6. -80% -80% -80% -80% -80% -50% -50% -50% -50% -50% -30% -30% -30% -30% -30% -20% -20% -20% -20% -20% -10% -10% -10% -10% -10% εbench δ bench Scenario A WorstCase Return -94.10 0.1 0.91% -4.1 0.56% -13.1 0.05 0.73% -33.45% -32.57% -4.37% -47.48% 14.3 0.16% -55.5 0.10 0.1 0.2: Benchmark fund: Worst-case outcomes given scenarios A and B.1 0.13% -15.3 0.49% -87.1 0.07% -15.84% -15.09% Scenario B WorstCase Return -94.94% -31.66% 2.37% -18.3 0.24% 0.36% -72.48% -87.04% -44.26% 0% -47.5 0.1 0. Hveem 41 (103) .05 0.3 0.22% -32.05 Table 4.1 0. Comparison with competition Return Index Months 48-96 / 50-98 p.94% -43.7 0.5 0.78% 11.35% -62.58% 10.80% -63.7 0. εbench and δ bench .1 0.

which the investor cannot lose more than if it starts its investment horizon at quarter 48. it is this eﬀect which also creates the capital protection property. Notable is that it takes longer time for the SPF than the benchmark fund to appreciate after the market turns bust.4. since it does not short bonds to reallocate its portfolio. thus not providing a limit where it cannot lose more capital. bigger problems to consider than its position in the SPF. It is reasonable to demand that the SPF should be less sensitive to market drops of −20% p.1. Since the fund rebalances its portfolio at month 48 a sense of upper loss limit. Table 4. Thus the investor loses a “maximum” (in sense of the underlying) of ε + δ which is applicable to both the SPF and benchmark fund. Thus the desired weights of εSPF and δ SPF should provide better capital protection against downside risk than the benchmark fund. By inspecting Table 4. such as in the −80% drop scenario (Table 4. as the tables indicate. which is very desirable as seen in Figure 4. Next follows an analysis regarding the results and an interpretation regarding their implications for a portfolio manager. It is not a huge diﬀerence between the worst-case outcomes for diﬀerent extreme scenarios (−30 to −80% drops) when regarding the SPF (Table 4. Hveem 42 (103) .a. Thus a decline with 50% may erase almost all value attained in the options.2 it is easy to notice that the adequate levels for εSPF and δ SPF are 0. Hence the criteria εSPF + δ SPF < εbench + δ bench must hold if the SPF should have a higher capital protection for declines of the scale 30-50%. The SPF is less sensitive to Scenario B if it is only followed by a correction in the market (return of −10% to −20% p. The Capital Protection Property The structured products fund provides. one for the upside (δ+ ) and one for the downside SPF SPF (δ− ).). By increasing the downside limit δ− the fund has a higher tolerance for holding shortSPF term debt than when having a symmetrical δ SPF . thus making it less sensitive to a market crash. or more (it is diﬃcult to ﬁnd portfolio weights for the fund that would be less sensitive at a −10% market drop2 . is created. Notable is that the results are quite similar if there is a drop of −80% per month between months 50-51. This implies that if the fund experiences a total market crash (−100%) the investor still holds its position in bonds. the results for this scenario are disclosed in Appendix C1 .1 shows that the SPF is in general more sensitive to a large drop in the stock market after an extreme increase without having the possibility to rebalance the portfolio (Scenario B) than to Scenario A. while the market linked assets would by worth zero. since the value of the options does not decrease a lot more in the −80% case than in the −30% case. 4. One interesting property that is not disclosed in the tables is that the fund can limit its SPF downside signiﬁcantly using two diﬀerent δ SPF . M. This is due to that the SPF does not suﬀer a lot of the decline between months 50-51 and the portfolio is rebalanced at month 51 to lower levels.1 (notable is that these boundaries also provides a better worst-case outcome 1 The scenarios considered in the appendix are worse. On the other hand the portfolio does not beneﬁt so much of the rebalancing possibility during month 51 in the case of a huge decline between months 50-51.3 and δ bench = 0. as seen in Table 4.2). most certainly.Chapter 4. It is important to notice that the SPF appreciates quite a lot when the market starts to go up again.1). but if those scenarios would occur the investor would have. since it takes a few quarters to increase the participation rate. Important to notice is that as δ− increases it will be harder for the SPF to appreciate after a bear market since it accepts higher levels of short-term debt and does not reallocate in a high extent to structured products.1).15 respectively 0.a. But the value invested in options is convex and more volatile than investing in the underlying.6 Analysis and summary As mentioned earlier an adequate benchmark fund is a fund which has εbench = 0.1 and 4. The SPF has a greater upside for huge increases than the benchmark fund (Figure 4. (in general for the same levels of ε and δ as the benchmark fund) a much higher safety in extreme downturn markets (−80% drop).14.1 and 4.14). Table 4.2 shows that the benchmark fund on the other hand suﬀers to a high extent when the index has huge negative returns since it reallocates every three-month to the index. 2 Since the price of the options is more volatile than for stocks and indices.

−15% market crash. M. δ SPF = 0.15. δ bench = 0. −30% market crash.14: Illustrative example of the scenarios and movements of the SPF 1 (Portfolio 7) and the benchmark fund for εSPF = 0. εbench = 0. (f ) Scenario B.1. (h) Scenario B.4. 140 Price Underlying NAV Fund NAV Benchmark 140 Price Underlying NAV Fund NAV Benchmark 120 120 100 100 80 80 $ 60 $ 60 40 40 20 20 0 0 50 100 150 0 0 50 100 150 Month Month (c) Scenario A.1. Hveem 43 (103) . −80% market crash. Comparison with competition 140 Price Underlying NAV Fund NAV Benchmark 140 Price Underlying NAV Fund NAV Benchmark 120 120 100 100 80 80 $ 60 $ 60 40 40 20 20 0 0 50 100 150 0 0 50 100 150 Month Month (a) Scenario A.4. 140 Price Underlying NAV Fund NAV Benchmark 140 Price Underlying NAV Fund NAV Benchmark 120 120 100 100 80 80 $ 60 $ 60 40 40 20 20 0 0 50 100 150 0 0 50 100 150 Month Month (e) Scenario A. Figure 4. −30% market crash. (d) Scenario B. −15% market crash. −80% market crash.3. 140 Price Underlying NAV Fund NAV Benchmark 140 Price Underlying NAV Fund NAV Benchmark 120 120 100 100 80 80 $ 60 $ 60 40 40 20 20 0 0 50 100 150 0 0 50 100 150 Month Month (g) Scenario A. −10% market crash. −10% market crash. (b) Scenario B.

It is important that the fund manager reallocates some of the return to either cash or shortterm debt.7 Additional stress test The stress test is conducted to ensure that the SPF 1 yields a higher degree of capital protection than a benchmark fund given the attained levels of εSPF and δ SPF also for other scenarios than M.14). except in the products that are about to mature (especially during bull markets) and newly issued products during bear markets.14). the fund’s goals etc.3). Thus a SPF should have the parameters εSPF = 0. and when minimizing the absolute downside it is recommended to avoid overweighting the newly issued products during bear markets and products maturing the upcoming period. The fund should avoid investing in structured products that have maturity in the nearest future.25 the fund is more aggressive and can take advantage of the momentum gained during bull markets (as well as increasing the possibility of a higher return after bear markets. When the manager does not rebalance the portfolio investors may invest in a fund with a high risk. The fund manager should after periods of deprecation rebalance (if any existing) holdings from short-term debt to the structured products to increase the optionality (in this case the proportion of options) in the portfolio to avoid that new investors will suﬀer from the fund’s previous decline (Section 4. δ SPF = 0. Thus the fund manager should. 4.1 to provide better capital protection than a benchmark fund that has εbench = 0.4). This implies that the fund locks in some of its previous return as it appreciates. To summarize the previous sections: it is important to diversify the SPF over a numerous set of assets (Section 4. there are several constraints that should be made such that the risk of the portfolio is held under control. [26]). since these are very volatile and in practice almost only a bet with a high theta (cost of making the bet). due to the volatility of the option payoﬀs. It is. as the SPF value increases. Hveem 44 (103) . thus increasing the risk.2 or 0. It is also wise to limit the proportion invested in each structured product (to diversify adequately). [35]). Next follows an additional stress test to assure that these levels also provide better outcomes than the benchmark fund under simulated scenarios. by increasing εSPF to 0.4. These are just some of the possible boundaries. 1979.2 and 4.1. The proportion held in options should be restricted as done in this section.3 and δ bench = 0. when trying to minimize the downside risk. that allocates amongst diﬀerent structured products. Thus when formulating an optimization algorithm. and the appropriate levels of εSPF and δ SPF have been attained. imperative that the fund manager understands the risk of investing in structured products in a sophisticated portfolio. allocate broadly over the spectrum of available products. It is important to contemplate regarding what the fund is trying to accomplish and what tradeoﬀs diﬀerent portfolio choices imply. Thus the level of εSPF needs to be decided in comparison with the competition and how the SPF is managed.3). It is also important to rebalance the portfolio after a decline to give the previous investors the possibility to “win the losses back” as people prefer this re-investment strategy due to get-evenitis (Shefrin. thus protecting the previous proﬁts. The SPF 1 has been stressed for the predeﬁned scenarios. with diﬀerent maturities.Chapter 4. hence maintaining the ratio between the capital allocated in options and bonds at a reasonable level (Section 4. and how the capital protection is aﬀected by the market returns and the reallocations. Examples of outcomes for Portfolio 7 using diﬀerent SPF and δ SPF with diﬀerent scenarios are disclosed in Figure 4.4). In a downturn market investing in newly issued products increases the value held in options in the portfolio. It is not only important to rebalance the portfolio after a series of appreciations to decrease the risk of the fund.15. The Capital Protection Property for a −15% market crash as depicted in Figure 4. Figure 4. the fund manager decides the appropriate limits depending on what the SPF is trying to attain in capital protection sense and in relation to competition.14. as shown in this study. 2002. These eﬀects are really desirable and in line with human behavior to rebalance due to loss-aversion (Kahneman and Tversky. It also protects new investors from investing in a (too) high proportion of options in relation to the holdings in bonds. Also the fund should during a bear market avoid overweighting newly issued products (Section 4.

16.35%. since the model should be stressed towards negative outcomes. Thus the strategy for the SPF described above (using Portfolio 7) is applied with the Nasdaq OMXS30 Index as underlying between 1996-01-02 − 2010-09-30. For tenors over one year quoted SEK swap rates are used.59%.5% p.15 and the speciﬁed outcomes) shows that SPF 1 is a better investment alternative than the benchmark fund.2 2000 2000 1500 1000 1000 1000 500 500 500 0 −0. which should yield better worst-case outcomes than the benchmark fund according to the study conducted above.a. 4. Thus the result (Figure 4.8 1 1. which is optimal for risk-averse investors.0105 and µ = 0 on a days basis. mean: (b) Benchmark: min: -24.4 0. The study is conducted through investigating 500. max: 105. The risk-free rate is assumed to equal the STIBOR.2 0 0. This is a good proxy for the interest rates for ELNs in the Nordic market where the bonds are issued by local investment banks. Figure 4.15b).a. (c) Underlying: min: -80.r.2 1. notable is that SPF 1 is superior in extreme worst-case scenarios as well. notable is that it is not important exactly what the parameters are since the study searches for patterns. Comparison with competition Histogram over the SPF:s 36 month (overlapping) returns 3500 3000 Histogram over the Benchmark funds 36 month (overlapping) returns 4000 3500 2500 3000 Histogram over the Underlying 36 month (overlapping) returns 3000 2500 2000 2500 Outcomes Outcomes 1500 Outcomes −0. the choice of these parameters should depend on the goals of the fund and thus determine the risk-level. εSPF and δ SPF it is interesting to investigate how the fund would have performed in the past through a backtest. thus the STIBOR rate reﬂects the rate of the zero-coupon bonds for ELNs issued in Sweden in SEK.8 1 1.a.3%. kurtosis is also displayed on the positive side and not the negative side.15: Histograms over 36 months returns for the competing funds and the stress test. available on Bloomberg.45%. Figure 4.4 0 −1 1500 0 −0.4.2 0 0.6 Return 0.15.15 and δ SPF = 0. The same type of Monte Carlo simulation as performed in Section 4.1. and the volatility used when pricing the options is 20. is discloses in Appendix C (Figure C. Higher εSPF and δ SPF implies a higher downside risk and the historical worst-case scenario is best for the most restrictive choice of εSPF and δ SPF .15a and 4. The following parameters are also used: rf = 6% p. max: 455.15a indicates that SPF 1 has a positive skewed distribution.1). thus the daily log returns for the index are modeled with a student’s t-distribution with ν = 5. The interest rates for the zero-coupon bonds are given by the STIBOR (Stockholm Interbank Oﬀered Rate). which indicates that SPF 1 has the most desirable distribution of the three and the most limited downside.01%. The drift equals to zero. Thus as mentioned earlier. mean: 15. σ = 0.3 is used.4 −0. max: 127.05%. The volatility used is the realized volatility over every last three-month period. for diﬀerent εSPF and δ SPF .26%. thus it may aspire to investors who are looking for a decent upside and a limited downside. Hveem 45 (103) .2 0.6 0. Notable is also that SPF 1 has a higher upside than the benchmark fund (comparing Figure 4.4. M.19%.8 Backtesting A backtest means that a modeled is tested on historical data and evaluated on that basis. mean: 6.000 36-month periods and the 36 months return.4 0 1 2 Return 3 4 5 (a) SPF 1: min: -23.2 0.4. since this is a quite conservative choice and a common rate to use as the risk-free rate (a higher rf will yield a lower return for the SPF). The results for a stress test where µ equals −30% p. the ones speciﬁed above. Now that it is concluded how a SPF should be constructed w. and certainly the index when considering downside risk. The backtestresults for diﬀerent portfolios are disclosed below.19%. The result is disclosed in Figure 4. Thus the next subsection focuses on a backtest for the SPF and how it would have performed historically. Thus SPF 1 has εSPF = 0.69.4 Return 0.t.

based on this index. SPF 4.10.16: Backtest results: OMXS30 vs SPF 1 (Portfolio 7) and Benchmark. δ SPF = 0.25. SPF 4. δ SPF = 0.2.17: Backtest results: OMXS30 vs SPF 1 (Portfolio 3) and Benchmark. Thus it is not a perfect benchmark and a more adequate benchmark is disclosed in Chapter 6.30%.05.95%. which is not including dividends and a benchmark fund. 400 350 300 400 350 300 $ 250 200 150 100 1996 $ 250 200 150 100 1996 1998 2000 2002 2004 2006 2008 2010 2012 1998 2000 2002 2004 2006 2008 2010 2012 Year Year (c) εSPF = 0. Figure 4. The big diﬀerence lies in their allocation schemes for the structured products.2. Figure 4. δ SPF = 0. the total return version only exists since 2002. (b) Nasdaq OMXS30 NAV Fund NAV Benchmark εSPF = 0. 400 350 300 Nasdaq OMXS30 NAV Fund NAV Benchmark 400 350 300 Nasdaq OMXS30 NAV Fund NAV Benchmark $ 250 200 150 100 1996 $ 250 200 150 100 1996 1998 2000 2002 2004 2006 2008 2010 2012 1998 2000 2002 2004 2006 2008 2010 2012 Year Year (a) Nasdaq OMXS30 NAV Fund NAV Benchmark εSPF = 0.16c: −9.17% and for the index −67.1. 400 350 300 Nasdaq OMXS30 NAV Fund NAV Benchmark 400 350 300 Nasdaq OMXS30 NAV Fund NAV Benchmark $ 250 200 150 100 1996 $ 250 200 150 100 1996 1998 2000 2002 2004 2006 2008 2010 2012 1998 2000 2002 2004 2006 2008 2010 2012 Year Year (a) Nasdaq OMXS30 NAV Fund NAV Benchmark εSPF = 0. δ SPF = 0.05.25.16a: −2.15.15. M.15%. This standard index version is used in this chapter to be able to cover two market crashes. The Capital Protection Property The time horizon considering capital protection is of three years as mentioned earlier. δ SPF = 0.16d: −16. The worst-case outcome over all overlapping three years periods for the benchmark is −14.10. SPF 4. Notable is that the comparison is done with the OMXS30.1. (b) Nasdaq OMXS30 NAV Fund NAV Benchmark εSPF = 0.1.79%. (d) εSPF = 0. δ SPF = 0. 400 350 300 400 350 300 $ 250 200 150 100 1996 $ 250 200 150 100 1996 1998 2000 2002 2004 2006 2008 2010 2012 1998 2000 2002 2004 2006 2008 2010 2012 Year Year (c) εSPF = 0. Hveem 46 (103) . δ SPF = 0. δ SPF = 0. SPF 4. The results indicate that the SPFs have the desired properties.Chapter 4.1.16b: −8. but it will not be covered as extensively as SPF 1 since SPF 2 is mostly of theoretical interest. The next subsection is regarding an alternative to the SPF 1 described above.1. (d) εSPF = 0.13%.1. Thus most of the portfolios had a better worst-case return than the benchmark fund for the OMXS30 index between 1996 to the third quarter of 2010.

3. M. The ˜ weight vector amongst the options is denoted woptions and is in this case given as follows. Thus option Portfolio 3 will be used as a base to calculate the adequate portfolio weights for the structured products.9 Structured products fund 2 The structured products in SPF 1 were allocated using a ﬁxed weights scheme. given as.19 indicate that SPF 2 does not only provide a better protection during the market crash but also a better future potential when the market starts to appreciate and turns into a bull market. Important to notice is that this section is very theoretical and has less practical applications than SPF 1. The results for Scenario A and B are given in Table 4. The fund reallocates its portfolio in the same manner as for SPF 1 (using the short-term debt etc. the underlying. since the portfolio weights between two time periods often would change signiﬁcantly. even though the total value held in options are under control. thus it is reasonable that SPF 2 should beat SPF 1 in terms of worst-case outcome w. the only diﬀerence is that w diﬀers.3 it is easy to see that the SPF 2 has much better worst-case outcomes including every scenario and every selection of εSPF and δ SPF than SPF 1.4.r.3. SPF 2 is quite theoretical since the ratio options:bonds may decline signiﬁcantly for some products if the market turns bust. A temporary weight 1T wtemp The rest of the allocation scheme is exactly the same as in Section 4. Figure 4. The same analysis is conducted for this fund construction as for SPF 1. ξ. (t) i Ot i i βt +Ot . Hveem 47 (103) .). through the limits of εSPF and δ SPF . Thus inﬁnitesimal positions in some products and huge positions in others may be necessary to attain the desired level of options held in each tenor. The same products are available as for SPF 1. ˜ wt = wtemp (t) .3 for the SPF 1. Thus to minimize the risk of movements in the underlying the fund should allocate the portfolio based rather on the options itself than based on the structured products. Thus it is impossible to guarantee that the fund will hold the desired amount of options from the diﬀerent tenors in comparison to each other using SPF 1. if i > j i oi t ˜ thus w is given as. This implies that the scheme may be adequate in times of normal market movements.4.3.t. j its index and oi = t vector wtemp (t) is created.1 and 4. Comparison with competition 4. the only thing that is diﬀerent is that the weights are attained through a diﬀerent scheme. if i = j temp wi (t) = j woptions ot . This means that the portfolio may overweight certain options signiﬁcantly. This means that the allocation algorithm calculates the weights for the structured products based on the proportion of options:bonds in the products and the desired relation of concentration amongst the options. The optimal portfolio weights that minimized the risk of movements in the underlying were discovered in Section 4. let ξ be the ﬁrst non-zero element of woptions . The downside risk should be minimized for movements in the underlying. since the value of options is more volatile than the whole structured product. This scheme would also generate massive transaction costs. which indicated the weights allocated in the certain structured products. The Portfolio 3 was the best choice to minimize downside risk with in Section 4. By comparing the results in Table 4.4.4.14 and 4. Now consider the same setting as for the SPF 1. but is disrupted in stressed scenarios. if i < j 0. which is quite realistic. woptions = 0 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 1 11 T .

2 400 350 300 Nasdaq OMXS30 NAV Fund NAV Benchmark 400 350 300 Nasdaq OMXS30 NAV Fund NAV Benchmark $ 250 200 150 100 1996 $ 250 200 150 100 1996 1998 2000 2002 2004 2006 2008 2010 2012 1998 2000 2002 2004 2006 2008 2010 2012 Year Year (a) Nasdaq OMXS30 NAV Fund NAV Benchmark εSPF = 0. Hveem 48 (103) .05 (b) Nasdaq OMXS30 NAV Fund NAV Benchmark εSPF = 0.1 (d) εSPF = 0.18: Backtest results: OMXS30 vs SPF 2 (Portfolio 3) and Benchmark. implications etc. M. are applicable to SPF 2 in the exact same way as SPF 1.1 400 350 300 400 350 300 $ 250 200 150 100 1996 $ 250 200 150 100 1996 1998 2000 2002 2004 2006 2008 2010 2012 1998 2000 2002 2004 2006 2008 2010 2012 Year Year (c) εSPF = 0.Chapter 4.2.1 Figure 4.10. Hence only the results are disclosed for the interested reader.15. δ SPF = 0. The Capital Protection Property In the SPF 2 case it is possible to choose εSPF = 0. The same analysis. The following section discusses the terminology capital protection and how it should be used further in this thesis. δ SPF = 0.20 and still have a lower downside risk than the benchmark fund. δ SPF = 0.25. δ SPF = 0.

−30% market crash. δ bench = 0. 140 Price Underlying NAV Fund NAV Benchmark 140 Price Underlying NAV Fund NAV Benchmark 120 120 100 100 80 80 $ 60 $ 60 40 40 20 20 0 0 50 100 150 0 0 50 100 150 Month Month (e) Scenario A. δ SPF = 0. −80% market crash. (d) Scenario B.3. −15% market crash. εbench = 0.4. −30% market crash. (b) Scenario B. −80% market crash.4. (h) Scenario B. Figure 4.15.19: Illustrative example of scenarios and movements of the SPF 2 (Portfolio 3) and benchmark fund for εSPF = 0. 140 Price Underlying NAV Fund NAV Benchmark 140 Price Underlying NAV Fund NAV Benchmark 120 120 100 100 80 80 $ 60 $ 60 40 40 20 20 0 0 50 100 150 0 0 50 100 150 Month Month (g) Scenario A. M. Comparison with competition 140 Price Underlying NAV Fund NAV Benchmark 140 Price Underlying NAV Fund NAV Benchmark 120 120 100 100 80 80 $ 60 $ 60 40 40 20 20 0 0 50 100 150 0 0 50 100 150 Month Month (a) Scenario A.1. Hveem 49 (103) . −10% market crash. 140 Price Underlying NAV Fund NAV Benchmark 140 Price Underlying NAV Fund NAV Benchmark 120 120 100 100 80 80 $ 60 $ 60 40 40 20 20 0 0 50 100 150 0 0 50 100 150 Month Month (c) Scenario A. (f ) Scenario B.1. −15% market crash. −10% market crash.

88% 5.1 0.1 0. The Capital Protection Property Return Index Months 48-96 / 50-98 p.20 0.05 0.17% -8.30% -2.1 0.15 0.72% 1.20 0.05 0.79% -4.05 1 0.82% 0.20 0.1 0.76% -5.a.25 0.11% 0.10 1 0.3 0.25 0.89% -10.51% -14.00% 1.15 0.15 0.10 1 0.16% -8.76% -3.1 0.79% -11.3 0.3 0.70% -2.05 0.77% -1.93% -0.80% -6.25% -3.69% 2.58% 0.47% -3.05 1 0.61% -21.60% 0.20% 2.15 0. M.3 0.17% -8.24% -2.39% -9.51% -3.89% -10.05 0.3: SPF 2: Worst-case outcomes for the SPF 2 based on Portfolio 3 option weights given scenarios A and B.25 0.Chapter 4.3 0.15 0.15 0.80% 4.1 0.1 0.79% -11.88% -5.15 0.74% 4.49% -1.90% 2.05 1 0.48% -9.15 0.1 0.67% -20.1 0.1 0.61% -1.89% -4.08% -11.26% 1.17% -3.1 0.15 0.1 0.00% -0.66% -12.90% -4.29% Scenario B WorstCase Return -30.15 0.05 Table 4.74% -4.11% -8.48% -3.48% -9.88% -5.88% -1.25 0. εSPF and δ SPF .1 0.1 0.20 0. -80% -80% -80% -80% -80% -80% -80% -50% -50% -50% -50% -50% -50% -50% -30% -30% -30% -30% -30% -30% -30% -20% -20% -20% -20% -20% -20% -20% -10% -10% -10% -10% -10% -10% -10% εSPF δ SPF Scenario A WorstCase Return -11.20% -10.24% -10.54% -7.98% -1.20 0.22% -2.76% -18.1 0.1 0.1 0.67% 0.08% -6.24% -9. Hveem 50 (103) .10 1 0.36% 7.07% -6.10 1 0.05 0.1 0.25 0.05 1 0.16% 1 0.1 0.25% -15.10 1 0.60% 3.1 0.10% -2.69% -1.42% -14.

if the investors hold their position during this time horizon they will receive a certain level of capital protection.5. alternative deﬁnitions of capital protection. Note that these alternative deﬁnitions are for capital protection. Cons The point of using this deﬁnition is to quantify the worst-case outcome. a certain percentage of the investment should be guaranteed. and many investors mix up the terminologies (Chuan.4. The aim of the portfolio allocation is that the worst-case scenario should exceed the invested amount X0 times the degree of capital protection κ (determines the degree of capital protection. 1979. Alternative deﬁnitions of capital protection 4.. Still.5 Alternative deﬁnitions of capital protection Since it is not possible to construct a capital guaranteed SPF it is imperative to discuss the usage of the terminology capital protection. i where XT is the portfolio value at time T (end of the investment horizon) for the ith scenario. which indicates that the portfolio will not have a negative return.1 Deﬁnition 1 . It is absolute in the sense that it cannot be distorted and it is not possible to argue that a placement is capital guaranteed if it may have a negative return (still not taking the credit risk into consideration). i.. [37]). Thus when claiming that a fund is capital protected it should have a limited downside risk. not guarantee. The terminology capital protection on the other hand is more diﬀuse and ﬂexible. [26]). The capital protection property is regarding a certain time horizon. it is quite easy to have the same associations to the word protection as the word guarantee. These deﬁnitions are evaluated regarding their pros and cons and which deﬁnition is best to use. i. N . [12]). Thus while it is not possible to call the fund capital guaranteed it is essential that the fund uses the terminology capital protected in its communication with the investors to appeal to their loss-aversion (Kahneman and Tversky. 4. The investor can be quite conﬁdent that it will not lose more capital than the given level over the time horizon. which means that the portfolio’s performance should be Monte Carlo simulated with a certain number of scenarios.. N. The terminology capital guarantee is something absolute.e. which generates a worst-case outcome for the portfolio. . the assumptions regarding the market model and M. By using the original deﬁnition and modify it slightly it may be possible to construct a fund which fulﬁlls the requirements for an alternative deﬁnition of capital protection. What does it mean that an investment is capital protected? Capital protection means.e. i min XT > κX0 i i = 1. 2008. It is also possible to use a sophisticated algorithm to solve the linear optimization problem as given in Section 2.5.Absolute lower bound This deﬁnition is a worst-case scenario based deﬁnition. Hveem 51 (103) . ensuring a well-deﬁned solution to the problem. that the investor is guaranteed a certain minimum repayment of the invested amount in the end of its investment term (time horizon). The focus lies on two alternative deﬁnitions of capital protection. can be any positive number). Pros This is probably the most concrete deﬁnition of capital protection since the worst outcome simulated actually would exceed the minimum acceptable level of return. Hence the portfolio is considered as capital protected if. according to UBS (2010.2. but the worst-case outcome will always be based on the scenarios. This leads us to the purpose of this section.3.

CVaR This deﬁnition is based on CVaR.000. CVaR is the risk measure that will be used further on in this thesis. how many scenarios are needed? Choosing between N = 10. “if it goes bad. it is a coherent risk measure and the result converges as the number of scenarios grows.000 scenarios. the yield curve etc. For more information regarding the deﬁnition of CVaR please consult Section 2. Another advantage is that the linear optimization problem for CVaR is well deﬁned and easy to solve using the simplex method. Cons The negative aspect is that it does not provide an absolute deﬁnition of capital protection.3. Also. which means that the CVaR for the portfolio at some quantile α must be less than (1 − κ) X0 .g. A negative side of this deﬁnition is also that the risk measure is only saying something about the worst-case outcome. the options.5. The focus on minimizing the worst-case outcome may impose a high risk for the investor. It is also a risk measure that can capture the risk of heavy tails and skewness. The Capital Protection Property how the underlying is modeled. thus eliminating the possibility to gain when the market goes up).000 or 100. due to its major beneﬁts. since it discloses. It is more robust. so how is it possible to determine the adequate number of scenarios to use. Also it may be a more adequate deﬁnition than the worst-case deﬁnition. The portfolio weights will in general diﬀer for the two diﬀerent risk measures.95 CVaR is the expected value of the loss for the 5% worst outcomes. Note that it is possible to use the alternative deﬁnition 1 in the same manner as with alternative deﬁnition 2 in the upcoming scenario optimization. 50.2 Deﬁnition 2 . Thus the portfolio is considered to be capital protected if. by solving the minimum regret optimization problem instead. since they measure diﬀerent things. 4. which one is adequate? The problem is that when increasing the number of scenarios. leading to a suboptimal portfolio. much more than using a worst-case outcome deﬁnition. Hveem 52 (103) . e. For example by using the quantile α = 0. if κ = 1 the portfolio may end up in only investing in bonds and almost nothing in options. how bad does it get?” and not only for the worst-case scenario. since the worst-case scenario would for most portfolios converge to −100% in return given that N → ∞? Another problem with this deﬁnition is that it can be extremely hard to ﬁnd a portfolio that actually fulﬁlls this requirement (depending on the choice of κ.1. not anything about the remaining unfavorable outcomes. for a determined level of capital protection κ. the number of extreme outliers are also increased. if the CVaR is positive it does not guarantee that there are no negative outcomes.Chapter 4. Thus it is possible to capture a lot of the risk using CVaR. CVaRα (Lw ) = E [Lw | Lw ≥ VaRα (Lw )] ≤ (1 − κ) X0 Pros A positive side of this deﬁnition is that it is more realistic (in sense that a portfolio may actually fulﬁll the requirement even for high values of κ) than the worst-case deﬁnition. M.

Thus by using Black-Scholes formula and implied volatility the correct price is attained. When comparing the total return index with the standardized version it is easy to see that the total return index yields a higher return of approximately 3. The implied volatility is the volatility that should be used in Black-Scholes formula so that it gives the fair price of the option. there are no systematic diﬀerences for the prices of ELNs when using either implied volatility or realized volatility (Wasserfallen and Schenk. thus it is based on 52 weekly observations of the index log returns. dependence on the chosen market model in this chapter.3. in Section 3. where dividends are reinvested in the index. 5.a. which implies that it is hard to ﬁnd appropriate historical data such that the price of the options can be modeled through implied volatility (surfaces). The thesis is regarding ELNs and as discussed earlier. thus the components that must be modeled are: the yield curve. Hult and Lindskog.0% p. 53 . It is important to notice that the liquidity of options on OMXS30 is quite poor. since the OMXS30 index is not including dividends. the underlying and the volatility. in the Black-Scholes model.2 Index There exists countless of models describing the movements of a stock or an index. which implies that the expected dividend yield is 3. This thesis studies Equity-Linked Notes. [39]). It is of great importance that the risk factors are modeled in a way such that they reﬂect the underlying risk factors. Probably the most popular model is to describe the movements of the underlying as a geometric brownian motion as done by Black and Scholes (1973. 2006.1. There exists a total return version of OMXS30. as far as possible. These risk factors determine the fair price of the fund. 2009. It is also common to model the returns by using for example normal variance mixtures (Aas. [1]. Thus the volatility will be modeled as the realized volatility during one year.0% p. The problem with Black-Scholes formula is that it does not always price the options correctly due to the misspeciﬁcation that the price of an asset follows a geometric brownian motion. 5.1 Volatility and option pricing It is important that the options in this study can be priced through a closed-form expression. The reason why the previous study has been so extensive is to avoid. The dividend rate of the index must be estimated. using continuous compounding during the observed time period (since 2002 when it ﬁrst was started to 15th of November 2010). The Black-Scholes formula for option pricing will therefor be used for pricing the plain vanilla options. since the study involves the usage of Monte Carlo simulation and a simulation based pricing model would thus be too computer intensive. 1996.a. [23]). [9]).CHAPTER 5 Modeling Financial Assets Modeling the ﬁnancial assets is one of the most important issues in portfolio optimization and allocation problems. Thus it may reduce the misspeciﬁcation and error in pricing on the Nordic market to use realized volatility instead of implied volatility. which is available on Bloomberg. since it provides a fast and adequate pricing model.

Thus the PCA should result in four signiﬁcant factors. Modeling Financial Assets Another popular model is proposed by Heston (1993) in [20] where he proposes a model including stochastic volatility that results in a closed-form expression for option pricing. Not only the underlying should be modeled but also the yield curve. This choice of model guarantees that the sampling of scenarios will be parsimonious. Since one of the most appropriate ways to model the yield curve is by PCA and since PCA also is a good method to describe stock/index returns it is a natural choice to use PCA for the two combined. STIBOR is the daily interbank oﬀered rate in Stockholm.Chapter 5.riksbank. 1991. The Swedish Riksbank publishes daily quotes of the STIBOR on their homepage http://www. also it is important to maintain the correlation in the data. The correlation between OMXS30 and the Swedish yield curve is negative for shorter tenors and positive for tenors above one year (also the absolute value of the correlation is much higher for longer tenors). thus it also reﬂects the yield on the zero-coupon bonds that are issued by the local investment banks. since this is a common choice as the risk-free rate when pricing in the Swedish market. which will be enough to explain most of the dependence structure in the data.g. in the Swedish market the STIBOR is not independent of the stock index OMXS30. 1400 1200 Nasdaq OMXS30 Index 1000 800 600 400 1998 2000 2002 2004 2006 2008 2010 Year Figure 5. enough to describe the market movements. 5. Hence the STIBOR is an appropriate rate both for the risk-free rate and the zero-coupon rate.4 A PCA model for the yield curve and index A four-factor model is. it is desirable to have a model that is easy to sample from. The correlation matrices for daily. An adequate method to use is principal component analysis (PCA) as described in Section 2.e. Thus it is better to use weekly log returns for the simulation since the dependence structure amongst the data is not contaminated as much as the daily log returns by noise. In the described market model the STIBOR will be used as the risk-free rate. 5. Since the aim of this chapter is to model scenarios. steepness and curvature (Litterman and Scheinkman.2 together with the index to maintain the dependence structure. be described by three dominant factors in a three-factor model. The correlation is in general higher for weekly returns than for daily returns during the observed time period. principal components. i. M. these are the level. Hveem 54 (103) . since the yield curve is described by three factors and the index (mainly) by the remaining factor. thus it is necessary to consider the dependence structure between the index and the yield curve. [29]). in general.1: The closing index levels of OMXS30 between January 1996 and November 2010. Therefor the yield curve will be modeled as described in Section 2. weekly and yearly log returns between OMXS30 and the SEK rates between 1M to 3Y are disclosed in Appendix D. for the risk factors so that the scenarios can be evaluated in an optimization problem.3 Yield curve It is as mentioned earlier a good choice to use PCA to model the yield curve changes since a yield curve can. There exists signiﬁcant dependence between a yield curve and its market’s corresponding index.se. e. as mentioned earlier.2.

57 0. which is more than enough. The volatility is modeled by using the one year realized volatility. 9M.4. By inspecting Figure 5.94% of the variation in the data.00 84.82 94. 6M. STIBOR has the tenors 1M.4.81 9. PC 1 2 3 4 5 6 7 8 9 Explained variation [%] 68. [29]) as well as by Barber and Copper (2010. also the factor loading for the index is almost zero. 12M and the swap rates the tenors 2Y and 3Y. 2M. [6]) that most PCAs for yield curves can with only three PCs describe more than 95% of the variation (notable is that by removing the index the ﬁrst three factors explains 97. The STIBOR is used as the risk-free rate up to one year and above one year the SEK swap rates are used. Hveem 55 (103) . Also it seems more adequate for the observed data sample to model with log returns than absolute changes. which are weakly stationary.44 97.94 98.72 0. 5.4. since these are in some sense already log returns.1 Data set The OMXS30 from the beginning of 1996 until the middle of November 2010 is used for the simulation. Thus the index and the yield curve nodes are modeled with log returns.30 0. the data set ranges from the beginning of 1996 until the middle of November 2010. The M. 3M.89 100. thus there is a total of nine PCs. the ﬁrst four PCs explain 97.00 16.5. The index and the yield curve must be rewritten on a stationary form since they are not stationary.50 0.22 99. A linear interpolation is performed for products with a time to maturity that does not coincide with these tenors.1: PCA results: Variation explained by each principal component and total explained variation (cumulative explained variation).37 0. it is possible to prevent negative interest rates by this procedure.11 Cumulative explained variation [%] 68.00 Table 5. This is in line with the results found by Litterman and Scheinkman (1991.68% of the variation).2 PCA results A PCA is performed on the whole data set. 5. A PCA model for the yield curve and index 8 7 6 [%] 5 4 3 2 1 1998 2000 2002 2004 2006 2008 1M 2M 3M 6M 9M 12M 2Y 3Y 2010 Year Figure 5.59 99. Even though it may seem strange to use log returns for the yield curve.65 99. Notable is that the four PCs describe more than 95% of the variation.62 3. which is very important.3 it is easy to see that the ﬁrst PC’s factor loadings are all positive and can be interpreted as the level change.2: SEK STIBOR and Swap rates between January 1996 and November 2010.

This means sampling random dates (with equal probabilities) that are used to extract the corresponding factor scores.8 −1 OMXS30 1M 2M 3M 6M 9M 1Y 2Y 3Y Risk Factor Figure 5. To conduct the study the data between January 1996 and June 2010 is divided into overlapping three years intervals with a distance of six months. even though the factor loadings exhibit some sample dependence.2 0 −0. performing the PCA on daily returns generates less stable results. The third PC corresponds to the index since the factor loading is close to one for OMXS30 and small for the rest of the factors. which is enough to retrieve a good estimation of the factor loadings. The advantage of using historical simulation is that the scenarios have happened in the M. Modeling Financial Assets Principle Components 1 0. explaining changes in OMXS30 and the yield curve up to three years. samples of factor scores should be drawn from the empirical distribution. To investigate if the model is adequate the factor loadings and explanatory power of overlapping PCAs must be studied through time and investigated for stability. The PCA is computed on weekly returns in this section. Concluded: the factor loadings are relative stable through time and stable enough to serve for simulation. Thus the four PCs explain most of the variation in the data sample regardless of which sample period selected.4. Hveem 56 (103) . The PCs’ factor loadings for the overlapping periods are disclosed in Figure 5. It is important that the factor loadings are stable through time if the distribution is used for sampling (since the estimation is used as a proxy for the future).3: Factor loadings for the ﬁrst four principle components. Thus this gives approximately 150 observations. the factor loadings have the same structure for all samples.4 PC 1 (Level) PC 2 (Steepness) PC 3 (Index) PC 4 (Curvature) Factor Loadings 0. PC 1 has the smallest variation. the ﬁrst PC explains between 59% and 74% of the variation and the other PCs exhibit similar patterns.8 0. The results are disclosed in Table 5.2 −0. The fourth PC corresponds to curvature since products in the middle tenors have positive factor loadings when the others have negative. Notable is that the explanatory power for the whole period was 97. second PC corresponds to steepness since its factor loadings change sign moving to longer tenors. and replace the sample before the next draw. Thus to simulate new PCs.94% and that the explanatory power in the model is higher during the latter part of the time period.4 −0. 5.48%.3 Historical simulation Historical simulation is a non-parametric model that draws samples from the empirical distribution with replacement. The factor loadings have been estimated during a period of approximately ﬁfteen years.6 −0.6 0.4.Chapter 5.2. On the other hand the total explanatory power of the four PCs is always above 93. with lower explanatory power. while the other three have higher levels of variation.

73 97.59 24.52 2.58 19.00 8.53 18.95 94.75 8.47 10.77 62.50 Cumulative [%] 97.84 67.57 71.51 74.48 93.91 2.79 61.53 4.5 0.80 60.88 58. A PCA model for the yield curve and index Start Jan 1996 Jul 1996 Jan 1997 Jul 1997 Jan 1998 Jul 1998 Jan 1999 Jul 1999 Jan 2000 Jul 2000 Jan 2001 Jul 2001 Jan 2002 Jul 2002 Jan 2003 Jul 2003 Jan 2004 Jul 2004 Jan 2005 Jul 2005 Jan 2006 Jul 2006 Jan 2007 Jul 2007 Jan 1996 End Dec 1998 Jun 1999 Dec 1999 Jun 2000 Dec 2000 Jun 2001 Dec 2001 Jun 2002 Dec 2002 Jun 2003 Dec 2003 Jun 2004 Dec 2004 Jun 2005 Dec 2005 Jun 2006 Dec 2006 Jun 2007 Dec 2007 Jun 2008 Dec 2008 Jun 2009 Dec 2009 Jun 2010 October 2010 PC 1 73.24 7.68 4.38 20.92 3.16 10.55 10.94 Table 5.93 10.05 17.38 61.96 12.76 16.62 PC 4 3.27 61.66 4.00 98.08 4.09 7.91 11.85 9.80 66.62 4.4: Factor loadings for the ﬁrst four PCs for 24 overlapping time periods between January 1996 and June 2010.68 59.59 93.44 4.87 93.34 65.71 12.5 −1 OMXS30 −1 OMXS30 1M 2M 3M 6M 9M 1Y 2Y 3Y Risk Factor Risk Factor Principle Component 3 1 1 Principle Component 4 Factor Loadings 0 Factor Loadings 1M 2M 3M 6M 9M 1Y 2Y 3Y 0.64 9.07 96.10 11.79 98.85 2.5 −1 OMXS30 −1 OMXS30 1M 2M 3M 6M 9M 1Y 2Y 3Y Risk Factor Risk Factor Figure 5.59 4.09 2.17 9.18 69.20 2.96 16.77 23.61 72.48 1.2: Variation explained by each principal component and total explained variation (cumulative explained variation) for 24 overlapping time periods between January 1996 and June 2010.51 93.70 2.19 64.5 0 −0.54 10.5 0. each line corresponds to one sample.74 10.58 18. M.42 21.30 11.98 65.42 69.22 98.62 64.18 11.30 97.12 97.10 97.74 97.11 99.74 9.85 17.99 3.01 23.81 PC3 9.52 97.84 14.21 9.90 97.21 20.94 98.26 65.84 17.42 4.95 66.58 7. Hveem 57 (103) .89 68.52 5.35 99.17 9.89 2.99 3.5 −0.09 96.69 3.24 63.43 98.15 9.81 8.28 7.51 20.90 4.5.46 93.5 −0.66 11.00 PC 2 11.59 62.37 96.5 0 −0.88 3.35 67.96 20. Principle Component 1 1 1 Principle Component 2 Factor Loadings 0 Factor Loadings 1M 2M 3M 6M 9M 1Y 2Y 3Y 0.88 16.4.01 3.64 10.96 12.34 21.38 98.

5. not yearly or monthly returns.0296 2.4 Monte Carlo simulation The idea of a Monte Carlo simulation is to ﬁt a parametric distribution to the observed data. Modeling Financial Assets past. random samples are drawn from the parametric distribution to generate new data. The following parameters for the M.00 1. they are realistic and reﬂect the empirical distribution. Thus in the case of the PCA a parametric distribution should be ﬁtted to the observed factor scores.6. On the other hand the problem is that there is no possibility of implementing new scenarios. The sample should also be rescaled with its historical mean and standard deviation to attain the ﬁnal sampled data.k k.00 0. Hveem 58 (103) . instead of using e. 4 ri.1763 0.j .g. as shown in Figure 5. A relative good ﬁt is actually attained for this sample with a multivariate t-distribution. t-location scale µ ˆ σ ˆ ν ˆ 0. k.8734 0. copulas.00 0. But the model is not restricted in this setup since the return over a three-year period is generated from weekly returns.8189 8. but it is not as good as the t-copula model as disclosed further down.k + εi. Also it is really simple to implement and use historical simulation. or other graphical illustrations such as histogram-ﬁts and scatter plots. thus the model is restricted to the outcomes of the past. 9 εi. It is easy to see that the PCs can be described by t-distributions with diﬀerent levels of freedom. By using QQ-plots. Scatter plots indicate that the four PCs come from a spherical distribution and that there exists dependence between the samples (scatter plots have elliptical shapes) as seen in Figure 5.j = k=1 QT Fi.Chapter 5.8880 4. This means that if no extreme events have happened in the past the probability of generating an extreme scenario is zero. location and scale. Multivariate t-distribution If ν (degrees of freedom) for the marginal distributions (in this case the individual PCs) are close to each other it is possible to use a multivariate t-distribution.j ∼ N 0. thus simulating three-years returns using weekly data does not put any serious restrictions on the simulation.00 0. The ith sample of the jth variable’s return using the ﬁrst four PCs in a historical simulation is given as.4189 0.7824 PC 1 2 3 4 Table 5.2976 2.j where.4.j . The model would be limited when studying only weekly returns. it is possible to illustrate and identify appropriate parametric distributions. 5.3: The ﬁt for a student’s t-distribution to each marginal distribution (factor scores) for the ﬁrst four PCs attained through a maximum likelihood estimation. Var k=5 QT F•.

4 0.1 0.2 0.25 0. A PCA model for the yield curve and index Principal Component 1 0.2 (b) PC 2 Principal Component 4 Factor Scores PC 4 Student’s t−distribution 0. M.4 0.25 0.6 0.5. 15 15 10 8 10 10 6 4 5 5 2 PC 2 PC 3 0 0 PC 4 −30 −25 −20 −15 −10 −5 0 5 10 15 20 0 −2 −5 −5 −4 −6 −10 −10 −8 −15 −35 −30 −25 −20 −15 −10 −5 0 5 10 15 20 −15 −35 −10 −35 −30 −25 −20 −15 −10 −5 0 5 10 15 PC 1 PC 1 PC 1 10 8 6 4 2 10 6 8 6 4 2 2 4 PC 3 PC 4 0 −2 −4 −6 −8 −10 −15 0 −2 PC 4 −10 −5 0 5 10 15 0 −2 −4 −6 −8 −6 −10 −5 0 5 10 15 −10 −15 −10 −5 0 5 10 −4 PC 2 PC 2 PC 3 Figure 5.5 1 0.05 0 −10 −8 −6 −4 Data −2 0 2 4 (a) PC 1 Principal Component 3 0.3 0.5: Histograms for the distribution of factor scores (PC 1-4).1 0.1 0.6 Factor Scores PC 3 Student’s t−distribution 1.35 Factor Scores PC 1 Student’s t−distribution 0.4 0. the red line illustrates the pdf of a t-distribution. the t-distribution seems to be an adequate parametric distribution for the PCs’ marginal distributions.15 0.15 0.2 Density 0.4.05 0 −30 −25 −20 −15 Data −10 −5 0 5 0.3 Density 0.2 0 −4 −3 −2 −1 Data 0 1 2 3 0 −3 −2 −1 0 1 Data 2 3 4 5 6 (c) PC 3 (d) PC 4 Figure 5.35 0.2 0.45 Principal Component 2 Factor Scores PC 2 Student’s t−distribution 0.6: Scatter plots of pairwise PCs and corresponding factor scores for the empirical (observed) samples. Hveem 59 (103) .3 Density 0.8 Density 0.

draw samples of F and Fε and plug in the sampled values into Equation 5.0617 as above. ˆ Σ= µ = .1201 0 It is common to assume that the residual term is multivariate normal distributed.0195 0 0 0 0 0 0.2170 0.j i.1680 −0. For the observed data set the stochastic vector is approximately multivariate t-distributed. The marginals described in Table 5. which has the ˆ following dependence structure. 2. F ∼ tν (0.337 0 0 0 0 0 0 0.0000 −0. Take the inverse of each cumulative ˆ marginal distribution on the samples to attain the new centered and rescaled samples.5770 0 0 . ˆ ν = 3. If all the marginals had the same levels of freedom it would be the exact same model as with a multivariate t-distribution.4985 1.2170 −0.1 above. 4 9 ri. ˆ 0 0 0 0. The multivariate t-distribution is a normal variance mixture and hence easy to sample from e.1202 0. Hveem 60 (103) .0000 To sample from the copula with t-marginals sample from a multivariate t-distribution with ν = 3. The best ﬁt (using MATLAB’s copulaﬁt) is a t-copula with ν = 3.0617.1680 0.0030 . Modeling Financial Assets multivariate t-distribution are attained using a maximum likelihood estimation. use the cumulative distribution for a univariate tν variable on the sample.1202 1. Thus by using a Copula it is possible to maintain the marginal distributions intact and use the dependence structure from another distribution.3301 0 0 0 0 0.j = k=1 QT Fi.g. ˆε = Σ 0. Σε ) . and.2326. This is not a necessary condition and the residual term ε can be rewritten as a matrix (factor loadings) times a stochastic vector (vector of the remaining PCs). Thus. ρ= ˆ −0.1) where.4985 −0. but as seen in Table 5.0733 0.0841 1. MATLAB has an inbuilt function to perform the operation.8622.0154 0 0 0 0 0 0 0 0 0 0.0841 0. Σ) .Chapter 5. using a multivariate t-distribution (in this setting).j k=5 Q T Fε . multiply these samples with each dispersion coeﬃcient and add the mean to attain the ﬁnal factor score sample.3 this is not the case.0000 0. M. Hence to perform a Monte Carlo simulation.0000 −0.0082 0 0 0 0.0099 0 0 0 0. ˆ ˆ the samples are now uniformed samples from the tν copula. νε = 2. ˆ Fε ∼ tνε (0.0733 .3 will be used for the ﬁrst four PCs. Thus an adequate choice is to use t-marginals and a t-copula since the dependence structure is spherical and similar to the one sampled from a t-copula.k + k. 1. Copula The marginal distributions for the factor scores are approximately t-distributed as seen above. k.k (5.

t-location scale µ ˆ σ ˆ ν ˆ 0.00 0.00 0. notable is that the multivariate t-distribution does not provide a bad ﬁt at all.1179 0.1260 1.0658 1.00 0. A PCA model for the yield curve and index 15 15 10 8 10 10 6 4 5 5 2 PC 2 PC 3 0 0 PC 4 −30 −25 −20 −15 −10 −5 0 5 10 15 20 0 −2 −5 −5 −4 −6 −10 −10 −8 −15 −35 −30 −25 −20 −15 −10 −5 0 5 10 15 20 −15 −35 −10 −35 −30 −25 −20 −15 −10 −5 0 5 10 15 PC 1 PC 1 PC 1 10 8 6 4 2 10 6 8 6 4 2 2 4 PC 3 PC 4 0 −2 −4 −6 −8 −10 −15 0 −2 PC 4 −10 −5 0 5 10 15 0 −2 −4 −6 −8 −6 −10 −5 0 5 10 15 −10 −15 −10 −5 0 5 10 −4 PC 2 PC 2 PC 3 Figure 5.9 shows that the copula provides a much better ﬁt for the marginals than the multivariate t-distribution.0000 . The error term can be modeled either through a multivariate t-distribution or any other parametric distribution such as a copula model.1405 2.1380 0.1661 0. less than 2. By ﬁtting a t-copula with t-marginals the following result is retrieved.7 and 5.0183 0.8 with Figure 5.5% of the total variance). when comparing Figure 5.1133 0.2901 1.0000 0.1837 0.6872 0.0000 0.6.5897 0.1080 2.0832 2.1338 1.4: Parameters for the residual factor scores attained through a maximum likelihood estimation of t-distributions.4.0692 3.1837 0. M.1661 0.8771 0.1179 0. It does not matter a lot which distribution is chosen since the error explains the remaining noise in the data (which is small.1133 −0. ˆ Figure 5.1380 0.4963 0.8576 PC 5 6 7 8 9 Table 5. just not as good as the copula model.0000 0.7: Scatter plots of pairwise PCs and corresponding factor scores sampled from the multivariate t-distribution.1260 −0. ρ = ˆ ε 1.0658 0.1338 0.2901 0. ν ε = 3. The copula provides also a better dependence structure.00 0.0183 0. Thus it is easy to conclude from the plots that the copula provides a better ﬁt.00 0.0808 2.0000 0.4124.5. Hveem 61 (103) .

M. Thus a historical simulation is used further in this thesis since a sensitivity analysis shows that the results are in practice identical using both. Modeling Financial Assets 15 15 10 8 10 10 6 4 5 5 2 PC 2 PC 3 0 0 PC 4 −30 −25 −20 −15 −10 −5 0 5 10 15 20 0 −2 −5 −5 −4 −6 −10 −10 −8 −15 −35 −30 −25 −20 −15 −10 −5 0 5 10 15 20 −15 −35 −10 −35 −30 −25 −20 −15 −10 −5 0 5 10 15 PC 1 PC 1 PC 1 10 8 6 4 2 10 6 8 6 4 2 2 4 PC 3 PC 4 0 −2 −4 −6 −8 −10 −15 0 −2 PC 4 −10 −5 0 5 10 15 0 −2 −4 −6 −8 −6 −10 −5 0 5 10 15 −10 −15 −10 −5 0 5 10 −4 PC 2 PC 2 PC 3 Figure 5. reducing the sample time in many cases from several hours to minutes. Hveem 62 (103) . 156 return samples with replacement out of 781 observed weeks provides 781156 diﬀerent combinations. The advantage of using historical simulation instead of the parametric simulation is that it is much faster to use. thus limiting the outcomes to the empirical distribution. This is particularly the case since weekly return data is used to simulate three years trajectories.Chapter 5. which converges towards inﬁnity. The two models would diﬀer much more if the return for just one week should be simulated. Hence an appropriate choice is to use either the copula model with Monte Carlo simulation or historical simulation. When sampling three-years returns from weekly returns it is still possible to generate almost an inﬁnite number of combinations using the empirical distribution.8: Scatter plots of pairwise PCs and corresponding factor scores sampled from the t-copula with t-marginals. When sampling one week’s data only one draw is required. the result does not diﬀer a lot in practice.

4.9: QQ-plots of the empirical marginal distributions against the multivariate t-distribution’s marginals (to the left) and the copula’s marginals (to the right). Hveem 63 (103) . A PCA model for the yield curve and index Principal Component 1 20 15 10 10 5 Principal Component 1 Multivariate t−distribution Margin Quantiles Copula Margin Quantiles −30 −25 −20 −15 −10 −5 0 5 10 0 0 −5 −10 −15 −20 −25 −10 −20 −30 −40 −30 −50 −35 −35 −35 −30 −25 −20 −15 −10 −5 0 5 10 Empirical Quantiles Empirical Quantiles Principal Component 2 8 6 4 2 0 −2 −4 −6 −8 −10 −10 6 4 2 Principal Component 2 Multivariate t−distribution Margin Quantiles Copula Margin Quantiles −5 0 5 0 −2 −4 −6 −8 −10 −12 −10 −5 0 5 Empirical Quantiles Empirical Quantiles Principal Component 3 5 4 3 2 1 0 −1 −2 −3 −4 −5 −5 4 3 2 Principal Component 3 Multivariate t−distribution Margin Quantiles Copula Margin Quantiles −4 −3 −2 −1 0 1 2 3 4 1 0 −1 −2 −3 −4 −5 −5 −4 −3 −2 −1 0 1 2 3 4 Empirical Quantiles Empirical Quantiles Principal Component 4 8 8 Principal Component 4 Multivariate t−distribution Margin Quantiles 6 6 4 Copula Margin Quantiles −2 0 2 4 6 8 4 2 2 0 0 −2 −2 −4 −4 −4 −4 −2 0 2 4 6 8 Empirical Quantiles Empirical Quantiles Figure 5.5. M.

.

where it was previously found that the products closest to maturity have the worst outcomes of equivalent option 65 . which types of products that are better to invest in than others. All structured products in this chapter are 100% capital guaranteed and follows the dynamics given in Chapter 3. the strike as well as the price of the product itself at issuance. where a modiﬁed Korn and Zeytun framework is proposed. i. CVaR will be considered for α = 0. which is a relative theoretical concept. The chapter covers three diﬀerent types of portfolio choices: ﬁxed portfolio weights.e. i. rolling portfolio weights and dynamic portfolio weights.1 Fixed portfolio weights Fixed portfolio weights is the setup where product one always is the product closest to maturity and product twelve the product with the longest time to maturity (since it exists only twelve products). it is interesting to investigate which products are more optimal than others to allocate in. The portfolio will be allocated using the CVaR algorithm described in Chapter 2 with the modiﬁcation that no binary variables will be used (no ﬁxed transaction costs). since they turn the problem into a mixed integer linear problem. which is not desirable. the fund is allowed to allocate a maximum of 30% in each asset. The diﬀerent alternatives are evaluated in relation to a benchmark fund and the most beneﬁcial fund construction is investigated. Only proportional transaction costs are considered since ﬁxed transaction costs transforms the linear CVaR optimization problem into a mixed integer linear problem. The study indicates that the most imperative factor that aﬀects the return is in fact not the portfolio selection itself but instead the transaction costs. thus risk measures such as standard deviation are inappropriate (as described earlier). The results indicate that the most beneﬁcial setup is the dynamic portfolio weights scheme. thus the structured products have a time to maturity of three years at issuance. The products closest to maturity are associated with the highest risk. where there is always twelve capital guaranteed structured products available. The ﬁrst portfolio allocation strategy/scheme that is described is the ﬁxed portfolio weight strategy. which increases the complexity of the problem a lot.3. The return is used as target function since this problem is better deﬁned than having the return as a constraint (needs less iterations via the simplex method).CHAPTER 6 Portfolio Optimization The fund should yield a high return as well as limit the risk under CVaR.3 are in line with the results found in this section. CVaR is used as risk measure since the distribution of the structured products is skewed and exhibits kurtosis. Since the ﬁxed weights concept is used. when considering tenors. In this chapter.95. Thus the study disregards transaction costs initially to investigate the optimal choice in a perfect market setting. The results presented in Section 4. The result should be seen as a guideline for optimal portfolio choice. CVaR. the bond’s notional amount equals the price of the underlying.e. thus it is similar to the portfolios used in Section 4. the investment horizon is three years. 6. the fund is allowed to rebalance the portfolio every quarter. regarding the tenor.

5 2 4 6 8 10 12 14 16 Portfolio CVaRα = 0.5 18 17. Thus to attain robust results 100 eﬃcient frontiers are simulated for diﬀerent risk levels (CVaR) as constraint and the average of these weights are calculated to attain robust portfolio weights.1) the assumption that the risk of movements in the underlying is the most prevalent risk factor can be conﬁrmed. The results are in general the same.e. since the ﬁrst three years are needed to generate relevant price data as the price process is path dependent. The CVaR algorithm described in Chapter 2 is applied to produce a robust eﬃcient frontier. Thus the eﬃcient frontier would look diﬀerent if an investor would want to invest in structured products today since the current yield curve.000 six-year trajectory samples based on the model described in Chapter 5.3 was correct. would look like this.000 scenarios for the eﬃcient frontier (in a timely manner on a regular PC). i. In this section. which reﬂects the intrinsic frontier in a better manner. The return over the last three years of each sample trajectory is considered.95 [%] Figure 6. mostly imposing just higher expected return and risk. in the long run the eﬃcient frontiers etc. while retrieving a more smooth return distribution for newly issued products as proven earlier (even though the interest rate was ignored in the previous chapters). It is also important to notice that a more general concept is investigated in this section since the previous three-year trajectories are simulated. It is important to understand that the result may not look exactly the same for all markets. Figure 6. Transaction costs have been disregarded and the restriction that it is only allowed to invest a maximum of 30% in each asset is in place. strikes and participation rates are known. Notable is that the expected return is higher for products with higher risk. Since similar results are retrieved when including the eﬀect of interest rate changes (as seen in Table 6. the result is disclosed in Figure 6. it is not possible to use a lot more than approximately 1. representing a low yield case will be used. Portfolio Optimization portfolios. i. not necessary all the time. Since the optimization is computer intensive. Thus yielding a more robust eﬃcient frontier. thus it seems to be adequate to use a CVaR optimization scheme to attain a smooth eﬃcient frontier. Investing in the products closest to maturity imposes an “all or nothing situation” for the options.Chapter 6. Thus it is reasonable to assume that the investment strategy proposed earlier in Chapter 4 is still adequate.e. the highest risk is associated with products that are closest to maturity and the newly issued products have a lower risk as disclosed in Table 6.1 indicates that the expected return increases as the risk increases. Thus the hypothesis that was presented in Section 4. The analysis is performed via 100.1. the yield curve that is in the beginning of the six-year period. In the appendix the case with an arbitrary yield curve is disclosed.5 Expected Portfolio Return [%] 19 18.95. Efficient Frontier Fixed Portfolio Weights 20 19. The last three years observed price data on the market is not used as start data since the analysis should be conducted independent of the current market. Hveem 66 (103) . Instead diﬀerent start levels for the yield curve will be considered.1. and the frontier M.5 17 16.1: Eﬃcient frontier for the optimization choice using the ﬁxed portfolio weights setting. index level. α = 0. the yield curve as of 15th November 2010. Hence the results in this chapter are mainly general results.

95 7.67 9.40 9 4.47 3.48 Table 6.6.23 10 16.1.39 12 10.73 Weights. since it discloses how robust the weights are.2 and the portfolio would receive the expected return according to Figure 6. a fund should allocate mainly in the last four .2: The table describes the robustness (standard deviation) of the weights for a portfolio allocation using 1.14 2 14.08 6.05 8.41 3 19. So what does this imply? Since each sample has diﬀerent paths up till the start of investment a high standard deviation implies that it is important to not use a generalized allocation scheme. as Table 6.91 5.2 indicates.50 1. than the other way around.46 1 20.71 3 12. which is important to notice.04 5. From the distribution and CVaR results for the individual products it is reasonable to assume that lower levels of CVaR allocate most of the capital to the products with the longest time to maturity (which is the case).2.07 18.71 17. depending on its risk tolerance (and transaction costs).95 [%] 5 6 7 8 9.95 of 3.95 0.000 scenarios for three diﬀerent levels of CVaR are used. It is important to analyze the standard deviation of the portfolio weights. The risk level and expected return will be very dependent on the transaction costs.01 9.94 3 8. Consider transaction costs of 2% (2% in proportional transaction costs each time the fund sells or buys an asset) and a portfolio which holds equal weights in the longest tenors w9 = w10 = w11 = w12 . sell of assets after a number of quarters elapsed.95 4% 8% 12% 1 0.93 16.37 0.68 10 4.57 10 11.42 9. The portfolio has without transaction costs a CVaR0.03 7. disclosed in Figure 6.7 6.81 12 3. CVaR0. Table 6. are attained. Hence depending on the desired maximum risk level the portfolio weights should be chosen according to Figure 6.69 5. Expected return [%] 4 5 6 7 8 9 19. The portfolio weights are approximately equally robust for the low risk scenarios and for high-risk scenarios.82 10.73 8. and if it is necessary to use multiple samples to attain a representative solution to the optimization problem.06 4.12 17.14 11 3.1. The sensitivity of the portfolio weights is higher with less sample scenarios.01 2. standard deviation [%] 5 6 7 8 9 0.72 2. Also it is more adequate to use CVaR as a constraint and maximizing the return as the target.ﬁve products to minimize the CVaR for the portfolio.30 16. CVaR0. Fixed portfolio weights is quite similar the frontier for a classic Markowitz optimization problem with standard deviation as risk measure. 1 18.51 Product Number. Thus the frontier portfolios (weights).80 6.04 7. since it does not only require less iterations but also generates more stable portfolio weights.2 shows that there exists signiﬁcant standard deviation for the portfolio weights. Thus following the path of capital guarantee.62 3.75 2 0.37 9. since the weights shift much depending on the scenarios.07 3.92 11 15. It should also.25 8.33 8.80 4 9.00 11.63 4 10.1: The table describes the CVaR at the 95% conﬁdence level and expected return for the diﬀerent products over the three-year period using the ﬁxed portfolio weights setting.000 scenarios and for three diﬀerent CVaR levels using the ﬁxed portfolio weights setting.94 6.62% and M. 100 samples of 1.50 Table 6. Next the sensitivity of each scenario’s portfolio weights is analyzed. one medium risk and one high-risk level.69 1.74 12 15. one low risk.11 11 11. But as the maximum CVaR allowed increases the portfolio will allocate in shorter products to yield a higher expected return (since the expected return increases almost linear over the twelve assets).92 1. with slightly more robust portfolio weights for the high-risk level.52 Product Number. Hveem 67 (103) .88 2 19.

attain portfolio weights that are desirable when taking in consideration transaction costs. ...k w1 + .95.k .w .52% and an expected return of 11. random start yield curve...k k = 1.2 11. The weight held in asset 12 one quarter is approximately held in asset 11 the next quarter and so on.k = 1 + RT..75% including the transaction cost.. On the other hand a portfolio holding equal amounts invested in products twelve throughout seven has a CVaR0. S + β + zk ≥ 0. The optimal weights are investigated in a setting of a low initial yield curve and the setting of an arbitrary. which can approximate the induced transaction costs of a portfolio with ﬁxed weights.2 5. Including the transaction costs it will have a CVaR0. by solving the optimization problem below.2 12.2 14. k=1 w.2 8. the results are disclosed in Figure 6. an alternative CVaR algorithm with proportional transaction costs is introduced.3. . α = 0... S k = 1. It is in general most M. Thus the fund can. − wN i = 1.16%.29% and an expected return of 9.. given a positive return. w RT.. To solve how to allocate in an optimal way.98% over three years. N − 1 i = 1.95 of 7. including transaction costs..2 9.. N −1 zk ≥ 0.2 7. Portfolio Optimization 100 90 80 70 60 Weight [%] 50 40 30 20 10 w1 0 4.Chapter 6. Hveem 68 (103) .95 of 9. .k wN . + wN = w1 + wN − ± wi ≥ 0.2 6..2: Portfolio weights for the eﬃcient frontier using the ﬁxed portfolio weights setting..2 10.95 [%] Figure 6.w .2 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w12 Portfolio CVaRα = 0.2 13..z.β max + − such that: β+ 1 S (1 − α) S zk ≤ C. k=1 w 1 N 1 + RT. an expected return of 15. 1 S S w 1 + RT... when minimizing the CVaR. N The optimal portfolio weights are investigated for the cases of no transaction costs and three diﬀerent levels of transaction costs. S k = 1. wi ≥ 0. N wi + i=1 i=1 + + + − − T Ci wi + T Ci wi + T CN wN = 1. . Thus the problem must be modiﬁed to ﬁnd a solution including transaction costs. . + 1 + RT. + − wi = wi+1 + wi − wi .

minimizing CVaR0. thus leading to that the fund allocates to more products.3: Optimal portfolio weights for the ﬁxed portfolio weight scheme for diﬀerent proportional transaction costs. The diﬀerent risk levels for the products are derived from the participation rate. It is preferable to invest in product twelve since its time to maturity for the ﬁrst roll equals the investment horizon. The reason why the return of the eﬃcient frontier is almost equal for all risk levels is that almost all products have the same expected return.95 given the requirement of a positive return. since they are over a three-year horizon almost equivalent products in return sense. the same as in the beginning of the previous section. The risk associated with the products is less for products that are close to roll over than products in the middle spectrum. The transaction costs will not impact the portfolio choice itself since all products are associated with the same degree of transaction costs (when the roll is performed).4 shows that the expected return increases as the risk increases.2. Thus this rolling product is a perpetual product (similar to the bond roll) where transaction costs occur when the investor rolls over to the new product (thus each product will over a three-year period have equal transaction costs in percentage).3 indicates that the products that are closer to their ﬁrst roll over are more risky than products that have a long time to their ﬁrst roll. A CVaR optimization algorithm.6. The probability of attaining a higher participation rate is higher for products that roll over later. The CVaR algorithm with CVaR as constraint and maximizing the expected return as the target function is performed 100 times (for robustness) to produce a robust eﬃcient frontier. Hveem 69 (103) .2 Rolling portfolio weights Rolling portfolio weights is the setup where product one is a product that has one quarter left to maturity and is rolled over at maturity to a new equivalent capital guaranteed structured product with three years to maturity. A higher participation rate (due to often a higher yield) imposes a higher risk. Product two has two quarters to maturity before it is rolled over to a new capital guaranteed structured product. On the other hand the risk associated with options is higher when the time to maturity is smaller. Figure 6. and so on. 100 90 80 70 w1 w2 w3 w4 w5 w6 w7 w8 w9 w 10 100 90 80 70 w1 w2 w3 w4 w5 w6 w7 w8 w9 w 10 Weight [%] 60 50 40 30 20 10 0 0 1 2 3 Weight [%] 60 50 40 30 20 10 0 0 1 2 3 w11 w12 w11 w12 Proportional Transaction Costs [%] Proportional Transaction Costs [%] (a) Low yield (b) Generalized yield Figure 6.4. Table 6. but only in a very small extent. the result is disclosed in Figure 6. thus providing intrinsic capital guarantee. Rolling portfolio weights optimal to allocate in the ﬁrst four products. 6. The expected return is almost similar for all risk levels. is performed for the ﬁxed portfolio weights. and as the transaction costs rises it is less optimal to rebalance the portfolio often. thus transaction costs are not considered in this section. Thus the products that have the longest time until their roll will M. according to the algorithm described above.

25 2 11.83 3 7.19 11.12 8.06 11 17.54 17.72 12 0.95 Weights.78 6.86 11 5. which is reasonable as disclosed in Figure 6.59 7.95 [%] 4 5 6 7 8 9 10.65 12.71 standard deviation [%] 6 7 8 9 4. 1 9. Thus the average participation rate over the investment horizon is higher for products with 1-9 quarters to the next roll.72 12.87 10 8.94 7. but they will only go through a few quarters (the ones with the least risk) until the investment horizon ends with this high participation rate.73 9.98 Product (months until next roll).00 12 17.34 Table 6.8 3 4 5 6 7 8 9 Portfolio CVaRα = 0.89 12.77 8.81 Product (months until next roll).68 10. CVaR0.43 3 17.03 12 4.44 11.60 17.95 [%] Figure 6.31 9.69 5.3.87 17.89 4 5. Expected return [%] 4 5 6 7 8 9 10 17.5.84 17.88 17.83 2.86 10.58 17.Chapter 6. to attain lower levels of CVaR.66 7.4: The table describes the robustness (standard deviation) of the weights for a portfolio allocation using 1.82 17.64 17.63 10.57 17.86 Expected Portfolio Return [%] 17. 5 4.04 2 10. imposing a higher risk.05 12.86 5.57 2 18.01 3 11. Hveem 70 (103) . Also these products will beneﬁt of the previous roll before the high participation rate that with a high probability had a low participation rate.15 10 6.4: Eﬃcient frontier for the optimization choice using the rolling portfolio weights setting.81 17. CVaR0.40 13.000 scenarios and for three diﬀerent CVaR levels using the rolling portfolio weights setting.69 Table 6.43 9. risk receiving higher participation rates in the end. while the risk proﬁle of M.95 4% 6% 8% 1 9.98 2.73 11 10. The expected return will in the long run be very similar for all rolls. Portfolio Optimization Efficient Frontier Rolling Portfolio Weights 17. Investigating the portfolio weights for the robust eﬃcient frontier shows that the portfolio should. The standard deviation result is similar to the result in the previous section.35 12.4. which indicates that it is important as a portfolio manager to not use a generalized scheme for the allocation of the portfolio since the portfolio weights depends a lot on the previous path.30 12. The robustness of the weights is also studied through investigating the standard deviations of the weights as previously.89 7.28 1 18.76 12.62 13. α = 0. The risk levels and expected returns for the twelve diﬀerent rolls are given in Table 6. the result is presented in Table 6.22 5.83 17.89 8.34 13.95. allocate to the products that have a lower individual CVaR.32 7.57 17.3: The table describes the CVaR at the 95% conﬁdence level and expected return for the diﬀerent products over the three-year period for the diﬀerent rolls.85 17.

the only thing that diﬀers is approximately a parallel shift. Thus when investing with the rolling weight scheme.95 [%] Figure 6. Thus the distributions of the two. and the rest of the capital in a basket of bonds with time to maturity between three months to three years (approximately equally weighted. of 19. which would decrease the expected return of these products.5 results for diﬀerent benchmark funds are disclosed.6 6.6.95 the portfolio will actually have an expected return of 17.82% over three years (excluding transaction costs.75%. It is better to invest with rolling portfolio weights than the ﬁxed weight scheme if the expected return it yields is satisfactory. which would increase the CVaR and decrease the expected return). rolling). Thus by requiring a maximum of 3.95 of 10. over the three years investment horizon. which also induces much higher transaction costs. the products diﬀers.95 of 9.3 Benchmark fund The analysis would be less relevant without a comparison with a relevant benchmark. The high standard deviation of the portfolio weights implies that it is very important to evaluate the portfolio depending on the observed paths (yield curve.6 1 w 2 w 3 w 4 w 5 w 6 w 7 w 8 w 9 w 10 w 11 w 12 Portfolio CVaR α = 0. The results indicate that it is hard for the SPF to compete with the benchmark fund including dividends (in terms of return) since the benchmark has a much higher expected return given the same risk level. thus it is imperative that the diﬀerent portfolio constructions are compared with a benchmark fund. This yields an expected return.3. Consider the benchmark fund described earlier. a portfolio’s risk and return is only relevant in a certain context. plus minus 10%.6. As described earlier in this paper.49%. In Table 6.6 8. with or without dividend. which places 30% of its capital in the index. 6. it is better to allocate to the newly issued products (as stated in the previous chapters and sections).6 7. are of course almost identical. thus the SPF must be benchmarked against alternatives at the Swedish market. Thus it is better to invest in products with a long time to the next roll. participation rate. Benchmark fund 100 90 80 70 60 Weight [%] 50 40 30 20 10 w 0 4.5: Portfolio weights for the eﬃcient frontier using the rolling portfolio weights setting. Notable is that transaction costs have been disregarded.88% for the same risk level.6 5. The distribution of the fund returns is given in Figure 6.12% and a CVaR0. Thus the SPF must compete with the benchmark portfolio given low risk. compared to the ﬁxed weights which only yields an expected return of 16.95. Observed market data for the Swedish market has been used throughout the analysis in this chapter. When disregarding the expectation of the dividends the portfolio instead receives an expected return of 16.80% in CVaR0.81% including the expectation of 3% annual dividends in the index and a CVaR0. index level etc) since they will impact both the expected return and the risk proﬁle. α = 0. M. Hveem 71 (103) .

and the results show that it is only possible for the SPF to compete against other low risk funds to yield a higher expected return.16 17.75% 2 2.5: The table discloses the expected return and CVaR for diﬀerent levels of index weights over the three-year period for a benchmark fund. The fund manager has the possibility today to invest in twelve structured products. Thus it is desirable to investigate the portfolio choice for the whole three-year time period to avoid excessive transaction costs (since this is the time to maturity of the longest product that exists).5 (a) Incl.05 4.13 Index allocation 10% 20% 30% 40% Table 6.20 13. εbench = 0. This is also in line with the idea of the SPF. Table 6.25 19. 6. which is that it should be a low risk fund. product two is the product with two quarters until maturity and so on as previously. It is assumed that product one is the product with one quarter until maturity.5 CVaR0.95 11. Thus these types of funds would be able to compete with the benchmark fund in the low risk spectrum.5 1 1.5 Benchmark Fund 300 0 0. Thus when considering to allocate the portfolio over a three-year period the fund manager needs to take in count how the payoﬀ of the products that mature before the end of the investment horizon should be reinvested.50 8.49% 1.4 Dynamic portfolio weights The allocation problem is quite complex for the SPF.5 0 −0.5 0 0. Korn and Zeytun (2009.43 3. Hveem 72 (103) . hence there exists twelve products (the products are also standard). where the investor’s investment horizon is longer than the time to maturity. It is necessary to consider the full three-year time horizon since a portfolio allocated on a three-month basis would be suboptimal and induce high transaction costs.95 = 9. Expected Return CVaR0.3. Korn and Zeytun proposes the following modiﬁed multiple M. The next section will investigate the last allocation scheme.Chapter 6.6: Histograms over the sampled three-year returns for the benchmark fund. where only one of the products does not mature before the end of the investment horizon.35 Excl.5 2 2.95 10. [27]) propose a method to solve optimal investment problems with structured products under CVaR constraints.12 10.40 -1. Incl.88 14. which is also the case for the ﬁxed portfolio weights.75 19. Portfolio Optimization Benchmark Fund 700 450 400 600 350 500 300 400 250 200 150 200 100 100 50 0 −0.72 23. where the rolling weights portfolio generates higher expected returns relative to the risk.5 1 CVaR0. the dynamic portfolio weights allocation scheme. 6.1. Dividends.16 -1.38 16.95 = 10.5 indicate that the SPF can compete initially (without transaction costs) in the low risk spectrum. dividend Figure 6. Dividends. dividend (b) Excl.66 15.3 and 6. Expected Return CVaR0.

k .. given diﬀerent start values in the optimization. where each intermediate rebalancing point creates a new loop. with twelve possible products to allocate within. Also.k is the return of asset l for the remaining time period (time until the end of the investment horizon from time j). N − 1 k = 1.. The optimization problem is modiﬁed by setting constraints on νlj .z. thus relative few assets..k + β + zk ≥ 0.. and for robust weights it would take years to solve the optimization problem..k = w1 RT. M. + wN −1 RT..ν RT.. wi ≥ 0.6.. j..k + . ν ≥ 0. the optimization problem is not always convex. such that the problem is easier to solve.. β+ 1 S (1 − α) S zk ≤ C. N − 1 where νlj is the percentage of the product j’s payoﬀ that is allocated to asset l after maturity. . w1 + . Thus the modiﬁed algorithm is as follows.ν 1. the proportional k transaction costs constraints and budget constraints can be added in the exact same way as for the regular CVaR optimization problem.1 Modifying the Korn and Zeytun framework The algorithm described by Korn and Zeytun is adequate for a limited number of intermediate time periods..k = ΠN . S − 1. Thus each product that matures is reinvested using the weights νlj .. since the iterations and as a consequence the computing time grows exponentially. 6. Note that the last product does not mature prior to the investment horizon N thus RT. .β max such that: w.. Hveem 73 (103) . Eleven outer optimization loops. .wN = 1. The problem with this setting is that it is no longer a linear optimization problem. not minutes or hours.ν N RT.ν N −1... Thus it cannot be solved with only a simplex procedure. . Dynamic portfolio weights period CVaR allocation algorithm....k k = 1. S i = 1. j k = 1. thus resulting in diﬀerent solutions.ν.. Concluded: it is not possible to use this algorithm to solve the allocation problem in this setting. . l rT −j. j = 1.4. Let νlj be the weights that minimize the CVaR for the particular product j including the reinvestment.. . N j = 1. 1 S S w. + νN 1 + rT −j.ν RT. S w..ν RT.k . are needed since the portfolio has eleven intermediate rebalancing points.4..k = 1 + Πj k j j+1 j N νj+1 1 + rT −j..k + . thus reducing the total problem to only solving twelve linear optimization problems. It is not feasible to allocate with this sophisticated optimization algorithm in this setting since the time to attain a solution would be months. k=1 w. k=1 zk ≥ 0. Πj is the return of asset j from the start of the investment period until k its maturity and k corresponds to the scenario. The optimization problem above is without transaction costs.. It must instead be solved by creating optimization loops.k + wN RT.

5 4 4. Repeat ii until the full price trajectories for all products are known. Thus the big diﬀerence between these two models is that the modiﬁed Korn and Zeytun framework is dynamic and minimizes the risk of the reinvested capital and allocates based on that knowledge. Portfolio Optimization i. and not just use a static allocation scheme. since it can generate higher expected returns to the same risk level. N −1 rT −(N −2).7 is attained. but also lower return).Chapter 6.5 7 7.k and complete the full return trajectory for product N − 1. since the products with the longest time to maturity have the lowest risk. most structured products in this setting have similar expected returns. ii. iv. Without taking in count transaction costs the robust eﬃcient frontier in Figure 6. The expected return and CVaR for this setup are disclosed in Table 6. thus the solution will in many cases be similar to the one of rolling portfolio weights. Solve the CVaR optimization problem with minimizing CVaR as target and the constraint of a positive return for the period between N − 1 to N to attain ν N −1 . α = 0.46 17.53 17.51 Expected Portfolio Return [%] 17.5 6 6. it is not the same (it generates lower risk. especially for the lower risk segments. The modiﬁed Korn and Zeytun algorithm allocates even more in the newly issued products. Most of the reinvested capital will be placed in the newly issued product.iv 100 times to attain robust portfolio weights. to choose the portfolio weights depending on the previous M. The results attained through this allocation scheme are much more beneﬁcial than the previous two allocation schemes. but since the returns are path dependent. the risk proﬁle will be diﬀerent.5 8 Portfolio CVaRα = 0.95. iii. N − 3 to N and so on (in this case eleven optimization problems).7.47 17.6.52 17. Thus the solution is in practice very similar to the one of rolling portfolio weights.95 [%] Figure 6.48 17. The standard deviation result indicates that it is important to use the scheme actively. depending on the previous price data. The expected return does not increase a lot as the portfolio increases its risk tolerance.7: Eﬃcient frontier for the optimization choice using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework.5 17. Hence even though the solution in this case is extremely similar to the rolling weights problem. v.49 17. The standard deviation of the weights are analyzed in the same way as in previous sections and disclosed in Table 6. Thus it is this allocation scheme that should be used in relation to the benchmark portfolio. Repeat steps i . thus between N − 2 to N . Solve the well-deﬁned CVaR optimization problem between 1 and N to attain initial weights w (since all scenario trajectories are known). Efficient Frontier Dynamic Portfolio Weights 17.5 5 5. Notable is that this is for the low yield setting and the solution to the problem is not always the same. Sample N trajectories from the model described in Chapter 5 to attain returns for the structured products.45 3 3. Hveem 74 (103) .

71 7. as disclosed in Figure 6.54 3 9.06 13.26 10.37 6.68 11 17.74 12.98 0.8 4.80 1 17.52 12.8 5.04 2 17. α = 0.28 9.96 3 17.25 17.31 17. Expected return [%] 4 5 6 7 8 9 10 17.88 CVaR0.8. In this chapter the yield curve from middle of November 2010 on the Swedish market is used as an initial yield curve of the six-year period.28 6.95 [%] 4 5 6 7 8 9 10. Notable is that the algorithm is best for low risk strategies.39 12 0.66 Product.46 7.36 Table 6.22 4. Weights.6: The table describes the CVaR at the 95% conﬁdence level and expected return for the diﬀerent products over the three-year period using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework .33 2.95 [%] Figure 6.31 12. since they both have high expected return and low risk.00 12 17. since it minimizes the risk of the reinvested capital.85 Table 6.27 17.22 7.45 Product.24 9.81 12.42 10 12.42 11 10.7: The table describes the robustness (standard deviation) of the weights for a portfolio allocation using 1.62 3 10.20 17.64 2 10.6.36 12.31 standard deviation [%] 6 7 8 9 3.34 12.68 4. Hveem 75 (103) . 6.61 12 4.8 6. Dynamic portfolio weights 1 8.23 2 9.67 8.74 10 6. 100 90 80 70 60 Weight [%] 50 40 30 20 10 w1 0 3. Therefor the algorithm allocates mainly in products 1-3 and 10-12.95. which is the goal of this thesis.20 13.87 9.24 17.60 8.95 4% 6% 8% 1 11.97 13. When analyzing the portfolio weights for the diﬀerent risk levels it is easy to notice that the products with a long time to maturity are very attractive.24 12.66 9.4. CVaR0. months until maturity.8 7.49 12. 5 3.2 Dependence of the initial yield curve It is important to note that the risk level and level of expected return are signiﬁcantly dependent on the initial yield curve.74 7. paths. Also the products that are close to maturity have a higher expected return and a lower risk than the products in the middle spectrum.87 4 1. Thus the results are M.05 13.24 17.8 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w12 Portfolio CVaRα = 0.44 4.38 11 4.8: Portfolio weights for the eﬃcient frontier using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework.06 6.4.48 8.000 scenarios and for three diﬀerent CVaR levels using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework. months until maturity.

since it imposes less risk for the investor in the SPF (lower participation rate) at the same time as competing mixed funds actually have increased risk. it is possible to see that the low risk strategy of ﬁxed portfolio weights cannot compete with the low risk strategy of the rolling portfolio weights. Since most products have approximately equal expected return the only diﬀerence is that the arbitrary yield curve case has a higher expected return and a slightly diﬀerent risk level. This can be derived from the fact that increasing the limit for CVaR allowed does not necessary increase the expected return. It is interesting to investigate the standard deviation for the portfolio weights given the previous price paths. Hveem 76 (103) . Notable is also that the ﬁxed portfolio weights strategy will in most cases impose a lot more transaction costs than the other strategies. Portfolio Optimization dependent on this initial yield curve. it is deﬁnitely a good choice to invest in the SPF. Concluded: the SPF is more beneﬁcial for investors in a market state of a low yield. Also this scenario has a much lower minimum risk level for the CVaR. Notable is that the results are very similar to each other. This is what the thesis is looking for.9 discloses the closing prices of OMXS30 during the three-year time period prior to the day of investment and the robust eﬃcient frontier for the investment choice according to the modiﬁed Korn and Zeytun algorithm. The ﬁxed portfolio weights scheme gains a much higher expected return given the same risk levels for the arbitrary yield curve case. A high standard deviation for the portfolio weights indicates that the optimal portfolio weights are depending on the previous paths. For the benchmark fund the expectations on the dividends in the future are really important. The expected return is signiﬁcantly higher for this scenario than for the generalized case described above. Therefor the optimal portfolio choice with an investment horizon that starts the 16th of November 2010 in the Swedish market will now be studied. Almost the exact same risk levels are attained in the rolling portfolio weights setting to a much higher expected return in the arbitrary yield case (compared to the low yield case). The same patterns of weights for diﬀerent risk levels are shown for the arbitrary yield setting. Figure 6. In Appendix E the corresponding ﬁgures and tables are disclosed where an arbitrary start yield curve is used instead to attain unbiased results. Thus using a strategy where the fund always buys structured products close to maturity is the only way to trigger a higher expected return in the SPF. thus a low risk strategy in one setting can be expected to also be a low risk strategy in the other setting.Chapter 6. thus actually reducing the risk and increasing the expected return. 6. thus reducing its attractiveness. The expected return increases (due to the higher yield) but the risk of the index does not increase. if the SPF can generate a lower risk for the same level of expected return as existing funds. than for the SPF.3 Observed paths There exists in general an optimal choice for the portfolio weights given that the portfolio is resampled.4. Notable is also that by inspecting the eﬃcient frontier for the ﬁxed portfolio weights. The optimal portfolio weights are very similar to each other for both low-risk and high-risk strategies. On the other hand the standard deviation of the portfolio weights was quite high which indicates that it is important to choose the portfolio weights actively. which on the other hand increases the risk dramatically. By comparing the alternatives of rolling and dynamic portfolio weights against investing in a competing benchmark portfolio it is possible to notice that there are levels where it is more beneﬁcial to invest in the SPF. but on the other hand there are portfolios which impose higher expected returns than for the rolling and dynamic weights in the ﬁxed portfolio weight scheme. the thing that diﬀers the most is the level of expected return and CVaR. and it is important for the portfolio manager to understand the impact of having diﬀerent expectations for the dividends. The results are a little bit diﬀerent when comparing the benchmark fund with a low yield curve against an arbitrary yield curve. By analyzing the standard deviation of the portfolio weights it is possible M.

but the products that have standard deviation are the products that the portfolio allocates in.70 2.00 0.8 6.000 scenarios is enough to determine the optimal portfolio weights.00 0.00 0.00 0.00 10 0.82 12 0.8 3.82 0.00 0.00 0.00 0.00 5.60 0.96 4.00 0.00 standard deviation [%] 6 7 8 9 0.57 3 0.8 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w12 Portfolio CVaRα = 0. Hveem 77 (103) .00 2 1.52 1.6 26.00 0.8: The table describes the robustness (standard deviation) of the weights for the portfolio allocation using 1.8 5.00 0.6 Expected Portfolio Return [%] Nasdaq OMXS30 27. but the amount allocated in each asset diﬀers slightly.8 2.00 0. The result indicates that the speciﬁed setting (known paths) has a much lower standard deviation for the portfolio weights than the generalized investment scheme.94 0.4 27. The standard deviation is higher for the scenarios with higher risk. CVaR0.95 [%] Figure 6. Dynamic portfolio weights Efficient Frontier Dynamic Portfolio Weights 1100 27.10: Portfolio weights for the eﬃcient frontier using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework for 16th of November 2010.2 27 26. 5 0.4 1000 900 800 700 600 2008 2009 2010 Index 1 2 3 4 5 6 7 8 Year Portfolio CVaRα = 0.00 0.000 scenarios and for three diﬀerent CVaR levels using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework for 16th of November 2010.9: Observed price paths of OMXS30 during the three years leading up to 16th of November 2010 and the eﬃcient frontier using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework for 16th of November 2010.6. On the other hand the quality of the result is still enhanced when resampling the portfolio weights (as most often).4.00 0.34 3.00 0.10.85 4 0. Thus it is the same assets that are more or less most optimal in each optimization.00 0. 100 90 80 70 60 Weight [%] 50 40 30 20 10 w1 0 1.25 8.00 0.8 26.95 [%] Figure 6.10 5. α = 0.95.8 4. to assess if 1.8. the result is disclosed in Table 6.8 7.00 Weights. The weights are quite similar as in the generalized M. The portfolio weights for the eﬃcient frontier for the optimal investment choice as of the 16th November 2010 are disclosed in Figure 6.00 Table 6.00 11 0.95 1% 4% 7% 1 1.00 0.000 simulations to attain relative stable portfolio weights. Thus the result indicates that it is reasonable to use 1.

The risk of the SPF increases as the yield increases. Thus the SPF’s expected return is 21.Chapter 6.a. but they diﬀer slightly (adjusted for the particular scenario).5% or even 0. In Table 6. Also the SPF is a lot less risky than the benchmark funds at the 30% and 40% levels. in general. Concluded: the whole idea of a structured products fund is very dependent on the transaction costs and it is only possible for the fund to compete with other funds.a. which implies that the benchmark portfolio’s risk decreases as the yield increases. The SPF is compared with a benchmark fund. Hveem 78 (103) . It is not unreasonable that big funds (or investment banks) have the possibility to invest to as low transaction costs as 0. which explains the higher expected return for both the SPF and benchmark portfolio in the arbitrary yield setting. all assets have the same level of transaction costs. since the participation rate increases.9). As seen in the previous section. The transaction costs occur when an asset is bought or sold.95 of 7.5% in proportional transaction costs to be more attractive than the benchmark fund investing 30% of its capital in the index. With a higher yield the bond portfolio gains a higher return. less sensitive to yield changes)).3% p. In the low yield case the SPF provides. but the expected returns. thus this will be conducted given the constraint of a positive return. Table 6. 6. for the three-year tenor. when studying the arbitrary yield case.32% (as shown in Table 6. Thus a fund manager must be aware of these characteristics and how they impact the portfolio’s expected return and risk. for buying as well as selling.83%. which includes dividends from the index (the most realistic benchmark). if M. Both the setting with a low start yield curve (three years prior to the initial investment horizon) and the setting with an arbitrary start yield curve are investigated.08%. Fixed transaction costs are disregarded to avoid transforming the problem into a mixed integer linear problem.5% p. in the three-month tenor and 2. such as mixed funds. a CVaR0.95 for diﬀerent benchmark funds and SPFs are disclosed. CVaR at diﬀerent levels varies of course. both when the asset is sold and when a new asset is bought. Thus it is imperative to adjust the portfolio optimization depending on the previous observed paths. Thus the benchmark fund has both a higher expected return and a lower risk than the SPF. In the arbitrary yield setting: the SPF has higher risk associated with it than the benchmark fund at 20% level (assuming equal transaction costs for the two funds).9 the expected return and CVaR0. has a CVaR0. Assume that the proportional transaction costs for structured products are approximately 3% and for the benchmark portfolio 1%. The low yield setting reﬂects the setting of having a yield of approximately 1. as well as the underlying assumptions. Portfolio Optimization case. lower risk to a higher expected return than the benchmark fund with an aim of 20% index investments (when assuming the same transaction costs level). Thus rebalancing a portfolio induces transaction costs. thus it would be an irrational decision to invest in the SPF. Thus the SPF must in the arbitrary yield case have as low as 1.25%. The same patterns are exhibited for a huge variety of data sets. for diﬀerent participation levels as well as for diﬀerent levels of proportional transaction costs. Notable is that the risk of the bond part decreases (a bond’s delta is increasing in yield (higher yield.5 Transaction costs This section focuses on the dynamic portfolio weights setting and how transaction costs impact the risk and return levels in comparison to the benchmark fund. It is important that the reader takes care and understands that these levels are dependent on the data that is used in the modeling. On the other hand the SPF has a higher expected return and lower risk than the 30% benchmark fund.95 of 6.9 indicates that the SPF can compete against diﬀerent benchmark funds during diﬀerent market settings. The results indicate that transaction costs is one of the most important concepts for the fund manager to consider in its portfolio.36% and the benchmark fund which allocates 30% of its assets in the index has an expected return of 23. the most desirable setting for the dynamic portfolio weights is to minimize the portfolio’s risk.

70 5.95 for diﬀerent levels of proportional transaction costs for the SPF and benchmark funds over a three-year investment horizon for both the low yield setting and the arbitrary yield setting.43 3.56 11.25 4.66 17.44 20.94 6.95 [%] 17.37 5.18 13.20 Benchmark 40% Arbitrary yield Expected Return [%] CVaR0.08 13.25 0.32 16.90 21.95 [%] 27.32 5.50 3.6.04 5.18 11.00 0.61 12.50 1.08 6.50 3.28 7.50 1.58 19.39 14.70 13.50 1.35 23.17 24.00 1.84 11.00 TC 0.48 10. Transaction costs TC [%] 0.00 2.85 11.93 9.99 1.98 15.10 16.00 1.28 22.41 26.95 [%] 19.44 5.09 12.00 26.22 4.97 6.25 0.95 [%] 15.53 6.69 10.23 20.50 2.09 17.64 27.10 Benchmark 30% Low yield Expected Return [%] CVaR0.25 21.82 21.99 26.00 0.05 5.19 3.07 Benchmark 30% Arbitrary yield Expected Return [%] CVaR0.50 2.59 6.75 25.25 15.69 18.50 3.00 2.20 14.73 23.51 0.50 2.36 Benchmark 20% Arbitrary yield Expected Return [%] CVaR0.82 3.00 TC 0.68 8.72 19.59 25.25 0.09 18.57 22.00 1.14 18.00 2.64 16.56 14.02 20.26 3.54 14.78 3.50 3.95 [%] 28.97 18.95 [%] 24.43 24.51 16.36 2.21 4.00 0.9: The table discloses the expected return and CVaR0.82 2.56 13.00 1.16 24.18 0.61 24.30 9.00 0.15 Table 6.75 6.58 1.89 20.95 [%] 20.19 14.79 17.65 4.79 23.67 6.13 Benchmark 20% Low yield Expected Return [%] CVaR0.11 3.93 15.04 23.50 1.41 5.81 0.00 2.95 [%] 23.50 8.13 17.87 15.50 2.41 16.5.00 3.00 SPF Dynamic Weights Low yield Expected Return [%] CVaR0.07 16.00 TC 0.58 15.97 7.44 Benchmark 40% Low yield Expected Return [%] CVaR0.73 4.40 12.78 SPF Dynamic Weights Arbitrary yield Expected Return [%] CVaR0.91 13.16 12.88 14.46 27.41 23. M.62 17.13 20.56 5.26 8.29 26.70 14.25 0.78 12.26 3.83 7.08 10. Hveem 79 (103) .59 21.30 21.32 22.69 2.

Chapter 6. Portfolio Optimization

the fund keeps down the transaction costs. If the SPF can have the same level of transaction costs as the benchmark funds it will be able to generate a higher expected return to a lower risk. Notable is that as the interest increases the ability for the SPF to compete with the benchmark portfolios decreases (in terms of low risk). Thus it is very important that the fund manager understands how its choices aﬀect the return and risk of the fund. The most beneﬁcial investment setup to use is the modiﬁed Korn and Zeytun framework. The scheme usually implies that the fund allocates most of its capital in the three newest products (the products with the longest time to maturity) and almost all of the remaining capital in the product closest to maturity, since these can soon be reallocated in the newly issued products.

6.6

Summary

How can the results found in Chapter 4 and 6 be combined into an investment strategy? It has been established that the products with the longest time to maturity have, in general, the lowest CVaR (and expected return). Chapter 4 showed that it is important (if the fund manager wants to minimize the risk) to limit the level of options held in the portfolio. Thus when competing with a benchmark fund the fund manager could by allowing between 5-25 percent of the portfolio value to be allocated in options provide a better protection against downside risk. The investors are more or less guaranteed, by limiting the option portfolio in this manner, that a market crash will not erase more than 25% of the fund’s portfolio value. Rebalancing in this manner implies that the fund gains capital protection on its previous gains, since it locks them in by investing in bonds. Obviously by minimizing the risk in this manner the potential return is also reduced. Section 4.4 indicated that the fund should not overweight products that have been issued the last quarter during downturn markets and also avoid products close to maturity to attain protection to the worst-case outcomes. Minimizing the worst-case outcome imposes a slightly diﬀerent portfolio than minimizing CVaR. Hence it is very important for the investor/portfolio manager to understand how the diﬀerent decisions aﬀect the risk proﬁle. Investing only in products close to maturity would provide a higher expected return and impose a higher risk. This paper is not stating that any option is better than the other; it is only presenting the impact of diﬀerent portfolio options and goals. The other main investment alternative is to minimize CVaR with the whole investment universe available as done in this chapter, thus allocating most of the capital in the newly issued products since they have the best risk proﬁle in CVaR sense. When comparing the results of the three constructions (ﬁxed, rolling and dynamic) it is concluded that dynamic portfolio weights is the best choice in risk adjusted return sense (generates lower risk to the same return levels). The modiﬁed Korn and Zeytun framework will not overweight newly issued assets during downturn markets since it is minimizing the risk, thus rather investing in the products with a low ratio options:bonds. The downside with this strategy is that it may induce relative high transaction costs. The constraint of holding a maximum of 25% value in options can be added, thus forcing the portfolio manager to reduce the worst-case outcome risk at the same time as the CVaR (but notable minimizing the CVaR will in most cases avoid allocating too much capital in the options). Diﬀerent portfolio managers will make diﬀerent choices regarding which alternative is best, none is better than the other since they are constructed to restrict diﬀerent events. Transaction costs will be a central question no matter which investment strategy is chosen. The results are only relevant in relation to the investment environment, thus it is necessary that the SPF has the possibility of being more attractive than the benchmark fund, either at risk or return. Notable is that transaction costs are more important for the expected return and risk level than the portfolio choice itself. Thus given high transaction costs it is impossible for a SPF to compete with a benchmark fund with low transaction costs. Hence if the reader is considering to start a SPF it should focus on the transaction costs, it is impossible to outrun the mixed funds available at the market in the long run without limiting the transaction costs. M. Hveem 80 (103)

6.7. Backtesting

6.7

Backtesting

This section investigates how the modiﬁed Korn and Zeytun algorithm would have performed historically. The model is backtested from the beginning of 2002 to the middle of 2010 (data for OMXS30 total return is only available since 2002 from Bloomberg). In each time step the simulation is based on the last four year’s data (log returns), to capture approximately a market cycle. Proportional transaction costs at the levels 0%, 1%, 2% and 3% are considered. The CVaR is minimized given the constraint of a positive return. Notable is that the scheme is only rebalancing matured capital, to avoid as much unnecessary transaction costs as possible. The results are compared with the benchmark fund described earlier (with 30% market participation), which is based on the total return version of OMXS30, and with the actual OMXS30 total return version. Figure 6.11 shows that the SPF had the desired characteristics during the ﬁnancial turmoil during 2002 and during 2008-2009, imposing a high level of capital protection. Also the SPF did not actually yield any return during the latter part of 2009 and 2010, rather negative return. This is due to that the portfolio is rebalanced quite signiﬁcantly during the last quarters (imposing negative returns due to transaction costs) and the volatility drops signiﬁcantly, which implies that the value of the options decreases. The interest rate movements also impact the return during these quarters negative, the increase in the yield results in lower bond prices.

160 140

Nasdaq OMXS30 Total Return NAV Fund SPF Benchmark

160 140

Nasdaq OMXS30 Total Return NAV Fund SPF Benchmark

[%]

100 80 60 2003 2004 2005 2006 2007 2008 2009 2010

[%]

120

120 100 80 60 2003 2004 2005 2006 2007 2008 2009 2010

Year

Year

**(a) TC = 0%, SPF total return 51.73%.
**

Nasdaq OMXS30 Total Return NAV Fund SPF Benchmark

**(b) TC = 1%, SPF total return 43.35%.
**

Nasdaq OMXS30 Total Return NAV Fund SPF Benchmark

160 140

160 140

[%]

100 80 60 2003 2004 2005 2006 2007 2008 2009 2010

[%]

120

120 100 80 60 2003 2004 2005 2006 2007 2008 2009 2010

Year

Year

(c) TC = 2%, SPF total return 36.73%.

(d) TC = 3%, SPF total return 29.25%.

Figure 6.11: Backtesting results for the dynamic portfolio weights SPF against a benchmark mixed fund with 30% participation rate and OMXS30.

The transaction costs impact the return signiﬁcantly (notable is that the benchmark fund suﬀers in approximately the same extent). The problem with investing using a structured products fund is that the structured products mature and the capital must be reinvested in new products, imposing new transaction costs. Thus if the fund instead invests in products that have a longer time to maturity it is possible to both reduce the risk (since products with a longer time to maturity have lower risk) and the transaction costs. A backtest is also conducted for the ﬁxed portfolio weights strategy using the algorithm from Section 6.1 and in Figure 6.3, which takes in count the transaction costs, thus allocating mostly in the newly issued products, the result is disclosed in Figure 6.12. This strategy generates a higher return than the dynamic portfolio weights strategy during the time period, which is more due to a coincidence than an optimal portfolio choice, since this is just one observation. The ﬁxed portfolio weights setting generates higher transaction costs than the dynamic portfolio weights setting. It is really important to not rebalance the whole portfolio every quarter, instead the fund should mainly rebalance capital that has matured, in a setting with high transaction costs. Rebalancing the whole portfolio every quarter would in the dynamic portfolio weights setting, M. Hveem 81 (103)

Chapter 6. Portfolio Optimization

160 140

Nasdaq OMXS30 Total Return NAV Fund SPF Benchmark

160 140

Nasdaq OMXS30 Total Return NAV Fund SPF Benchmark

[%]

100 80 60 2003 2004 2005 2006 2007 2008 2009 2010

[%]

120

120 100 80 60 2003 2004 2005 2006 2007 2008 2009 2010

Year

Year

**(a) TC = 0%, SPF total return 63.64%.
**

Nasdaq OMXS30 Total Return NAV Fund SPF Benchmark

**(b) TC = 1%, SPF total return 45.86%.
**

Nasdaq OMXS30 Total Return NAV Fund SPF Benchmark

160 140

160 140

[%]

100 80 60 2003 2004 2005 2006 2007 2008 2009 2010

[%]

120

120 100 80 60 2003 2004 2005 2006 2007 2008 2009 2010

Year

Year

(c) TC = 2%, SPF total return 33.81%.

(d) TC = 3%, SPF total return 30.60%.

Figure 6.12: Backtesting results for the ﬁxed portfolio weights SPF including proportional transaction costs against a benchmark mixed fund with 30% participation rate and OMXS30.

with 3% in transaction costs, yield 15% in return instead of approximately 29%. Thus the portfolio manager should assume a type of buy and hold strategy and mainly reallocate matured capital. If the fund assumes a buy and hold strategy the portfolio manager should be aware of that it still needs to rebalance the whole portfolio sometimes, e.g. when the option portfolio value exceeds its allowed boundary. Thus it is recommended that, in a setting of high transaction costs, the portfolio manager uses the modiﬁed Korn and Zeytun framework to initially allocate its capital, and to reallocate the matured capital. The importance of using a more active portfolio strategy is higher when the possibility of diversifying in several structured products with diﬀerent underlying exists.

M. Hveem

82 (103)

the thesis uses the risk measure CVaR. The modiﬁed Korn and Zeytun framework is used to minimize the risk (CVaR) given the constraint of the expected return. On the other hand as the index starts to decrease (during bear markets) the portfolio should underweight newly issued products (products with the longest time to maturity) since they are risky products in downturn markets. which is referred to as the modiﬁed Korn and Zeytun framework. thus attaining as much capital protection as possible. Hence the thesis instead proposes a more sophisticated allocation algorithm. The study is conducted with both low 83 . The study shows that the concept should not be disregarded and that it is possible to construct a SPF that is competitive. The capital that is gained when assets are sold should be reinvested in the newly issued products since they have in general the lowest risk associated with them. The fund should sell of the products when they are close to maturity (a quarter or two left). It is also common to describe a fund’s risk characteristics via CVaR. to minimize the downside risk. A SPF with high transaction costs will not have the possibility of beating the competing mixed funds at either risk or return. CVaR is used as a risk measure since it captures both skewness and kurtosis and is adequate to use in scenario based optimization. The thesis ﬁnds that a portfolio manager has mainly two investment schemes to allocate the portfolio with. thus making it an adequate risk measure to benchmark through. since these assets exhibit a lot of risk especially if they are in the money. The thesis instead studies the concept of alternative capital protection and ﬁnds that a fund can be capital protected at a certain level for a given risk measure. On the other hand the scheme can induce a high-risk portfolio during unwanted market states. The study indicates that it is not possible to attain a portfolio based on structured products that is capital guaranteed.CHAPTER 7 Conclusions The thesis has studied the concept of a structured products fund (SPF) and how it should be constructed to be competitive in risk sense. Thus it is impossible to control the expected return of the investment as well as the risk level. Thus the fund needs low transaction costs (in level with the competing funds) in order to not increase the risk and decrease the return in a too high extent. This framework is very beneﬁcial and creates low risk portfolios with high expected return. The study focuses a lot on the issue of capital guarantee since it is desirable that the SPF has the same characteristics as a capital guaranteed product. The framework that is presented in detail in Section 6. Transaction costs aﬀect the expected return and risk level for the SPF in a high extent.90 less than for example 0% can claim that the portfolio is capital protected on a 100% level with a conﬁdence level of 90%. This scheme is used to both restrict the risk and the induced transaction costs. since it does not adjust to changes in the market. given that the transaction costs are low. The beneﬁt with this scheme is that it is easy to implement and that it serves as a reasonable investment choice during most time periods.4 shows that it is possible to generate portfolios that have a higher expected return than competing funds to a lower risk. The ﬁrst investment scheme is a type of rolling investment scheme where the portfolio manager invests broadly in almost all the available products (preferably equal weights). Thus a fund that has a CVaR0. Thus the study indicates that the most important factor for a portfolio manager to consider is the transaction costs.

no matter which scheme is used. A Black-Litterman approach to the optimization problem is left for future studies where it. to avoid downside risk for its investors. The big advantage by using the modiﬁed Korn and Zeytun framework is that it adjusts for these discrepancies and that the portfolio manager does not need to modify the allocation model if it modiﬁes its investment universe. can provide superior return due to subjective views on the market. Only one underlying for the ELNs has been used and the results can be generalized without serious loss of generality to the multi-dimensional case. Thus it is very important to use the boundaries of allowed proportion options:bonds especially when combining diﬀerent types of ELNs. for a SPF. The results indicate that a SPF can be a superb investment vehicle for investors searching for low risk alternatives with a limited downside. The thesis studies ELNs with a time to maturity of three years and the results can be generalized to products with a longer time to maturity at issuance (newly issued products. in combination with the multi-underlying case.Chapter 7. A recommended level is that the SPF is allowed to allocate a maximum of 25% of its capital in options (within the structured products) with a target level of 15%. The results show that a SPF is most beneﬁcial during times of low yield since the risk of the SPF increases a lot as the yield goes up. Conclusions yield and arbitrary yield start trajectories. The thesis ﬁnds that it is important. in CVaR sense). Allocating according to the modiﬁed Korn and Zeytun framework will in most cases lead to allocating in the best available portfolio (given the constraints) in every time step since it can be used for speciﬁc paths (adjust for the current market data). The thing that diﬀers for a capital guaranteed ELN with a time to maturity of three years at issuance and one with a longer time to maturity (or shorter) is the ratio options:bonds. M. less risk). since it generates well-balanced portfolios in risk and return sense. since it is a rather complex issue. Hveem 84 (103) . hence reducing the downside risk. It is strongly recommended to use the modiﬁed Korn and Zeytun framework to allocate the portfolio with. They also show that it is imperative that a portfolio manager understands how the portfolio allocation aﬀects the characteristics of the fund.e. Thus a newly issued ELN with ﬁve years to maturity has a higher risk than a newly issued ELN with three years to maturity since it has a higher proportion options. The portfolio should be rebalanced if this limit is exceeded. that the portfolio is not allowed to hold more than a certain degree of value attained in options (a limit). Thus the main result is: yes it is possible to construct a fund based on structured products that is competitive and exhibits a type of capital protection (i.

48 (5). 25 (2). 275-309.. Exorcising Ghosts of Octobers Past. (2007. J. Relationship Between Implied Volatility Indexes and Stock Index Returns. [9] Black. Liu. (2002). (2006). J. Wiley Interdisciplinary Reviews: Computational Statistics. Journal of Portfolio Management. The Journal of Finance. H. A. C.. Multivariate Stock Returns Around Extreme Events: A Reassessment of Economic Fundamentals and the 1987 Market Crash. 81 (3). Wall Street Journal.F.. G. Journal of Economics and Finance. Capital Protected is not Capital Guaranteed. Principal component analysis of yield curve movements. [5] Bank For International Settlements.. Mean-Variance and Scenario-Based Approaches to Portfolio Selection.. [16] Glot. (1992). [15] Frazzini. (1973). R. & Scholes. 5). NYU Working Paper. Y. S.. [11] Browning. & Li. doi:10. Mallett. (2000). The Generalised Hyperbolic Skew Student’s t-distribution. & Litterman. [7] Björk. Discussion Paper in Finance 2000-01. (2009).1007/s12197-010-9142-y. 26 (7). Principal Component Analysis.. Basel: BIS [6] Barber. L. 4 (2). G. [14] Fisher. No.com/capital-protected-is-not-capital-guaranteed/ [13] Deng. The Journal of Portfolio Management. T. T. 28-43. (1999). 1-47.. M. October 15). F.. & Copper. Basel II: International Convergence of Capital Measurement and Capital Standards: A Revised Framework. Journal of Banking & Finance. [8] Black.ifa-sg. [4] Alexander.. & Persand. K. 1505-1518. 92-100. L. Li. F. G. 10-22. [17] Grinold. [12] Chuan. Coleman. C1. Journal of Financial Econometrics. C. (2008. (2010). The Disposition Eﬀect and Underreaction to News. Minimizing CVaR and VaR for a portfolio of derivatives. S. August. p. (1999). R. M. (2010). Arbitrage Theory in Continuous Time (3rd ed. FIN-99-071.Bibliography [1] Aas. 61 (4). (2006). Structured Products In the Aftermath of Lehman Brothers. The Pricing of Options and Corporate Liabilities. (2006). J. Oxford: Oxford Finance. 85 . [10] Brooks. E. & McCann. Working Papers Securities Litigation & Consulting Group. S. [3] Acerbi. 31 (3). 2017-2046. University of Reading.. C. Journal of Banking & Finance. (2009). T. Global Portfolio Optimization. & Williams.. 583-605. 30 (2). Financial Planning Central: http://www.. Journal of Political Economy. A. (2005). 2. (2006). 637-654. Value at Risk and Market Crashes. Financial Analysts Journal. P. R. C.). Spectral measures of risk: A coherent representation of subjective risk aversion. 1-16. [2] Abdi.

). T. R. Stockholm: Kungliga Tekniska Högskolan. Simsek. Structured Product Glossary. M. J. My News Desk: http://www. Lesniewski. Conditional Value-at-Risk for General Loss Distributions.... A.. S.. 327343..com/se/pressroom/swedbank/pressrelease/ view/swedbank-stoerst-paa-indexobligationer-366818 [33] Rockafellar. Evaluating Long-Term Performance of Structured Products. [36] Soto. [28] Krokhmal. (2010). K. (2010). & Uryasev. [24] Hult.com/treasury/structured-products/wmsglobal/struc_prod_glossary.-J. H. 77-91. (2010. R. R.). & Woodward. Derivatives (5th ed.Bibliography [18] Hagan. E. H. T. London: Risk Books.). D. New Jersey: Prentice Hall.an Empirical Study of SEB Equity-Linked Notes.. H. & Tversky. Portfolio selection. [35] Shefrin. Hammarlid. Palmquist. J. 47 (2). K. (2002). 2001-5. [23] Hult. (2002). 54-61. (2009). 11-27. Available online at: http://ssrn. S. Hveem 86 (103) . EDHEC Risk and Asset Management Research Centre.com/abstract=985404. (2002). D. Working Paper No. The Journal of Fixed Income. [19] Hansen. Portfolio optimization with conditional value-at-risk objective and constraints. Econometrica. P.hsbcnet. [31] Martellini.C. O. Wilmott Magazine September 2002. P.. Lindskog. (1991). M. & Scheinkman. J. Optimization. Part I. 84-108. L. G. [26] Kahneman. & Lindskog. [29] Litterman.. 263-291.. H. F.. [25] Johansson. 4 (2).html [22] Hull. (1993). L. & Goltz. S. HSBC: http://www. Managing Smile Risk. [30] Markowitz. Structured forms of investment strategies in institutional investors portfolios. P.. ISE Dept. & Lingnardz. (2004) Portfolio Construction and Risk Budgeting (2nd ed. B. Evaluation of Equity-Linked Notes . Lecture notes in Risk management. (2005). Swedbank störst på indexobligationer. C. Prospect Theory: An Analysis of Decision under Risk.. (1952). 291-304. (2001). Stockholm: Royal Institute of Technology. & Rehn. S. 58 (3). Stockholm: Royal Institute of Technology. D. (2009). [20] Heston. Technical report.. Lund: Lund University. (1979). (2010. Kumar. Journal of Finance. 6 (2).. (2004) Using Principal Component Analysis to Explain Term Structure Movements: Performance and Stability. Journal of Risk. Oxford: Oxford University Press. Beyond Greed and Fear (2nd ed. [27] Korn. 1 (1).. A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and Currency Options. The Review of Financial Studies.. P. Common Factors Aﬀecting Bond Returns. A.. H. [34] Scherer. & Lärfars. Options. & Zeytun. Futures. Solving optimal investment problems with structured products under CVaR constraints. & Uryasev. [21] HSBC. (2010). 7 (1). Risk and portfolio analysis: principles and methods. [32] My News Desk.mynewsdesk. (2002). and Other. F. January 29). S. September 14).

[38] Uryasev. 37-43. Portfolio Insurance for the Small Investor in Switzerland. (2010. (1996). W. Financial Engineering News. Banking terms.com/1/e/gcc/bankingterms. UBS AG: http://www. Hveem 87 (103) .ubs. The Journal of Derivatives. C. 14. P.html. 3 (3). & Schenk. M. [39] Wasserfallen. (2000). 1-5. Conditional Value-at-Risk: Optimization Algorithms and Applications. October.Bibliography [37] UBG AG.. 12).

.

Appendices 89 .

.

120 Volatility NAV Fund 100 X: 24 Y: 92. 91 . −10. −7.APPENDIX A Naive fund constructions .76 [%] 60 [%] 0 5 10 15 20 25 60 40 40 20 20 0 0 0 5 10 15 20 25 Quarter Quarter (c) Scenario 5.72%.24%. 0%. return between Q12-Q24.87%.Additional scenarios Naive fund construction 1 250 Price Underlying NAV Fund 200 120 Price Underlying NAV Fund 100 X: 24 Y: 100 Start of investment 80 150 Start of investment 100 X: 24 Y: 92. (b) Scenario 4. (d) Scenario 6. Figure A.1: Scenarios 3-6.28 120 Start of investment 100 Interest Rate NAV Fund Start of investment 80 80 X: 24 Y: 89.13 60 $ $ 20 25 40 50 20 0 0 5 10 15 0 0 5 10 15 20 25 Quarter Quarter (a) Scenario 3. −7.

.

2 0 0.5 0 −0.APPENDIX B Option portfolios .2 −0.37%. −19.3 −0.2 0.1 0 0. 93 .2 1500 150 1000 400 100 500 200 50 0 −1 −0.Additional portfolios w9 = w10 = w11 = w12 = w13 = w14 = w15 = w16 = w17 = w18 = w19 = 1 0 0 0 0 1 4 0 1 0 0 0 1 4 0 0 0 1 10 0 0 0 1 10 0 0 0 0 0 0 1 10 0 0 0 1 10 1 8 0 0 0 0 0 0 1 10 1 8 1 8 0 0 0 1 10 1 8 0 0 1 1 10 1 8 0 0 0 T T T 1 10 1 8 1 10 T T T T 3 10 T T T 1 10 T 0 1 4 0 1 4 1 8 1 8 0 1 4 0 1 4 0 1 4 0 1 4 0 0 1 4 1 10 1 8 0 0 1 4 3 10 0 0 1 4 3 10 0 0 1 4 0 0 0 1 8 1 8 0 0 0 1 8 1 8 0 0 0 1 8 1 8 0 0 0 1 8 1 8 0 0 1 8 0 0 1 8 0 0 1 8 0 0 0 1 8 0 1 8 0 1 8 0 1 8 0 0 0 0 Option portfolio average monthly returns (36 months periods) 1200 2500 Option portfolio average monthly returns (36 months periods) 300 Option portfolio average monthly returns (36 months periods) 1000 2000 250 800 200 Outcomes Outcomes 600 Outcomes −0.1 0.1: Histograms over the outcomes for option portfolios 9-11.6 −0.4 0 −0.05 0. Figure B. (b) Portfolio 10.05 0 0.1 −0.4 −0. the worst-case average 36 month return is given in the caption.4 −0. −100%.1 0.15 −0. −47.02%.15 Average return per month Average return per month Average return per month (a) Portfolio 9.8 −0. (c) Portfolio 11.2 −0.

25 40 20 0 −0.1 0 0.15 Outcomes 150 150 100 100 50 50 0 −0.05 0 0.Appendix B.1 0.05 0 0.05 0 0.05 0.25 −0.05 0.05 0 0.25 −0.15 100 80 60 80 60 40 20 0 −0.2 −0.15 −0.15 Average return per month Average return per month (e) Portfolio 16.93%.15 Average return per month Average return per month (c) Portfolio 14.25 −0. M.1 −0. Option portfolio average monthly returns (36 months periods) Outcomes Outcomes −0. −20.05 0.15 −0.46%.1 −0.1 0.05 0 0. Hveem 94 (103) .1 −0.55%.47%.4 0 −0.15 −0.1 −0.1 0.44%.15 −0.04%. Option portfolio average monthly returns (36 months periods) 600 250 (b) Portfolio 13.3 −0. −24.2 0 −0.05 0.15 −0.05 0.2 0.15 150 100 50 0 −0.2 −0.05 0 0. Option portfolio average monthly returns (36 months periods) 500 200 400 Outcomes 300 Outcomes −0.05 0.3 −0.74%.25 Outcomes −0.15 Average return per month Average return per month (a) Portfolio 12. −26. Figure B.3 150 100 200 50 100 0 −0. Option portfolios .2 −0.15 Average return per month Average return per month (g) Portfolio 18.05 0 0. Option portfolio average monthly returns (36 months periods) 250 200 200 Outcomes −0.1 0.1 0. −39. the worst-case average 36 month return is given in the caption. (h) Portfolio 19. Option portfolio average monthly returns (36 months periods) 160 140 120 100 180 160 140 120 (f ) Portfolio 17.1 −0.2 −0.1 0.05 0.2 −0.1 −0.15 −0.94%. −24.1 0. Option portfolio average monthly returns (36 months periods) 250 (d) Portfolio 15. −20.2 −0.2: Histograms over the outcomes for option portfolios 12-19. −20.1 −0.15 −0.25 −0.1 0.2 −0.Additional portfolios Option portfolio average monthly returns (36 months periods) 180 160 140 120 Option portfolio average monthly returns (36 months periods) 250 200 Outcomes 100 80 60 40 20 0 −0. −20.

40% -32.02% -26.90% -8.15% -47.23% -73.20 0.20 0.1 0.1 0.85% -57.Additional scenarios Scenario C & D Return Index Months 49-96 / 51-98 p.20 0.10% -79.05 0.1 0.1 0.23% -71.47% -51.1 0.20 0.22% 2.76% -64. εbench and δ bench .05 0.26% -60.3 0.a.70% -23.57% -29.88% -63.10 0.10% -8.1 0.86% -50. 95 .05 0.10 0.7 0.69% Scenario D WorstCase Return -98.3 0.1 0.10 0.1 0.7 0.75% -19.7 0.05 Table C. -80% -80% -80% -80% -80% -50% -50% -50% -50% -50% -30% -30% -30% -30% -30% -20% -20% -20% -20% -20% -10% -10% -10% -10% -10% εbench δ bench Scenario C WorstCase Return -97.1 0.05% -29.52% -69.5 0.33% -11.1 0.26% -71.56% -87.7 0.12% -94.10 0.13% -43.5 0.1: Benchmark fund: Worst-case outcome given scenarios C and D.APPENDIX C SPF vs Benchmark .20 0.02% -32.3 0.58% -83.70% 0.27% -14.36% -91.1 0.91% -42.01% -78.3 0.05 0.92% -71.10% 0.1 0.1 0.96% -12.1 0.1 0.1 0.82% 2.60% -10.7 0.10% -91.95% -18.5 0.21% -32.57% -57.10 0.1 0.3 0.11% -25.1 0.5 0.1 0.32% -78.09% -52.83% -37.53% -40.34% -62.5 0.41% -51.77% -77.1 0.

28% 1 0.87% -25.14% -26.25 0.62% -64.05 1 0.42% 1.27% -63.10 1 0. return 30% p. Scenario D Months 0-48.65% -1.43% -64.24% -53.a. return -80% p.20 0. Months 48-49.35% -20.1 0.a.79% -20.09% -36.32% 3.10 1 0.35% -11.95% -63. return 50% p.65% -11.3 0.1 0.62% -55.82% -3.31% -62. Months 96-144.1 0.76% 0.20 0.26% -26.05 Table C.28% -30.45% -3.41% -24. return 30% p.91% Scenario D WorstCase Return -61. Months 48-50. Months 50-51.3 0.26% -15.2: SPF 1: Worst-case outcomes for the SPF 1 Portfolio 7 given the scenarios C and D.a.15 0.Appendix C.61% -14.1 0.03% -30.25% -27.58% -10.3 0.1 0.20% -3.10 1 0.10 1 0.1 0.a.1 0.1 0.47% -37. return -X% p.15 0.30% -25. return 30% p.26% -17.3 0.05 1 0. return -X% p.1 0.3 0. Months 98-144. return 30% p.26% -24.00% -32.69% -62.1 0.15 0.a.72% -64.91% -55.1 0.20 0. return -80% p.1 0. Months 51-98.1 0.m.a.Additional scenarios Scenario C Months 0-48.05 1 0.18% -16. εSPF and δ SPF .25 0.94% 3.07% -6.05 1 0.1 0.25 0.m.10 1 0.91% -64.a.20 0. -80% -80% -80% -80% -80% -80% -50% -50% -50% -50% -50% -50% -30% -30% -30% -30% -30% -30% -20% -20% -20% -20% -20% -20% -10% -10% -10% -10% -10% -10% εSPF δ SPF Scenario C WorstCase Return -20.65% -20.15 0.1 0.25 0.671% -25.95% -54.46% -54.09% -54.49% -51.1 0.25% -10.31% -52.41% -5.12% -50.22% -35.1 0. Months 49-96.48% -31.15 0. M.1 0.84% -17. SPF vs Benchmark . Hveem 96 (103) .1 0.10% -32.67% -17.a.49% -61. Return Index Months 49-96 / 51-98 p.20 0.17% -51.1 0.25 0.

87% -24.92% -9.46% -15.63% -12.05 Table C.10 1 0.28% -23.15 0.1 0.1 0.5% -1.02% -24.68% -15.46% -20.92% -13.20 0.05 1 0.15 0.10 1 0.25 0.15 0.05 0.78% -27.11% -26.69% -12.33% -14.15 0.92% -5.43% -20.20 0.00% Scenario B WorstCase Return -57.77% -13.15 0.15 0.1 0.71% -20.62% -9.82% -16. -80% -80% -80% -80% -80% -80% -50% -50% -50% -50% -50% -50% -30% -30% -30% -30% -30% -30% -20% -20% -20% -20% -20% -20% -10% -10% -10% -10% -10% -10% εSPF δ SPF Scenario A WorstCase Return -26.84% -26.49% 0.05 1 0.05 0. Hveem 97 (103) .28% 5.05 1 0.3: SPF 1: Worst-case outcomes for the SPF 1 Portfolio 3 given scenarios A and B.90% -35.25 0.53% -11.98% -14.94% -29.20 0.19% -13.05 0.45% -22.88% -40.1 0.11% -19.05 1 0.a.10 1 0.10 1 0.12% -9.10 1 0.1 0.15 0.1 0.35% 1 0.1 0.1 0.25 0.1 0.56% -8.15 0.51% -26.15 0. M.94% -0.05 0.36% -5.01% -16.71% -9.21% -12.00% 3.1 0. εSPF and δ SPF .91% -8.25 0.1 0. Scenario A & B Return Index Months 48-96 / 50-98 p.96% -16.05 0.72% -3.53% -4.31% -1.19% -8.79% -13.Portfolio 3.1 0.1 0.59% -31.1 0.71% -12.40% -16.35% -5.15 0.87% -18.53% -16.20 0.78% -37.38% -0.44% 0.20 0.77% -49.1 0.25 0.15% -19.

1 Return 0.Appendix C. max: 45. Nasdaq OMXS30 NAV Fund NAV Benchmark (b) εSPF = 0. SPF vs Benchmark .15. max: 63.2 Return 0.15.59%.5 −0.1 0 0.4 0.6 0. for underlying instead of 0% p. δ SPF = 0. min: −37.4 0.5 0 Return 0.4 −0.3 −0.Additional scenarios Stress testing µ = −30% p.24%. Hveem 98 (103) . min: −20.10.87%.2 0 0. as in Section 4. δ SPF = (b) Benchmark: εbench = 0.8 1 2000 1000 1000 0 −0. mean: −7. Nasdaq OMXS30 NAV Fund NAV Benchmark 400 350 300 400 350 300 $ 250 200 150 100 1996 $ 250 200 150 100 1996 1998 2000 2002 2004 2006 2008 2010 2012 1998 2000 2002 2004 2006 2008 2010 2012 Year Year (c) εSPF = 0.03%.25.3 0. (c) Underlying: min: −92.1: Histograms over 36 months returns for the competing funds and the modiﬁed stress test Backtesting 400 350 300 Nasdaq OMXS30 NAV Fund NAV Benchmark 400 350 300 Nasdaq OMXS30 NAV Fund NAV Benchmark $ 250 200 150 100 1996 $ 250 200 150 100 1996 1998 2000 2002 2004 2006 2008 2010 2012 1998 2000 2002 2004 2006 2008 2010 2012 Year Year (a) εSPF = 0.5 1 1.05. 0. mean: δ bench = 0. Figure C. (d) εSPF = 0.1.a.2 0. mean: −56.2 −0.3. δ SPF = 0.70%.1.82%. M. 8.1.01%.2: Backtest results: OMXS30 vs SPF 1 (Portfolio 4) and Benchmark.a. δ SPF = 0.4.83%.2.. Figure C. max: 129.60%.1.5 (a) SPF 1: εSPF = 0.7 Histogram over the SPF:s 36 month (overlapping) returns 3500 2500 Histogram over the Benchmark funds 36 month (overlapping) returns 4000 3500 2000 3000 2500 1500 2500 Outcomes 2000 1500 1000 500 500 500 0 −0.1.4 0 −1 Histogram over the Underlying 36 month (overlapping) returns 3000 Outcomes 1500 Outcomes −0. δ SPF = 0.

0193 0.0000 Correlation matrix between OMXS30 and SEK yield curve.9864 0.9149 1.8916 0.9563 1.1741 0.5972 0.7048 (Jan 1996 .8471 0.8016 2M -0.6855 0.1029 1M -0.8422 0.8191 0.7048 1.0000 0.3837 0.7399 0.9610 0.8229 0.4317 0.3915 0.6438 0.9610 0.9563 0.1233 0.9626 1.9608 0.6677 0.0000 9M -0.1899 0.3837 0.8897 0.9777 1. 1Y 2Y 3Y 0.7473 0.8849 0.9362 1Y 0. Index 1M 2M 3M 6M Index 1.0193 0.5965 monthly log returns 9M 1Y -0.9608 1. Index 1M 2M 3M 6M Index 1.8429 0.9578 0.7850 0.3914 0. 2Y 3Y 0.8634 0.0012 1M -0.8559 0.6677 0.7226 0.0080 -0.9513 0.9149 0.0000 0.8947 6M 0.5646 1.0935 -0.4488 0.8693 0. 1Y 2Y 3Y -0.5317 0.4981 1.9553 6M -0.1900 0.2382 0.1320 0.0000 0.0504 1.2666 0.9214 2Y 0.1041 -0.8325 0.0000 9M -0.4488 0.0000 0.9362 1.4783 0.2382 0.9800 1Y -0.8965 0.1233 0.9711 1.8849 0.1741 0.3657 daily log 9M 0.8415 0.8897 3M -0.8223 0.9578 1.0000 0.4100 1.8970 1.7473 0.8471 0.9214 0.0211 0.4981 0.2802 0.3610 0.9711 1Y -0.9864 0.0000 0.7497 0.8191 0.5317 0.4783 0.1041 1.7226 0.4410 0.1328 0.5900 0.4993 0.3765 0.9134 0.9166 0.9777 0.0757 0.1320 -0.7850 0.3657 0.1900 0.4100 returns (Jan 1996 .0000 0.9626 2Y 0.8971 0.4993 weekly log 9M -0.Nov 2010).0000 0.9412 0.0000 0.0000 -0.6684 1.7678 2M -0.8415 2M -0.4317 0.0000 0.7399 3Y 0.0117 0.0000 0.0484 -0.4975 0.9800 0.4410 0.6684 0.1121 0.8971 0.0458 -0.1899 0.3166 0.8422 3M -0.2802 0.9345 1.3166 0.0484 0.6855 0.3070 0.7940 0.9258 0.4487 0.0000 0.4975 0.4912 0.0070 0.0000 0.8559 0.8325 0.0000 0.3914 0.4487 3Y 0.9932 0.7678 0.0000 0.5965 0.0000 0.0023 -0.7497 0.APPENDIX D Correlations between the market index and the yield curve Correlation matrix between OMXS30 and SEK yield curve.3378 0.1029 0.0120 0.0000 -0.8947 1.8970 0.8229 0.0000 0. Index 1M 2M 3M 6M Index 1.0000 0.1891 0.3610 0.8429 0.9932 1.0000 0.9345 6M -0.0070 0.0000 99 .0211 1M -0.2666 0.9431 0.3378 0.7628 0.0458 0.0012 0.0504 -0.Nov 2010).0000 -0.8016 0.5972 0.Nov 2010).5965 3Y 0.4912 0.8965 0.0000 0.9431 1.6438 0.0023 0.8223 0.3070 0.0000 Correlation matrix between OMXS30 and SEK yield curve.6419 0.5646 returns (Jan 1996 .7628 0.0935 0.9553 1.0757 0.3915 0.8634 0.0000 9M 0.3765 0.5965 0.7940 0.8693 0.0117 0.1328 -0.1891 0.7791 0.7791 0.0120 0.9166 3M -0.8916 1.9513 2Y 0.7081 0.5900 0.7081 0.0080 1.9134 0.9258 0.6419 0.9412 1.1121 0.

.

19 9 7. 100 90 80 70 60 Weight [%] 50 40 30 20 10 w1 0 6.14 Product Number.76 26.91 Table E.8 12.38 12 6.76 3 16. thus the initial yield curve is drawn randomly from the observed sample.1: Portfolio weights for the eﬃcient frontier using the ﬁxed portfolio weights setting. 101 .17 10 6.8 14 15 16 17 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w12 Portfolio CVaRα = 0. The initial yield curve is drawn arbitrary between January 1996 to November 2010.99 27.95 [%] 5 6 7 8 12.95.1: The table describes the CVaR at the 95% conﬁdence level and expected return for the diﬀerent products over the three-year period using the ﬁxed portfolio weights setting.64 11 6.68 12 25.2 7.73 2 18.63 1 22.6 11.27 3 28.Arbitrary start yield curve This appendix uses an arbitrary start yield curve.94 25.75 10 25.53 11. α = 0.58 1 28.45 4 14.4 9.21 9. Fixed portfolio weights Product Number.20 26.33 25.58 11 25.APPENDIX E Optimization problem .95 [%] Figure E.4 8.20 10.78 2 28. CVaR0. Expected return [%] 4 5 6 7 8 9 27.6 10.

Dividends.43 28.97 12 0 12 27.Appendix E.4 5.31 20. Dividends.95.09 -3.13 24.6 4.2 6. 100 90 80 70 60 Weight [%] 50 40 30 20 10 w1 0 4.76 26.89 1 7.72 14.27 2 26.3: The table discloses the expected return and CVaR for diﬀerent levels of index weights over the three-year period for a benchmark fund.83 7.57 23.89 Table E.41 14.37 5.95 17. Expected Return CVaR0.49 13.67 26.41 Index allocation 10% 20% 30% 40% Table E.76 20. M. Expected return [%] 4 5 6 7 8 9 10 26.8 5.Arbitrary start yield curve Rolling portfolio weights Product (months until next roll).4 7.99 11 26.46 Excl.41 11 9.96 14.18 13.46 1. Expected Return CVaR0.95 16.73 26. Benchmark fund Incl.2: Portfolio weights for the eﬃcient frontier using the rolling portfolio weights setting.81 0.33 14.68 1 26.83 10 12.87 3 11.17 18. α = 0.64 26.18 11. CVaR0.95 3 26.64 26.95 [%] 4 5 6 7 8 9 13.25 -3.95 [%] Figure E.2: The table describes the CVaR at the 95% conﬁdence level and expected return for the diﬀerent products over the three-year period using the rolling portfolio weights setting.6 7 7.13 Product (months until next roll).63 26. Hveem 102 (103) .8 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w12 Portfolio CVaRα = 0.8 6.95 2 10. Optimization problem .

95 [%] Figure E.53 14.88 Product. CVaR0. months until maturity.00 12 26.70 3 26. α = 0.85 1 26.8 w2 w3 w4 w5 w6 w7 w8 w9 w10 w11 w12 Portfolio CVaRα = 0.69 2 12.8 7.46 26.01 1 9.58 3 13.34 11 7.95 [%] 4 5 6 7 8 9 14.52 10 10.35 26.31 26. months until maturity.95.74 12 0.63 14.35 26.4: The table describes the CVaR at the 95% conﬁdence level and expected return for the diﬀerent products over the three-year period using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework .8 8.41 26.8 6. M.3: Portfolio weights for the eﬃcient frontier using the dynamic portfolio weights and the modiﬁed Korn and Zeytun framework.34 2 26. Hveem 103 (103) .14 12. 100 90 80 70 60 Weight [%] 50 40 30 20 10 w1 0 5.Dynamic portfolio weights Product.52 26.45 14.00 13. Expected return [%] 4 5 6 7 8 9 10 26.61 Table E.70 11 26.