Basic concepts in

Portfolio Analysis

Wolfgang Marty

(3. Version 18.07.2011)

Table of contents

1. Introduction

2. Return analysis
2.1 Interest rate, return and compounding
2.2. Return contribution and attribution
2.3. Time weighted rate of return (TWR)
2.4. The money weighted rate of return (MWR)
2.5. TWR attribution versus MWR attribution

3. Risk measurement
3.1. Absolute risk
3.2. Value at Risk
3.3. Relative risk
3.4. Risk decomposition on asset level
3.5. Risk decomposition with factor analysis

4. Performance measurement
4.1. Investment Process and Portfolio Construction
4.2. Portfolio optimisation
4.3. Absolute Risk adjusted measures
4.4. Capital asset pricing model
4.5. Relative Risk adjusted measures

5. Investment controlling
5.1. Investment controlling and the investment process
5.2. Building blocs and set up of investment controlling
5.3 The benefits of GIPS reporting
5.4. Some critical issues in performance attribution

Index

References
Preface

While getting acquainted with a series of topics relating to

the banking industry I realized that scientific methods and
terminologies describe many aspects of financial markets. Many
notions that are used in finance have a scientific background. In
the daily activity of the finance industry, it is often not even
realized that mathematical concepts are applied. The use of
mathematics ranges from basic arithmetic manipulations such
adding positions on an account or calculating interest payments,
to sophisticated problems such that investing times series or
portfolio optimization. In my view the connection between
science and financial mathematics is particularly apparent in
portfolio analysis. The illustration of this fact is one of the
scopes of this book.
In traditional asset management industry I was confronted
with statements like the following: ‘I am a practitioner and I do
not need the abstraction of a theory.’ Nothing could be further
from the truth. Using portfolio analysis, I intend to explain why
concepts that are based on mathematics can enhance the
decision making process of the participants in the financial
community. Furthermore the approach we describe here
alleviates the unification of the terminology and provides a
concise use on the notions.
Financial markets are inundated by data. In other words
they are produced literally continuously. In portfolio analysis, the
amalgamation and reconciliation of data is important. We try to
answer the following question: What information is relevant in
portfolio analysis? We intend to explain the figures that asses a
specific portfolio.
In this book the basic notions in portfolio analysis are
introduced by a mathematical notation. For more advanced
concepts and ideas we limit our exposition to intuitive
descriptions. The book does not only contain traditional material
but also current topics in portfolio analysis. We also refer to the
relevant literature.
Acknowledgments

This book is based on several courses and seminars held at

CS business school. The transformation of a series of
presentations in a lecture book is a not to be underestimated
challenge. Many discussions throughout my daily activities
consisted substantially to the here presented material. I am
particular grateful to Dominik Büsser and Petra Illic for revising
my first daft of this book.
I would like to give my thanks to my employer Credit Suisse
and in particular to the following colleagues in alphabetical
order: Rolf Bertschi, Ralph Häfliger, Dr. Stefan Illmer, Thomas
Kellersberger, Kurt Oberhänsli, Elias Mulky and Thomas
Widmer. Their thoughts and ideas added great value to various
parts of this book.
The continuing discussions with Professor Janos Mayer
from the University of Zurich helped me concerning
mathematical aspects of this book. In addition, I could establish
an excellent platform with Yuri Shestopaloff for exchanging
thoughts on different issues in performance measurement.
Conventions

This book consists of 5 chapters. The chapters are divided

into sections. (1.2.3) denotes formula (3) in Section 2 of
Chapter 1. If we refer to formula (2) in Section 2 of Chapter 1
we only write (2) otherwise we use the full reference (1.2.3).
Within the chapters, definitions, assumptions, theorems and
examples are numerated continually, e.g. Theorem 2.1 refers to
Theorem 1 in Chapter 2. They are numerated continually.
The end of a theorem, lemma or example is marked with ◊ .
Square brackets [ ] contain references. The details of the
references are given at the end of the book.
 1

1. INTRODUCTION

A portfolio is a set of investments. The return of a portfolio

is probably the most important information for the investor. In
most cases the investor has an intuitive perception of the
return. If the value of the portfolio rises the return is positive and
if the value of the portfolio declines the return is negative.
However, there is no unique method for calculating the return.
The discussion of different approaches for return calculation is
subject of this book.
Generally speaking prices are provided by market or by
theoretical considerations. In the following we assume that the
prices of the different investments are available and as a
consequence it is possible to calculate the return. Returns can
be calculated from past prices in the market place (ex post),
from actual prices or from projected prices (ex ante) (see Figure
1.1).
The return depends of the input data and the more
frequent the data is available the more meticulously the return
of the investments and the portfolio can be monitored.

ex post ex ante

past future

P-1 P0 P1

Time axis

Now - 1 year now now + 1 year

The setting for portfolio analysis Figure 1.1

There are two main different concepts for measuring the

return. The first concept starts with the evaluation of the
portfolio, parts of a portfolio or an investment at different points
 2

of time. The return is then calculated while time elapses, i.e. it

always refers to a time span. A value for a return is always
accompanied by a time span. For instance, a return over
multiple years is usually annualized. This concept neutralizes
external cash flows in the portfolios. Thus, it is appropriate for
comparing portfolio managers, i.e. it is the base for peer
analysis. The unit of the return is percentage. This concept is
subject of Section 2.3 entitled Time weighted rate of return
(TWR). Examples are e.g. the total return of an investment,
profit and loss considerations.
The second concept is based on an arbitrage
relationship, i.e. a condition which avoids a situation with a
profit without risk. We refer here to the saying: ‘There is no free
lunch in the financial market.’ The corresponding return
measurement is based on the value of portfolio, parts of a
portfolio or an investment at one point in time. This concept
includes cash flow and it is thus the return from the client’s
perspective. Profit and Loss are also considered. This concept
is subject of the Section 2.4 entitled Money weighted rate of
return (MWR). The Internal rate of return (IRR) equation is
such an example (see e.g. Section 2.5 and Section 2.8) of the
MWR approach. TWR and MWR are applied on an absolute and
a relative basis. We discuss in section 2.5 the relationship
between TWR and MWR.
With the development of portfolio analysis risk
considerations have become more influential. Here we assess
the risk of a portfolio by the elementary measure standard
deviation used throughout in statistics in other words we
quantify the risk by a value for the fluctuation of the return.
Intuitively this means that if the return is constant there is no
risk and the bigger the risk value is the bigger the uncertainty of
the returns is.
As will be showed in Chapters 2 and 3 the return and the
risk can be considered on an absolute and on a relative basis.
The tracking error originally stems from Control theory and is
a relative risk measure. We will see that the concept of returns
 3

is easier to understand than to the concept of risk. Returns are

additive, i.e. when referring the weights of the investments then
the sum of the return of the investment adds up to the returns
of the portfolio. Adding risks of investments obeys a
generalization of the theorem of Pythagoras which states that
under the assumption that a and b are rectangular the sum of
a2 and b2 is equal to c2 (Figure 1.2).

b
a

c
A geometrical interpretation: Adding risk Figure 1.2

The following statement explains the important notion of

performance:

Performance encompasses risk and return.

Remark 1.1: In physics, power is introduced through work

by unit of time and the transition from finance to physics is
reflected by performance to power, by return to work and by
risk to time.

Performance deals with the past, the present and the

future as seen in Figure 1.1. The time axis and the Price Pi
(i = −1, 0, 1) of an asset in the past on the left side and
potential outcome in the future on the right side. In the past we
deal with statistics whereas in the present and the present we
are confined to applying models providing forecasts and
projections. The calculations have to reflect cash flows such as
dividends from equities or coupons from bonds and liabilities
 4

that are typically important for pension funds. Many assets do

not have a price history. Estimations to their future behaviour
can solely depend on future scenarios and a statistical price
analysis is not possible. Such examples are Initial Public
Offering (IPO), recently built house or bonds just launched in
the market.
The expost return analysis is a time series analysis and an
exante return is the same as a return forecast. The expost
risk analysis of a portfolio is a time series analysis of the
returns of the portfolio, i.e. all past changes in the portfolio are
reflected in the risk figure. An exante risk analysis, however,
is based on the actual portfolio and the following mathematical
question:

• What are the main drivers of the risk in the portfolio?

A set of factors intends to explain the risk of a portfolio. The

choice of the factors requires intense research. They are
modelled by time series and do not depend of the portfolios
under investigation. They are estimated by the underlying
market universe of the portfolio. The factors are objective
estimations of the considered markets and the portfolio is
subjective. Thus a risk model can also be used for risk
management. Updating the risk model is a slow process, i.e.
updating once a month is sufficient. Up to our knowledge the
data is mostly based on monthly frequency and in some cases
on daily frequency.
We proceed by summarizing the main topics addressed
here:
1. We introduce some basic notions for describing and
assessing properties of portfolios.
2. The decomposition of the portfolio into its constituents is
discussed.
3. The portfolio is investigated on an absolute basis and
relative to a reference portfolio.
 5

The following definition brings the notion of return and risk

together.

Definition 1.1: Portfolio Analytics is concerned with

quantifying the sources of the return and assessing the risk of a
portfolio. It does not only measure the evolution of the wealth
over a certain time period but also provides a comprehensive
discussion of the performance of specific portfolios.

We proceed with the following remarks:

Remark 1.2: In portfolio analytics we consider mostly
variance or covariance as risk measures. They are introduced in
Chapter 3.
Remark 1.3: We distinguish between performance
measurement on a portfolio level and performance
measurements on segment or on constituent level and on a
multi period level. We intend to discuss the complete portfolio
analysis problem.
Performance attribution is an important part of Portfolio
Analytics. The performance attribution is essentially the
decomposition of a real number, i.e. 5 = 1 + 4 or 5 = 2 + 3. In
the arithmetic just considered the operation usually goes from
right to left. In performance attribution in the operation goes
from left to right. The return and the risk number of a portfolio
is decomposed. However, what is paramount is that this
decomposition reflects the requirement of the client, portfolio
managers and the adviser. In this book we focus on the return
attribution. On the risk side we will only give an introduction in
decomposing to the risk.
We present a performance attribution to the TWR and
MWR. IRR with multi solutions we illustrate by easy examples.
Performance Attribution for TWR is common practise, but the
approach presented here for the MWR is new.
Chapter 4 introduces a holistic approach from a portfolio’s
view, i.e. return and risk considerations are simultaneously
taken in account. We start with the part portfolio construction in
 6

the investment process. We describe the key ingredient for the

portfolio construction. The material presented here is basic and
traditional but it can be considered as summary of voluminous
text books. We show by an example how constraints can
change optimal portfolios. In Chapter 5 we focus on the part
investment controlling in the investment process. We compare
many portfolios and give an introduction in Global Investment
Performance Standard (GIPS). We illustrate a performance
review.
 7

2. RETURN ANALYSIS

2.1. Interest rate, return and compounding

Retail accounts are facilities that give the clients of a bank the
possibility to deposit their money. The client lends money to the
bank whereas the bank borrows money from the client. Money on
accounts is also called a retail deposit. The bank will recompense
their clients by paying interest rate. Such accounts bear no or
only little interest rate. Low interest rates can be justified by the
fact that the costs of the banks for the infrastructure, staff and
paperwork are substantial. In the following we present the basic
interest calculation. The Beginning value BV and the End value EV
of a money account are related to the interest rate r:

EV = BV (1+ r). (2.1.1)

BV and EV are amounts in US Dollar ($). However, our

considerations in the following are applicable to any reference
currency. In relationship (1) the time span between BV and EV is
not specified. Assuming that an investor puts a 100$ on an
account and gets 200$ after 10 years and another investor earned
starting from 100$ to 200$ in a year. Interest rates are usually
quoted on a yearly or annual basis. However, other time spans like
days, months or 6 months (semi-annual) can also be appropriate in
specific situations. It is seen that an interest rate is always referred
to a time spam. It is not related to a point on the time line.
Sometimes interest rate is always called a growth rate.
In most cases the relationship (1) is applied in 3 versions. First
if we know BV and r, EV can be calculated. BV is the present
value und EV is the future value. The future value is on the right
side of the present value on the time axis (Figure 1.1).
 8

Example 2.1: If BV = 100$ and r = 5% then

EV = 100$ . (1 + 0.05) = 105$.

Secondly if EV and BV are known and if we assume than the

investor starts from an initial lump sum BV >0 then the solution r
of (1) is called the return expressed by

EV − BV EV
r= = − 1. (2.1.2)
BV BV

We divide the profit or loss (EV − BV) by the invested capital BV.
EV cannot be negative as a complete loss translates in EV = 0, i.e.
that is we have

EV ≥ 0,

i.e. by (1)

r ≥ −1.

Example 2.2: BV = 100$ and EV = 200$ then

EV − BV 200$ − 100$
r= = = 1,
BV 100$

i.e. in percentage

r = 100%.

◊
 9

Thirdly if EV and r with r > −1 is given we have

EV
BV = .
1+ r

By using the discount factor d defined by

1
d= (2.1.3)
1+ r

we find

BV = d EV.

Broadly speaking interest rates with the accompanying

discount factors translate money amounts though time.

Definition 2.1: Compounding is the reinvestment of the

income to earn more income in the subsequent periods. If the
income and the gains are retained within the investment vehicle or
reinvested, they will accumulate and contribute to the starting
balance for each subsequent period’s income calculation.

In the following we look at an arbitrary but fixed time base

period. We proceed by assuming that the interest is distributed
twice and is reinvested by the same interest rate at the middle of
the period. We consider

EV2 = ⎡⎢ BV ⋅ ⎛⎜ 1 + ⎞⎟ ⎤⎥ . ⎛⎜ 1 + ⎞⎟
r r
⎣ ⎝ 2 ⎠⎦ ⎝ 2⎠

If interest is distributed n-times per period we find

 10

n
⎛ r⎞
EVn = BV ⎜1+ ⎟ . (2.1.4)
⎝ n⎠

We define e by the limes

n
⎛ 1⎞
lim ⎜1 + ⎟ = e.
n→ ∞ ⎝ n⎠

The number e appears in many contexts in science and is called

after its founder Leonard Euler. The 14 digit numerical value is

e = 2.71828’18284’5905.

The faculty n! is defined by

n! = n (n − 1) (n − 2)…..

For calculating e numerically it is favourable to use the series

1 1 1 1 1
e = 1+ + + + + +….
1! 2 ! 3 ! 4 ! 5 !
and find

n
⎛ r⎞ r
lim ⎜1 + ⎟ = e
n→∞ ⎝ n⎠

which yields

n
⎛ r⎞
EV∞ = BV lim ⎜1 + ⎟ = BV er. (2.1.5)
n→ ∞⎝ n⎠
 11

We see that the end values EV1, EV2,…., EV∞ are increasing
with the compounding frequency. EV∞ is the Ending value using
continuously compounding as the money is reinvested
momentarily. The above formulae are assumed that the beginning
value BV, the interesting rate r and the compounding assumption
are given and the ending values EV1, EV2,…., EVn,…, EV∞ are
computed.

Example 2.3: For r = 0.05 (= 5%) annually we get in

decimals and BV = 100$

EV2 = 105.06250 (Semi annual),

EV4 = 105.09453 (Quarterly),
EV∞ = 105.127110 (Continuous).

If the beginning value BV, the ending value EV and the

number of compounding annually are given the underlying interest
is based on (4)

⎛ EV ⎞
r(n) = ⎜⎜ n − 1⎟⎟ n .
⎝ BV ⎠

Consistency with (2) yields r = r(1). If the beginning value BV and

the ending value EV are given the underlying return is found based
on (5)

EV
r∞ = ln ,
BV

where ln is the natural logarithm, i.e. the logarithm for the basis e.
 12

In the following we look at N base periods with n = 1 in the

above formulae. We distinguish different ways of compounding.
Simple or arithmetic simple return means that the interests are
adding

EV = BV . (1 + N . r). (2.1.6)

We use simple interest calculation if the earned income is

withdrawn from the account. Geometric compounding is mostly
used in practice

EV = BV . (1 + r)N. (2.1.7)

For N = 1 (6) is the same as (7). Note that the time unit is year.
Then the continuous versions, i.e. the formulae for arbitrary t are

EV1(t) = BV . (1 + t . r), resp. (2.1.8a)

EV∞ (t) = BV . (1 + r)t. (2.1.8b)

We illustrate (8) with r = 0.01 (= 10%)

r (8a) (8b) (8b)-(8a)

0 1.000 1.000 0.0000000
0.1 1.010 1.010 0.0004234
0.2 1.020 1.019 0.0007551
0.3 1.030 1.029 0.0009942
0.4 1.040 1.039 0.0011399
0.5 1.050 1.049 0.0011912
0.6 1.060 1.059 0.0011471
0.7 1.070 1.069 0.0010070
0.8 1.080 1.079 0.0007697
0.9 1.090 1.090 0.0004343
1 1.100 1.100 0.0000000
 13

Remark 2.1: (8a) and (8b) are the same for t = 1. We

consider (8a) and (8b) for t ∈ [0, 1]. For no compounding we have

EV(t) = BV(1 + tr), ∀ t ≥ 0

and for continuous compounding we have

EV(t) = BV er1t . (2.1.9)

For t = 1 there follows from (8a) and (8b)

r1 = ln(1+ r).

thus

r1 t = ln(1 + r) t, ∀ t ≥ 0

and we conclude

EV(t) = BV eln(1+r ) t = BV . (1 + r)t, ∀ t ≥ 0.

In (8a) there is no compounding and in (8b) we have continuously

compounding. A similar formula can be derived for arbitrary
compounded period.

Remark 2.2: Considering semi-annual compounding by

r
(1 + )2 (2.1.10)
2

is faulty as the time is scaled by 2. In other words a semi-annular

interest stream of 1$ at the end of the year is
 14

r r
(1 + )(1 + r1)0.5 − 1 +
2 2

where r1 is a reinvestment assumption of the first cash flow. This

formula assumes that the proceeds are paid out after the first half
year and the interest is r1 for the second half of the year.

⎛ r r⎞
EV = ⎜ (1 + )(1 + r1)0.5 + ⎟ BV .
⎝ 2 2⎠

Example 2.4: We consider a semi-annual Bond with Coupon

C = 8%, yield to maturity r = 4% and time to maturity t = 0.5
years. Referring to discount factor (3) the coupon stream in the
first year based on (8b) and (10) after half a year is

C C
2 = 3.922323, 2 = 3.921569
(1 + r )0.5
1+
r
2

and after one year is

C C
2 = 3.846154, 2
2 = 3.698225
1+ r ⎛ r ⎞
⎜1 + ⎟
⎝ 2 ⎠

We note that they are not the same. The size of the difference
depends of the value for C and r.

◊
 15

Furthermore continuous compounding

EV = BV . erN

is mostly used in the academic literature because it allows to

formulate theoretical contexts in an elegant and efficient way. The
continuous version for arbitrary time t is

EV = BV . ert.
 16

2.2 Return Contribution and Attribution

In the following we start by a set of investments like equities

or bonds.

Definition 2.2: A portfolio is a set of investments.

Building blocs of the analysis of a portfolio are

investments. On the time axis we consider a beginning portfolio
value PB and an end portfolio value PE.

Definition 2.3: The simple or arithmetic rate of return rp

of a portfolio P is

PE − PB PE
rP = = − 1, (2.2.1)
PB PB

i.e. it is measured as the change of the portfolio value relative

to its beginning value over a prespecified time span in the
past.

Remark 2.2: We denote by R1 the set of real numbers. By

multiplying PB and PB by the scalar λ ∈ R1 we have

PE − PB λ ⋅ PE − λ ⋅ PB
rP = = .
PB λ ⋅ PB

Thus we conclude that rP is invariant by λ ∈ R1. Using

1
λ=
PB

we conclude that we can assume without loss of generality

PB = 1. Thus rP does not depend on the size of the portfolio.
 17

Remark 2.3: (2) and (9) express the same. However, (1) is
applied much more often than (2) and (9) in prominent in portfolio
analysis as PE and PB are measured in the market place followed
by calculating the return.

Example 2.4: A portfolio moved from 80 to 100. What is its

return in decimals and in %?

100 − 80
r= = 0.25 which is in percentage equal to 25%.
80

In this formula we assume that there is no cash flow between

PB and PE. Furthermore we did not specify how long it takes to
achieve this return. Is this the return over a day, a month or a
year?

Definition 2.4: The investment universe is given by the

securities the portfolio manager is allowed to invest in .

In the following we consider a portfolio of n stocks with Price

PBj, 1 ≤ j ≤ n, PEj, 1 ≤ j ≤ n and number of units Nj. The net
asset value of the portfolio is

n n
PB = ∑ N j ⋅ PB j , PE = ∑ N j ⋅ PE j , resp. (2.2.2)
j =1 j =1

which yields by (1)

 18

n n
∑ N j ⋅ PE j − ∑ N j ⋅ PB j
j =1 i =1
rP = n
=
∑ N j ⋅ PB j
j =1
(2.2.3a)
n n (PE j − PB j )
∑ N j ⋅ (PE j − PB j ) ∑ N j ⋅ PB j ⋅ PB j
j =1 j =1
n
= n
.
∑ N j ⋅ PB j ∑Nj ⋅ PB j
j =1 j =1

By defining for j = 1, 2,….,n the weights

N j ⋅PB j
wj = n
, j = 1, 2,….,n. (2.2.3b)
∑ Ni ⋅PBi
i =1

and with the return rj of an asset j

PE j − PB j
rj = , j = 1, 2,….,n. (2.2.3c)
PB j

By inserting (3b) and (3c) in (3a) we find that the (absolute)

return rP of a portfolio P is the weighted average of the return of
the return:
n
rP = ∑ w jr j = w1 r1 + w2 r2 + ….. + wn rn. (2.2.4)
j =1

The terms wj rj, 1 ≤ j ≤ n are called the return contribution. They

are the contribution of the return of the asset j to the return of the
portfolio whereas the return rj allows the comparison of the returns
 19

between the assets j. They are nonweighted or unweighted

returns.

Remark 2.4: With (3b) we introduce the weights at the

beginning of the period. From (3b) there follows

n n N j ⋅ PB j
∑wj = ∑ n
= 1. (2.3.5a)
j =1 j =1
∑ Ni ⋅PBi
i =1

In portfolio theory this condition is called the budget

constraint. If there is no short selling we have the nonnegativity
constraint

wj ≥ 0, j = 1, 2,….,n. (2.3.5b)

Form (5) it follows

0 ≤ wj ≤ 1, j = 1, 2,….,n.

Example 2.5: We consider the following price movement of 3

stocks A, B and C:

Stock Beginning End Return

A 120 160 33.3%
B 100 120 20.0%
C 30 50 66.6%

Table 2.1

The returns of the stocks follow from a beginning value and an

ending value. Further we assume the weights
 20

w1 = 15%, w2 = 25%, w3 = 60%.

The return contribution shows

Absolute
Stock Weights Return
Contribution
A 15% 33.3% 5%
B 25% 20.0% 5%
C 60% 66.6% 40%
Portfolio return 50%

Table 2.2

We see that the elements or the basis of the return analysis

are the returns of the individual securities in the portfolio. In an
absolute return contribution analysis the return is decomposed
according to a breakdown of the investment universe. Often the
return is broken down into the return of the different countries or
the different industries in the investment universe.

Definition 2.5: A segment is a set of investments of the

investment universe.

We illustrate (4) by considering a breakdown of the universe

into two segments. One part consists of the first m securities with
returns r1,….,rm and in the second part we have the n − m
securities with returns rm+1,….,rn. With the abbreviations

m n
W1 = ∑ w j , W2 = ∑ w j ,
j =1 j = m +1
 21

w1r1 w 2r2 w mrm

R1 = m
+ m
+ ...... + m
,
∑wj ∑wj ∑wj
j =1 j =1 j =1
w m+1rm+1 w m+ 2rm+ 2 w nrn
R2 = n
+ n
+ ........ + n
∑wj ∑wj ∑wj
j =m +1 j = m +1 j = m +1

(4) is the same as

rP = W1 ⋅ R1 + W2 ⋅ R 2 . (2.2.6)

It is seen that (6) is the special case n = 2 in (3). R1 and R2 are

the return of the 2 segments. If, for instance, R1 > R2 the return of
segment 1 is higher than the return of segment 2. The overall
return of the portfolio is then the weighted sum of the returns of
the two segments. We summarize as follows

The return is linear, i.e. the return of a portfolio is equal to the

sum of the weighted returns of its investments.

In return contribution it is assumed that we only measure

the return of a single portfolio manager or in other words it is
a one-dimensional breakdown of the return.
It is given for instance by the theme of a fund or by the
contract between the client and an asset management firm. If the
portfolio manager’s goal is to maximize the absolute return, the
return analysis has to account for this. Here we have an important
principle in return analysis or more general in performance
measurement. The analysis of the performance has to account for
the investment strategy of the portfolio manager. Given appropriate
risk consideration in an absolute return strategy the portfolio
manager tries to maximizes the absolute return of his or her
 22

portfolio. We see that the elements of the return analysis are the
returns of the individual securities in the portfolio. In an absolute
return contribution analysis the return is decomposed according to
a breakdown of the investment universe in segments. Often the
return is broken down into the return of the different countries or
the different industries in the Investment Universe. For evaluation
the return of an absolute contribution we need:

Definition 2.6: A risk-less asset is defined as an investment

that has no capital loss over a predetermined period and the risk-
free return rf is the return that we can earn on such an
investment.

Typical examples of risk-less assets are Teasury-Bills or

Certificates of deposits. They are deemed to be risk-less as the
issuers of these money market instruments are guaranteed to
repay the money the investor has invested and to pay the interest
rate due.

Definition 2.7: The excess return is the difference

between the return of a portfolio rP and the risk-free return r f

rP − rf.

If we consider a relative return analysis the return is

measured against a reference portfolio.

Definition 2.8: A reference portfolio is called a

benchmark portfolio or shortly a benchmark .

In the finance industry there are two kinds of Benchmarkes.

For equity or fixed income portfolios most investors consider
industry standards benchmarks. In the equity world the most
important index providers are MSCI and FTSE and in the fixed
 23

income area names like J.P. Morgan, Citigroup, Barclays and

Merrill Lynch are market leaders in producing indices. For
balanced portfolios, i.e. Portfolios that consist of different asset
classes, investors mostly use tailor-made benchmarks. In most
cases the benchmark is a mix of equity, bond and money market
indices. The proportion of the different asset classes is subject to
discussions between the client and the portfolio manager. It is
important to realize that the performance of the portfolio manager
is measured against the Benchmark. Thus a relative return
contribution has to be considered. Relative return measures the
contributions of the securities in the portfolio relative to the
contributions of securities in the Benchmark.
We proceed by adopting the following notation for a
Benchmark portfolio. We consider n investments with Beginning
Price BBi, 1 ≤ i ≤ n, and end prices BEi, 1 ≤ i ≤ n with Mi units.
The asset value of the portfolio is

n n
BB = ∑ Mi ⋅ BBi , BE = ∑ Mi ⋅ BEi , resp.. (2.2.7a)
i =1 i =1

Similar to (3b), with

M j ⋅ BB j
bj = n
,1 ≤ j ≤ n (2.2.7b)
∑ Mi ⋅ BBi
i =1

we have the weights of the benchmark securities in

benchmark. The return of the Benchmark (Portfolio) is

n
rB = ∑ b j rj . (2.2.7c)
j =1
 24

Definition 2.9: With (4) and (7) the arithmetical

relative return ARR is the difference

ARR = rP − rB .

With (7) we have for the relative return

n
ARR = ∑ ( w j − b j ) r j . (2.2.8)
j =1

This decomposition is called Brinson-Hood-Beebower (BHB).

From

n
0 = ∑ (w j − b j )
j =1

we have

n
0 = ∑ (w j − b j ) ( − rB )
j =1

and by adding (8) there follows

n
ARR = ∑ ( w j − b j ) (rj − rB ). (2.2.9)
j =1

This decomposition is called Brinson-Fachler (BF). From (7) and

(9) it is seen that return analysis is not a unique process and in
fact in many practical situations (9) is used instead of (7). In the
decomposition if term of the sum in (7) is positive the portfolio
manager contributed positively to the relative return otherwise
 25

negatively. Thus (9) allows a meaningful discussion of the relative

return. Summarizing, in return contribution we analyze the portfolio
return in view of different investment universes. It is a one
dimensional process.
If the portfolio manager is allowed to invest in assets outside
the benchmark we have bi = 0 for these assets and (4), (8) and
(9) can also be used for the return analysis. However, assets
outside the benchmark contribute with the absolute return to the
relative return of the portfolio. From the Portfolio managers view,
investments outside the benchmark are thus considered to be
risky.

Example 2.6: We consider the same 3 stocks as in Example

2.5 and the universe of the portfolio and the benchmarks are the
same with

w1 = 15%, w2 = 25%, w3 = 60%

in (3a) and

b1 = 25%, b2 = 25%, b3 = 50%

in (7) (see Table 3). Inputs are in bold. Table 2.3 shows two
decompositions of the relative return 1.00% of the portfolio versus
the benchmark. In the last line is the return of the portfolio
−1.50% and the benchmark 2.50%, i.e. we find that the portfolio
outperforms the benchmark by 1.00%. The value added indicates
that stock A and C are underperforming the benchmark and B is
outperforming the benchmark. However, Stock B has a neutral
position and does not contribute to the relative return. As A is an
underweighting position versus the benchmark, the investing
decision is favourable, that is the contribution is positive, namely
2.00% on an absolute basis (BHB) and 1.75% on a relative basis
 26

(BF). As C is an overweighting position versus the benchmark, the

investing decision is unfavourable, that is the contribution is
negative, namely -1.00% on an absolute basis (BHB) and -0.75%
on a relative basis (BF).The following table shows the BHB and
BF decomposition:

Value Under/ Contri- Contri-

Portfolio Benchmark added Over Bution Bution
rj − rB weight BHB BF

Re-
weight weights
turn

A -20% 15% 25% -17.5% -10% 2.00% 1.75%

B 30% 25% 25% 32.5% 0 0.00% 0.00%

C -10% 60% 50% -7.5% 10% -1.00% -0.75%

Re-
-1.50% -2.50% 1.00% 1.00%
turn

Table 2.3

If we consider a capitalization weighted benchmark, the

benchmark weights are the capitalization weights of the securities
in the benchmark portfolio. However, from (7) it is seen that they
can be chosen differently from capitalization weights. It is seen that
in nature the benchmark portfolio is the same as the investment
portfolio and the elements of the return analysis are the securities
in the investment universe.

We proceed by explaining the analysis of the Money

managers’ performance. We attribute the return to the appropriate
decision makers and to the different steps of the investment
 27

process. We analyse different factors of the investment process.

As in the previous section we distinguish between an absolute and
relative attribution:

Definition 2.10: Return Attribution is the decision oriented

decomposition of the return.

Here we only confine to an explanatory illustration for absolute

attribution. The absolute attribution is no used widely. However, as
the following example shows the formulae already get intricate.
Base for the attribution is:

• Segmentation of the investment universe and (or) the

benchmark universe
• Multi level investment process

Example 2.7: We consider the Example 2.6 exposed in Table

2.3 and we assume the Universes of the portfolio and the
benchmark are the same and consists of stock A with weight
w1 = 15%, stock B with weight w2 = 25% and stock C with
weight w3 = 60%,. We introduce a two step investment process
and 3 Portfolio Managers (PM).
• PM1 decides on the Country allocation (W1, W2),
• PM2 decides on the portfolio in Country X (w1, w2),
• PM3 decides on portfolio in Country Y (w3).

We ask what the return of the individual managers is. We have

a two level decision. First the PM1 (country manager) decides on
W1 = w1 + w2 and W2 = w3. Referring to the notation in (7) the
country benchmarks rX, rY are
 28

b1 b2
rX = r1 + r2 , rY = r3 .
b1 + b2 b1 + b 2

Using the data from Example 2.6 we find

b1 b2
= 0.5, = 0.5, rX = 25% and rY = 10%.
b1 + b 2 b1 + b 2

PM1’s return is measured by

⎛ b1 b2 ⎞
rPM1 = (w1 + w 2 ) ⎜⎜ r1 + r2 ⎟⎟ + w 3 r3 = −4%.
⎝ b1 + b2 b1 + b2 ⎠

rPM1 is the return arising from investing in country X and Y. PM1

has taken in Country X overweight position and this was a wrong
decision. It is called the asset allocation effect. The PM2
(Country X) decides on w1 and w2 and his return is

⎛ b1 b2 ⎞
rPM2 = w1 r1 + w 2 r2 − (w1 + w 2 ) ⎜⎜ r1 + r2 ⎟⎟ =
⎝ b1 + b2 b1 + b2 ⎠

= 4.5% − 0.4 . 5% = 2.5%.

rPM1 is the return from deciding on the securities in X. The PM2

has taken the right decision as he has taken an underweight in
Stock A and an overweight in Stock B. It is called the stock
picking effect.
If the PM3 (Country Y) is not allowed to buy stocks outside
the benchmark there is no decision to be taken for PM3. We see
the definition of the universe is crucial in return attribution. If PM3
had more choices it could influence the return. Moreover, it is not
clear that the allocation is an asset allocation effect or a sector
 29

effect (chapter issues in performance). PM1 and PM3 decisions

coincide. In the following section we will see that decision makers
can have common influence on part of the return.

In the return attribution it is assumed that the investment

process consists of different steps and different decision makers.
It is a multi step process. Again absolute and relative return
attribution can be considered. In the following we focus on
relative return attribution that consists of two layers. As in the
return contribution a breakdown of the Investment Universe (e.g.
Countries or Industries) is considered. In most cases Investors are
interested in a country and (or) industry breakdowns. However,
also classifications according to the security capitalization in the
equity market or duration buckets in the fixed income area are
possible.
In the following we distinguish between arithmetic attribution
(addition) and geometric attribution (product).

a. The arithmetic attribution

In the following Wj, 1 ≤ j ≤ n, are the weights of the portfolio

part with return Rj, 1 ≤ j ≤ n, and Vj, 1 ≤ j ≤ n, is the weight
which is in most cases the market capitalization of the
corresponding benchmark segment with return Bj,1 ≤ j ≤ n.
Following Definition 2.9 the arithmetic relative return is

∑ Wj ⋅ R j −∑ Vj ⋅ B j = ∑ (Wj ⋅ R j − Vj ⋅ B j )
n n n
ARR = (2.2.10)
j =1 j =1 i =1

and consider the identity

 30

Wj . Rj − Vj . Bj =
(2.2.11)

(Wj − Vj ) . Bj + (Rj − Bj ) . Vj + (Wj − Vj ) . (Rj − Bj )

(see Figure 2.1). This identity allows the discussion of the relative
return.
We proceed by the first part of the sum in (10) we refer to
Brinson-Hood-Beebhower (see (7)) and introduce by A BHB i in the
following the asset allocation effect A jBHB in segment j based
on

A BHB
j = (Wj − Vj ) . Bj.

By referring to Brinson-Fachler (see (8)) we denote by A BF

i in the
following the asset allocation effect A jBF in segment j based
on

A BF
j = (Wj − Vj ) (Bj − rB).
.

We note that we can A BHB

j consider as the special case when the
benchmark return vanishes in A BF j . The asset allocation effect is
the overweight or underweight of the segment of the breakdown
of the benchmark. However, the weight of the constituents within
the segments stays the same. A BHB j > 0 ( A BHB
j < 0) says
whether the asset manager decisions to overweight or
underweight the segment adds positively (negatively) to the return
of the portfolio. A BF BF
j > 0 ( A j < 0) says whether the asset
manager decisions to overweight or underweight the segment
adds positively (negatively) to the relative return of the portfolio
 31

versus the benchmark. As in (8) on portfolio level A BHB

j and A BF
j
add up to the same:

n n n
Atot = ∑ A BH
j = ∑ ( Wj − Vj )(B j − rB ) = ∑ A BHB
j . (2.2.12a)
j =1 j =1 j =1

It measures the return of the portfolio manager that takes the

asset allocation decisions relative to the considered breakdown of
the investment universe into account. There are two possibilities
for a positive asset allocation effect namely an overweight of that
segment followed by an outperformance or an underweight
followed by an underperformance of the segment relative to the
benchmark. For a negative asset allocation effect an analogous
statement holds. Stock selection Sj effect in segment j
defined by

Sj = (Rj − Bj ) . Vj
.

Ri
portfolio
weights

Wi
benchmark
Vi

return
Bi
Brinson-Hood-Beebower Figure 2.1
 32

measures the effect of the security selection within the

universe. We can measure the overall stock selection in the
portfolio or the stock selection within a specific segment.
In the relative performance attribution there is a third
component of the return called interaction Ij effect in
segment j defined by

Ij = (Wj − Vj ) . (Rj − Bj ).

This interaction effect is a part of the performance that

cannot be uniquely attributed to a particular decision maker.
For instance if the asset allocation effect and the stock select
effect is positive then the overall return is not barely the sum
of the two components but returns cumulate and as a
consequence the return is bigger. It is important to calculate
and to account for the interaction effect but it is a very
subjective decision which decision maker is the owner of the
interaction effect.
The asset allocation effect, the Stock selection effect and
the interaction effect are called the management effect .
With

Wj . Rj − Vj . Bj = A BHB
j + S j + Ij

and

n n
S tot = ∑ S j , Itot = ∑ I j (2.2.12b)
j =1 i =1

we find by (10), (11)

 33

n n n
ARR = ∑ Wj ⋅ R j − ∑ V j ⋅ B j = ∑ ( A j + S j + I j ) =
j =1 j =1 j =1

n n n n n n
∑ A BHB
j + ∑Sj + ∑Ij = ∑ A BF
j + ∑Sj + ∑Ij =
i =1 j =1 j =1 i =1 j =1 j =1

Atot + Stot + Itot.

We note that in general:

Wj . Rj − Vj . Bj ≠ A BF
j + Sj + Ij

and summarize

The BF decomposition does not hold on segment level but on

portfolio level the relative return is equal to the three
management effects. The BHB decomposition holds on
segment level and on portfolio level.

We conclude the discussion on the BF and the BHB

decomposition of the arithmetic relative return. The return
attribution is a reverse problem, i.e. the return is known and
observable and the decomposition of return is not unique. In the
following examples we illustrate different decomposition based on
different requests of a user of a return attribution system.

Example 2.8: We expand the portfolio in Example 2.6 by

stock D. The input in Table 2.4 in bold:
 34

Value Under/ Contri-

Portfolio Benchmark added Over Bution
rj − rB weight BHB

Re-
weight weights
turn

A -20% 15% 25% -19.5% − 10% 2.00% 1.95%

B 30% 25% 25% 30.5% 0 0.00% 0.00%

C -60% 10% 20% -59.5% − 10% 6.00% 5.95%

D 30% 50% 30% 30.5 20% 6.00% 6.10%

Re-
13.5% − 0.5% 14% 14%
turn

Table 2.4

We consider the 2 segments. A segment consists of Stock A and

B and another segment consists of Stock C and D. The following
Table 2.5 shows the absolute return of the benchmark and the
portfolio:

Segment j Wj Rj Vj Bj

1 40% 11.25% 50.00% 5.00%

2 60% 15.00% 50.00% −6.00%

Total 13.50% −0.50%

Table 2.5
 35

Table 2.6 illustrates (10) - (12):

Relative
BHB BF S IA
Return
2.000% −0.500% −0.550% 3.125% −0.625%

12.500% −0.600% −0.550% 10.500% 2.100%

Total 14.000% −1.100% −1.100% 13.625% 1.475%

Table 2.6

This example shows that all investment decisions exempt the in

investment in Stock B are favourable and the portfolio
outperformances the benchmark by 14.00%.

Example 2.9 (two investment processes): We assume

rP − rB = 2.0%

over a month. We decompose the relative return of a balanced or

multi asset class portfolio in two different ways.
The first decomposition assumes a three step investment
process which starts with the benchmark definition, asset
allocation and stock picking. Doing such an analysis assumes also
that the asset allocation as well as the stock picking each is done
by one decision maker. Following (11) the report of an attribution
system might find over a month for the Asset allocation effect with
return 2.0%, the Stock picking effect with return −0.7% and for
the interaction effect the return −0.3%. Then the decomposition is

2.0% = 3.0% − 0.7% − 0.3%.

 36

In the second decomposition of the relative return we consider

an investment process that is a more complex decision making
process. We assume that the investment process consists of six
steps and that we have the following returns: benchmark
definition, the internal benchmark selection with a return 0.3%, an
asset allocation effect with a return 1.0%, fixed income asset
allocation with a return −0.5%, an equity asset allocation with a
return 0.8% and portfolio implementation or stock picking with a
return 0.7%. The interaction has a return −0.3%. Then we have

2.0% = 0.3% + 1.0% − 0.5% + 0.8% + 0.7% − 0.3%.

Looking at the figures, in the first decomposition one would

conclude that the balanced account manager is a poor stock
picker but if the second investment process would be correct, on
the contrary the balanced account manager would be a good stock
picker. Neglecting the real investment process and not reflecting it
in the performance attribution potentially causes wrong
interpretation and can lead to wrong conclusions.

Example 2.10 (two different benchmarks): We consider a

European equity mutual fund which a formal external benchmark is
the MSCI Europe. We assume that over a period the relative return
2.0% of the portfolio versus MSCI Europe is decomposed in the
asset allocation effect with a assumed return 3.0%, a the stock
picking effect with return -0.7% and the interaction effect -0.3%,
i.e.

2.0% = 3.0% − 0.7% − 0.3%.

 37

In order to measure this mutual fund against the index MSCI

Europe and to decompose its excess return is not appropriate if
one wants to measure the quality of the asset manager.
The product management of the mutual fund company
positioned this mutual fund internally as a growth product with the
internal benchmark of a European growth index. To get the right
picture the set up of the performance attribution has to be
changed in a way that the excess return versus the external
benchmark is split in four effects: a) benchmark selection with a
return 1.8%, b) asset allocation effect with a return −0.5% c) a
stock picking effect with a return 1.0% and d) an interaction with a
return −0.3%. We have the following decomposition

2.0% = 1.8% − 0.5% + 1.0% − 0.3%.

Such a return decomposition ensures that the contribution of the

positioning of the product and the contribution of the asset
manager can be isolated and assessed. In our case the signs of
the asset allocation and stock picking effects changed just due the
fact that we changed the relevant benchmark to the European
growth index.

Example 2.11 (two different segmentations):

Performance attribution software is normally quite flexible in setting
up the analysis, so that the performance analysts can choose
between decomposing the return by using different segments such
as countries or sectors. Depending on the chosen segment to
decompose the return the performance attribution may come up
with different management effects. We assume an asset manager
of a European equity account and consider the following
decomposition of the relation return 2.0% into asset allocation
 38

effect, stock selection and interaction by different segmentations

of the benchmark MSCI Europe. With a sector approach

2.0% = 3.0% − 0.7% − 0.3%

and with a country approach

2.0% = −0.5% + 2.7% − 0.2%

we come up with different management effects. Setting up the

performance attribution in a wrong way may lead to a wrong
interpretation and to wrong conclusions. This again shows that
setting up the performance attribution according to the investment
process is essential to get a meaningful analysis and appropriate
feedback into the investment process.

In (2.2.2) is assumed that

o the portfolio is invested in a single currency. An illustration is a
fixed portfolio with bonds that comprises only of one currency
or
o the return arising of different currencies is not of interest.
Equity of internationally active companies does not relate to a
single currency and a currency attribution is not requested.

b. The geometrical attribution

Definition 2.11: The geometrical relative return is

1 + rP
GRR = − 1. (2.2.13)
1 + rB
 39

Remark 2.5 (absolute return of the portfolio): If rB = 0

then

GRR = ARR = rP.

Remark 2.6 (Interpretation of GRR): Based (12) we find

1 + rP r −r
GRR = −1 = P B =
1 + rB 1 + rB

PE − PB BE − BB
−
PB BB .
BE − BB
1+
BB

According to Remark 2.2 we scale PB or BB such that we

can assume PB = BB and it follows:

PE − BE
PB PE − BE PE
GRR = = = − 1.
BE BE BE
PB

We conclude that

o GRR can be expressed by the end value of the portfolio

and the benchmark more specifically the GRR is the rate
of return of the portfolio ending value and the benchmark
ending value.

Example 2.12: We suppose that difference the portfolio and

the benchmark is
 40

ARR = 5%.

Firstly this can be found with rp = 10%, rB = 15%. This yield

GRR = 4.35%.

Secondly assuming rp = 40%, rB = 45% we find

GRR = 3.45%.

The arithmetic relative return is the same for both cases. The user
of the geometrical relative return can argued that with a higher
benchmark return a higher portfolio return is easier to achieve that
a lower portfolio return with a lower benchmark return. The
geometrical relative return is relative to benchmark return.

Remark 2.7: From Example 2.11 we see that GRR measures

the outperformance of the performance relative to the return of the
benchmark.
Segment

Geometric Linking

Geometric Linking Decomposition

Time

T
Figure 2.2
 41

We proceed with the decomposition on segment level.

The asset allocation effect for segment j, j = 1,....,n is

defined by

⎛1 + B j ⎞
Aj = (Wj − Vj) . ⎜⎜ − 1⎟⎟ , (2.2.14a)
⎝ 1 + rB ⎠

the stock selection effect is defined by

⎛1+ R j ⎞ 1+ B j
Sj = Wj ⎜ .
− 1⎟ . (2.2.14b)
⎜ 1 + B j ⎟ 1 + rS
⎝ ⎠

and the notional portfolio with returns rS is defined by

N
rS = ∑ Wj B j (2.2.14c)
j =1

and the geometrical relative return for segment j

1 + GRRj = (1 + Aj) (1+ Sj).

We proceed by
n n ⎛1 + B j ⎞
A tot = ∑ A j = ∑ j j ⎜ 1+ r
( W − V ) ⎜ − 1⎟⎟ =
j =1 j =1 ⎝ B ⎠

n ⎛ rB − B j ⎞
∑ j j ⎜⎜ 1 r ⎟⎟ =
( W − V )
j =1 ⎝ + B ⎠

1 n
∑
1 + rB j =1
( Wj − Vj )(B j − rB ) =
 42

rS − rB 1 − rS
= −1
1 + rB 1 + rB

and
n n ⎛ ⎛ 1+ R j ⎞ 1+ B j ⎞
S tot = ∑ Si ∑ Wj ⋅ ⎜ ⎜
= − 1⎟ ⋅ ⎟=
⎜ ⎜ 1 + B ⎟ 1 + r ⎟
j =1 j=1 ⎝⎝ j ⎠ S ⎠

n ⎛ ⎛ R j − B j ⎞ 1+ B j ⎞
∑ Wj ⋅ ⎜⎜ ⎜⎜ 1 + B ⎟⎟ ⋅ 1 + r ⎟⎟ =
j =1 ⎝⎝ j ⎠ S ⎠

rP − rS 1 + rP
= − 1.
1 + rS 1 + rS

From (13) we have

1 + rP 1 + rP 1 + rS
1 + GRR = = .
1 + rB 1 + rS 1 + rB

We define the asset allocation effect of for the portfolio

1 + rS
Atot = −1
1 + rB

and the selection effect of for the portfolio

1 + rP
Stot = −1
1 + rS

and have

GRR + 1 = (1 + Atot) (1 + Stot)

 43

Note: We do not propose any relationship between GRR and

GRRj, j = 1,....,n.

c. The currency attribution

The return of the portfolio depends of the currency and the

return of the portfolio can be expressed in different currency. In
the following we discuss the relation between the different returns.

Definition 2.12: The base currency relates to the investor’s

accounting currency.

Definition 2.13: The local currency relates to the currency

in which the investment is made.

We start with an investment value PB in the base currency.

We assume an investment in a market or an investment in the
market whose return rL is not measured in the base currency, but
in the local currency.

Definition 2.14: The currency exchange rate that translate

at time t from local currency, base currency, resp. to the base
currency, local currency, resp. is denoted by ec,t, ê c,t , resp.

We have

1
ê t,C = .
e t,C

The return rC from the currency exchange rate is

 44

local currency

eB,c eE,c

PB base currency PE
rC =

Evaluation in different currencies Figure 2.3

eC,E − e C,B eB,C

= −1 (2.2.15a)
eC,B eE,C

1 1
−
êC,E êC,B ê C,B
rc = = − 1. (2.2.15b)
1 ê C,E
ê C,B

The end value of the investment PE is

êC,B (1 + rL )
PE = PB .
êC,E,

By (15b) the total return rT in the base currency is

 45

ê C,B (1 + rL )
PB − PB
PE − PB ê C,E
rT = =
PB PB

ê C,B (1 + rL )
= − 1 = (1 + rL) (1 + rC) − 1,
ê C,E

i.e.

rT = rL + rC + rC rL.

The expression (15) is independent on PB. We see that we have

geometric linking of the market and the currency return.

Example 2.13: We consider a CHR investor that buy a bond

denominated in Danish kroner and Polish Zloty. We want to
calculate the return in CHF. The bonds have the following prices
for at the beginning PB, DKK = 111.110, PB, POL= 111.160 and the
end PE,DKK = 111.946, PE,POL = 111.049. In Table 2.7 and Table
2.8 bold figures. In Table 2.8, the exchange in Polish Zloty and
Danish Kroner are given against US Dollar. With the value of Table
2.7 against the Swiss Franc we find for example

6.06270
5.2470 = .
1.15535

t1 t2
CHF/US 1.15535 1.07835

Table 2.7
 46

US US CHF CHF
t1 t2 t1 t2
DKK 6.06270 6.08235 5.24750 5.64149
POL 3.32075 3.37715 2.87424 3.13177

Table 2.8

1. For the Danish Bond we have by (12b)

5.24750
rDKK = − 1 = −0.06984
5.64149

i.e. the CHF has a negative return because the Danish Crone is
inflating. (13) yields

111.946 5.24750
rT = − 1= −0.07121.
111.110 5.64149

2. For the Polish Bond we have

2.87424
rPOL = − 1 = −0.08223,
3.13177

i.e.

111.049 2.87424
rT = − 1 = −0.08315.
111.160 3.13177
◊

We proceed by considering the currency attribution of a

segment j with a unique currency

rT,j = rC,j + rM,j + rC,j rM,j,1 ≤ j ≤ n

 47

∑ w jrT, j = ∑ (w j (rM, j + rC, j + rC, jrM, j )) =

n n
rT,P =
j =1 j =1

n n n
∑ w jrM, j + ∑ w jrC, j + ∑ w jrM, jrC, j .
j =1 j =1 j =1

If the investments in the base currency are in the segment 1 we

have r1,C = 0, i.e.

n n n
∑ w jrM, j + ∑ w jrC, j + ∑ w jrM, jrC, j .
j =1 j=2 j=2

The total return of the portfolio consists of three parts. The first
and the second part are the sum of the market return and the
currency return of the segment. The third part cannot be assigned
to the currency return and the market return.
 48

2.3. The time weighted rate of return

a. Absolute return measurement

We consider

t0 = 0,....,tN = T (2.3.1)

as time points and corresponding cash flows C1,....,CN−1 on the

time axis and

k−1rP,k (2.3.2a)

denotes the return of the portfolio between tk−1 and tk,

k = 1,....,N (see Figure 2.4).
C1 Ck CN−1

t0 = 0 t1 t2 tN-1 tN = T
The time axis and cash flows Figure 2.4

Based on the beginning portfolio values PBk and the ending

portfolio values PEk+1 we calculate the return k−1rP,k by

PEk − PBk −1
k−1rP,k = , k = 1,...,N. (2.3.2b)
PBk −1

In order to measure the return between time 0 and time k we link

the returns multiplicatively by
 49

0rP,k = (1 + 0rP,1) . (1 + 1rP,2) ...... (1 + k−1rP,k) − 1, k = 1,...,N (2.3.2c)

where 0rP,k is the return of the portfolio P from 0 to k.

Furthermore it is assumed in (2c) that the returns are
compounded at the end of each sub period. As we multiply we
allude to an area as geometrical object and the formula is called
geometrically linked returns.

Definition 2.15: For equidistant knots tk = k, k = 0,..., N = T,

we consider the geometric mean return r̂P

r̂P = N (1 + 0 rP,T ) − 1. (2.3.3a)

It is seen when compounding with r̂p over the time period we get
r̂p by (3a), i.e.

( )
r̂P = N 1 + r̂p − 1. (2.3.3b)

Definition 2.16: For annual data over N years the

annualized return ar̂P is defined by (3b). For monthly date
over N months the annualized return ar̂P

( )
a r̂P = 12N 1 + r̂p .

Remark 2.8: We see that we have here by (3) two different

attributions along the time axis. The first attribution is according to
(2c) and the second is described by (3) where the returns over
0rP,K is uniformly distributed or attributed with r̂p defined by (3b).

We note that the returns in (2) are commutative, i.e. the

returns stay unchanged by considering for example r1 in period 2
 50

and r2 in period 1. The commutative law of the returns and the

independence of the size of the portfolio by geometrically linking
are illustrated by the following example.

Example 2.14: a. We assume N = 2 with t0 = 0, t1 = 1 and

t2 = 2 in (1) and PB0 = 1$, PB1 = 100$, 0rP,1 = 5%, 1rP,2 = 7% in
(3). Then we have with (3b)

PE1 = PB0 (1 + 0rP,1) = 1$ (1 + 0.05) = 1.05$

and

PE2 = PB1 (1 + 0rP,2) = 100$ (1 + 0.07) = 107$.

(2c) yields

0rP,2 = (1 + 0rP,1) . (1 + 1rP,2) − 1 = 0.1235.

There is an inflow C1 = 100 − 0.07 = 98.93 at t1 =1. It is seen

that C1 that nor infer the calculation of 0rP,2.

b. We assume N = 2 with t0 = 0, t1 = 1 and t2 = 2 in (1) and

PB0 = 100$, PB1 = 1$, 0rP,1 = 5%, 1rP,2 = 7%. Then we have with
(3b)

PE1 = PB0 (1 + 0rP,1) = 100$ (1 + 0.05) = 1.05$

and

PE2 = PB1 (1 + 0rP,2) = 1$ (1 + 0.07) = 107$.

(2c) yields

0rP,2 = (1 + 0rP,1) . (1 + 1rP,2) − 1 = 0.1235.

 51

There is an outflow C1 = 1 – 107 = –106 at t1 =1. It is seen that

C1 that nor infer the calculation of 0rP,2.
We see that there is a cash inflow in a. and a cash outflow in
b. at time 1, although the result is the same in both cases. We see
that the cash flows are reinvested by 1rP,2. In a and b we find by
(3a) for the annualized return

r̂P = N (1 + 0 rP,T ) − 1 = 2 (1 + 0 rP,2 ) − 1

= 2 (1+ 0.1235 ) -1 = 0.059953.

c. We assume N = 2 with t0 = 0, t1 = 0.5 and t2 = 1 in (1) and

PB0 = 1$, PB1 = 100$, 0rP,1 = 5%, 1rP,2 = 7% in (3). Then we
have with (3b). Following case a above we find

0rP,1 = (1 + 0rP,1) . (1 + 1rP,2) − 1 = 0.1235.

d. We assume N = 2 with t0 = 0, t1 = 0.5 and t2 = 2 in (1) and

PB0 = 100$, PB1 = 1$, 0rP,1 = 5%, 1rP,2 = 7%. Then we have with
(3b). Following case b above we find

0rP,1 = (1 + 0rP,1) . (1 + 1rP,2) − 1 = 0.1235.

We see that in both cases c and d we have

r̂P = 1 (1 + 0 rP,1) − 1 = 0.1235.

We conclude that the return does not dependent the cash flow but
is dependent from the underlying time span.

◊
 52

The mean defined in the following definition does not respect

compounding. It is used extensively in Chapter 4 on risk.

Definition 2.17: For equidistant knots tk = k, k = 0,..., N = T,

we consider the arithmetic mean return rP of a portfolio P . It is
calculated by the sum of the returns in (2a) divided by the number
N of returns in (2a):
1 N
rP = ∑ k −1rP,k . (2.3.4)
N k =1

By expanding (2.2.2) we denote by Nj,k the units and Pj,k the

prices of asset j at the time tk, 0 = 1,….,N − 1. With

n n
PBk = ∑ N j,k Pj,k , PEk = ∑ N j,k Pj,k +1
j =1 j =1

there follows that (2a) is equivalent to

n
∑ N j,k Pj,k +1
j =1
1 + k−1rP,k = n
.
∑ N j,k Pj,k
j =1
With (2b) we have by chain linking for k = p,....,q, 0 ≤ p < q ≤ T
n
q-1 ∑ N j,k Pj,k +1
j =1
1 + prP,q = ∏ n
p =k
∑ N j,k Pj,k
j =1

and for p = 0 and q = N we find

 53

N-1
∑ N j,k Pj,k +1
j =1
0rP,T = ∏ n
− 1. (2.3.5)
k =0
∑ N j,k Pj,k
j =1

At the end of each period, we divide by the portfolio value and

observe the price movement of the securities in the portfolio over
the next period. In other words we neutralize the portfolio value
and the number N of shares outstanding remains unchanged over
one time period.

Definition 2.18: (5) is called the time weighted rates of

return calculation (TWR).

As potential changes in the money invested in the portfolio

does not affect the return. We simply look at relative changes over
multiple time periods. Time weighted rates of returns are used by
the index providers. Most market indices are calculated on a daily
basis, i.e. the time unit is days. The value of the market portfolio is
calculated at the end of each business day and formula (5) is
evaluated. The returns are published by data providers. It is well
known that indices can be rebased at each point in time. This
again reflects the fact that (5) is independent of the portfolio size,
i.e. if we consider an index with base currency $ and rebase the
portfolio to 100 we observe the development of an initial portfolio
of 100$.
Let us assume

N i,k = λk . Ni,k−1 , λk ∈ R1 for k = 1,….,N − 1 (2.3.6)

and consider (5) over two periods [λk−1, λk] and [λk, λk+1],
0 ≤ p < k < q ≤ T and apply (6) twice
 54

n n
∑ N j,k -1 Pj,k ⋅ ∑ N j,k Pj,k +1
j =1 j =1
(1 + k−1rP,k)(1 + krP,k+1) = n n
=
∑ N j,k -1 Pj,k -1 ⋅ ∑ N j,k Pj,k
j =1 j =1
(2.3.7)

n n
∑ N j,k Pj,k +1 ∑ N j,k -1 Pj,k +1
j =1 j =1
n
= n
= (1 + k−1rP,k+1).
∑ N j,k -1 Pj,k −1 ⋅ λk ∑ N j,k -1 Pj,k −1
j =1 i =1

By applying (6) to the denominator and using (7) repeatedly we

have
n
∑ Ni,T Pi,T
i =1 −1
0 rT = n
∑ Ni,0 Pi,0
i =1

and consequently
PVT − PV0
0 rT = ,
PV0

i.e. the return reduces to the return based on the change of the
portfolio value at time 0 and T. The portfolio values at the time t1 to
tK-1 do not affect the return as intermediate Portfolio value are
dropping.
We summarize:
 55

Theorem 2.1: If (6) holds then the return of a portfolio over

multiple periods is equal to the geometrically linked returns of the
portfolio over the sub periods.

Proof: The assertion follows by applying (7) to the interval

[p, q], 0 ≤ p < q + 1 ≤ T consecutively.

We distinguish two cases:

Case 1:
n n
∑ N j,k −1 Pj,k = ∑ N j,k Pj,k , k = 1,….,N − 1 (2.3.8)
j =1 j =1

This case includes the case λk = 1, k = 1,….,N − 1 in (6) in other

words as the number of securities are held constant the buy and
hold strategy is reflected. However, case 1 leaves open the
possibility of a rebalancing within the portfolio and the portfolio is
restructured (see Figure 2.5). By evaluating (7) for the buy and
hold strategy and the rebalanced portfolio the value added from
restructuring the portfolio can be measured. In both case the
return is only determined by the initial value and the terminal value
of the portfolio.
Furthermore this case is important in understanding the
computation of Total Return Indices. If an index reinvests overnight
or instantaneously it is assumed that a cash flow such as dividends
or coupons stay in the index, however it is distributed between the
securities of the index, i.e. the portfolio value is unchanged, but
the cash flow is distributed such that the ratio of the market
capitalization is unchanged.
 56

Theorem 2.2: If (8) holds then the return of a portfolio over

multiple periods is equal to the geometrically linked returns of the
portfolio over the sub periods.

Proof: The assertion follows by considering the interval [p, q],

0 ≤ p < q + 1≤ T and hence we have by considering (7) over two
periods [λk−1, λk] and [λk, λk+1]
n n
∑ N j,k -1 Pj,k ⋅ ∑ N j,k Pj,k +1
j =1 j =1
(1 + k−1rP,k)(1 + krP,k+1) = n n
=
∑ N j,k -1 Pj,k -1 ⋅ ∑ N j,k Pj,k
j =1 j =1
n
∑ N j,k Pj,k +1
j =1 PEk +1
n
= = (1 + k−1rP,k+1).
PBk −1
∑ N j,k -1 Pj,k -1
j =1

◊
Case 2: There exists a k, k = 1,…., N − 1 with

n n
∑ Ni,k −1 Pi,k ≠ ∑ Ni,k Pi,k . (2.3.9)
i =1 i =1
N i ,k P i, k N i, k P i, k + 1
Ch an ges o f Unit s

N i ,k- 1 P i ,k -1 N i ,k - 1 P i , k

C h a n g e s o f P ric e s

Chain linking at a seam Figure 2.5

 57

This case reflects a change in the portfolio value due to an external

cash flow as we divide in (5) by the portfolio value. However, the
cash flow does not affect the return calculation as the return
depends only the movement between the knots tk, k = 0,…., N.
Taking the argument further chain linking return measurement
does not take into account the timing of external cash flows. This
property underpins the importance of this methodology in the index
industry. The finance industry is only interested in objective market
movements. For the difference of the return from a client’s or a
portfolio manager’s perspective we refer to the Section 2.5.
Furthermore this case is important in understanding the
computation of Total Return Indices. If an index reinvests overnight
or instantaneously it is assumed that a cash flow such as dividends
or coupons stay in the index, however it is distributed between the
securities of the index, i.e. the portfolio value is unchanged, but
the cash flow is distributed such that the ratio of the market
capitalization is unchanged.

Theorem 2.3: If (9) is holds then following is implied: If the

external cash flow in tk
n n
C= ∑ N j,k −1 Pj,k − ∑ N j,k Pj,k
j =1 j =1
is distributed with (6) the return over [λk-1, λk+1] is independent of
the portfolio value in tk otherwise the return over [λk-1, λk+1] is
dependent on the value
n
∑ N j,k −1 Pj,k
j =1
and the allocation of the external cash flow C in Nj,k,1 ≤ j ≤ n in tk.
 58

Proof: The assertion follows from Theorem 2.1 or from

n n
∑ N j,k -1 Pj,k ⋅ ∑ N j,k Pj,k +1
j =1 j =1
(1 + k−1rP,k)(1 + krP,k+1) = n n
.
∑ N j,k -1 Pj,k -1 ⋅ ∑ N j,k Pj,k
j =1 j =1 ◊

b. Relative return measurement

Let tp, tk and tq, p < k < q denote times on the time axis
(Figure 2.4). The following formulae are valid on a portfolio level.
Based on (2c) we consider the geometrically linked return between
the time from tp to tk and from tk to tq of the portfolio.

prP,q = (1 + prP,k)(1 + krP,q) − 1 = prP,k + krP,q +prP,k . krP,q (2.3.10a)

and the benchmark

prB,q = (1 + prB,k)(1 + krB,q) − 1 = prB,k + krB,q +prB,k . krB,q. (2.3.11b)

Definition 2.19: The term p r•,k ⋅ k r•,q with • either a portfolio

or a benchmark is called cross term.

Definition 2.20: The arithmetical relative returns pARRq,

pARRk, kARRq, p < k< q, p = 0,....,N, k = 0,....,N, q = 0,....,N is
defined by

pARRq = prP,q − prB,q,

pARRk = prP,k − prB,k,

kARRq = krP,q − krB,q.

 59

For pARRq and p < k < q we find

pARRq = prP,k + krP,q + prP,k . krP,q − prB,k − krB,q −prB,k . krB,q =

prP,k − prB,k + krP,q − krB,q + prP,k . krP,q −prB,k . krB,q = (2.3.12)

pARRk − kARRq − prP,k . krP,q − prB,k . krB,q.

We proceed by introducing two different decomposition of the

relative arithmetical return for two periods. We refer to [1,
Bonafede] and [7, Feibel].
Based on (12) the first decomposition of the cross terms is

prP,k krP,q
.
− prB,k . krB,q =

prP,k krP,q
.
− prP,k . krB,q + prP,k. krB,q − prB,k . krB,q =

prP,k (krP,q − krB,q) + krB,q (prP,k − prB,k),

thus by (12) again we have

pARRq = (1 + krB,q) . (prP,k − prB,k) + (1 + prP,k) . (krP,q − krB,q),

i.e.

pARRq = (1 + krB,q) . pARRk + (1 + prP,k) . kARRq. (2.3.13a)

We see in (13a) that in addition to the sum of the arithmetical

relative return the relative return in the first period is compounded
with the total benchmark return in the second period and the
relative return of the second period in compounded with the total
portfolio return in the first period. We refer to [9, Feibel, page
278].
 60

Based on (12) the second decomposition of the cross

terms is

prP,k krP,q
.
− prB,k . krB,q =

prP,k krP,q
.
− prB,k . krP,q + prB,k . krP,q − prB,k . krB,q =

krP,q (prP,k − prB,k) + prB,k (krP,q − krB,q),

thus by (12)

pARRq = (1 + krP,q) . (prP,k − prB,k) + ( 1 + prB,k) . (krP,q − krB,q),

i.e.

pARRq = (1 + krP,q) . pARRk + (1 + prB,k) . kARRq. (2.3.13b)

We see in (13b) that in addition to the sum of the arithmetical

relative return the relative return in the first period is compounded
with the total portfolio return in the second period and the relative
return of the second period in compounded with the total
benchmark return in the first period.
We proceed by an example that illustrates (13):

Example 2.15: We consider two periods with t0 = 0, t1 =1

and t2 = 2 in (1) with returns 0 rP,1 = 2%, 0 rB,1 = 10%,
1rP,2 = 10%, and 1rB,2 = 2%. We find

0ARR2 = 0.

(13a) yields the decomposition

0 = − (1 + 0.1) . 8% + (1 + 0.1) . 8%

and (13b) yields the decomposition

 61

0 = − (1 + 0.02) . 8% + (1 + 0.02) . 8%.

This example shows two decompositions of the return over time.

The attribution of the return over time is not unique and depends
of reinvestment assumptions.

pARRq is in general critically dependent on the compounding

knots between p and q (see Theorem 2.3). This property is
encountered by the absolute return (see (2.1.6) and (2.1.7)) and
the relative return. As exposed so far the value for pARRq is not
necessary unique.
In the following we assume that compounding pattern is given
and we respect all compounding times tk, k = 0,....,N. We use the
iterative character by (13) by q = k + 1. In addition we respect all
compounding time tk.

pARRk+1 = (1 + krB,k+1) . pARRk + (1 + prP,k) . kARRk+1 (2.3.14a)

resp.

pARRk+1 = (1 + krP,k+1) . pARRk + (1 + prB,k) . kARRk+1. (2.3.14b)

We continue by considering the segmentation of the

portfolio over time (Figure 2.5). The weights change in time but
are fixed at each time knots. By adapting the notation from
(2.2.10) for the segment level we consider k = 0,....,N − 1

n n
k rP,k +1 = ∑ Wj,k R j,k , k rB,k +1 = ∑ Vj,k B j,k . (2.3.15)
j= 0 j=0
 62

Definition 2.21: The arithmetical relative return kARRj,k+1,

k = 0,....,N − 1 in segment j is defined by

k ARR j,k +1 = W j,k Rj,k − V j,k Bj,k. (2.3.16)

Based on the first decomposition (14a) we consider the

iteration for p ARR1j,k with respect to k

1
p ARR j,k +1 =
(2.3.17a)
. 1 .
(1 + krB,k+1) p ARR j,k + (1 + prP,k) kARR j,k+1

with p = 0, 1,....,T − 2, k = p + 1, p + 2 ....., T − 1 and kARRj,k+1

is given by (15). The initial values are

1
p ARR j,p +1 = pARRj,p+1 = Wj,p . Rj,p − Vj,p . Bj,p
(2.3.17b)
p+1ARRj,p+2 = Wj,p+1 . Rj, p+1 − Vj, p+1 . Bj, p+1.

Reinvestment assumption: The return on p to k is

compounded with the relative return of segment j in the second
period form k to k + 1 and the segment return in the second
period is compounded with the relative return of the first period.

Based on the second decomposition (14b) we consider the

iteration for p ARR 2j,k with respect to k

2
p ARR j,k +1 =
(2.3.18a)
(1 + krP,k+1) . p ARR2j,k + (1 + prB,k) . kARR j,k+1
 63

with p = 0, 1,....,T − 2, k = p + 1, p + 2, ....., T − 1 and

kARRj,k+1is given by (15). The initial values are

2
p ARR j,p +1 = pARRj,p+1 = Wj,p . Rj,p − Vj,p . Bj,p
(2.3.18b)
. .
p+1ARRj,p+2 = Wj,p+1 Rj, p+1 − Vj, p+1 Bj, p+1.

Reinvestment assumption: The segment return from k to

k + 1 is compounded with the relative return in the second period
form p to k and the segment relative return in the second period is
compounded with the total return of the benchmark in the first
period.

The iterations (17) and (18) introduce p ARRij,k +1, k > p,

i = 1, 2. They calculates the arithmetic relative return if a new data
points kARRj,k+1 is available. The following Lemma ensures that
they are a decomposition of pARRk+1.

Lemma 2.1: We claim that

n
∑ p ARRij,k +1 = pARRk+1, i = 1,2 (2.3.19)
j =1

where is calculated by (17), (18), respectively.

Proof: We start by the right side from (19) and combining

with (17a), (18a) we have

n
∑ p ARR1j,k +1 =
j =1
n n
1
(1 + krB,k+1) .
∑ p ARR j,k + (1 + prP,k) .
∑ k ARR j,k +1 ,
j =1 j =1
 64

n
∑ p ARR2j,k +1 =
j =1
n n
2
(1 + krB,k+1) .
∑ p ARR j,k + (1 + prP,k) .
∑ k ARR j,k +1 , resp.
j =1 j =1

We adopt a proof by induction and start with k = p + 1. The

assertion follows from (13) and (14). For general k the assertion
follows from the induction assumption, (13) and (14).

Example 2.16 (continued): We consider again two periods

with t0 = 0, t1 =1; t2 = 2 with returns 0 rP,1 = 2%, 0 rB,1 = 10%,
0 rP,1 = 10% and 1rB,2 = 2% and assume in addition the portfolio
and the benchmark consists of 2 segments with the following
data:

W1,0 = W1,1 = W2,1 = W2,2 = V1,0 = V1,0 = V1,0 = V1,0 = 0.5;

R1,0 = R1,0 = R1,0 = R1,0 = 2%;

B1,0 = B1,0 = B1,0 = B1,0 = 10%.

(13a) yields the decomposition

1
0 ARR j,2 = 0, j = 1, 2,

i.e.
0 = (1 + 0.1) . 0.5 (2% − 10%) + (1 + 0.1) . 0.5 (10% − 2%)

and (13b) yields the decomposition

 65

2
0 ARR j,2 = 0, j = 1, 2,
i.e.

0 = − (1 + 0.02) . 8% + (1 + 0.02) . 8%.

This example shows two decompositions of the return over time.

We proceed by investing the management effects over

multiple periods by adopting the approach of Brinson-Hood-
Beehower. We have a set of management effect in each segment
and at each time point.

Definition 2.22: The asset allocation effect kAj,k+1,

k = 0,....,N − 1 in segment j is defined by
Segment

Management effect

Time

T
The full portfolio return problem (management effect) Figure 2.6

kAj,k+1 = (W j,k − V j,k ) . Bj,k. (2.3.20a)

The stock selection effect kSj,k+1, k = 0,....,N − 1 in segment j

is defined by
 66

kSj,k+1 = (Rj,k − Bj,k ) . V j,k (2.3.20b)

The interaction effect kIj,k+1, k = 0,....,N − 1 in segment j is

defined by

kIj,k+1 = (W j,k − V j,k ) . (Rj,k − Bj,k). (2.3.20c)

Based on the decomposition in (27) and (31) we consider the

iteration for the management effect

1
p A j,k +1 = (1 + krB,k+1) . p A1j,k + (1 + prP,k) . kA j,k+1 (2.3.21a)

1
p S j,k +1 = (1 + krB,k+1) . p S1j,k + (1 + prP,k) . kS j,k+1 (2.3.21b)

1
p I j,k +1 = (1 + krB,k+1) . p I1j,k + (1 + prP,k) . kI j,k+1 (2.3.21c)

2
p A j,k +1 = (1 + krB,k+1) . p A 2j,k + (1 + prP,k) . kA j,k+1 (2.3.21d)

2
p S j,k +1 = (1 + krB,k+1) . p S2j,k + (1 + prP,k) . kS j,k+1 (2.3.21e)

2
p I j,k +1 = (1 + krB,k+1) . p I2j,k + (1 + prP,k) . kI j,k+1 (2.3.21f)

with p = 0, 1,....,T − 2, k = p + 1, p + 2 and the initial values are

given by (20).

Lemma 2.2: Based on (21) we claim that

n n n
i i
pARR k+1 = ∑ p A j,k +1 + ∑ p S j,k +1 + ∑ p Iij,k +1 , i = 1,2.
j =1 j =1 j =1
 67

Proof: We adopt a proof by induction and start k = p + 1. We

consider for i = 1,2
n n n
i i
pARRp+2 = ∑ p A j,p + 2 +∑ p S j,p + 2 + ∑ p Iij,p + 2 =
j =1 j =1 j =1

n
∑ ( p A ij,p + 2 +p Sij,p + 2 +p Iij,p + 2 ) .
j =1

The assertion follows from (13), (20) and (21). For general k the
assertion follows from induction assumption (13), (20) and (21).

Definition 2.23: For the geometrical relative return pGRRq

pGRRk, kGRRq, p < k< q, p = 0,....,N, k = 0,....,N, q = 0,....,N we
define
1 + p rP,q
pGRRq = − 1,
1 + p rB,q

1 + p rP,k
pGRRk = −1,
1 + p rP,k

1 + k rP,q
kGRRq = −1.
1 + k rB,q

As a consequence of the definition we have

1 + p rP,q (1 + p rP,k )(1 + k rP,q )

1 + pGRRq = = =
1 + p rB,q (1 + p rB,k )(1 + k rB,q )
 68

(1 + pGRRk)(1 + kGRRq).

We see that the geometrical relative return has no cross term.

Example 2.17: We consider two periods with returns

0 rP,1 = 2%, 1rP,2 = 10%, 0 rB,1 = 10% and 1rB,2 = 4%. For the
geometrical linked return we have for the portfolio

1.02 * 1.10 − 1 = 1.122 − 1 = 0.22 or 22%

and for the benchmark

1.10 * 1.04 − 1 = 1.144 − 1 = 0.144 or 14.4%.

The geometrical relative return is

1.122
− 1 = −0.01923 or –1.923%.
1.144

◊
 69

2.4. The money weighted rate of return

In this section we concentrate on methods for calculating rates

of return that take cash flows into account. We consider

N −1 Ck PVT
PV0 = ∑ tk
+ (2.4.1a)
k =1(1 + r ) (1 + r ) T

where PV0, PVT, respectively is as in the previous section initial,

terminal value of the portfolio, respectively and Ck are the cash
flows at time tk, k = 0, 1, 2,...., N with t0 = 0, tN = T. The
solutions of equation (1a) are called internal rates of returns
(IRR). We denote the solutions of (1a) with IR.

The fundamental properties are:

• (1a) is based on the arbitrage condition that the value today is
equal to the discounted cash flow in the future.
• The investment assumption is that the cash flows are
reinvested by the internal return.

(1a) has in general not an explicit solution, however, there are

several numerical methods like for instance the Secant, Newton-
Raphson or modified Newton-Raphson methods for computing a
solution of the equation for r.
Starting with an initial approximation for the value of IR, more
accurate approximations for IR are then computed. In [22,
Shestopaloff] different methods for computing IR are compared. In
numerical analysis it is well known for instance that the Secant
method has the advantage that we do not need any derivatives
although the convergence speed is slower than by applying the
Newton-Raphson method.
 70

Remark 2.5 (Invariance of the solution IR): The

1 1
multiplication by or of (1a) leaves the solution of (1a)
PVT PV0
invariant. Thus without loss of generalization we can assume

N −1 Ck 1
PV0 + ∑ t
= , 0 < t1 <....< tN = T (2.4.1b)
k =1 (1 + r ) k (1 + r ) T

or

N −1 Ck PVT
1+ ∑ tk
= , 0 < t1 <....< tN = T. (2.4.1c)
k =1(1 + r ) (1 + r ) T

More generally the solutions of (1) is invariant under multiplication

by λ ∈ R1 (see Remark 2.3).

Remark 2.6 (Normalization of the solution IR to [0,1]):

~ t
By considering the transformation tk = k , k = 0, 1,...., N in (1a)
T
we have

N −1 Ck PVT ~ ~
PV0 + ∑ ~
t
=
1+ r
, 0 < t1 <....< tN = 1. (2.4.1d)
k =1(1 + r ) k

1
With d = in (1a) and with the solution IR1 of (1d) we
1 + IR
1
consider d1 = . Then we find with
1 + IR1

d 1 = dT
that d1 satisfy (1d)
 71

t
N −1 Ck N −1 ~ N −1 − k
− tk
PV0 + ∑ ~ = PV0 + ∑ C k d1 = PV0 + ∑ C k d1 T =
k =1 (1 + r ) tk k =1 k =1

N −1
PV0 + ∑ C k d − tk = dT = d1.
k =1

Furthermore we have

IR1 = (1 + IR)-T − 1. (2.4.1e)

The basic idea of equation (1a) is to discount all cash flows at

a point in time. In (1a) we discount to t = 0. It is seen that the IR is
independent on the time and does only depend on the cash flows.

Definition 2.24: The notion money weighted rate of

return (MWR) refers to a method for assessing the return
reflecting the timing and size of cash flows.

Thus IRR method is a type of a MWR. It is based on a time

arbitrage condition. Furthermore if there are no cash flows, the
solution of (1a) is equal to 0r1 in (2.3.5), i.e. the time weighted rate
of return is equal to the money weighted rate of return. The
difference of TWR and MWR is called the timing effect as the
timing of the cash flows explains the difference between the two
concepts (see Section 2.5b).
As solving (1a) is mathematically challenging we concentrate
on the analytics of (1a). It is seen that (1a) is a nonlinear equation
for the unknown r. By considering the discount factor (1.1.3) we
consider the function P defined

N −1
P(d) = PVT dT − PV0 − ∑ Ck ⋅ dk . (2.4.2)
k =1
 72

We proceed by illustrating (2)

Example 2.18 (Different signs of the cash flow): We

consider one cash flow and equidistant knots, i.e.

t0 = 0, t1 = 1, t2 = 2.

From (1a) we have

PV0 = C1 d + PVT d2.

Hence

0 = PVT d2 + C1 d − PV0. (2.4.3a)

The solutions d1/2 are

- C1 ± (C1)2 + 4PV0PVT
d1/ 2 = , (2.4.3b)
2PVT
i.e.

1
IR1/ 2 = − 1. (2.4.3c)
d1/ 2

We note that in this special case the equation (2a) can be solved
explicitly and has two solutions. We consider

a) PV0 = 2$, C1 = 4/3$, PV2 = 1$ in (1b)

or

PV0 = 1$, C1 = 2/3$, PV2 = 0.5$ in (1c).

 73

The case represents a bond, i.e. there is an investment at the

beginning and two out flows.

PV 0

C1 PV 2 t[years]
0 1 2

Cash flow positive Figure 2.6

Then (3b) yields

d1 = 0.8968 and d2 = −2.2301

and by (3c) the solutions for the internal rate return are

IR1 = 0.11506, IR2 = −1.4484.

Figure 2.7 depicts the function (2) with the choices of the
parameter considered here
3
2
1
0
-3 -2 -1 -1 0 1 2 PVT=1
-2 PVO=1
-3
-4
-5
-6

Minimum negative-cash negative (Bond) Figure 2.7

 74

b) PV0 = 2$, C1 = −4/3$, PV2 = 1$ in (1b)

or

PV0 = 1$, C1 = −2/3$, PV2 = 0.5$ in (1c).

PV 0
C1

PV2 t[years]
0 1 2

Cash flow negative Figure 2.8

Then (3b) yields

d1 = 2.230139 and d2 = −0.89681

and by (3c) the solutions for the internal rate return are

IR1 = −0.55160, IR2 = −2.11507.

In a) (b)) the minimum is positive (negative). In both case the y axis

is negative and there is a unique positive solution. We see that the
minimum of the Figure 2.8 to 2.9 change from negative to
positive. This example can be solved explicitly because the knots
are equidistant and we have only a cash flow.
 75

0
-2 -1 -1 0 1 2 3
POT=1.0
-2 P0T=0.5

-3

-4

-5

-6

Minimum positive - Cash positive Figure 2.9

In the following we assume T = 1, 2,.... and t1 = 1,...,

t2 = 2,..., tN−1 = N − 1. Then (1) is a polynomial. Follow the
fundamental theorem of algebra the equation P(d) = 0 has in
general T zeros or solutions that can be complex. We denote by C
the set of complex numbers and i is the unit of the imaginary
numbers.
We consider the special case where there are no cash flows
during the considered time span from 0 to T. For Ck = 0,
k = 1,...., N − 1, the condition P(d) = 0 is then equal to

PVT
dT = ,
PV0

Hence

PV0
dT =1
PVT
 76

and consequently
PVT
d1,....,T = T exp(2πi/k) ∈ C, k = 1,....,T,
PV0
PVT
where T is the positive root. By (1.1.3) we find for the
PV0
internal rate of rate

1
IR1,....,T = − 1 ∈ C.
dT

In T is odd we have a real positive zero d and T − 1 complex zeros

d whereas if T is even we have a positive and a negative real zero
d. In addition we have T − 2 complex zeros d. If there is no cash
flow MWR is equal TWR. TWR is not a special case of MWR when
the cash flow is zero. As illustrated in Section 2.5b TWR measure
the return in view of the portfolio (see also 2.13). MWR measure
the return in view of the client.

Example 2.19: We assume T = 2 and no cash flow at T = 1

and

PV0 = 99$, PV2 = 120$.

in (1). Then
120
d2 =
99
with

120 120
d1 = = 1.1010, d2 = − = −1.1010.
99 99
 77

We have
IR1 = 0.1010, IR2 = −2.1010.

From a business aspect IR1 = 0.1010 has to be considered as

IRR.

Negative real zeros can also be found in cases with 2 periods

and one cash flow by solving the quadratic equation explicitly. To
sum it up it is seen that the solution of (1) is not unique.
As we are assuming that the interest rates are positive it is
seen from Figure 2.10 that we are interested in varying r from −1
to infinity and as a consequence d is supposed to be in the interval
[ − ∞ ,−1] and [1, ∞ ]. In Figure 2.10 we illustrate the discount
factor z and z3.
In the following we assume T in R1. As interest rates are small
real numbers we develop the discount factors in a neighborhood of
0. More precisely, in the theory of complex analytic the discounts
factors are called meromorphic functions. They have a singularity
in r = −1. Furthermore for r < 1 the Taylor expansion

1 a a a
t
= 1 + 1 r + 2 r 2 + .... + k r k + ..... , t ∈ R1 (2.4.4)
(1 + r) 1! 2! k!
with
k
ak = ∏ (− 1)k ( t + j − 1)
j=1
holds. For t = 1 we find¨

1
= 1 − r + r 2 + .... + ( −1)k ⋅ r k + .....
1+ r
 78

1
d values

0
-5 -3 -1 1 3 5
-1

-2

-3

-4
r values

Discount factors Figure 2.10

and for k = 1 we have with (4)

a1 = − t .

With T = 1 and 0 < tk < 1, k=1,....,N − 1 in (1) we find by

1
multiplying with
1+ r

N −1 Ck
(1 + r) ⋅ PV0 = ∑ t −1
+ PV1 .
k =1 (1 + r) k

We consider the first order approximation and have by (4) with

t = tk − 1
 79

N −1
(1 + r) ⋅ PV0 = ∑ Ck (1 − r(t k − 1)) + PV1 .
k =1

Such an approximation can be good or bad in nature. The

accuracy of the approximation depends on the magnitude of the
increments r and Ck. The smaller the interest rates, the time period
and the cash flows are the better the series converges or in other
words the better the first order approximation is. In addition, the
approximation is better the more terms in the Taylor Series (4) of
(1) are reflected. We solve

N −1
PV1 − ∑ Ck − PV0
r= k =1 . (2.4.5a)
N−1
PV0 + ∑ Ck (1 − t k )
k =1

r is not IR because our derivation contains the approximation by

the Taylor Series (4). By replacing Ck by −Ck, k = 1,....,N − 1 we
have:

Definition 2.25: The Modified Dietz return MD is defined

by

N −1
PV1 − ∑ Ck − PV0
MD = k =1 . (2.4.5b)
N −1
PV0 + ∑ Ck (1 − t k )
k =1

In the denominator we have the sum of the value of the initial

portfolio and the time weighted cash flows. In the following
Original Dietz return OD the cash flow are centered in the
middle of the interval
 80

Definition 2.26: The Original Dietz return OD is defined

by:
N −1
PV1 − ∑ Ck − PV0
OD = k =1 . (2.4.6)
N −1
PV0 + ∑ 0.5 ⋅ Ck
k =1

Example 2.20 (Compare Example 2.14): a. We assume

N = 2 with t0 = 0, t1 = 0.5 and t2 = 1 in (1) and PV0 = 1$,
PV1 = 107$, C1 = 98.95$. Then we have with (3b)

PV1 − C1 − PV0 107 − 98.95 − 1

MD = OD = = = 0.1397.
PV0 + 0.5 ⋅ C1 1 + 0.5 ⋅ 98.95

b. We assume N = 2 with t0 = 0, t1 = 0.5 and t2 = 1 in (1) and

PV0 = 100$, PV1 = 1.07$, C1 = –104. Then we have with (3b)

PV1 − C1 − PV0 1.07 + 104 − 100

MD = OD = = = 0.1056.
PV0 + 0.5 ⋅ C1 100 − 0.5 ⋅ 104

◊
Definition 2.27: The Profit and Loss PL is defined by

N−1
PL = PV1 − ∑ Ck − PV0
k =1

and the average investment capital AIC of a portfolio is defined

as follows
 81

⎡PL
⎢ IR , IR ≠ 0
AIC IR =⎢ N −1
, (2.4.7a)
⎢PV + ∑ C (1 − t ), IR = 0
0 k k
⎣⎢ k =1

N−1
AICMD = PV0 + ∑ Ck (1 − t k ) , (2.4.7b)
k =1

N−1
AIC OD
= PV0 + ∑ 0.5 ⋅ Ck . (2.4.7c)
k =1

Lemma 2.1 (AIC is continuous): With (7a) we have

N−1
IR
lim AIC
IR → 0
= PV0 + ∑ Ck (1 − t k ) .
k =1

Proof: We consider an Portfolio with beginning Value PV0,

ending value PVT and cash flows Ck ≠ 0, k = 1,...., N − 1, N ≥ 2
over the time unit interval. The IR is the solution of

N-1
PV0 (1+ IR) + ∑ Ck (1 + IR )1- t k = PVT.
k =1

We subtract the total cash flow

N-1 N-1 N-1

1- t k
PV0 (1+ IRi) + ∑ C k (1 + IR ) − ∑ Ck = PVT − ∑Ck .
k =1 k =1 k =1
By
N-1
PL = PVT − PV0 − ∑ Ck
k =1
 82

we have

N-1 N-1
1- t k
.
PV0 IR + ∑ C k (1 + IR ) − ∑ Ck = PL
k =1 k =1

and the average invested capital we find

∑ Ck ((1 + IR )1- t k )
N-1
−1
IR k =1
lim AIC = PV0 + lim .
IR → 0 IR → 0 IR

The above relationship shows that the average invested capital

is equal to the initial value corrected by difference of the cash
flows and the discounted cash flow to the invested start point
divided by the internal rate of return. In the absence of cash flow
the average invested capital is equal to the beginning value of the
portfolio. For IR → 0 we have by the rules of hospital

N-1
MD
AIC = PV0 + ∑ Ck (1 − t k ) .
k =1
◊

If we neglect second and higher terms in the Taylor series (4)

we get for the difference of the discount factor and the first order
approximation t

1
− (1 − t ⋅ r).
(1 + r) t
 83

This difference is called the second order effect. In Figure

2.11 it is illustrated that the error term grows with the power t
reflecting the time and the interest rate r.
Again considering equation (1d) in Figure 2.12 we evaluate
(1) between r = 0 and r = 1.6 for a bond with 4 coupon 0.2 and a
last coupon including the face value 1.2. Then for r = 0 the sum of
the Face Value and the Coupon amounts to 2.0. It is seen if a
Price PV0 lies between 0 and 2, a numerical procedure will
compute the corresponding internal rate of return. Also for r = 0.2
the price of the bond PV0 is equal to 1. Furthermore it is seen that
the discount factor are monotonic decreasing functions and so is
the sum of discount factors and as a consequence there is a
unique solution r for a given price between 0 and 2. Cases where
the cash flows are not positive throughout are subject to further
research.

Example 2.21 (Sign change of interest, Interval [0,2]):

We consider one cash flow and equidistant knots, i.e.

t0 = 0, t1 = 1, t2 = 2.

We assume

PV0 = 60$, C1 = 40$, PV2 = 30$.

By (2) the solutions for the quadratic polynomial are

d1 = 0.8968 and d2 = −2.23014

and the solutions for the internal rate return are

IR1 = 0.1151, IR2 = −1.4484.

 84

0.08
0.07
s e co n d o rd e r e rro r

0.06
t=1
0.05
t=2
0.04 t=3
0.03 t=4
0.02 t=6
0.01
0.00
00
01
02
04
05
06
07
08
10
11
12
13
14
16
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
0.
interest rate

Second order effect Figure 2.11

2
1.8
1.6
1.4
1.2
PV0

1
0.8
0.6
0.4
0.2
0
0.00 0.50 1.00 1.50
r value

Convexity of a Bond (C = 2%) Figure 2.12

 85

We assume

PV0 = 60$, C1 = 30$, PV2 = 30$.

Then we have by (2)

IR1 = 0.00, IR2 = −1.50.

We assume

PV0 = 60$, C1 = 20$, PV2 = 30$.

By (2) the solutions for the quadratic polynomial are

d1 = 1.2573 and d2 = −1.5907

and the solutions for the internal rate return are

IR1 = −0.2057, IR2 = −1.6287.

In all three cases the solution that can be identified as an interest

rate is unique although the quadratic equation has two distinct
solutions.

From the example above it can be seen that from PV0 < C1 + PV2
resp. PV0 > C1 + PV2 it follows IR1 > 0 resp. IR1 < 0.

Example 2.22 (Sign change of interest, Interval [0,1]):

As in the previous example we consider one cash flow

PV0 + C1 dα = PV1 d, 0 < α < 1

 86

α PV0 C1 PV1 IRR AIC(IRR) MD AIC(MD) OD AIC(OD)

0.25 30 20 60 21.51% 46.50 22.22% 45.00 25.00% 40.00
30 30 60 0.00% 52.50 0.00% 52.50 0.00% 45.00
30 40 60 -15.63% 63.99 -16.67% 60.00 -20.00% 50.00
0.5 30 20 60 25.36% 39.44 25.00% 40.00 25.00% 40.00
30 30 60 0.00% 45.00 0.00% 45.00 0.00% 45.00
30 40 60 -19.57% 51.09 -20.00% 50.00 -20.00% 50.00
0.75 30 20 60 27.66% 36.15 28.57% 35.00 25.00% 40.00
30 30 60 0.00% 37.50 0.00% 37.50 0.00% 45.00
30 40 60 -22.46% 44.52 -25.00% 40.00 -20.00% 50.00

Table 2.7

but opposite to Example 2.9 the time intervals are non equidistant,
and do not have length 1.
We calculate the figures in bold. The values for the IRR are
computed by the Excel Routine XIRR. However, for the value
25.36% (α = 0.5) we found by using (1e) and (7a)

0.2536 = (1 + 0.1196) . (1 + 0.1196) − 1,

60 − 20 − 30
AICIR = = 39.44.
25.36

By (5)

PV1 − C1 − PV0 10
MD = = = 22.22%,
PV0 + (1 − t k )C1 45

by (7b)

AICMD = 30 + (1 − 0.25) . 20 = 45.

By (5)

PV1 − C1 − PV0 60 − 40 − 30 10
OD = = =− = −20%,
PV0 + 0.5 C1 50 20
 87

by (7c)
1.
AICOD = 30 + 40 = 50.
2

Example 2.23 (Transition from 2 real solutions to no real

solution by continuing changing the cash flow): We assume

PV0 = −30$, C1 = 80$, PV2 = −30$.

Then the quadratic equation for the internal rate return yields
the solution

IR1 = −0.54858, IR2 = −1.21525.

We assume

PC0 = −30$, C1 = 60$, C2 = −30$.

Then we have

IR1 = 0.00, IR2 = 0.00.

We assume

C0 = −30$, C1 = 40$, C2 = −30$.

Then there is no real solution. We see that the solution is behaving

different in all tree cases.

◊
 88

2.5. TWR attribution versus MWR attribution

a. Absolute Decomposition

In this section we consider a portfolio with n constituents at

discrete points in time tk, k = 0,...., K with t0 = 0 and tK = T. For
the value B PV0 of the portfolio and the value BPVj,0, j = 1,...., n of
its segments at the beginning t0 = 0 of the time period considered
we have
n
PVB0 = ∑ PVB j,0 (2.5.1a)
j =1
and for the value PV E T of the portfolio and the value PVEj,T,
j = 1,...., n of its segments at the end tK = T of the time period we
have
n
PVET = ∑ PVE j,T . (2.5.1b)
j =1
Remark 2.5: Contrary to (2.3.5), (2.3.7), resp., (1) specifies
the time the portfolio is considered. In (1) the products in (2.3.5),
(2.3.7), resp. are summarized to values of segments, i.e. therefore
this notation allows the investigating of not only investments like
equity or bonds but also cash and cash flows.

Remark 2.6: The units of the equation are currency. For our
exposition here we choose the US$.

The cash flows PCj,k, j = 1,...., n, k = 1,...., N − 1 with the

segment j of the portfolio P is composed of internal cash flow
PICj,k and external cash flow PECj,k with

PCj,k = PICj,k + PECj,k (2.5.1c)

 89

S egment

PCji,k

Time

The full portfolio return problem (Cash Management) Figure 2.12

and
n n
PCk = ∑ PC j,k = ∑ (PIC j,k + PEC j,k ) .
j =1 j =1
We start with an initial investment PVB0 and consider a series of in
and out flows PCk, k = 1,…, N − 1 in the portfolio followed by a
terminal value PVET

IRtot = IR(PVB0, PC1,…., PCN-1, PVET) (2.5.2)

is the solution of an equation that equals the initial investment and

the present value of the future cash flows including the terminal
value. If there are no cash flows IR is equal to TWR. In the case of
cash flows the difference of TWR and IR is reflected in the Timing
effect and the management effect as described in the following
Section 2.5b.
 90

Example 2.24 (Difference between TWR and MWR

[14, Illmer]): We assume N = 2, t1 = 1, t2 = 2, PVB0 = 200$ and

0rP,1 = 0.25, 1rP,2 = −0.2

in (2.3.2). Then

PVE1 = 200$ (1 + 0.25) = 250$.

We distinguish 3 cases:

a)
PIC1 = 0, PEC1 = 0,

i.e. by (1c)

PC1 = 0.

By (2.3.5) we have for the TWR

PV1 ⋅ PV2 250 ⋅ 200
0 r2 = −1= −1=
PV0 ⋅ PV1 200 ⋅ 250

(1 + 0.25)(1 − 0.2) − 1= 0.

We have
PL = 0

and the solution of (2.4.1) is

d1/2 = 1,
 91

i.e. for MWR we find

IR1/2 = 0.

b)

PIC1 = 0, PEC1 = 50,

i.e. by (1c)

PC1 = 50.

By (2.3.5) we have for the TWR

PV1 ⋅ PV2 250 ⋅ 240
0 r2 = − 1= − 1 = 0.
PV0 ⋅ PV1 200 ⋅ 300
We have

PL = −10$.

and the solution of (2.4.1)

d1 = 1.144675, d2 = −0.72801

i.e. we find for MWR

IR = −0.12639

c)

PIC1 = 0, PEC1 = −50,

 92

i.e. by (1c)

PC1 = −50.

By (2.3.5) we have for the TWR

PV1 ⋅ PV2 250 ⋅ 160
0 r2 = − 1= −1=0
PV0 ⋅ PV1 200 ⋅ 200
and we have

PL = 10$

and the solution of (1) is

d1 = 0.848386, d2 = −1.47339,

i.e. for MWR we find

IR = 0.178709.

In all tree cases the TWR vanishes as TWR is insensitive to cash

flows after the first period and the MWR, however, reflects the PL,
i.e. selling in a bear market is a good decision and buying in a bear
market is a bad decision. In addition MWR equals TWR if there is
no cash flow. The discussion between MWR and TWR is in the
next section.

With PVBj,k we denote the Portfolio Value of segment j,

j = 1,…., n, at the begining of period k, k = 0,…., N − 1, and with
PVEj,k the Portfolio Value of segment j, j = 1,…., n at the end of
 93

period k = 1,…, N. Then the weights Wj,k of segment j at the

beginning of period k are
PVB j,k
Wj,k = n
, k = 0,…., N − 1, j = 1,…., n,
∑ PVB j,k
j =1
and for the return Rj,k over the period [k, k + 1], k = 0,…., N − 1
of segment i we have
PVE j,k +1 − PVB j,k
R j,k = , k = 0,…., N − 1, j = 1,…., n.
PVB j,k
As a consequence we have for the market value of the portfolio
PVEj,k+1 at time k + 1 in segment j:

PVEj,k+1 = (1 + R j,k) PVBj,k. (2.5.3)

Furthermore PCj,k denotes the portfolio cash flow in or out of

segment j at time k given by

PCj,k = PVEj,k − PVBj,k, k = 1,…., N − 1. (2.5.4)

The IRj in segment j is the given by

IRj = IR(PVBj,0, PCj,k, k = 1,…., N − 1, PVEj,T) =

(2.5.5)
IR(PVBj,0, Rj,k, Wj,k, k = 0,….,N − 1).

The total Profit and Loss PLtot over the investment period [O, T] is
equal to the sum of the PLj, j = 1,…., n of the individual
segments: n
PL tot = ∑ PL j . (2.5.6)
j =1
 94

Definition 2.25: Assuming that there exists a solution IRj ≠

0, IRtot ≠ 0, resp. the average invested capital for segments
AICIR IR
j , AIC tot , resp. is
PL j
AICIR
j = , (2.5.7a)
IR j

PL tot
AICIR
tot = , resp.. (2.5.7b)
IR tot
Using the average invested capital for the internal rate of
return we have
n
AICIR
tot IR tot = ∑ AICIRj IR j ,
j =1
thus we find a decomposition of IR
n AICIR
j
IRtot = ∑ IR
IR j . (2.5.8)
j =1 AIC tot
Definition 2.26: By defining the return contribution RCj by
AICIR
j
RC j = * IR j (2.5.9)
AICIR
tot
we have
n
IR tot = ∑ RC j . (2.5.10)
j =1

The IR of the portfolio is equal to the return contribution defined

by (10).

We proceed by discussing three special cases:

 95

Case 1 (No cash flows): If PCj,k = 0, j = 1,...., n, k = 1,....,

N − 1, K ≥ 2 and T = 1
PL j PVE j,T − PVB j,0
AICIR
j = = = PVBj,0
and IR j IR j

PL tot PVE tot,0 − PVB tot,T

AICIR
tot = = = PVB0.
IR tot IR tot
Thus by (1a)
n
AICIR
tot = ∑ AICIR
i .
i =1
Case 2 (IR = 0): We consider an segment j with beginning
Value PVj,0, ending value PVj,T and cash flows PCj,k ≠ 0,
k = 1,...., N − 1, N ≥ 2 over the time unit interval. The IRj is the
solution of
N-1
PVj,0 (1+ IRj) + ∑ PC j,k (1 + IR j )t k = PVj,T
k =1
We subtract the total cash flow
N-1 N-1 N-1
PVj,0 (1+ IRj) + ∑ PC j,k (1 + IR j )t k − ∑ PC j,k = PVj,T − ∑ PC j,k .
k =1 k =1 k =1
By
N-1
PLj = PVj,T − PV j,0 − ∑ PC j,k
k =1
we have
N-1 N-1
PVj,0 . IRj + ∑ PC j,k (1 + IR j )t k − ∑ PC j,k = PLj
k =1 k =1
and assuming IRj ≠ 0 for the average invested capital we find
 96

∑ PC j,k ((1 + IR j ) tk )
N -1
−1
PL j
AICIR
j = = PV j,0 + k =1 .
IR j IR j
The above relationship shows that the average invested capital
is equal to the initial value corrected by difference of the cash
flows and the discounted cash flow to the invested start point
divided by the internal rate of return. In the absence of cash flow
the average invested capital is equal to the beginning value of the
portfolio. For IRj → 0 we have by the rules of hospital
N-1
AICMD
j = PVj,0 − ∑ PC j,k t k .
k =1
We conclude that if IRj = 0, j = 1,...,n there follows IRtot = 0 and by
(1)
n
AICIR
tot = ∑ AICIR
j .
j =1

Case 3 (No compounding, Modified Dietz) We are

concerned with the modified Dietz formula on portfolio level
N −1 n
PVT − ∑ ∑ C j,k − PV0
k =1 j =1
MD tot = N−1 n
PV1 + ∑ ∑ C j,k t k
k =1 j =1
and on a segment level
N−1
PVj,T − ∑ C j,k − PVj,0
MD j = k =1 , j = 0,1, 2, ....,n
N −1
PVj,T + ∑ C j,k t k
k =1
we find
 97

N −1
n
PVj,0 − ∑ Ck t k
k =1
MD tot = ∑ N-1 n
⋅ MD j .
j =1
PV0 + ∑ t k ∑ C j,k
k =1 j =1

By considering the first term of the product in the above sum is the
ratio of average invested capital of the individual assets and the
average invested capital of the whole portfolio and the second
term is the Modified Dietz of the individual assets. By using the
abbreviation
N−1
PVi,0 − ∑ Ck t k
w MD
j = N-1
k =1
n
, j = 1,2,…., n.
PV0 + ∑ t k ∑ C j,k
k =1 j =1
It seen that the Modified Dietz of the whole portfolio is
decomposed into the weighted average of the Modified Dietz of
the individual assets:
n
MD tot = ∑ w MD
j MD j .
j =1
By using the definition of the average invested capital for the
modified Dietz we find

n
AICMD
tot = ∑ AICMD
j .
j =1

We note that this decomposition has a weighting scheme that

does not add up to one as we have in general there is no
relationsship between
n
AICIR
tot and ∑ AIC tot
j .
j =1
 98

Summarizing by (6), (7) and (8), the IR is decomposed such

that the decomposition is consistent with the profit and loss of the
portfolio.
In the following we use only the AIC for the internal rate of
return and suppress the superscript in AIC throughout.

Example 2.25: We consider a portfolio with initial value

PV0 = 100$ over two years. In the first year the value of the
portfolio goes to 130$ and in the second year the value goes to
218$. The portfolio consists of two segements the first one has a
initial value of PV1,0 = 60$ and the second one has an initial value
of PV2, 0 = 40$. In the first period the return of the first, second
segment resp. is

R1,0 = 25%, R 2,0 = 37.25%, (2.5.11a)

resp. and in the second period the returns are

R1,1 = 100%, R2,1 = 40%, resp.. (2.5.11b)

There is no external cash flow, i.e. PEC1 = 0, PEC2 = 0 (see (1c))

and there is an internal cash flow of

PC1,1 = PIC1,1 = −15$, PC2,1 = PIC2,1 = 15$ (2.5.11c)

in the middle of the 2 year period.

We proceed by the following calculations. Based on (3) and
(11) we have
 99

PVE1,1 = (1 + 0.25)*60$ = 75$

PVE2,1 = (1 + 0.375)*40$ = 55$.

Furthermore by (4)

PVB1,1 = 75$ − 15$ = 60$,

PVB2,1 = 55$ + 15$ = 70$.

Again with (3) we have

PVE1,2 = (1 + 1.0) * 60$ = 120$,

PVE2,2 = (1 + 0.4) * 70$ = 98$,

PVE2 = PVE1,2 + PVE2,2 = 218$.

With (6), (7) and (8) we have

PL1 = 45, PL2 = 73, PLtot = 118,

IR1 = 54.5, IR2 = 38.9, IRtot = 47.6.

75
AIC1 = = 1.3761,
54.5
43
AIC 2 = = 1.1053,
38.9
118
AIC tot = = 2.4789,
47.6
The return contributions (9) are
 100
AIC1 75 * 47.6
RC1 = IR1 = = 30.2542,
AIC tot 118

AIC 2 43 * 47.6
RC2 = IR2 = = 17.3457,
AIC tot 118
RC1 + RC2 = 47.6.

b. Relative Decomposition

In this section we discuss the impact of the client on the

overall performance of an account or in other words we ask the
question: what is the performance of an asset manager and what
part of the overall performance originates from the client’s
decisions?
We want to highlight some of the differences between the two
return calculation methodologies:
• The time weighted rate of return (TWR) is the return of an
account that measures the return in such a way that the return
figures do not depend on changes in the invested capital
o TWR measures the return from the asset manager’s
perspective (assuming that he has no control over external
cash flows)
o TWR allows a comparison with a benchmark as well as with
peers and the competitors
o The calculation, the decomposition and the reporting of TWR
is common practice in the asset management industry
o The presentation of TWR is one of the main principles of the
GIPS standards
 101

• The money weighted rate of return (MWR) measures the

account return in such a way that the return figures depend on
changes in the invested capital
o MWR measures the return form the client’s perspective
(assuming that the client has control over external cash
flows)
o MWR allows no comparison with peers and the competitors,
however, with an appropriate adjusted benchmark
o The calculation, the decomposition and reporting of MWR is
not common practice in the asset management industry
o MWR is not addressed by the GIPS standards in detail.

The last mentioned point leads very often to the fact that the
performance of an asset management account is very often
reported to the existing client only from the asset manager’s
perspective. The asset manager argues always that they have to
report their performance and not the one from the client –
neglecting the impact of the (external) cash flows.
In the following we discuss in addition the return attribution
from a client’s point of view and illustrate the decomposition of the
money weighted rate of return (MWR) and its relationship to the
return attribution based on the time weighted rate of return
concept.
As mentioned above today it is common practice in the asset
management industry to calculate and to report the time weighted
rate of return (TWR) on a total portfolio level to existing as well as
to prospective clients. The TWR is insensitive to changes in the
money invested in the account and therefore allows a comparison
of the account return across peer groups and against a benchmark
or an index. This property might be the main reason why it is also
common practice to analyze and to decompose the TWR and not
the MWR of the account. However, there is also a need for
calculating and decomposing the MWR because it is the MWR
which covers the timing effect of cash flows into or out of the
 102

account. The MWR reflects the timing of cash flows and is

consistent with the absolute profit and loss of the account. Due to
these characteristics the MWR is the true return from a client’s
point of view if it is the client’s decision to invest money into or to
withdraw money from the account.
As it is not common practice to run a performance or return
attribution from a client’s perspective which covers the timing
effect of (external) cash flows, in the following we illustrate an
intuitive procedure to decompose the MWR of an account on the
total portfolio level. The basic idea of this approach is to
decompose the MWR of an account such that the most important
investment decisions from a client’s perspective are reflected:

• The “benchmark effect”, which is the return contribution due to

the decision to invest the initial money into a specific benchmark
strategy and which is equal to the benchmark return over the
investment period,
• The “management effect”, which is the return contribution due
to the decision to change the asset allocation and stock
selection of the account relative to the benchmark over the
investment period, and
• The “timing effect”, which is the return contribution due to the
decision to change the money invested in the benchmark
strategy and in the active asset allocation of the account over
the investment period.

Figure 2.13 illustrates the so called decision-oriented return

decomposition approach and shows the relationship between the
MWR and the TWR of an account where the TWR of an account is
the sum of the benchmark effect and the management effect. In
addition, for investment periods with no (external or non-
discretionary) cash flows the MWR is equal to the TWR and there
is no timing effect. The different return contributions can be further
decomposed according to the specific investment decisions. For
 103

example, the management effect is often split up into the asset

allocation, stock selection and currency management effect and
the timing effect can be decomposed in the effect from changing
the money invested in the benchmark strategy and in the effect
from changing the money invested in the active asset allocation.

Account MWR

Benchmark effect
+ Management effect
+ Timing effect

Account TWR

Overview of the decomposing approach Figure 2.13

The return contributions are calculated using the benchmark return

and the MWR as well as the TWR of the relevant account. The
TWR of the account is calculated assuming no cash flows but
considering the active asset allocations over the investment period.
In measuring the benchmark return it is also implicitly assumed
that no money is invested into or is withdrawn from the benchmark
portfolio. On the opposite, the MWR of the account reflects not
only the active asset allocations but also - and often more
importantly - the timing effect of the (external) cash flow decisions.
After calculating the overall returns of the benchmark and the
account, the benchmark effect equals the benchmark return, the
management effect is the difference of the TWR of the account
and the benchmark return and the timing effect is the difference
between the MWR and the TWR of the account.
In order to isolate the timing effect completely it is necessary
to calculate not only the “true” TWR but also the “true” MWR. The
 104

“true” TWR is not affected by any (external) cash flow. It is best

practice to calculate the “true” TWR on a daily basis and then to
link the daily returns geometrically over the investment period. On
the opposite the “true” MWR covers the total timing effect of all
(external) cash flows and is calculated using the internal rate of
return methodology over the whole investment period. Using
instead an approximation method often results in a residual return
component relative to the fictitious “true” return (TWR or MWR)
whose missed evidence may lead to misleading feedback into the
investment process.
This section illustrated why neither the MWR calculation nor
the MWR decomposition should be neglected but rather
incorporated into the performance reporting and evaluation process
or investment controlling. Not considering the MWR concept and
with this ignoring the timing effect of the (external or non-
discretionary) cash flows bears the risk of misinterpretation and of
wrong feedback into the investment process. The MWR concept
adds value and is by no means outdated. Moreover, all participants
are encouraged to remember the basics in order to get back the
lost insights of the MWR concept and should reintroduce the MWR
concept to the area of performance measurement as well as to the
area of performance attribution.
In (2.2.9) and (2.2.10) we decompose the relative return of a
portfolio. Essentially the formula is valid for a single period, i.e. the
weights are given at the beginning of the investment period and it
is assumed that there is no cash flow. In the following section we
introduce and discuss the case of the IR Attribution. We show that
the concept of the MWR is more general than the TWR approach.
In a Portfolio the weights are given at the begining of a period,
however, at the end of a period they are different due to market
changes. Similar to the previous section we consider a benchmark
with n constituents at discrete points in time tk, k = 1,...., K with
 105

t0 = 0 and tK = T. For the value of the benchmark value BBV0 and

its segments BBVj,0 , j = 1,...., n at t0 = 0 we have
n
BBV0 = ∑ BBVj,0
j =1
and for the value of the portfolio value EBVT and its segments
EBVi,T, j = 1,...., n at tK = T we have
n
EBVT = ∑ EBVj,T .
j =1
The cash flow is composed of internal cash flow IBC and external
cash flow EBC
n n
BCk = ∑ BCi,k = ∑ (BIC j,k + BEC j,k ) .
j =1 j =1
The weights of the benchmarks are given by
BVBi,k
Vi,k = n
∑ BVBi,k
i =1
and for the return in a period we have
BVE j,k +1 − BVB j,k
B j,k = .
BVB j,k
As a consequence we have for the market value of the
benchmark at time k + 1 in segment j:

BVEj,k+1 = (1+Bj,k) BVBj,k (2.5.12)

Furthermore BCkj denotes the derived portfolio cash flow,

Benchmark cash flow, respectively in or out of segment i at time k.
There is
 106

PCj,k = PVEj,k − PVBj,k and BCj,k = BVEj,k − BVBj,k (2.5.13)

Definition 2.27: A Notional (or reference) Portfolio is a

sythetic portfolio against a portfolio is measured with a passive
Strategy.

Definition 2.28: Starting with the portfolio or the benchmark

EVj,k ( •,•) , EVtot,k ( •,•) , k = 1,2,….,K, resp. (2.5.14a)

denotes the ending value using for the first argument W or B and
for the second argument R or M.
We note
n
∑ EVj,T (•,•) = EVtor,T (•,•) , k = 1,2,….,K. (2.5.14b)
j =1
We introduce the following four versions of IR and the
corresponding Profit and Loss:

1. (Portfolio)

IRj(R, W) = IR(Rj,k, Wj,k, k = 0,….,K − 1) =

(2.5.15a)
IR(PVBj,0, PCj,k, k = 1,…., N − 1, EVj,T(R,W))

is the IRi using the returns and the weight of the portfolio. (5) and
(15a) are the same. In addition we have
K −1
PLj(R, W) = PVBj,0 − ∑ PC j,k − EVj,T(R,W) (2.5.15b)
k =1
where PLj in (6) is the same as the left side of (14b).
 107

2. (Notional Portfolio)

IRj(R, M) = IR(Rj,k, Mj,k, k = 0,…., N − 1) =

(2.5.15c)
IR(PVBj,0, BCj,k, k = 1,…., N − 1, EVj,T(R,M))

is the IR using the returns of the portfolio and the weight of the
benchmark and
N −1
PLj(R, M) = PVBj,0 − ∑ BC j,k − EVj,T(R,M), (2.5.15d)
k =1
3. (Notional Portfolio)

IRj(B, W) = IR(Bj,k, Wj,k, k = 0,…., N − 1) =

(2.5.15e)
IR(BVBj,0, PCj,k, k = 0,…., N − 1, EVj,T(B,W))

is the IR using the returns of the benchmark and the weight of the
portfolio and
N −1
PLj(B, W) = PVBi,0 − ∑ PC j,k − EVj,T(R,W) (2.5.15f)
k =1
4. (Benchmark)

IRj(B, M) = IR(Bj,k, Mj,k, k = 0,…., N − 1) =

(2.5.15g)
IR(BVBj,0, BCj,k, k = 0,…., N − 1, EVj,T(B, M))

is the IR using to the returns of the benchmark and the weight of

the portfolio and
N −1
PLj(B, M) = PVBj,0 − ∑ PC j,k − EVj,T(B,M). (2.5.15h)
k =1
 108

By (2), (4) and (15a) we have IRtot = IRtot(R, W) and by the left
side in (6) we have PLtot = PLtot(R, W). Similary to (15c) - (15h)
we introduce the notation IRtot(R, M), PLtot(R, M), IRtot(B, W),
PLtot(B, W), IRtot(B, M) and PLtot(B, M).

Definition 2.29: Referring to Definition 2.9, we define the

average invested capital AIC j ( •,•) , resp. for the first argument
R or B and for the second argument W or M by
PL j ( •,•)
AIC j ( •,•) = , (2.5.16a)
IR j ( •,•)

PL tot ( •,•)
AIC tot ( •,•) = , resp.. (2.5.16b)
IR tot ( •,•)
We proceed by the identity

X − Y=
(2.5.17)

Z − Y + V − Y + X – Z – V + Y,

∀X∈ R1, ∀Y∈ R1, ∀Z∈ R1, ∀V ∈ R1.

We apply (17) to (compare (2.2.10))

AICj(R, W) − AICj(B, V) =

AICj(B, W) − AICj(B, V) + AICj(R, V) − AICj(B, V) +

AICj(R, W) − AICj(B, W) − AICj(R, V) + AICj(B, V),

AICtot(R, W) − AICtot (B, V) =

 109

AICtot (B, W) − AICtot (B, V) + AICtot (R, V) − AICtot (B, V) +

AICtot (R, W) − AICtot (B, W) − AICtot (R, V) + AICtot (B, V).

and by using the notation for IR in (15)

AIC j (R, W) AIC j (B, M)
IR j (R, W ) − IR j (B, V) =
AIC tot (R, W ) AIC tot (B,M)

AIC j (B, W) AIC j (B,M)

IR j (B, W) − IR j (B, V) +
AIC tot (B, W ) AIC tot (B,M)

AIC j (R,M) AIC j (B,M)

IR j (R, V) − IR j (B, V) +(2.5.18)
AIC tot (R,M) AIC tot (B,M)

AIC j (R, W) AIC j (B, W)

IR j (R, W) − IR j (B, W ) −
AIC tot (R, W ) AIC tot (B, W )

AIC j (R,M) AIC j (B,M)

IR j (R, V) + IR j (B, V) .
AIC tot (R,M) AIC tot (B,M)
We proceed by
n
PLtot ( •,•) = ∑ PL j (•,•) . (2.5.19)
j =1
Hence

∑ (PL j (R, W ) − PL j (B, W )) =

n
PLtot (R,W) − PLtot(B,W) =
j =1

∑ (PL j (B, W)
n
− PL j (B, V) + PL j (R, V) − PL j (B, V) + (2.5.20)
j =1
PL i (R, W) − PL i (B, W ) − PL i (R, V) + PL i (B, V) ).
 110

With (19)
1 1 n

AIC tot ( •,•)

Ptot ( •,•) = ∑ PLi (•,•) .
AIC tot ( •,•) j =1

Hence with (14) we find

n PL j (R, W ) PL j (B, V )
IRtot(R, W) − IRtot(B, V) = ∑( AIC tot (R, W )
−
AIC tot (B, V )
) =
j =1

n ⎛ AIC j (R, W ) AIC j (B, V ) ⎞

∑ ⎜⎜ AIC IR j (R, W ) − IR j (B, V ) ⎟⎟ =
j =1⎝ tot (R, W ) AIC tot (B, V ) ⎠
n ⎛ AIC j (B, W) AIC j (B, V)
∑ ⎜ AIC (B, W ) j
⎜ IR (B, W) − IR j (B, V) +
j =1⎝ tot AIC tot ( B, V )

AIC j (R,M) AIC j (B,M)

IR j (R, V) − IR j (B, V) + (2.5.21)
AIC tot (R,M) AIC tot (B,M)

AIC j (R, W) AIC j (B, W)

IR j (R, W) − IR j (B, W ) −
AIC tot (R, W ) AIC tot (B, W )

AIC j (R,M) AIC j (B,M) ⎞

IR j (R, V) + IR j (B, V) ⎟⎟ .
AIC tot (R,M) AIC tot (B,M) ⎠

Definition 2.30: The return contribution RCi ( •,•)

AICIR
j ( •,• )
RC j ( •,•) = ∗ IR j ( •,•). (2.5.22)
AICIR
tot ( •,•)
We decompose the relativ return by

IRtot(R,W) − IRtot(B,V) = Atot + Stot + Itot. (2.5.23a)

 111

The Asset Allocation Atot effect is given by

[ ]
n n
Atot = ∑ A j = ∑ RC j (B, W ) − RC j (B, V ) . (2.5.23b)
j =1 j =1
The Stock Picking effect Stot is given by

[ ].
n n
Stot = ∑ S j = ∑ RC j (R, V ) − RCi (B, V ) (2.5.23c)
j =1 j =1
The Interaction Itot is given by
n
Itot = ∑ Ii =
i =1 (2.5.23d)

∑ [RC j (R, W ) − RC i (R,M)] − [RC j (B, W ) − RCi (B, M)] .

n

j =1

1. External cash flow

This cash flow can be injected by the client. If we have
external cash flows the value of the IR is not equal to the TWR and
the return of the client perspective is not equal to the return
achieved by the portfolio manager.

2. Internal cash flow

This cash flow might be tipically caused by a change of a
different asset allocation pursued by the portfolio manager or due
to the rebalancing of the benchmark.

We pursue different strategies:

1) If we have no external cash flows in the portfolio, benchmark,
resp.
n n
∑ IPCki = 0, ∑ IBCki = 0 , resp.,k = 1,…,N − 1
i=0 i=0
 112

2) Buy and hold in portfolio, benchmark, resp. k = 1,…,N − 1

IPCki = 0 , IBCki = 0 , resp..

We note that 1) follows 2).

Example 2.25 (continued): We consider a Benchmark

with initial value BBV0 = 100 $ over two years. In the first year the
value of the portfolio goes to 170$ and in the second period the
value goes to 284$. The benchmark consists of two segment. The
first one has a value of BVB1,0 = 20$ and the second one of
BVB2,0 = 80$. In the first period the return of the first, second
segment is
B1,0 = 150%, B2,0 = 50%, (2.2.24a)

resp. and in the second period the return is

B1,1 = 20%, B2,1 = 120%, resp.. (2.2.24b)

There is no external cash flow, i.e. BEC1 = 0, BEC2 = 0 and there

is an internal cash flow of

BIC1 = 40$, BIC2 = −40$ (2.2.24c)

in the middle of the 2 year period.

We discuss the relative return portfolio versus benchmark and

proceed first with the calculation for the benchmark. Based on
(12) and (24) we have

EV1,1(B,M) = BVE1,1 = (1 + 1.50)*20$ = 50$,

 113

EV2,1(B,M) = BVE2,1 = (1 + 0.50)*80$ = 120$,

EV1(B,M) = BVE1 = BVE1,1 + BVE2,1 = 170$.

Furthermore by (13)

BVB1,1 = 50$ + 40$ = 90$,

BVB2,1 = 120$ − 40$ = 80$.

Again with (12) and (22) we have

EV1,2(B, M) = BVE1,2 = (1 + 0.2) * 90$ = 108$,

EV2,2(B, M) = BVE2,2 = (1 + 1.2) * 80$ = 176$,

EV2(B, M) = BVE2 = BVE1,2 + BVE2,2 = 284.

Adapting the notation (14) and (20)

48
AIC1IR (B, B) = = 0.7940,
52.9
136
AICIR
2 (B, B) = = 1.7918,
75.9
184
AICIR
tot (B, B) = = 2.6861.
68.5

AIC1IR 48 * 68.5
RC1(B, B) = IR1(B, B) = = 17.8695,
AICIR
tot 184

AICIR
2 136 * 68.5
RC2(B, B) = IR2(B, B) = = 50.6304,
AICIR
tot 184
 114

RC1(B, B) + RC2(B, B) = 68.5.

The argument ( •,•) yields the following numerical results:

(R, W) in (13a) and (13b)

i Cash flow PL(R, W) IR(R, W)
1 − 60 −15 120 75 54.5
2 − 40 15 98 43 38.9
tot −100 0 218 118 47.6
Table 2.4
(R, M) in (13c) and (13d)
i Cash flow PL(R, M) IR(R, M)
1 −60 -15 162 117 77.3
2 −40 15 165 110 85.2
tot −100 0 327 227 80.8
Table 2.5
(R, W) in (13e) and (13f)
i Cash flow PL(R, W) IR(R, W)
1 −20 40 130 70 73.8
2 −80 -40 98 58 38.4
tot −100 0 228 128 50.9
Table 2.6
(B, M) in (13g) and (13h)
i Cash flow PL(B,M) IR(B,M)
1 −20 40 108 48 52.9
2 −80 -40 176 136 75.4
tot −100 0 284 184 68.5
Table 2.7
 115

We note that the IRR in the Tables 4 to 7 are non weighted.

The relative consideration portfolios versus benchmarks yields the
following results. For the PL we find according (46)

75 − 48 = 27,
−43 − 136 = −93,
118 − 184 = 66.

Asset Stock Interaction Total

Allocation Picking Effect
1 22 69 -64 27
2 -78 -26 11 -93
tot -56 43 -53 -66
Table 2.8

By the Tables 4 to 7 we find for the difference of the portfolio and

benchmarks

−0.2087 = 0.476 − 0.685.

For the effects we find according to (21) – (23)

Asset Stock Interaction Total

Allocation Picking Effect
1 0.1001 0.2379 -0.2139 0.1240
2 -0.2753 -0.2247 0.0573 -0.3328
tot -0.1752 0.1230 -0.1565 -0.208
Table 2.9

◊
 116

3. RISK MEASUREMENT

It is the task of every performance measurement department

to calculate return figures. This is without doubt important
information for the investor because the return reports of the
achievement with a specific portfolio. However, a return figure is
just one realization of the portfolio in the past. A further question is
how this return is achieved. Here risk considerations are
requested. It can well be that two portfolios have the same returns
but the risks are significantly different. Return calculations are
deterministic and risk relates to randomness. Risk calculation uses
methods from statistics and from the theory of probability.
Generally returns are the result of precise calculations and risk
figures are rather estimations that are based on a model. We see
that different risks can assume different characteristics and
occurrences.
In this chapter we focus on risk and in the following chapters
we discuss return and risk together, i.e. the performance is
discussed.
The term risk was originally already used in the 16th century by
Italian merchants for danger or hazard. The expression refers to
potential losses and damages in context of a company or an
enterprise. Risks are pitfalls that have to be avoided.
The notion risk is used in many areas and contexts. A general
definition can be as follows: Risk is the potential for an occurrence
of an undesired negative consequence of an event. In the
American Hermitage Dictionary risk is defined as the possibility of
suffering losses or losses due to an event that will quite probably
occur.
In financial markets, different risks are distinguished. We
proceed with the description of some types of risks.
With market risk we refer to risks that are common to an
entire class of assets or liabilities. The value of investments may
decline over a given time period simply because of economic
 117

changes or other events that impact large portions of the market.

Asset allocation and diversification can protect against market risk
because different portions of the market tend to underperform at
different times. Market risk is also called systematic risk.
Credit risk is usually an important issue in bond investments.
It reflects the possibility that a bond issuer can default by failing to
repay principal and - or interest in a timely manner. Bonds issued
by a government or an authority, for the most part, are immune
from default since Governments can print more money. Bonds
issued by corporations are more likely to be defaulted on, since
companies can go bankrupt. Credit risk is also called default risk.
Liquidity risk refers to the fact that an investor can have the
difficulty of selling an asset. Unfortunately, an insufficient
secondary market may prevent the liquidation or limit the funds
that can be generated from an asset. Some assets are highly liquid
and have low liquidity risk (such as a stock of a publicly traded
company) while other assets are highly illiquid and are highly risky
(such as a house).
The call risk is reflected by the cash flow resulting from the
possibility that a callable bond will be redeemed before maturity.
Callable bonds can be called by the company that issued them,
meaning that bonds have to be redeemed to the bondholder
usually so that the issuer can issue new bonds at a lower interest
rate. This forces the investor to reinvest the principal sooner than
expected, usually a lower interest rate.
Political risk is the risk of changes in a country’s political
structure or policies caused by tax laws, tariffs, expropriation of an
asset or restriction in repatriation of profits for example, a company
may suffer from loss in the case of expropriation or tightened
foreign exchange repatriation rules or from increased credit risk if
the government changes policies to make it difficult to pay
creditors.
 118

Legal risk is a description of the potential for loss arising from

the uncertainty of legal proceeding, such as bankruptcy and
potential legal proceedings.
The currency risk refers to the fact that business operations
and the value of investment are affected by changes of currency
rates. The risk usually influences business but it can also affect
individual investors who make international investments. It is also
called exchange rate risk.
Inflation risk assesses the fact that the purchasing power of
the currency shrinks the value of assets. Inflation causes money to
decrease in value at some rate and does so whether the money is
invested or not.
The financial risk is the aggregation of the risks described
above. In financial markets a security whose return is likely to
change much is said to be risky and a security whose return does
not change much is said to carry little risk. Generally we can say
that equity has higher risk than a money market instrument. In the
following we discuss the risk of a portfolio in view of some of the
financial risks discussed above.
 119

3.1. Absolute Risk

We introduce some notions from descriptive statistics. The

overall idea is to merge a series of data into one value. We start by
a return series of the portfolio P

rP,k, k = 1, 2,…., N (3.1.1)

of periodic returns, i.e. rP,k is the return between tk-1 = k − 1,

tk = k, k = 1,..., N = T.

Remark 3.1 (about collecting representative statistical

samples): The statisticians’ "camps" argue about the problem for
a long time. Ideally, we need a uniform and representative general
assembly of samples. Assuming that the statistical model we
chose is right and adequate to considered process, then, the
method of collecting samples is defined by the chosen statistical
model. Note that we always make some assumptions, explicit or
implicit, about the nature of our process when we choose methods
of collecting samples. Some people may not realize this, when
they collect samples, but it is impossible to pick up the method for
collecting samples without such assumptions. A typical assumption
is that the sample is independent and identically distributed.
The correlation or comovement time is assumed inferior to the
periodic time of the sample. It dictates the periodicity of the sample
(Figure 3.1). Further equidistant knot underpins the independency
of the sample. So, the method of collecting samples is defined by
several factors: the nature of the process or phenomenon we
study, our model, our measurement tools and our purposes.

Average is a generic term for statistics describing the middle

of the data set. The geometric mean return defined in Definition
3.1 and the arithmetic mean return defined in Definition 3.2 are
examples of averages.
 120

Correlation function

Samples

Time

Equidistant knots Figure 3.1

Definition 3.1: A population is defined as all members of a

specified group.

Definition 3.2: The variance var(P) of a portfolio P is the

sum of the squared deviations from the arithmetic mean rP defined
by (2.3.4c) divided by the number of returns in (1):

1 N
var(P) = ∑ (rP,k − rP )2 (population variance). (3.1.2)
N k =1

Definition 3.3: The standard deviation std(P) of a portfolio

P is defined by

std(P) = var(P ) . (3.1.3)

Often the term absolute risk is used similarly with standard

deviation or particularly in finance with volatility.

Remark 3.2 (intuition): The standard deviation is a measure

of how widely the actual returns are dispersed from the arithmetic
mean return. r does not mean a lot if the variance is big. On the
other hand if var(P) = 0 the mean is equally to the return series
 121

(1). In [23, Taleb] these two cases (small or big var(P)) are called
Mediocristan or Extremistan.

Remark 3.3 (unit): The unit of the standard deviation and the
return is percentage or decimals. The standard deviation has the
same unit as the return series.

Example 3.1: We consider N = 1, 2, 3, ….

rP,k = 5%, k = 1, 3,…., 2N − 1,

rP,k = 15%, k = 2, 4,…., 2N.

Then we have

rP = 10% and std(P) = 5%.

Remark 3.4 (annualizing): The variance is additive, i.e. if we

have f, f = 0,1,2,…. variances per year that are independent and
identically distributed we can multiply by f but the standard
deviation, risk (volatility) is not additive. The variance is annualized
by f whereas the risk and volatility is multiplied by f .

As in science, the first attempt for analysing data in finance

like the return series (1) is the normal distribution. It is the
benchmark in the theory of distributions. It assumes that the N in
(1) and (2) is big and tends or converges to the mean μ and the
standard deviation σ. The normal distribution has the probability
density function
 122

( x −μ )
−
1 2σ2
f(x) = exp , ∀ x ∈ R1. (3.1.4)
2
2 πσ

It can be characterised by returns that are equally distant from

the mean return and have the same relative frequency of
observations. Most of the returns are close to the average mean
return and there are relatively few extremely high and low returns.
Sometimes, the normal distribution is called by the nickname bell
shape distribution. Statistically, the following statements hold:

• About 68.3% of the observed returns will be within the

range of one standard deviation above and below the mean
return.

• About 90% of the returns will be within ± 1.65 standard

deviations.

• About 95.4% of the returns will be within ± 2.0 standard

deviations.

• About 99.7%, i.e. almost all of the returns will be within

± 3.0 standard deviations.

If returns are normally distributed and linked to asset prices by

the natural logarithm, then the asset price follows a log normal
distribution with probability density function

(ln( x ) − μ )
−
1 2 σ2
f(x) = exp , ∀ x > 0.
2
x 2 πσ

This means that continuously compounded, normally distributed

returns and log normally distributed assets prices are consistent.
 123

Definition 3.4: A sample is a subset of a population.

We estimate statistically figures like variance, etc. by a sample

of a population, i.e. a specific aspect of a population is estimated.
The estimation of the historical mean is unbiased, i.e. the expected
value of the estimation is the average. If the average is unknown,
i.e. is estimated by (2), it can be shown that the estimator for the
variance (3) is biased, meaning there is a systematic error in the
estimation.
The unbiased estimator with N samples for the variance is

1 N
Evar(P) = ∑ (rP,k − rP )2 (empirical or sample variance).(3.1.5a)
N - 1k =1

Asymptotically, i.e. N → ∞ , the two estimators are the same. The

corresponding standard variation is then

Estd(P)= E var(P ) (3.1.5b)

For N → ∞ we have the transition from the statistics to the

theory of probabilities. Referring to (4), the normal distribution is
important in the theory of probabilities.
 124

3.2. The Value-at-Risk (VaR)

In 1993 the G30 (an influential international body consisting of

senior representatives of the private and public sectors and
academia) published a seminal report for the first so-called off
balanced sheet products, like derivatives, in a systematic way.
Value-at-Risk (VaR) as a market risk measure was born and
RiskMetrics set an industry-wide standard.
In highly dynamic world with round-the-clock market activity,
the need for instant market valuation of trading position (known as
marking-to-market) became a necessity. Moreover, in markets
where so many positions were written on the same underlying
managing risk based on simple aggregation of normal positions
became satisfactory. Banks pushed to be allowed to consider
netting effects, i.e. the compensation of long versus short
positions on the same underlying.
Today value at risk is probably the most widely used risk
measure in institutions and has made its way into Basel II capital-
adequacy. It is deterministic to asses the maximum loss. Value at
Risk is a straightforward extension form maximum loss. The idea is
simply to replace maximum loss by maximum loss which is not
exceeded with a high probability the so called confidence level α.

Definition 3.3: Quantile is a generic term for grouping when

sampling in descriptive statistics. Examples are quartiles = 4,
quintiles = 5, deciles = 10 or percentiles = 100.

In probabilistic terms, VaR refers to thus simply a quantile of

the loss distribution. When we are communicating the risk
associated with a portfolio or an investment it is useful to convert
the standard deviation in percentage into dollar terms.

Definition 3.4: Value at risk VaRα(P) of a portfolio P at

t = 0 with value PV is the estimate of maximizing dollar loss we
 125

could expect to experience, over the time horizon, with a stated

level of confidence α.

Thus input parameter for calculating value of risk figures are

• the level of confidence α

• an set of returns (1) with a frequency typically days or
months

Typical value for are α = 0.95 or α = 0.99 in market risk

management the time horizon is usually 1 or 10 days, in credit
management and operational risk management in usually one year.

Remark 3.5: The level of confidence is also used in

conjunction with the test of a hypothesis. When a hypothesis is
statistically tested, two types of errors can be made. The first one
is that we reject the null hypothesis while it is actually true, and is
referred to as a type I error. The second one, a type II error, is that
the null hypothesis is not rejected while the alternative is true. The
probability of a type I error is directly controlled by researchers
through their choice of the significance level α. When a test is
performed at the 5% level, the probability of rejecting the null
hypothesis while it is true is 5%. A confidence interval α means

• a probability 1 − α for an error of the first kind.

• a probability α for an error of the second kind.

So we see that there is a trade off between the two errors.

Remark 3.6: VaR has been fundamentally criticized as risk

measure on the grounds that is has poor aggregation properties.
 126

We proceed by describing three types of methodology for

calculating value at risk figures:

a. Historical VaR

We start with a frequency distribution or a histogram and

select the α% worst returns. (100 − α)% of the returns does not
fall under the value of the maximum of these worst returns.

Example 3.2: We assume that we have in (1) N = 100, i.e.

we consider 100 returns of a portfolio P with portfolio value PV an

r1,P < r2,P < r3,P < ..... < r100,P.

Then we have

VaRH
5% (P ) = r5,P PV

and

VaR1H% (P ) = r1,P PV.

We see that the general case shows some minor

complications like equal sign or the fact that the number of return
points is not divisible by the confidence level. Furthermore the
number of data points has to exceed a minimal number of points
for calculating VaRHα (P ) .
The methodology is solely empirically and does not use a
model. It is thus non-parametric.
 127

b. Parametric VaR

We start by assuming that the loss function in normal

distributed (see Figure 3.2). The short fall probability under the
normality assumption is

( x −μ )
z −
1 2 σ2
Φ( z ) = ∫ exp dx.
2
2 πσ −∞

Then VaRPz% (P ) is

VaRPz% (P ) = − (μ + z σ) . PV

For z1 = −1.64448 we have Φ(z1) = 0.05 and

VaRP5% (P ) = − (μ + z1σ) . PV

and for z2 = −2.3263 we have Φ(z1) = 0.01 and

VaR1P% (P ) = − (μ + z1σ) . PV.

μ, σ, resp. is estimated by (1), (5), resp..

c. Monte Carlo simulation

Starting point is the transformation of market conditions into

the change to portfolios. This method is calculation intensive and
depends on the distribution of the randomly applied markets
conditions. It can be preferably used for portfolios that have
strongly non-linear payoffs.
 128

Probability

The normal distribution Figure 3.2

 129

3.3. Relative Risk

Referring to (3.1.1), we proceed by considering a second

periodic return series

rP̂,k , k = 1, 2, ….,N (3.3.1)

of a portfolio P̂ .

Definition 3.5: The covariance cov(P, P̂ ) of the portfolios

P and P̂ is the sum of the products of the return (3.1.2) of P and
the return (1) of P̂ divided by the number N of returns:

1 N
cov(P, P̂ ) = ∑ (rP,k − rP )(rP̂,k − rP̂ ) . (3.3.2a)
N k =1

Covariance is a statistical measure of the tendency of two

data series to move together. It measures the direction and the
degree of the association of the returns of the portfolios P and P̂ .
It is difficult to interpret as anything other than the average product
of the difference between the deviations of the portfolio returns
from the arithmetic mean in (2). Furthermore, it is also difficult to
use for portfolio comparison because it is impacted by the absolute
size of the returns.
Referring to the empirical Variance in (3.1.5) the same is valid
for the covariance

Ecov(P, P̂ ) =

(3.3.2b)
1 N
∑ (rP,k − rP )(r̂P,k − r̂P ) (empirical or sample covariance).
N - 1k =1
 130

Definition 3.6: The correlation corr(P, P̂ ) is defined by

E cov(P,P̂ )
corr(P, P̂ ) = . (3.3.3)
E var(P ) E var(P̂ )

Remark 3.7: We note that the covariance and variances used

in (3) is not influenced from the choice between sample or
population in (3.1.2), (3.1.5) resp. and (2).

Remark 3.8: By the Cauchy Schwarz in equality [8, Kreysig]

it can be shown that

−1 ≤ corr(P, P̂ ) ≤ 1.

This mathematical property is valid for any series (3.1.1) and (1).

Remark 3.9: The numerations of the entries in the formulae

do not influence the numerical values, i.e. any permutation of the
measurement leaves the numerical value unchanged. There are no
dependencies between the measurements.

Correlation normalizes the covariance to the Interval [-1,1] and

be used for direct comparison of different portfolios. It has no units
and is a scalar. A correlation close to 1 (-1) indicates that the two
time series move together (in the opposite direction). A correlation
close to 0, however, indicates that they are out of synchrony. A
correlation equal to 1 does not mean that the returns series in (1)
and (6) are the same.

Definition 3.7: The coefficient of determination R(P, P̂ ) -

squared or R2(P, P̂ ) is defined by
 131

(
R2(P, P̂ ) = Corr(P, P̂ ) . )2
R2(P, P̂ ) is the proportion of variability in the returns of the
portfolio P return that relates to the variability of the returns of the
Portfolio P̂ . It measures the degree of association of the portfolios
P and P̂ . A high R2(P, P̂ ) indicates that the portfolios P and P̂ are
probably exposed to similar risk exposure that are driving return.

We proceed by the tracking error. The notion was first

introduced in Control Theory. The tracking error is a measurement
of the relative risk of a portfolio versus the benchmark. The
benchmark portfolio is denoted by B with the periodic return series

rB,k , k = 1, 2,…., N.

Definition 3.8 (model independent): The tracking error

TE(1) is defined as follows

1 N 2
TE(1) = ∑ dk (3.3.4a)
N k =1

where

dk = rP,k − rP − rB,k + rB , k = 1, 2,…., N. (3.3.4b)

Lemma 3.1: If the tracking error TE(1) is zero, there follows

rP, k = rP + ck, k = 1,2,….,N,

rB, k = rB + ck ,k = 1,2,….,N

with
 132

N
∑ ck = 0.
k =1

Proof: By assumption of the Lemma

1 N 2
∑ dk = 0,
N k =1
hence

rP,k – rB,k = rP − rB , k = 1,2,….,N.

We assume that there exists a ck ∈ R1 such that

rP,k – rB,k = rP + ck − ck − rB , k = 1, 2,….,N.

The right side can be broken in different ways. The only way which
leads not evidently to the assertion of the Lemma is

rP,k = rP + c k, k = 1,2,….,N.

By summing up we have

N N
∑ rP,k = N (rP + ∑ ck )
k =1 k =1

and by (2.3.4), the Lemma is shown.

We proceed by illustrating the computation of the tracking

error
 133

Time Portfolio Value Benchmark Value

t0 P0 B0
t1 P1 B1
t2 P2 B2
t3 P3 B3
. . .
tN PN BN

Table 3.1

The returns in (4) can be calculated differently. We have 4

versions. We consider
1a)

Pk − Pk −1 B − Bk −1
rP,k = , rB,k = k , k = 1,….,N, (3.3.5a)
Pk −1 Bk −1

1b)

Pk B
rP,k = ln , rB,k = ln k , k = 1,….,N (3.3.5b)
Pk −1 Bk −1

for using (4). We proceed by

P
Q k = k , k = 1,….,N, (3.3.5c)
Bk
and consider

2a)
Q k − Q k −1
rP,k − rB,k = , k = 1,….,N, (3.3.5d)
Q k −1
 134

2b)

Qk
rP,k − rB,k = ln , k = 1,….,N, (3.3.5e)
Q k −1

for applying (4). By calculating the difference rP,k − rB,k with (5b)
and compare to the difference in (5d) we see than that they are
algebraically different. We note, however, that cases (5b) and (5e)
are the same since

Qk P P
ln = ln Qk − ln Qk-1 = ln k − ln k −1 =
Q k −1 Bk Bk −1

Pk B
ln Pk − ln Bk − ln Pk-1 + ln Bk-1 = ln − ln k .
Pk −1 Bk −1

We proceed with a second definition of the tracking error:

Definition 3.9 (Regression portfolio versus benchmark):

The tracking error TE(2) is defined by

TE(2) = std(P) 1 − ρ2 (P,B ) . (3.3.6)

Theorem 3.1: The tracking error TE(2) minimizes the

distance from the line that regresses the portfolio returns against
the benchmark returns.

Proof: The linear regression assumes that the values of one

the variable are at least partial dependent of the other. If the there
is a linear relationship between return of the portfolio and the
benchmark then the Tracking error is zero.
We proceed with the Ansatz
 135

rP,k = α + β rB,k + εκ, k = 1,....,N, N ≥ 2.

In this regression rP,k is the dependent variable and rB,k is the

independent variable. The method of the ordinary least squares
(OLS) yields

α = rP − β rB

with

cov(P,B )
β= . (3.3.7)
var(B )

We consider the error

εκ = rP, k − α − β rB, k = rP, k − rP − β (rB, k − rB )

and find

N n
∑ εk = ∑ (rP,k − rP − β(rB,k − rB ) ) =
2 2

k =1 k =1

N
(
∑ (rP,k − rP )2 − 2β(rB,k − rB )(rP,k − rP ) + β (rB,k − rB )2 .
2
)
k =1

With
N N
2β ∑ (rB,k − rB ) (rP,k − rP ) = 2β ∑ (rB,k − rB )2 .
2
k =1 k =1

We find (6) by using (7)

 136

var(B )
TE(2) = std(P) 1 − β2 = std(P) 1 − ρ2 (P,B ) .
var(P )

Remark 3.10: Theorem 3.1 is a time series analysis and in

3.4 we consider a cross sectional analysis, we regress over a
universe at a specific point of time.

Lemma 3.2: If the portfolio returns are a linear combination of

the benchmark return, then TE(2) = 0.

Proof: That follows from the fact that εk = 0.

◊

Lemma 3.3: TE(1) and TE(2) satisfy the inequality:

TE(2) ≤ TE(1).

Proof: Based on Theorem 3.1 we see TE(2) minimizes all

linear combination and following Lemma 3.1 TE(1) is such a
specific linear combination.

The notion tracking error stems from control theory and is

defined as measure of the difference between the desired output
from the measured output. Originally the aim was to achieve a
tracking error equal to zero. In passive asset allocation the investor
wants a vanishing tracking error. This endeavour is in line with the
original definition of the tracking error. As the expression says in
finance the error between the return of the portfolio and the
benchmark is measured. Today, however, the tracking error is also
 137

used in active §asset allocation. A tracking error significantly

differently from zero means that the portfolio manager has a
considerably deviation of his portfolio from the benchmark.
Portfolio 
Returns

Observation 6 Regression Line

Residual ɛ6

Residual ɛ5

Observation 5

Benchmark 
Returns

The regression Figure 3.3

The ex post risk of a portfolio P is based on its historical

returns (1). There are fundamental questions when using time
series:
• What is an optimal length of a particular time series to
be used for a performance analysis of a given
portfolio?

• What data frequency like daily, monthly should be

used?

In order to assess the quality of an asset manager is the

averaging of the asset allocation and the stocking the effect over
the total observation period. Normally the performance attribution
calculates a total figure for the whole reporting period for the
different management effects and does not show the management
effects over time for example on a monthly basis. In Figure 3.4 we
compared the management effects for the whole period with the
 138

ones on a monthly basis. The total figures indicate that the asset
manager was quiet a bad stock picker and a good asset allocater
during the reporting period. But this is the wrong conclusion
because the positive asset allocation effect was mainly generated
in the first two months of the reporting period and the monthly
asset allocation effect over the last seven months were constantly
negative. With respect to the stock picking effect there is a similar
situation but vice versa. The monthly stock picking effect was
negative over the first 8 months but constantly positive over the
last seven months. It is now up to further analysis to get more
insight into the figures and to come up with the “right” conclusions.

Stock picking effect

Asset allocation effect

Do the management effects vary over time? Figure 3.4

 139

3.4. Risk decomposition on asset level

The last section is based of the historical return series of a

portfolio. For instance a shift from equity to bond a year ago is
reflected in the analysis. The exposition in the last section is thus
called ex post risk analysis. In this section we decompose the risk
of a portfolio into segments (Definition 2.5) or individual
investments. The weights of the portfolio are as of today that why
the approach in this section is called ex ante risk analysis.
However, the returns of the invested investments or segments are
historical. The time series of the returns can provide estimations
for the variance and covariance.
We need for the following some concepts from a special
discipline in mathematics called linear algebra.
A real matrix A is defined by

⎡ a1,1 . . a1,n ⎤
⎢ ⎥
A= ⎢ . . ⎥
⎢⎣aN,1 . . aN,n ⎥⎦

where ai,j, i = 1,….,N, j = 1,….,n are real numbers, i.e. for short A
is an element in RNxn.
A real vector v is defined by

⎡ v1 ⎤
⎢.⎥
⎢ ⎥
v= ⎢ ⎥
⎢ ⎥
⎢.⎥
⎣⎢ v n ⎥⎦

where vj, j = 1,….,n are real numbers, i.e. for short v is an element
in Rnx1 = Rn. The Null vector v is defined by
 140

⎡0 ⎤
⎢.⎥
⎢ ⎥
v = ⎢ ⎥.
⎢ ⎥
⎢.⎥
⎣⎢0 ⎥⎦

The transpose AT of A is defined by

⎡ a1,1 a1,N ⎤
⎢ ⎥
⎢ ⎥
T
A = ⎢ ⎥,
⎢ ⎥
⎢ ⎥
⎣⎢a n,1 a n,N ⎥⎦

for short AT is an element in RnxN. The transpose vT of v is

defined by

vT = [v1 v n ],

for short vT is an element in R1xn = Rn.

Matrices are used for assessing the variance and the standard
deviation of a portfolio. The components of a vector are holdings
of the portfolio, benchmark, resp. or the returns of an investment
or a segment of the portfolio, benchmark, resp..
 141

We proceed by introducing the Matrix multiplication, i.e. the

multiplication of two Matrices

⎡ a1,1 . . a1,n ⎤
⎢ ⎥
A= ⎢ . . ⎥
⎢⎣aN,1 . . a m,n ⎥⎦

and
⎡ b1,1 . . b1,M ⎤
⎢ ⎥
B= ⎢ . . ⎥.
⎣⎢bn,1 . . bn,M ⎥⎦

We consider C = A . B where the element of C is defined

⎡ c1,1 . . c1,M ⎤
⎢ ⎥
C= ⎢ . . ⎥
⎣⎢cN,1 . . cn,M ⎥⎦

by
n
ci, j = ∑ ai,k bk, j , 1 ≤ i ≤ N, 1 ≤ j ≤ M.
k =1

Example 3.3: We consider the matrices

⎡ 1 2 3⎤ ⎡1 0 0⎤
A = ⎢0 1 2 ⎥ , B = ⎢2 1 0 ⎥
⎢ ⎥ ⎢ ⎥
⎣⎢0 0 1⎦⎥ ⎢⎣3 2 1⎦⎥

and calculate the product C = A . B

 142

⎡14 8 3⎤
C = ⎢ 8 5 2⎥ .
⎢ ⎥
⎢⎣ 3 2 1⎥⎦

Definition 3.11: A ∈ Rnxn is called positive definite if

n n
∑ ∑ ai, j v j v i > 0, ∀v j ∈ R1, ∀v i ∈ R1 (3.4.1a)
j =1 i =1

for v ≠ 0 is satisfied and A ∈ Rnxn is positive semi-definite if

n n
∑ ∑ a i, j v j v i ≥ 0, ∀v j ∈ R1, ∀v i ∈ R1. (3.4.1b)
j =1i =1
is satisfied.

On one side an investor could argue that there is no expost

risk because there are no unprecedented events in the past on the
other side another investor could have the stance that there is no
risk in the future as no event has materialized. We conclude that
risk to do with probability.
The idea of the exante risk model is based on the actual
weighting on the portfolio and the covariance matrix S of the
constituents, however, is based on historical data. In other words
the weights of the portfolio are subjective and thus dependent of
the investor and the historical behaviour of the market are
considered to be objective.
In the following we consider a portfolio P and a benchmark P
with n constituents Cj with returns

rP,k , rB,k , r j,k , 1 ≤ j ≤ n, 1 ≤ k ≤ N (3.4.2)

 143

at time tk. Here the frequency of the time point is arbitrary. In most
performance system monthly or daily data are used. Similarly to
(3.1.3) and (3.3.2) we introduce

1 N 2
var(Ci) = ∑ (ri,k − ri ) , i = 1,….,n (3.4.3a)
N k =1

and

1 N
cov(Ci, Cj) = ∑ (ri,k − ri )(r j,k − rj ) ,1 ≤ i, j ≤ n, i ≠ j (3.4.3b)
N k =1

The calculation of exante risk figures are based on these two

inputs. Exante Risk considerations can be perceived as risk
management by applying different weight schemes. With

1 N
rj = ∑ r j,k , 1 ≤ j ≤ N
N k =1

the input data is given

⎡r1,1 − r1 . . . r1,N − rn ⎤
⎢ . . ⎥
⎢ ⎥
⎢ . . ⎥
⎢ ⎥
R=⎢ . . ⎥ (3.4.4a)
⎢ . . ⎥
⎢ ⎥
⎢ . . ⎥
⎢rn,1 − r1 . . . rn,N − rn ⎥⎦
⎣

whereas
 144

⎡ r1,1 − r1 . . . . . . rn,1 − r1 ⎤
⎢ . . ⎥
⎢ ⎥
T
R =⎢ . . ⎥. (3.4.4b)
⎢ ⎥
⎢ . . ⎥
⎢⎣r1,N − rn . . . . . . rn,N − rn ⎥⎦

Lemma 3.3 (Variance Covariance matrix based on

return data): Considering (2) for the matrix

⎡ s1,1 . . s1,n ⎤
⎢ . . ⎥
S=⎢ ⎥
⎢ . . ⎥
⎢ ⎥
⎣sn,1 . . sn,n ⎦

defined by
1 T
S= RR (3.4.5)
N
we have

a) sjj = var(Cj, Cj), 1 ≤ j ≤ n,

sij = cov(Ci, Cj), i ≠ j ,1 ≤ i, j ≤ n.

b) S is positive semi-definite, i.e. S satisfies (1b).

Proof: a) The assertion follows from (1) and (3).

b) We multiply S by wT ∈ Rn and w ∈ Rn and find

wT A w ∈ R1, ∀ w ∈ Rn.
 145

With (5) and v ∈ Rn

v=wR

we have

1 T T 1
wT S w = w R R w = vT v, ∀ w ∈ Rn.
N N
By components
[v1 . . v n ]T ⎡ v1 ⎤
⎢ . ⎥
⎢ ⎥
⎢ . ⎥
⎢ ⎥
⎣vn ⎦

the right side is

v12 + ....... + v n2 ≥ 0 .

Thus the assertion (1b)

wT S w ≥ 0, ∀ w ∈ Rn

is shown.
◊

Remark 3.11: S is called the Risk matrix and is the starting

point for the portfolio optimisation theory (Chapter 4).

Example 3.2: We consider N = 3 data points, and n =1

segment and choose the vector
 146

⎡1 ⎤
w = ⎢2⎥
⎢ ⎥
⎢⎣3⎥⎦
and calculate wT w

[1 2 3 ]T ⎡1 ⎤
⎢2⎥ = 14.
⎢ ⎥
⎣⎢3⎥⎦
1 T
Then w w is the variance of an investment. We consider N = 1
N
data points and n = 3 segments and choose the vector

⎡1 ⎤
wT = ⎢2⎥
⎢ ⎥
⎢⎣3⎥⎦

and calculate wT w.

⎡ 1⎤ [1 2 3] ⎡ 1 2 3⎤
T

⎢2 ⎥ = ⎢2 4 6⎥ .
⎢ ⎥ ⎢ ⎥
⎢⎣3⎥⎦ ⎢⎣3 6 9 ⎥⎦

Evaluating the vector

⎡2⎤
⎢− 1⎥
⎢ ⎥
⎢⎣ 0 ⎥⎦
 147

we see that the matrix is only semi-positive definite. Then wT w is

the variance covariance matrix of a portfolio with 3 segments.

From the Cholesky decomposition [10, Neumaier] there

follows that the inverse of Lemma 3.1 is somehow also true: every
positive definite matrix can be derived from a not unique time
series. From the analysis of a real value function of a single
variable the element of S can be perturbed if S satisfies the strict
equality (1a). However, there is no statement about the magnitude
of perturbation. If S is semi definite the elements of S can in
general not be perturbed.

Theorem 3.2: We consider a portfolio P and benchmark B

with n constituents Cj, 1 ≤ j ≤ n with return history (2) and the
vector w for the weights wj for the portfolio and the vector b for the
weights bj for the benchmark, 1 ≤ j ≤ n. Then the following holds:

a) For the variance of the portfolio P we have

Var(P) = w TSw =

n n
∑ ∑ si, j w j w i =
j =1i =1

∑ ∑ w j w i cov( C j , C i ) + ∑ (w j ) var( C j ) =
n n n
2
j =1i =1 j =1
i≠ j
 148

[w1 . . wn ]T ⎡ s1,1 . . s1,n ⎤ ⎡w1⎤

⎢ . . ⎥⎢. ⎥
⎢ ⎥⎢ ⎥
⎢ . . ⎥⎢. ⎥
⎢ ⎥
⎣sn,1 . . sn,n ⎦ ⎢⎣wn⎥⎦

and the corresponding risk is

n n
std(P) = w TSw = ∑ ∑ si, j w j w i . (3.4.6)
j =1i =1

b) For the tracking error TE(1) defined in (3.3.4) we have

n n
T
TE(1) = ( w − b ) S( w − b ) = ∑ ∑ si, j ( w j − b j )( w i − bi ) =
j =1i =1

[w1 − b1 . . wn − bn ]T ⎡ s1,1 . . s1,n ⎤ ⎡ w1 − b1 ⎤

⎢ . . ⎥⎢ . ⎥
⎢ ⎥⎢ ⎥.
⎢ . . ⎥⎢ . ⎥
⎢ ⎥
⎣sn,1 . . sn,n ⎦ ⎣⎢wn − bn⎥⎦

Proof: From (2.2.4) and (3.1.2) we have

2
1 N 1 N ⎛n ⎞
var(P ) = ∑ (rP,k − rP )2 = ∑ ⎜⎜ ∑ w j (r j,k − rj ) ⎟⎟ =
N k =1 N k =1 ⎝ j =1 ⎠

1 N ⎛⎜ n n ⎞
∑ ⎜ ∑ ∑ w j w i (r j,k − rj )(ri,k − ri ) ⎟⎟ =
N k =1 ⎝ j =1i =1 ⎠
 149

n n 1 N
∑ ∑ w jwi ∑ (r j,k − rj )(ri,k − ri ) .
j =1i =1 N k =1

By Lemma 3.1 a) we have

n n n
var(P ) = ∑ ∑ w j w i cov( C j , C i ) + ∑ w j var( C j ) .
2

j =1 i =1 j =1
i≠ j

The proof of b) is the same as a).

Furthermore the correlation matrix is defined by

⎡ 1 . . ρ1,n ⎤
⎢ . . ⎥
ρ= ⎢ ⎥ (3.4.7)
⎢ . . ⎥
⎢ ⎥
⎣ρn,1 . . 1 ⎦

where

si, j
ρi,j =
si,i s j, j

Lemma 3.4: The correlation matrix (7) is positive semi-

definitive.

Proof: As S is positive semi-definite we have

n n
∑ ∑ si, j v j v i ≥ 0, ∀v j ∈ R1, ∀v i ∈ R1
j =1 i =1
 150

we consider
v̂ i
vi = ,
si,i

hence

n si, j
∑ si s j
v̂ j v̂ i ≥ 0, ∀v̂ j ∈ R1, ∀v̂ i ∈ R1
j =1

and conclude that the correlation matrix is also semi-positive

definite.

In the following we investigate the structure of the Risk matrix

and the correlation matrix.

Example 3.3: We consider a portfolio P with investment or

segments C1 und C2 and risk

var( C1) cov( C1, C 2 ) ⎞

S = ⎛⎜⎜ ⎟⎟ .
⎝ cov( C ,
1 2C ) var( C 2 ) ⎠

With matrices the risk of the portfolio is

⎛ var( C1) cov( C1, C 2 ) ⎞⎛ w1 ⎞

var(P ) = (w1 w 2 )⎜⎜ ⎟⎟⎜⎜ ⎟⎟ ,
⎝ cov( C ,
1 2C ) var( C 2 ) w
⎠⎝ 2 ⎠

i.e.

var(P) = (w1)2 var(C1) + 2w1 w2 cov(C1, C2) + (w 2 )2 var(C2).

 151

We consider 3 special cases:

1. cov(C1, C2) = var(C1) * var(C2) (ρ = 1)

var(P) = var(C1 + C2)

2. cov(C1, C2) = 0 (ρ = 0)

var(P) = var(C1) + var(C2)

3. cov(C1, C2) = var(C1) * var(C2) (ρ = −1)

var(P) = var(C1 − C2)

The example illustrates the Theorem of Pythagoras in Figure 1.2.

An eigenvalue λ ∈ R1 and an eigenvector v ∈ Rn of a quadric

matrix A ∈ Rnxn is defined by

A v = λ v.

For calculating the eigenvalue and eigenvalue we refer to a text in

linear algebra, see for e.g. [10, Neumaier].

Lemma 3.5: The matrix A ∈ Rnxn is positive definite only and

only if the eigenvalues of matrix are positive.

Proof: [10, Neumaier].

◊
 152

The following two theorems yield a further criterion for

checking that a matrix is positive definite.

Lemma 3.6 (Gerschgorin, weak version): Let A ∈ Rnxn be

an arbitrary quadric matrix with

n
α j = ∑ a i, j
j =1
j ≠1
then there is an eigenvalue λ of A and an index i such that

λ − a i,i ≤ αi

Proof: [16, Varga].

Lemma 3.7 (Gerschgorin, strong version): Let A ∈ Rnxn

be an irreducible matrix

n
αj = ∑ a ij .
j =1
j ≠1
Is λ an eigenvalue of A so is either

λ − a ii = α i
for at least one k or

λ − a ii < α i

for all k.
 153

Proof: [16, Varga].

Example 3.4: Let

⎡ 1 1⎤
⎢1 4 4⎥
⎢1 1⎥
S= ⎢ 1 ⎥.
⎢4 4⎥
⎢1 1
1⎥
⎣⎢ 4 4 ⎥⎦

Based of Gerschgorin the matrix is positive definitive, thus A

represents a correlation matrix. This matrix can be perturbed
without loosing the positive definiteness.

Example 3.5: We consider the matrix

⎡ 1 − 1 1⎤
S = ⎢− 1 1 − 1⎥ .
⎢ ⎥
⎢⎣ 1 − 1 1⎥⎦

For the following we refer to [10, Neumaier]. We illustrated the

calculation of the Eigenvalues. With

⎡1 0 0⎤
I = ⎢0 1 0 ⎥
⎢ ⎥
⎣⎢0 0 1⎥⎦
 154

we calculate the characteristic polynomial:

⎡1 − λ − 1 1 ⎤
P(λ) = det (A − λ ID) = ⎢ − 1 1 − λ − 1 ⎥ =
⎢ ⎥
⎣⎢ 1 − 1 1 − λ ⎥⎦

⎡1 − λ − 1 ⎤ ⎡− 1 1 ⎤
(1 − λ) det ⎢ ⎥ + det ⎢ ⎥
⎣ − 1 1 − λ ⎦ ⎣− 1 1 − λ ⎦

⎡ −1 1 ⎤
+ det ⎢ ⎥ = 0,
⎣1 − λ − 1⎦

where det stands for the determinant defined by

⎡a b ⎤
det ⎢
c d⎥ = ad − cd.
⎣ ⎦
Hence

P(λ) = λ2(λ − 3) = 0,

thus S is positive semi-definite and as a consequence has the

property of a correlation matrix.
For λ = 3 we have

⎡1⎤
v1 = ⎢− 1⎥
⎢ ⎥
⎣⎢ 1 ⎥⎦

and for λ = 0 we have

 155

⎡ 1⎤ ⎡0 ⎤
v2 = ⎢ 0 ⎥ , v3 = ⎢ 1⎥
⎢ ⎥ ⎢ ⎥
⎣⎢ 1⎥⎦ ⎣⎢ 1⎥⎦

for the eigenvectors.

Example 3.6: With

⎡ 1 − 1 − 1⎤
S = ⎢− 1 1 − 1⎥
⎢ ⎥
⎢⎣− 1 − 1 1⎥⎦

the vector
⎡1⎤
v = ⎢1⎥
⎢⎥
⎣⎢1⎥⎦
yields

vT S v = −3,

i.e. the matrix is not positive semi-definite. It is seen that 3 assets

cannot have mutually correlations −1.

◊
 156

3.5. Risk decomposition with factor analysis

A. Introduction
Factor models are a well-accepted way of reducing the
number of variable analyzing the return and risk of financial
investments. If we have a portfolio of n assets the number of
covariances, correlations, resp. is

n(n − 1)
,
2

i.e. there is quadratic growth with n.

We start by illustrating the concept of the factor analysis by
supposing a set of given data and we attempt to find a factor that
allows the reproduction data 1 and data 2. In Figure 3.5 Data 1
(Data 2) is approximately 1.5 (2.0) times the factor.

data 2 data 1

factor

Basic idea of factor analysis Figure 3.5

In factor analysis it is the aim to asses the main drivers of the

risk and to discuss the risk of a portfolio in term of sensitivity. If we
invest e.g. in Royal Dutch a potential factor analysis will include the
oil price as a factor and asses the sensitivity of the portfolio with
 157

Royal Dutch respect to oil price. We proceed by introducing the

following notations for 1 ≤ i ≤ N and 1 ≤ p ≤ M:
ri,t is the observed return of the security i at time t
(independent variable);

βi,p,t is the factor return of the security i at time t (dependent

variable);

fp,t are the factors to be estimated (slopes);

b: is a vector with returns that are known a priori

(deterministic) like currencies or the risk free rate.

There are different types of factor like

• Macroeconomic Factor
• Fundamental Factor
• Statistically significant independent factor

Here we do not describe the process of identification of a set of

factor and proceed by considering two types of factor analysis.

B. Cross-Sectional Model (Regression over securities)

The following basic linear relationship is only over a time
period k and has no relationship with the prior time periods. It
decomposes the return ri,k, 1 ≤ i ≤ N of securities of a
prespecified Universe and it is assumed. The characteristics of the
securities are assumed to be assigned linearly to the different
factors:

ri,k = βi,1,k f1,k + βi,2,k f2,k +….+ βi,M,k fM,k + b =

(3.5.1)
 158

M
∑ βi,p,t ⋅ fp,t + b
p =1

Obviously the relationship for

fp,t, 1 ≤ p ≤ M

cannot be satisfied exactly as M ≤ N in realistic situations,

however, there are techniques for minimizing the error ε

M
ri,t = ∑ βi,p,t ⋅ fp,t + b + ε
p =1

in a certain sense. It explains security returns via measured

sensitivities to systematic factors and estimated returns associated
with those factors. Changes in security characteristics are
captured and relates well to portfolio management. We consider

~ M
ri,k = ∑ βi,p,k ⋅ fm,k + b
p =1

as approximation of ri,k. For the return ~rP,k of the portfolio with n

securities we have

n
~
rP,k = ∑ w i,k ⋅ ~
ri,k + b.
i =1

We consider
 159

⎡ f1,1 − f1 . . . f1,N − fN ⎤
⎢ ⎥
⎢ . . ⎥
⎢ . . ⎥
⎢ ⎥
F=⎢ . . ⎥ (3.5.2)
⎢ . . ⎥
⎢ ⎥
⎢ . . ⎥
⎢f − f . . . fM,N − fN ⎥⎦
⎣ M,1 1

and consider the covariance matrix

⎡ t1,1 . . t1,M ⎤
⎢ . . ⎥
T=⎢ ⎥
⎢ . . ⎥
⎢ ⎥
⎣t M,1 . . t M,M ⎦

defined by
1 T
T= F F.
N

Following Lemma 3.1 T is positive semi-definite. Furthermore we

denote

⎡ βi1 ⎤
⎢ ⎥
⎢ ⎥
βi = ⎢ ⎥
⎢ ⎥
⎢ ⎥
⎣⎢βiM ⎥⎦
 160

The exposure of security i to the factor βi,p, 1 ≤ p ≤ M. Following

Theorem 3.2 the risk of a asset is the

M M
std(Ci) = βi T Tβi = ∑ ∑ t p,q βi,p βi,q
p =1iq =1

and covariance of

M M
cov(Ci, Cj) = β Tj Tβi = ∑ ∑ t p,qβ j,pβi,q
p =1q =1

⎡ β1,1 . . . β1,M ⎤
⎢ . . ⎥
⎢ ⎥
⎢ . . ⎥
⎢ ⎥
β = [β1 . . . βN ] = ⎢ . . ⎥
⎢ . . ⎥
⎢ ⎥
⎢ . . ⎥
⎢βN,1 . . . βN,M ⎥⎦
⎣

Theorem 3.3: We consider a portfolio P with a return series

(3.1.1) and n constituents Cj, 1 ≤ j ≤ n with return history (3.4.4)
and weights wj, 1 ≤ j ≤ n. Then the following holds:

a) for the variance of the portfolio P we have

T T
var(P) = w β T β w =
 161

n n M M
∑ ∑ w j w i ∑ ∑ t p,qβi,qβ j,q =
j =1 i =1 p =1 q =1
n n n
∑ ∑ w j w i cov( C j , C i ) + ∑ w j var( C j ) =
j =1i =1 j =1
i≠ j

T
⎡ β1,1 . β1,M ⎤ ⎡ w1 ⎤
⎢ . ⎡t . t M,1 ⎤ ⎡ β1,1 . . β1,n ⎤ ⎢ ⎥
. ⎥ ⎢ 1,1
. ⎥ ⎢ ⎥ (3.5.3a)
⎥⎢ ⎥ .
[w1 . . w n ] ⎢⎢
T ⎥ . . ⎥⎢ .
. . ⎥ ⎢ ⎢ . ⎥
⎢ ⎥ ⎢⎣t1,M . t M,M ⎥⎦ ⎢⎣βM,1 . . βM,n ⎥⎦ ⎢ ⎥
⎣⎢βn,1 . βn,M ⎦⎥ ⎣w n ⎦

and the corresponding risk is

n n M M
T T
std(P) = w β Tβw = ∑ ∑ w j w i ∑ ∑ t p,qβi,qβ j,q .
j =1 i =1 p =1 q =1

b) For the tracking error TE(1) we have

n n
TE(1) = ( w − b )T βT T β ( w − b ) = ∑ ∑ si, j ( w j − b j )( w i − bi ) =
j =1i =1

T
⎡β1,1 β1,M ⎤ ⎡ w1 − b1 ⎤
⎢ ⎥ ⎡ t1,1 tM,1 ⎤⎡ β1,1 β1,n ⎤⎢ ⎥
⎢ ⎥⎢ ⎥⎢ (3.5.3b)
[w1 − b1 wn − b1]T ⎢
⎢
⎥
⎥ ⎢ ⎥⎢ ⎥⎢
⎥.
⎥
⎢ ⎥ ⎢t1,M tM,M⎥⎦⎢⎣βM,1 βM,n ⎥⎦⎢ ⎥
⎣
⎢⎣βn,1 βn,M⎥⎦ ⎣wn − bn ⎦

The risk measures in (3.4.6) are called exante because there

only depend on the weights wj, 1 ≤ j ≤ n of the portfolio and the
benchmarks bj, 1 ≤ j ≤ n as of today are not dependents These
weights are subjective and the variance and the covariance in
(3.4.6) are objective, i.e. they are only dependent of the markets.
 162

One way is to estimate them from statistical techniques, but they

also can be derived from sheer economical considerations.

C. Times series (Regression over time)

Different to (1) we assume

βi,p = βi,p,t, 1 ≤ p ≤ M

i.e. the factors do not depend on t, thus we consider

M
ri,k = βi,1 fi,k + βi,2 f2,k + + βi,n fn,k + b = ∑ βi,p ⋅ fp,k + b,
p =1
i.e. we vary over time whereas in B. we vary on the universe.
Here we explain security return over time via measured return
associated with systemic factors and estimated sensitivities to
those factors. The classic example is the CAPM model. It is a one
factor model. The factor is also estimated by least regression
method. We will discuss it in the Chapter 4.

D. The principal component

An analysis aimed at finding a small number factor that
describes most of the variation in a large number of correlated
variables.

We proceed by illustrating the factor analysis to two typical

investment processes in portfolio management. Key is that
performance measurement mirrors the investment process. In
Factor analysis we assume an investment process where all
investment decisions are taken simultaneously and independently.

Example 3.6: We illustrate a typical equity investment

process (see Figure 3.4). We invest first in regions then in
sectors and followed by stock selection.
 163

Portfolio Return or Risk

Region A Region B

Sector 1 Sector 2 Sector 3 Sector 4

Stock 1 Stock 2 Stock 1 Stock 2 Stock 1 Stock 2 Stock 1 Stock 2

The equity investment process Figure 3.4

The equation (3.5.1) assumes the following form

M
k − rf,k = ∑ e i,p,k ⋅ fi,t +
ri,loc ∑ δind / sec ⋅ find / sec +
p =1 ind / sec

∑ δregion ⋅ fregion + ∑ δ country ⋅ βi,t ⋅ fcountry,t + εi,t

region country

th
where ri,lock is the local return of the i security rf,k is the risk-free
rate, ei,j,t is the exposure of the ith security to the jth factor, fj,t is the
factor return, βj,t is the exposure of the ith security to its country
factor and εj,t is the residual of the ith security with expected value
zero.
Exposure to region/country/industry group assumes either 0
or 1 depending whether or not the security is found in the region,
country or industry group. Regression over the equity universe
yields a line in the matrix F in (2).
The following table shows a specification of the factors
 164

Equity factors (Source Wilshire) Figure 3.5

Example 3.7: We illustrate a typical Fixed Income process

(see Figure 3.6). We first invest in currencies followed by durations
and then specific issues.
At t = 0 the price Pi,0 of the ith bond in local currency with time
to maturity ti,N is given by the total sum of the discounted cash
flow:

N
Pi,0 = ∑ CFi,k e −[r ( t k ) + s( t k )] t k
k =0

Here r is the spot rate and s is a spread to the spot rate at time tk,
0 ≤ k ≤ N. To calculate a return on this bond, it necessary to
express the price of the bond at some later small or instantaneous
time t
 165

Portfolio Return or Risk

Currency A Duration 1 Issue 1 Issue 4

Currency B Duration 2 Issue 2 Issue 5

Issue 3 Issue 6

A fixed Income investment process Figure 3.6

N
Pi,t = ∑ CFi,k e − [r ( t ) + dr ( t ) + ( s( t ) + ds( t )]( t k − t ) .
k =0

Following the work of Nelson Siegel for representing the shape of

the yield curve and the fact the dynamic of the yield curve can
describe by 3 basic movements it is proposed that dr(t) is seen by
an expansion of 3 factors around dr(t) = 0:

⎛ ⎛ t⎞
−⎜ ⎟ ⎞
⎛t⎞ ⎜1− ⎟
dr(t) = x1 + x2 ⎜1 − e ⎝ 7 ⎠ ⎟ + x3 t e ⎝ 7 ⎠ (3.5.4)
⎜⎜ ⎟⎟ 7
⎝ ⎠
 166

1.2

0.8
f1
0.6
f2

0.4 f3

0.2

0
0 5 10 15 20 25 30 35 40

Yield curves factor structure Figure 3.7

where x1, x2, x3 are coefficients for the exponentials in the above
expansion. We identify the factors

⎛ ⎛ t⎞
−⎜ ⎟ ⎞
⎛t⎞ ⎜1− ⎟
⎜ ⎟ t
f1(t) = 1, f2(t) = 1 − e ⎝ 7 ⎠ , f3(t) = e ⎝ 7 ⎠
⎜⎜ ⎟⎟ 7
⎝ ⎠

(see Figure 3.7). We see that f2 and f3 are zero at t = 0. At large t

f3(t) approaches 0 and, relative to f1, f2 dominates the yield whilst
at intermediate t (t = 7) f3 dominates f2 relative to f1. Consequently
f1 dominates the short part of the curves f2 affects the long end
and f3 impacts the middles part of the yield curve.
We decompose the change of the spread ds(t) linearly

dsi = ∑ χi, jλ j
j

is the spread factor and χi,j is the identity function which takes the
value 0 or 1 depending on whether the security is found in bucket.
 167

With (4) we find

Pi,t =

⎛ ⎛t ⎞⎞ ⎛ t⎞
⎜ −⎜ k ⎟⎟ ⎜1k − ⎟
− (r ( t k ) + x1 + x 2 ⎜1− e ⎝ 7 ⎠ ⎟+ x t
e⎝ 7 ⎠ + s( t ) + ∑ χ i ∂P dλ + ε ) t
⎜ ⎟ 3 k j ∂λ j i k
7 j j
N ⎜ ⎟
∑ CFi,k e ⎝ ⎠ .
k =0

Over time P is a function of P = P(x1, x2, x3, λj, t). The differential is

1 1 ∂Pi 1 ∂Pi 1 ∂Pi 1 ∂Pi 1 ∂P

dPi = dt + dx1 + dx 2 + dx 3 + ∑ i dλ j + εi .
Pi Pi ∂t Pi ∂x1 Pi ∂x 2 Pi ∂x 3 Pi j ∂λ j

We introduce the local return ri,loc

k of bond i at time k and the yield
ri,loc
k by holding the bond over time t

1 1 ∂Pi .
k − ri,k =
yield
ri,loc dPi − dt i
Pi Pi ∂t

The exposures are denoted by D1i,k, D2i,k, D2i,k and the factor
returns are fk,i, k = 1, 2, 3.

k − rt = D1i,k f k,1 + D2i,k fk,2 + D3 i,k f k,3 + ∑ λ j fλ j + εi

f
ri,loc
j

1 ∂Pi 1 N
D1i,t = = ∑ t k CFi,t k e − (r ( t k ) + s( t k )) t k ,
Pi ∂x1 x Pi k =1
1 = λ1 = 0

1 ∂Pi
D2 i,t = =
Pi ∂x 2 x2 =λ2 =0
 168

1 N ⎛ ⎛t
−⎜ k
⎞
⎟ ⎞
⎜ ⎝ 7 ⎠ ⎟ − ( r ( t k ) + s ( t k )) t k
∑t
Pi k =1 k ⎜1 − e ⎟ CF i, t k e ,
⎜ ⎟
⎝ ⎠

1 N ⎜⎛ t k ⎛⎜1− 7k ⎞⎟ ⎟⎞
t
1 ∂Pi ⎠ CF e − ( r ( t k ) + s( t k )) t k .
D3i,t = = ∑ t k ⎜ e⎝ ⎟ i,t k
Pi ∂x 3 x 3 =λ3 =0
Pi k =1 ⎜ 7 ⎟
⎝ ⎠
Similarly for the spread factor

1 ∂Pi N
− ( r ( t k ) + s( t k )) t k
SD1i,k, j = = χi, j ∑ t k CFi,t k e
Pi ∂λ j k =1
x1 = λ j = 0

Fixed Income factors (Source Wilshire) Figure 3.8

 169

The factor returns of the model are estimated with the regressing
of factor exposure against the return of a universe of bonds. They
are classified according Figure 3.8 and the regression function is

k − rf,k =
ri,loc

∑
χi,c∉Euro c∈countries χi,c (D1i,k f1,k,c + D2i,k f2,k,c + D3i,k f3,k,c ) +

⎛ ⎞
⎜ ⎟
χi,c∈Euro ⎜ D1i,k f1,k,Euro + D2 i,k f2,k,Euro + D3i,k f3,k,Euro + ∑ D1i,k fc,k,Euro ⎟ +
⎜ c∈Euro ⎟
⎝ Countries ⎠

+ ∑ χi,s,cSD1i,k fs,k,c + ∑ χi,q,cSD1i,k fq,k,c +

s∈Sector q∈qualities
c∈Currency c∈Currency

∑ χi,e,cSD1i,k fe,k,c + χic =USA ∑ A a,i,k fa,k + εi,k.

e∈qualities a∈other factors
c∈Currency

Regression over the bond universe yields a line in the matrix F in

(2).

We proceed by some conclude remark of the chapter.

1. A comparison or evaluation of expost versus expost Risk figure
is left to further research. It is seen from Examples 3.6 and 3.7
that exante absolute risk and exante tracking error is based the
choice of the factors in the factor analysis.
 170

2. It is seen that this chapter is based on the Definition 3.1 for

the variance and the Definition 3.5 for covariance. This
definition is basic and therefore often criticised in literature
because e.g.

• The centre around which standard deviation is calculated is

the arithmetical average historical return. However, prevailing
market condition can be substantially different and are also
thus also differentially forecasted from historical indications.

• Standard deviation treats positive and negative return

deviations from the mean with equal weight. As Standard is
perceived as risk, only negative deviations should be counted
as solely they are averse market movements for the investor.

Downside risk is an asymmetrical type of risk. A possible starting

point is the decomposition of the variance

1n 2 1 n 2 1 n 2
∑ (ri − r ) = ∑ (ri − r ) + ∑ (ri − r ) .
n i =1 n i =1 n i =1
ri ≤ r ri > r

Further exposition of down risk is beyond scope of this text and we

refer to the literature [9, Feibel]. However up to our knowledge
there are not studies on the superior performance of downside risk
against symmetrical risk measures.
 171

4. PERFORMANCE MEASUREMENT

4.1. Investment Process and Portfolio Construction

We start with considering the investment process depicted

in Figure 4.1. The discussion of return and risk in the previous
Chapters 2 and 3 is relevant to all part of the investment
process. We proceed with focusing on the different steps of the
investment process. The topic of this chapter is the asset
allocation, portfolio construction and rebalancing. The
evaluation of the investment strategy is in Chapter 5 (Investment
controlling).

Actual market data and

forecasts

Portfolio construction
Asset allocation process

Rebalancing
Constraint to
be respected

Portfolio analytics /
Performance analysis

Investment target, risk profile and benchmark

The investment process Figure 4.1

Definition 4.1: An asset class is a group of securities that

exhibit similar characteristics, behave similar in the marketplace
and are subject to the same laws and regulations. The three
main asset classes are equities (stocks), fixed-income (bonds)
and cash equivalents (money market instruments) (compare
www. investopedia. com).
 172

The asset allocation process is the decision oriented

process of investing in the different asset classes.
A typical investment strategy is reflected by the asset
allocation process followed by the portfolio construction. In
Example 2.7 the asset allocation process is the decision
marking process between equities markets in different countries
and the portfolio construction identifies individual equities. In
most situations the investment process is much more complex.
In this section we describe some approaches for the decision
making process of the pursued investment strategy which is
based on actual market data and forecasts. The definition of
the investment universe (Definition 2.4) and applicable
constraints on the portfolio and its constituents are:

• Liquidity needs
• Expected cash flow
• Investable funds (i.e. asset and liability)
• Time horizon
• Tax considered
• Regulatory and legal circumstances
• Investor preferences, prohibition, circumstances and unique
needs
• Proxy voting responsibility and guidelines

Running performance attribution implicitly assumes that the

asset manager has no guideline restriction for example with
respect to the minimum or maximum weight of a specific
security or asset class within the portfolio. If this assumption is
not true and there are restrictions in a way that the asset
manager is not able to invest according to the passive
investment alternative – the benchmark – the calculated
management effects may be misleading in a way that limiting an
investment may have a positive or negative impact on the
excess return. For example, there are equity indices which are
dominated by a specific security – assuming for instance 60%
of the index consists of this security – where a maximum weight
 173

limit of 10% is an implicit bet on this specific security by

predefining the minimum underweight of this security of 50%.
The question is who is responsible for the return contribution
due to this investment restriction. The same is true if there are
minimum limits for specific securities.
Portfolio construction is either

a) Creating a portfolio if there is new money to invest or

b) Rebalancing a portfolio if there exists already a portfolio.

a. Active investment strategy
We proceed with the following definition:

Definition 4.2: The objective of an active (passive)

investment strategy is to outperform (mimic) the benchmark.
Portfolios that try to outperform (mimic) the benchmark are said
to be actively (passively) managed.

An active investment process requires research of financial

markets, currencies, individual securities, yield curves etc. and is
thus much more costly than a passive investment process. An
active investment strategy attempts to take advantage of
changes in the risk premium and consequentially to respond
tactically to changing market conditions. We proceed by
describing approaches for analyzing and forecasting different
investments like bonds, equities, commodities or particular
financial markets.
In the following we give a tour d’horizon on the different
discipline for assessing financial markets. For more extensive
and in-depth information we recommend [16, Malkiel].
Referring to [16] the Firm-Foundation Theory and the
Castle-in-the Air Theory are the starting point of the following
two sections on fundamental research and technical research.
 174

Technical Fundamental
Research Research

Portfolio
Construction
Asset
Allocation

Political Economic
Research Research

Some different areas of financial market research Figure 4.2

In the following we give a tour d’horizon on the different

discipline for assessing financial markets. For more extensive
and in-depth information we recommend [16, Malkiel].
Referring to [16] the Firm-Foundation Theory and the
Castle-in-the Air Theory are the starting point of the following
two sections on fundamental research and technical research.

Fundamental research
Introduction and Terminology
For the investor in a stock of a company is vital to know the
‘right value’ of the common stock is. Fundamental research
strives to be relatively immune to the optimism and pessimism of
the crowed and makes a sharp distinction between a stock’s
market price and a theoretical price. There might be many such
theoretical prices and in literature they are called inner or
intrinsic value of a common stock. The Fundamental research is
based on the idea that the market price adjusts itself to the long
term to its intrinsic or true value, which depends on the
company and economic data.
In estimating the firm-foundation value of a security, the
fundamental’s job is to estimate e.g. the firm’s future stream of
earning, dividends, sales level, operating costs, corporate tax
 175

rate, depreciation policies and the costs of its capital

requirements.
Current deviations from the intrinsic value can be used
profitably, as the market will correct the valuation. Market
participants trade rationally and research is worthwhile. Overall
fundamental research is mostly company research, already
interconnected is the research of the industry.

Evaluation a stock: some first approaches

The most basic and easy understand model for evaluating a
common stock is the discounted cash flow model. It is
based on the arbitrage condition that the value of a common
stock today is the same as the present value of the future
dividends C1 and the value of the stock PV1 discounted to
today, hence we have

C1 PV1
PV0 = + .
1+ r 1+ r

By further assuming that the dividends Cj, j = 1,....,N are

known, iteration to N periods yields

N Ci
PV0 = ∑ j
+ PVN
j =1 (1 + r )

where PVN is the value of the stock after N periods. We see

that this is a version of the IRR equation considered in (2.4.1).
By assuming that the dividends are known for all future periods
we have

∞ Ci
PV0 = ∑ j
. (4.1.1)
j =1 (1 + r )
 176

A first question is whether instead of the dividends, the earnings

or earnings plus non-cash earnings should be discounted in (1)
[16].
As (2.1.1) in used in various versions, (1) is solved in
different ways. Firstly, PV0 is computed by using estimates on
the dividends and an appropriate discount rate. Secondly, the
idea is to calculate r out of the input parameter PV0, Cj, j =
1,....,n,..... An appropriate r can be calculated by the security
market line and the risk premium of the stock. Thirdly, the
equation is divided by the earning the left hand of the equation
would be the normal price earning at which the stock should
sell.
Here are some realization of the growth of Cj, j = 1,...., n,...
in (1):
1. Constant growth over an infinite amount of time
2. Growth for a finite number of years at a constant rate, then
growth at the same rate as a typical firm in the economy
from that point on.
3. Growth for a finite number of years at a constant rate,
followed by a period during which growth declines to a
steady-state level over a second period of years. Growth is
then assumed to continue at the steady-state level into the
indefinite future.

Technical research
Introduction and Terminology
Technical research is based on the graphical representation
of financial data. The value of technical analysis can be
expressed by the old Chinese proverb ‘A picture is worth ten
thousand words’. Technical research provides an abstract
access to financial market as every investment or a particular
financial market can be studied by graphical representations.
Examples for such investments are gold, currencies markets or
equities markets. For instance a graph can contain price, return
or other market data measured vertically in the graph and
 177

quantified by the unities on the vertical axis. The vertical axis

contains usually the time scale of the graph.
A chartist or a technical analyst seeks to identify price
patterns and trends in financial markets and attempt to exploit
these patterns. Although the study of price charts is primary,
technicians use also various further methods and approaches
for investing financial market. They are called technical
indicators. A typical example are moving averages MA(N)
defined by
1 N
MA(N) = ∑ rP,k .
N k =1

Referring to Figure 2.3 we assume here tN+1 = N + 1 and

MA(N) is a forecast at time N + 1. Furthermore for N = 1 we
have the simplest forecast model tomorrow is equal to today.
The idea is that the forecast of the future development
depends exclusively on the historical direction of prices and the
volume of trading. Most chartists believe that the market
movement is only 10 percent logical and 90 percent
psychological. Essentially it is assumed that financial markets
have a memory and it is possible to conclude their future
behavior from the past behavior.

Some properties of technical analysis are

• Based on strict rules and guidelines
• Provides objectivity in an investment process

• The theoretical framework can be applied immediately

• Applicable for any time horizon and market
 178

Description of some important concepts

Pattern
Elliot wave
Momentum
Rate of change
Trend
Moving average

Base concepts and their schematic representation Figure 4.3

We proceed with some basic technical indicators:

A trend is a line that connects at least two points. In an up
trend low points are connected. In a down trend high points are
connected. A trend, however, is more significant the more
points can be connected. ‘The trend is my friend until it breaks’
is a saying from technical analysis because future movements of
the financial markets are anticipated and invested accordingly.
In Physics the momentum is an important notion. It is the
increase and decreased of the speed of an object. A
momentum is a rate of the change. Momentum indicators give a
first signal before a trend is changing. It shows whether a trend
is accelerating or decelerating.
The Elliot wave principle is a form of technical analysis
that investor use to forecast trends in the financial market by
identifying extremes in investor psychology. Ralph Nelson Elliott
(1871-1948) developed the concept in the 1930. He proposed
that market prices unfold in specific patterns which practitioners
called Elliot waves or simply waves.
 179

The wave principle posits that collective investor psychology

moves from optimism to pessimism and back again in natural
sequence.
We proceed with the buy and sell behavior of a investor.
The experience shows that the investor is also too late when he
buys or sells. He buys when an up trend already is on his way
and he sells when a down trend is already partially realized in the
market place. An optimist is convinced to buy and a pessimist
mood is to sell.

Conviction = in the mood to buy

confidence
complacency

caution

concern

scorn Capitulation = in the mood to sell

Selling and Buying behavior Figure 4.4

The investor must learn when to buy when he is pessimistic and

is anxious and when to sell when he is optimistic and euphoric.

The most important goal of the technical analysis is detecting

buy and sell signals, i.e. to indicate when there is a turning
point.

Economic Research
We distinguish between Micro Economic and Marco
Economic Research. Fundamental research is an area that can
be perceived as Micro Economics whereas Marco economic
 180

deals with the economics of a country or a region. Economics

projections have an important impact on the performance of a
portfolio. A holder of a balance portfolio might switch from an
asset class to the other depending of the changing of the
economic situation. In a booming phase an investor should be
invested in equity. In declining markets, bonds are preferred,
and in a recession or a economic crisis commodity and cash or
money market like instruments are generally recommended (see
Figure 4.5).

Equities Bonds Commodity Cash

The investment cycle Figure 4.5

Here are some definitions of macro economic variables (see

www.investorwords.com).

Inflation is the rate at which the general level of prices for

goods and services is rising, and, subsequently, purchasing
power is falling. Central banks attempt to stop severe inflation,
along with severe deflation, in an attempt to keep the excessive
growth of prices to a minimum.
Stagflation occurs when the economy is not growing but
prices are increasing. This is not a good situation for a country.
This happened to a great extent during the 1970s, when world
oil prices rose dramatically, fuelling sharp inflation in developed
countries. For these countries, including the U.S., stagnation
increased the inflationary effects.
A general decline in prices is called deflation, often caused
by a reduction in the supply of money or credit. Deflation can be
caused also by a decrease in government, personal or
investment spending. The opposite of inflation, deflation has
 181

the side effect of increased unemployment since there is a

lower level of demand in the economy, which can lead to an
economic depression. Central banks attempt to stop severe
deflation, along with severe inflation, in an attempt to keep the
excessive drop in prices to a minimum.
Extremely rapid or out of control inflation is called
hyperinflation. There is no precise numerical definition to
hyperinflation. Hyperinflation is a situation where the price
increases are so out of control that the concept of inflation is
meaningless.
Interest rate is the amount charged, expressed as a
percentage of principal, by a lender to a borrower for the use of
assets. Interest rates are typically noted on an annual basis,
known as the annual percentage rate. The assets borrowed
could include, cash, consumer goods, large assets, such as a
vehicle or building. Interest is essentially a rental, or
leasing charge to the borrower, for the asset's use. In the case
of a large asset, like a vehicle or building, the interest rate is
sometimes known as the lease rate.
The Gross Domestic Product (GDP) is the monetary
value of all the finished goods and services produced within a
country's borders in a specific time period, though GDP is
usually calculated on an annual basis. It includes all of private
and public consumption, government outlays, investments and
exports less imports that occur within a defined territory that we
have

GDP = C + G + I + NX

where:
C is equal to all private consumption, or consumer spending, in
a nation's economy.
G is the sum of government spending.
I is the sum of all the country's businesses spending on capital.
NX is the nation's total net exports, calculated as total exports
minus total imports (NX = Exports – Imports).
The Gross National Product (GNP) is an economic
statistic that includes GDP, plus any income earned by residents
 182

from overseas investments, minus income earned within the

domestic economy by overseas residents.
The Consumer price Index (CPI) is a measure that
examines the weighted average of prices of a basket of
consumer goods and services, such as transportation, food
and medical care. The CPI is calculated by taking price changes
for each item in the predetermined basket of goods and
averaging them; the goods are weighted according to their
importance. Changes in CPI are used to assess price changes
associated with the cost of living. Sometimes it is referred to as
headline inflation.

The economic variables can be measured in the past, but

economists also publishing forecasts for them on a regular
basis (see Figure 1.1). Every major financial institution provides
forecasts and formulates a house view called economic outlook.
The forecasts are debated in media, but also applied to the
portfolios of the clients. A forecast is tied to a time horizon. We
distinguish between short term und long term forecasts. Usually
the maximal time horizon is one year. The accuracy of the
forecast is often not tested. The delivery of periodic forecasts is
one of the important tasks of economic research.

Example 4.1 (impact of a macro economic variable):

An inflation link bond protects the bond investor against the risk
of raising inflation.

A further area of economic research is the examination of

dependency between economic variables.
There is a substantial interest for business cycles (Figure
4.4. and 4.5) in economics [10, Gabisch, 13, Lorenz]. With t as
time variable, S as savings function, I as investment function, Y
as economic activity or real income of companies, K as capital
stock, α > 0 and δ > 0, the Kaldor Model [10, 13] is given by
ordinary differential equation
 183
dY
= α (I(Y, K) − S(Y, K))
dt (4.1.2)
dK
= I(Y, K) − δ(Y, K)
dt
and is a prototype of a dynamical system which generates
business cycles.
In the first equation economic of (2) economic activity is
seen to tend towards a level where savings and investments are
equal. A discrepancy between investments and savings induces
a change in the level of economic activity which proceeds until
the discrepancy is eliminated. If savings and investments are
linear functions of the level of activity, the economic system
would lead to hyperinflation with full employment, or a state of
complete collapse with zero employment [10]. As economic
systems in general do not behave in this manner, it is concluded
that savings and investment cannot be realistically characterised
in terms of linear functions. They have to be assumed nonlinear.
The second equation of (2) represents the accumulation of
capital stock. If there is no investment it is seen that the capital
stock is depreciating by a factor δ.
By using the Poincare Bendixson theorem it is shown in
[13] that (2) exhibits business cycles. The numerical
computation of business cycles is left to future research.

Political research
Investment decision depends also on political
considerations. Often an investor avoids a country because a
government is not stable enough. The buyer of government
bond can face the danger that the bond default or change its
conditions (Haircut). A caution portfolio manager decides only to
invest in countries with political stability.
Similarly before an election in a country a portfolio manager
might choose a portfolio close to the benchmark or a caution
portfolio manager decides only to invest in countries with
political stability. A left oriented government will augment its
 184

dents regardless of economic aspects whereas a right oriented

government purports more free financial markets.
Another investor might lend its money only to governments
that favor sustainable asset. These might be governments that
seek to reduce pollution and the accompanying emissions.
An investor might choose a country with a low tax bracket
and low regulations of financial markets.

Quantitative Research
Generally Data provide important information for the
investor. The extraction of relevant data is the task of
Quantitative Research applying the full apparatus of
mathematically techniques especially from statistics and the
theory of probability. Qualitative Research is accompanying
Quantitative Research. Qualitative research describes crucial
facts from financial markets. Often a conjecture from Qualitative
Research is confirmed. Quantitative Research encompasses the
development of Models. Modern Portfolio Theory belongs in
particular to quantitative research. It quantifies financial markets.
Modern Portfolio Theory needs both return and risk as input.
Asset Allocation and portfolio construction are forward looking
processes because they need forecasts. With fundamental and
technical research forecasts can be generated.
There are participants that beliefs that the research above is
useless as all inferences are always in the prices.
The random walk or efficient market theory refers to the
fact that security prices fully reflect all available information. This
is a very strong hypothesis. The efficient market hypothesis has
historically been broken into three categories each one dealing
with a different type of information:
With week information efficiency of a market is
expressed the fact that future stock prices cannot be predicted
on the basis of past stock prices.
With semi-strong information efficiency of a market is
indicated that even when using published available information
future prices are not predicable.
 185

The strong information efficiency of a market referred to

the fact that no information– not even unpublished development
– can be of use for predicting future prices, i.e. the current
market price reflects also the relevant non-public (insider)
information.

The stronger the information efficiency in the capital market

is the less interesting and less informative it is putting effort in
financial market research.

b. Passive investment strategy

A passive manager uses a range of different approaches –
full replication, stratified sampling, optimization techniques.
Under full replication we understand the investment in all
securities with the corresponding weight of the benchmark. In
stratified sampling the investor considers the constituents of
the benchmarks that influence the benchmark most. More
precisely, it allows the portfolio to match the basic
characteristics of the index without requiring full replication.
Characteristics of the benchmark like for instance the overall
duration, the convexity and maturity buckets of a fixed income
Benchmark are sequentially matched. Risk considerations such
as investigating correlation structures do not enter in the
portfolio construction. Optimization techniques intend to find
a tracking portfolio with fewer securities than the benchmark but
that has similar or even same characteristics as the benchmark.
In many cases the number of securities is fixed and a portfolio
with low tracking error and other desirable proprieties is
selected.
A portfolio tracking the benchmark has a tracking error
equals or almost equal to zero. The passive investment style
needs no forecasts and is therefore objective.
 186

4.2. Portfolio optimization

Portfolio optimisation is an approach for constructing

portfolios. The risk as introduced in Chapter 3 does not
interfered in the previous Section 4.1. Furthermore opposite to
for instance technical research and fundamental research
portfolio optimisation reflects the interaction between different
investments.
Harry Markowitz published his seminal work in portfolio
theory in the 1950s. Together with William F. Sharpe he
conducted a scientific study of portfolios invested in US equity
market in 1952. Harry Markowitz, William F. Sharpe and Merton
H. Miller received the Sveriges Riksbank Prize in Economic
Sciences in Memory of Alfred Nobel in 1990.
In financial markets many greedy investor wanted to earn
more money with less risk. Abnormality of a particular strategy
can be measured by Modern Portfolio Theory (MPT).
Portfolio optimisation is a part of MPT. In Section 4.3 we
continue with MPT. The basic ideas of MPT are still regularly
discussed in media see e.g. [17, Marty] and in literature e.g. [4,
Copland] and [23, Taleb]. For a comprehensive introduction in
MPT we refer to the text books [8, Elton] and [21, Sharpe].
We proceed by introducing the assets considered in this
section.

Remark 4.1: We distinguish between risk-less assets

and risky assets. Risk-less assets are introduced in Definition
2.6. As described in Section 3 an asset can bear different risks.
In the following we think of risk as degree of fluctuation of the
return expressed in volatility (see Definition 3.2). Risky assets
can assume significant volatility, i.e. the investor can experience
substantial gain and loss and they have not a volatility ‘close’ to
zero.

Typical risky assets are equities and MPT originates from

the investigation of equities.
 187

The problem of portfolio optimization under appropriate

condition is solved theoretically and optimal portfolios can by
numerically computed. They are many companies like Ibottson,
Barra and Wilshire that offer products with portfolio optimisation
software. As input they need the forecasts of the investors and
the risk is estimated by factor analysis (See Section 3.4).
However, its practical benefit is highly dependent on the
accuracy of the associated forecasts. Here lies the key, since
even forecast realizations that differ only slightly from
expectations can have a major impact.

Definition 4.3 (compare to [4, Copland page 169]): The

set of efficient or optimal portfolios is the set of mean-
variance choices from the investment opportunity set where for
a given variance (or standard deviation) no other investment
opportunity offers a higher mean return or equivalently for a
given mean return no other investment opportunity offers less
variance (or standard deviation).

Remark 4.2: The notion of a benchmark is introduced in

Definition 2.8. We see in the following that we can optimize with
respect to a money market rate (absolute optimization) or with
respect to a benchmark (relative optimization).

A mathematical formulation for the optimal portfolio problem

is as follows. We denote (see (2.2.3a)) by w the vector of the
weights for the portfolio

⎡ w1 ⎤
⎢ . ⎥
⎢ ⎥
w = ⎢ . ⎥, (4.2.1)
⎢ ⎥
⎢ . ⎥
⎣⎢ w n ⎥⎦

by b the vector of the weights for the benchmark

 188

⎡ b1 ⎤
⎢.⎥
⎢ ⎥
b= ⎢.⎥ (4.2.2)
⎢ ⎥
⎢.⎥
⎣⎢bn ⎥⎦

where

n
∑ bi = 1, bi ≥ 0, (4.2.3)
i=1

by p the vector of the weights for the positions

⎡ p1 ⎤ ⎡ w1 ⎤ ⎡ b1 ⎤
⎢.⎥ ⎢ . ⎥ ⎢.⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
p = ⎢ . ⎥ = ⎢ . ⎥ − ⎢ . ⎥, (4.2.4)
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢.⎥ ⎢ . ⎥ ⎢.⎥
⎣⎢pn ⎥⎦ ⎣⎢ w n ⎥⎦ ⎣⎢bn ⎥⎦

and by r the vector of returns (see (2.2.3b))

⎡ r1 ⎤
⎢.⎥
⎢ ⎥
r = ⎢ . ⎥. (4.2.5)
⎢ ⎥
⎢.⎥
⎣⎢rn ⎥⎦

Furthermore by referring to (2.3.4c) with the matrix S defined

by (3.4.5) the optimal portfolios w ∈ R1 are defined as those

a) that yield a maximum return μ ∈ R1 for a given risk σ ∈ R1,

i.e.
 189

μ = max wTr
w
under the condition

ABS1) absolute optimization

σ = wT S w

REL1) relative optimization

σ = pT S p

b) that minimize the risk σ ∈ R1 for a given return μ ∈ R1, i.e.

ABS2) absolute optimization

σ = min wT S w
w

REL2) relative optimization

σ = min pT S p
p
under the condition

wT r = μ

The portfolio optimal portfolio problem together with the

constraints
n
∑ wi = 1 (4.2.6a)
i =1
and

wi ≥ 0 (4.2.6b)
 190

are called the (Markowitz) Standard Model.

If only the budget constraint (6a) is assumed and no short
selling constraint is considered the optimisation model is
referred as Black Model [2, Bruce, page 159]. Optimal
portfolios are hence to be found on the efficient frontier (Figure
4.6 and 4.7).

Return
Asset A
Efficient Frontier

Inefficient Portfolios
Asset B

Risk

Absolute optimization Figure 4.6

Return

xEfficient Portfolio

Benchmark
Tracking Error

relative optimization Figure 4.7

 191

Remark 4.2: From (4) and (6a) follows for the positions

n
∑ pi = 0 .
i =1

We see that the positions can be positive or negative where with

(6b) we have only positive weights.

Remark 4.3: The relative optimization assumes the

benchmark portfolio is considered as the portfolio with risk zero
and therefore the reference portfolio. The absolute optimization
assumes only vanishing risk for specified correlation structures.

Remark 4.4: Positions are discussed in Table 2.3. We

consider three positions: An overweight, neutral and
underweight of securities versus the benchmark.

Remark 4.5: The problem ‘maximum risk’ is not tackled in

the framework discussed here.

Remark 4.6: The (Markowitz) Standard Model for two

assets can be solved explicitly and needs not a numerical
algorithm. We refer to the following Examples 4.2 and 4.4.

Example 4.2 (absolute Optimization with 2 assets):

We consider a Portfolio which consists of 2 assets A and B.
The covariance matrix is

⎡ 3 − 1 .5 ⎤
S= ⎢ (4.2.7)
⎣ − 1 .5 2 ⎥⎦

and the return is (5)

 192

⎡ 0 .7 ⎤
r = ⎢ ⎥. (4.2.8)
⎣ 0 .2 ⎦

The return of the portfolio is

μ = 0.7 w1 + 0.2 w2. (4.2.9)

By (3.4.5) and (3.4.6) in Theorem 3.2 the variance of the

portfolio is with (7)

Var(P) = w12 var(A) + 2 w1 w2 cov(A, B) + w22 var(B) =

(4.2.10)
2 2
3 w1 – 3 w1 w2 + 2 w2 .

In the following we assume the condition (6), i.e.

w2 = 1 − w1, w1 ∈ [0, 1]

and with yields by (9)

2μ = w1 + 0.4, μ ∈ [0.2, 0.7]

hence the parameterization of the weight is

w1 = 2μ − 0.4,
(4.2.11)
w2 = 1.4 − 2μ.

By vector

⎡ w1 ⎤ ⎡− 0.4⎤ ⎡1⎤
=
⎢ w ⎥ ⎢ 1 .4 ⎥ + 2μ ⎢− 1⎥ , μ ∈ [0.2, 0.7]. (4.2.12)
⎣ 2⎦ ⎣ ⎦ ⎣ ⎦

These weights also comprise portfolios that are not efficient.

They can are also be on the inefficient branch of the efficient
 193

frontier. In the risk σ return μrepresentation we find by (10) and

(11)
σ(μ) = 3 (2μ − 0.4)2 –

3 (2μ − 0.4)(1.4 − 2μ) + 2 (1.4 − 2μ)2 =

3 (−4μ2 + 3.6μ − 5.6) + 2 (1.96 − 5.6 μ + 4μ2) =

−14μ2 + 10.8μ − 17.8 + 3.92 − 11.2μ + 8μ2 =

−4μ2 − 0.4μ − 12.88.

In view of the absolute optimisation problem (see (ABS1) and

(ABS2)) we have uniqueness of the risk return relationship.
The maximum return portfolio is (ABS1) is

⎡ w1 ⎤ ⎡− 0.4⎤ ⎡ 1 ⎤ ⎡ 1⎤
=
⎢ w ⎥ ⎢ 1 .4 ⎥ + 1.4 ⎢− 1⎥ = ⎢0 ⎥ and μ = 0.7.
⎣ 2⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦

We proceed by the minimum risk portfolio (ABS2). By (6)

and (10) we have

Var(P) = 8w12 + 7w1 + 2.

The first derivative with respect to w1

∂Var(P)
= 16 w1 – 7.
∂w1

By requesting
∂Var(P)
=0
∂w1

we find
 194

⎡7⎤
⎢16 ⎥
⎡0.4375 ⎤
wg = ⎢ ⎥ = ⎢
⎢ ⎥ ⎣0.5625 ⎥⎦
⎢9⎥
⎣⎢16 ⎥⎦

and by (12)

μg = 0.41875.

The Standard Model, i.e. (6) is respected by

⎡ w1 ⎤ ⎡− 0.4⎤ ⎡1⎤
=
⎢ w ⎥ ⎢ 1 .4 ⎥ + 2μ ⎢−1⎥ , μ ∈ [0.41875, 0.7]
⎣ 2⎦ ⎣ ⎦ ⎣ ⎦

and the Black Model, i.e. only (6a) and not (6b) is respected by

⎡ w1 ⎤ ⎡− 0.4⎤ ⎡1⎤
=
⎢ w ⎥ ⎢ 1 .4 ⎥ + 2μ ⎢−1⎥ , μ ∈ [0.41875, ∞ ].
⎣ 2⎦ ⎣ ⎦ ⎣ ⎦

The short selling condition (6a) does affect the minimum risk
solution in this example.

If a matrix S ∈ Rnxn has a unique matrix S-1 ∈ Rnxn with

S-1S = SS-1 = I

then S-1 is called the inverse of S.

Lemma 4.1: If S is positive definite then there exists an

inverse Matrix S-1.

Proof: Following [10, Neumaier] we have

 195

SST = I.

The assertion follows from ST = S-1.

The minimum variance portfolio is discussed in detail in the

relevant literature [11, Kleeberg]. It offers a variety of interesting
characteristics. For example, it does not depend on forecasts.

Lemma 4.2 (global minimum risk portfolio): It is

assumed that S is positive definite. Furthermore the inverse of S
is denoted with S-1. With

⎡1⎤
⎢.⎥
⎢⎥
ID = ⎢.⎥
⎢⎥
⎢.⎥
⎣⎢1⎥⎦
the solution wg of

σ = min wT S w
w

for the Black model is

S −1 ID
wg = .
ID S −1ID

Proof: The Lagrange function is

L = wT S w + λ ID.

The derivative is
 196

∂L
= Swi + λ, i = 1,....,n
∂w i

and from the optimality condition we have

w = λS-1 ID

and from (6a)

1
λ=
ID S −1 ID

which yields the results. The minimal condition has to be

checked by the second derivative.
◊

In the following example S is only semi positive definite and

not positive definite.

Example 4.3 (Minimum risk portfolio): We consider the

correlation matrix
⎡ 1 − 1 1⎤
ρ = ⎢− 1 1 − 1⎥ .
⎢ ⎥
⎣⎢ 1 − 1 1⎥⎦

As shown in Example 3.5 this matrix has an Eigenvalue 0 with

multiplicity 2 and Eigenvalue 3. There is a two dimensional
Eigenspace to the Eigenvalue 0 with the basis

⎡ 1⎤ ⎡ 1⎤
v1 = λ1 ⎢ 1⎥ , λ1 ∈ R1, v2 = λ2 ⎢ 0 ⎥ , λ2 ∈ R1 (4.2.15)
⎢ ⎥ ⎢ ⎥
⎢⎣0 ⎥⎦ ⎢⎣ −1⎥⎦

and a one dimensional Eigenspace with Eigenvalue 3

 197

⎡ 1⎤
v3 = λ3 ⎢− 1⎥ , λ3 ∈ R1.
⎢ ⎥
⎣⎢ 1⎦⎥

For minimal risk portfolio we consider the Eigenspace of the

Eigenspace 0. In the Standard Markowitz model we have the
conditions
w1 ≥ 0, (4.2.16a)

w2 ≥ 0, (4.2.16b)

w3 ≥ 0 (4.2.16c)

and

w1 + w2 + w3 = 1. (4.2.17)
Let

v = v 1 + v2
where
⎡ w1 ⎤
v = ⎢w 2 ⎥
⎢ ⎥
⎢⎣ w 3 ⎥⎦

there follows from (15)

λ2 = 0,

hence (16c)

w3 = 0

with (17)

w1 + w2 = 2λ1,
 198

thus (17)

λ1 = 0.5

here is only one minimal risk portfolio with weight by (15)

⎡ w1 ⎤ ⎡0.5⎤
⎢ w ⎥ = ⎢ 0 .5 ⎥ .
⎢ 2⎥ ⎢ ⎥
⎢⎣ w 3 ⎥⎦ ⎣⎢ 0 ⎥⎦

In the Black model we have the condition

w1 + w2 + w3 = 1

there is only one minimal risk portfolio with weights

⎡ w1 ⎤ ⎡ 1 ⎤
⎢w ⎥ = ⎢ 1⎥ .
⎢ 2⎥ ⎢ ⎥
⎢⎣ w 3 ⎥⎦ ⎢⎣−1⎥⎦
◊

Example 4.4 (relative Optimization with 2 assets): We

again assume the return (9) and the risk (10). We consider the
benchmark weights (4) with n = 2

⎡ b1 ⎤ ⎡0.5⎤
⎢b ⎥ = ⎢0.5⎥ .
⎣ 2⎦ ⎣ ⎦

The return is the same as in (11) and the risk is

Var(P) = p12 var(A) + 2 p1 p2 cov(A, B) + p22 var(B) =

(4.2.13)
3 p12 – 3.0 p1 p2 + 2 p22.
 199

The REL1 is solved by

⎡ w1 ⎤ ⎡1.0 ⎤ ⎡ p1 ⎤ ⎡ 0.5 ⎤
⎢ w ⎥ = ⎢0.0 ⎥ , ⎢p ⎥ = ⎢− 0.5⎥ . (4.2.14)
⎣ 2⎦ ⎣ ⎦ ⎣ 2⎦ ⎣ ⎦

The return is μ = 0.7 and (13) with (14) yields Var(P) = 2.0.
The REL2 is solved by

⎡ w1 ⎤ ⎡ b1 ⎤ ⎡ 0 .5 ⎤ ⎡ p1 ⎤ ⎡0.0 ⎤
=
⎢ w ⎥ ⎢b ⎥ = ⎢ 0 .5 ⎥ , ⎢p ⎥ = ⎢0.0 ⎥ .
⎣ 2⎦ ⎣ 2⎦ ⎣ ⎦ ⎣ 2⎦ ⎣ ⎦

Thus from (8) we have μ = 0.45 and Var(P) = 0. The

Standard Model, i.e. (6) is respected

⎡ w1 ⎤ ⎡− 0.4⎤ ⎡1⎤
=
⎢ w ⎥ ⎢ 1 .4 ⎥ + 2μ ⎢−1⎥ , μ ∈ [0.45, 0.7]
⎣ 2⎦ ⎣ ⎦ ⎣ ⎦

and the positions are

⎡ p1 ⎤ ⎡− 0.9 ⎤ ⎡1⎤
=
⎢p ⎥ ⎢ 0.9 ⎥ + 2μ ⎢−1⎥ , μ ∈ [0.45, 0.7]
⎣ 2⎦ ⎣ ⎦ ⎣ ⎦

and the Black Model, i.e. only (6a) and not (6b) is respected is

⎡ w1 ⎤ ⎡− 0.4⎤ ⎡1⎤
=
⎢ w ⎥ ⎢ 1 .4 ⎥ + 2μ ⎢−1⎥ , μ ∈ [0.45, ∞ ]
⎣ 2⎦ ⎣ ⎦ ⎣ ⎦

and the positions are

⎡ p1 ⎤ ⎡− 0.9 ⎤ ⎡1⎤
=
⎢p ⎥ ⎢ 0.9 ⎥ + 2μ ⎢−1⎥ , μ ∈ [0.45, ∞ ].
⎣ 2⎦ ⎣ ⎦ ⎣ ⎦

In the risk σ return μ space we have

 200

σ(μ) = 3 (2μ − 0.9)2 –

3 (2μ − 0.9)(0.9 − 2μ) + 2 (0.9 − 2μ)2 =

4 (2μ − 0.9)2 = 16μ2 − 14.4 μ + 3.24.

Asset A and Asset B represent risk and return of two risky

securities such as equities. Figure 4.6 portrays the risk and
return on various portfolios comprising securities A and B.
In actual practice, however, portfolios comprising many risky
equities are optimized. In such cases, the efficient frontier looks
the same or at least similar, depending on the boundary
conditions and the input parameters.
Alongside risk and return, an additional important parameter
for portfolio optimization is correlation. This is a measure for the
movement of securities relative to each other. In mathematical
terms, the correlation must lie between the two extremes of –1
and +1. A and B are linked in three different ways in Figure 4.6,
corresponding to three different correlation values. For a
correlation of +1, the risk-return curve is a straight line linking
both securities. For a correlation of –1, the efficient portfolios
are at least partially linear, and one of the portfolios is risk-free.
This is on the return axis in Figure 4.6. And for a correlation
value between –1 and +1, A and B are linked by a curve;
optimization programs derive portfolios that are on the efficient
frontier.
What's key is that risk declines in line with correlation.
Figure 4.6 shows that only the portfolios above the minimum
variance portfolio are optimal. The part of the efficient frontier
below the minimum variance portfolio is described as the
"inefficient branch" in the relevant jargon.
What's clear is that optimization is one approach to
quantitative modeling available to the financial industry, allowing
forecasts to be expressed in the form of a concrete portfolio
 201

recommendation. Critics say that optimization is very easily

skewed by forecasting errors. If there is a benchmark,
optimization can be conducted in as stable a manner as
required; for, if the investor then buys that benchmark, he is
actually immune to portfolio optimization and the forecasts
made. The well-known approach of Black und Littermann is
essentially based on this concept.
Figure 4.6 and 4.7 are extreme simplifications. They merely
reflect the risk and return for different portfolios and assume
that the risk and return of the individual investments can be
defined. But it is precisely the forecasts and the estimates of
risk that are critical. It is also important that the optimal
portfolios move in line with changing forecasts and changing
market data. Hence, the optimal portfolio is constantly in a state
of flux and has to be optimized continuously. "Stability is the
enemy of optimality" is therefore often heard among specialists.
As mentioned before mathematics has solved the portfolio
optimization problem. However, the practical benefit of the
optimized portfolio is highly dependent on the accuracy of the
financial market forecasts. And it is precisely their verification
that forms the subject of many empirical studies, and offers the
basis for further development of Markowitz's original portfolio
optimization theory [6, Diderich, Marty], [19, Rustem, Marty].

The basic findings of MPT can be expressed as follows:

• Higher risk is always accompanied by higher average return.
• The interaction of the different investments in the portfolio
is reflected.

In the literature it was pretended that the efficient frontier is

continuously differentiable but in [25, Varös] it is shown that the
efficient frontier is in general only continuous by the following
example.
 202

Example 4.5 (non differentiability of the efficient

frontier): We consider in (3.4.5)

⎡ 3 3 − 1⎤
S = ⎢ 3 11 23⎥
⎢ ⎥
⎢⎣− 1 23 75⎥⎦

and (5)

⎡ 1⎤
r = ⎢3 ⎥ .
⎢ ⎥
⎣⎢5⎥⎦

An algorithm for computing optimal portfolios yields the

following results.

a) We assume conditions (6). For the global minimum variance

portfolio we find

μmin = μG = 1.2, σ2G = 2.8, wG = (0.95, 0, 0.05).

The maximal value is μmax = 5. The efficient portfolios consist of
the following pieces. Stability intervals are

μ ∈ [1.2, 1.5]: σ2(μ) = 5μ2 − 12μ + 10;

μ ∈ [1.5, 2.0]: σ2(μ) = μ2 + 1;

μ ∈ [2.0, 3.0]: σ2(μ) = 2μ2 − 4μ + 5;

μ ∈ [3.0, 5.0]: σ2(μ) = 10 μ2 − 48 μ + 65.

Except for μ = 3 is σ2(μ) differentiable on [1.2, 5.0]. σ2(μ) has

a kink at μ = 3. In Figure 4.8 we see the standard Markowitz
model.
 203

Efficient portfolio for the Standard Model Figure 4.8

b) We assume only condition (6a). For the global minimum

variance portfolio we get

μmin = μG = 0, σ2G = 1.0, wG = (2.0, −1.5, 0.5) and μmax = ∞ .

We have
μ ∈ [0, ∞ ]: σ2(μ) = μ2 + 1.

The efficient portfolios are along the straight line

⎡ 2 ⎤ ⎡− 0.75⎤ ⎡ 2 − 0.75μ ⎤
w(μ) = ⎢− 1.5⎥ + μ ⎢ 1.00 ⎥ = ⎢ − 1.5 + μ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣⎢ 0.5 ⎦⎥ ⎣⎢− 0.25⎦⎥ ⎣⎢0.5 − 0.25μ ⎦⎥

for μ ≥ 0.
 204

Figure 4.8 shows the numerical realisation of the Standard

Model and the Black Model.

Efficient portfolios for the Standard and the Black Model

Figure 4.9

With this example we have shown that the Standard Model can
behave significantly different from the Black model.

◊
 205

4.3. Absolute risk-adjusted measures

One aspect of portfolio analysis is the examination of the

portfolio along the time axis. The comparison of a specific
portfolio to similar portfolio is another aspect of portfolio
analysis, i.e. peer analysis as of now is the contents of the
section. Certainly the investor looks first at returns and then at
the risk taken. We proceed by considering measures that
consists of return and risk that is we introduce risk-adjusted
return measures.

Definition 4.6: We assume that a return series (3.1.1) is

given. The coefficient of variation CVA is the ratio of the
standard variation (3.1.2) to the arithmetic mean return (2.3.4c)

std(P )
CVA(P) = .
rP

We note that the denominator and the numerator in this

ratio have the same unit. The coefficient of variation reflects the
basic finding that for higher return we expect higher risk, i.e. if
Portfolio Managers 1 has achieved double as much return and
risk and a Portfolio Managers 2, they are having the same CVA
and their skills are the same.
If the return rP of the portfolio is positive that it can be said
qualitatively that a low CVA is desirable.

Example 4.6 (higher risk but lower coefficient of

variation): We consider a return series (3.1.1) with arithmetic
mean rP . We define for σ ∈ R1 and N = 1,….

rP,k = rP − σ, k = 1, 3,…., 2N − 1,

rP,k = rP − σ, k = 2, 4,…., 2N.

 206

Then we have

std(P) = σ

and

σ
CVA(P) = .
rP

In Example 3.1 we have rP = 10% and σ = 5% thus

1
CVA(P) = .
2

With rP = 10% and σ = 25% we find

2
CVA(P) = .
5

We see that the relation between the risks and the CVAs is
inverse.

We continue with an illustration from physics. We consider

two drivers who drove from A to B, C to D, resp.. We assume
that two drivers drove with constant speed. The question is:

Which driver is faster?

If the distance, the time, resp. is same the question can be

answers by the time, the distance, resp. the two drivers drove. If
the distances and the times are different we need to normalize
or gauge either by the distance, by the time, resp.. The speed
defined by the distance divided by time has to be used instead
 207

of distance (time), resp.. Here is the comparison to portfolio

analysis. If we want to compare two portfolios with both
different returns and risks.

Definition 4.7: The risk adjusted return RAR is the

inverse of CVA, i.e.

rP
RAR = .
std(P )

Following Definition 2.6 and similarly to (3.1.1) we denote with

rf,k, k = 1, 2,…., N

the risk-free return in period k, k = 1,….,N and following

Definition 3.2 the arithmetic mean return rf the considered
series of risk-free return is

1 N
rf = ∑ rf,k .
N k =0

As risky investments, risk-free investments enter also as

part in modern portfolio theory. The following risk-adjusted
return measure is named after Dr. William Sharpe. It reflects the
fact that the investor cannot earn a return above the risk-free
return without risk.

Definition 4.8: The Sharpe Ratio SR is the difference

between the annual arithmetic mean return and annual
arithmetic mean risk-free return divided by the annualized
deviation of the fund, i.e.

r −r
SR = P f .
std(P )
 208

Remark 4.3: CVA, RAR and SR are used for classifying a

set of portfolios. They map return and risk figures in the set of
the real numbers. If a portfolio manager is judged by SR, he
tried to maximize his SR because he attempts to achieve a high
return with little as possible risk. Even the return an equal the
SR can be different. If this return is positive the Portfolio
manager with less risk is superior to that with more risk. For
negative return this judgment is opposite which make no sense.
Strictly speaking the domain of applicability for SR is limited to
positive return. The same is valid for CVA and RAR.

4.4. The Capital Asset Pricing Model (CAPM)

The assumptions of the CAPM are [page 194, 4,

Copeland]:

• Investors evaluate portfolios by looking at the expected

returns and standard deviation of the portfolios over one-
period time horizon.
• Investors are never satiated, so when given a choice
between two otherwise identical portfolios they will choose
the one with the higher expected return.
• Investors are risk-averse, so when given a choice between
two otherwise identical portfolios they will choose the one
with lower standard deviation.
• Individual assets are infinitely divisible, meaning that an
investor can buy a fraction of a share if he or she so
desires.
• There is a risk-free rate at which the investor may either
lend or borrow money.
• Taxes and transaction costs are irrelevant.

In the following we considered one period tk = k,

tk+1 = k + 1, k = 0,….,N − 1 and portfolios that consist of risk-
free investments and risky investments (see Remark 4.1). As
 209

risky securities, risk-free investments are also part of modern

portfolio theory.
We return to the return and risk considerations in the
previous chapter. We distinguish between the active and the
passive investors. In the previous chapter an active investor was
considered as he could steer his return risk preference. A
passive investor does see any opportunities in market place and
clearing prices prevail. Every investor will hold a portfolio that
consists of a share of the market portfolio defined as follows:

Definition 4.9: The market portfolio MP with return rMP

is a portfolio consisting of an investment in all securities like
stocks, bonds, real properties and on where the proportion to be
invested in each security corresponds to its relative market
value. The relative market value of a security is simply equal to
the aggregate market value of the security divided by the sum of
the aggregate market value of all securities.

In principle the market portfolio is a thought experiment and

it is not possible to capture any single available asset in a
portfolio. My and you belongings are part of the market
portfolio. However, big universes like the MSCI world equity can
serve as a proxy.
A first market equilibrium principle yields based on two-fund
separation principle described in [4, Copland, page 181]:

Theorem 4.1: Each investor will have a utility maximizing

portfolio that is combination of the risk-free asset and a portfolio
(or fund) of risky assets that is determined by the line drawn
from the risk-free rate of return tangent to the investor’s
efficient set of risky assets. The portfolio that is on the tangent
to efficient frontier and on the efficient frontier is the market
portfolio.

Proof: We consider a portfolio consisting of the risk-less

asset RF and the market portfolio MP
 210

rP = w1rf + (1 − w1) rMP

and for the standard deviation of the portfolio we have

std(P)
1
( )
= (1- w1)2 var(RF) − 2 w1(1 − w1)cov(RF, MP) + w12 var(MP) 2 =

w1 std(MP).

As var(RF) = 0 and cov(RF, MP) = 0 we have

std(P )
w1 =
std(MP )

and
⎛ std(P ) ⎞ std(P )
rP = ⎜1 − ⎟ rf + rMP,
⎝ std(MP ) ⎠ std(MP )

hence

std(P )
rP = rf + (rMP − rf)
std(MP )

and in the risk return graph we have

rMP − rf
rP = rf + std(P).
std(MP )

Definition 4.10: This line specified in Theorem 4.1 is

called the capital market line (CML) (see Figure 4.6).
 211

Remark 4.3: The proof of Theorem 4.1 does respect any

constraints. Their impact on the CML is still subject to future
research.
CML
Return

Market portfolio

Efficient Frontier

Risk
The modern portfolio theory Figure 4.6

Remark 4.4: As expected approximations of the Market

portfolio are not efficient [5, Dalang R.,].

Remark 4.5: RAR are slopes on the efficient frontier and

the SR is the slope of the CML. We note that the investors have
the same Sharpe ratio

rMP − rf
, (4.4.1a)
std(MP )

i.e. the investor is risk neutral along the CML.

Theorem 4.2 (Capital asset pricing model (CAPM)):

We consider a Portfolio P that consists of a risky asset C1 with
return r1.Together with return of the risk-less asset rf and the
Market portfolio MP with return rM we have:

r1 − rf = β(C1,MP) (rMP − rf) (4.4.2a)

 212

where

cov( C1,MP )
β( C1,MP) = . (4.4.2b)
var(MP )

Proof: We consider a Portfolio P that consists of an asset

C1 with return r1 and a weight w1 and the market portfolio MP
considered as an asset. For the return of the portfolio we have

rP = w1r1 + (1 − w1) rMP

and for the standard deviation of the portfolio we have

std(P) =
1
( )
(w1)2 var(C1) − 2 w1(1 − w1)cov(C1, MP) + (1 − w1)2 var(MP) 2 .

We derive the return and the standard deviation of the

portfolio with respect to w1:

d rp
= r1 − rMP,
d w1

1
d std(P ) −
= (std(P )) 2 (2w1var(C1) +
d w1

2(1 − w1) cov(C1, MP) − 2 w1cov(C1, MP) +

2(1 − w1) var(MP)).

For the derivative in the return risk graph in the market portfolio,
i.e. w1 = 0 we find
 213

d rp
d rp d w1
= =
d std(P ) w d std(P )
1= 0
d w1
w1= 0

r1 − rMP
.
cov( C1,MP ) − var(MP )
std(MP )

As the slope has to be same the slope (1), hence

r1 − rMP r −r
= MP f ,
cov( C1,MP ) − var(MP ) std(MP )
std(MP )

thus

⎡ cov( C1,MP ) − var(MP ) ⎤

r1 − rMP = (rMP − rf ) ⎢ ⎥=
⎣ var( MP ) ⎦

⎡ cov( C1,MP ) ⎤ ⎡ cov( C1,MP ) ⎤

rMP ⎢ −1⎥ − rf ⎢ −1⎥
⎣ var( MP ) ⎦ ⎣ var( MP ) ⎦

which yields (2).

◊

Remark 4.6: β(P, MP) is called the CAPM beta. We

summarize the CAPM as follows

The excess return of a risky investment is equal to the CAPM

beta times the excess return of the market portfolio return.
 214

We conclude that the risk asset can be classified by the CAPM

beta.

The CAPM is one of the fundamental buildung blocks of

MPT. For a qualititive description we refer to [9, Feibel, page
191] and for a more quantitive formulation we refer to [4,
Copeland]. CAPM has the following properties:

• (2) is linear, i.e. (2) is also valid for a portfolio P with return

n
rP = ∑ w jr j = w1 r1 + w2 r2 + ….. + wn rn,
j=1

i.e.

rP − rf = β(P, MP) (rP − rMP) (4.3.3a)

where

cov(P,MP )
β(P,MP) = . (4.3.3b)
var(MP )

• The CAPM β(P,MP) is consistent with the least square β in

(3.3.9).

• For the market portfolio we have

cov(MP,MP ) var(MP )
β(MP,MP) = = = 1.
var(MP ) var(MP )

We note that there are also other portfolios that have

β(P, MP) = 1. These are tracking portfolios of MP, i.e.
Portfolios that are supposed to follow the market.
 215

• An investment that has no risk will earn the risk-free return,

where risk is measured by the volatility of returns.

• Two different types of risk causes the volatility in investment

returns. The first is the market risk of the investment,
which reflects the degree to which investment values vary
when the level of prices in the underlying market changes.
Volatility that affects the market as a whole is assumed to
correspond to changes in the underlying factors that
influence market prices in general. For example, if we
assume that stock valuations respond positively to increases
in the growth rate of the economy as a whole, we can say
that the economic growth factor is to some degree common
to all stocks. Economic growth, and other factors which
influence the value of market prices as a whole, are
systematic risk factors.

• Systematic risk factor are reflected in the market returns,

therefore, we can isolate the influence of systemtic factors
on an individual asset by observing market returns.

• All investment within the market are influenced by

systematic risk, but the degree of exposure to systematic
risks varies from asset to asset.

• Factors specific to the asset can generate volatility, for

example, a change of mangement within a company. This
type of volatility is unique to the asset, or unsystematic.

• Of the two components of the risk, those risks specific to

particular assets can be eliminated via portfolio
diversification. We assume that the particular risks of
different assets will offset each other in a diversified
investment fund.
 216

• Because the unique risk is diversifiable, the market does

not reward the asset with a premium for this risk. The
market rewards only the exposure to systematic risk. This is
because systematic risk cannot be diversified away and is
thus borne by every investor in the market.

• Investors expect a risk premium in exchange for bearing

this risk.

• Investment are awared a degree of return over the risk-free

rate, or a risk premium, based on the degree of market
risk. The reward to risk ratio is a linear function: For each
unit of systematic risk the asset is exposed to the market,
the market will reward the asset with one unit of excess
return.

• An investment that has risk exposure similar to the market

taken as a whole will have returns that vary in line with the
market. If an asset is less exposed to market factors, its
returns will vary to a lesser degreee as the market as a
whole. If the investment has greater exposure to systematic
factors, returns will vary to a greater degreee than the
market returns.

The graphical representation of the CAPM in the security

market line. In Figure 4.7 we have the β version and in Figure
4.8 the covariance version.
 217

rP

rMP
M

rf

β ( P , MP ) = 1 β(P,MP)
Figure 4.7

rP

rMP
M

rf

β ( P , MP ) = 1 β(P,MP)
Figure 4.8

We mention here the dawn back of the CAPM in [4,

Copeland, page 217], referring to [20, Ross].
1. The only legimate test of the CAPM is whether or not the
market portfolio (which includes all assets) is mean -
variance efficient.
2. If performance is measured relative to an index that is
expost efficient, then from the mathematics of the efficient
ser no security will have abnormal performance when when
mearued as a departure from the security market line.
3. If performance Is measured relative to an ex post inefficient
index, then any ranking of portfolio performance is possible,
depending on which inefficient index has been chosen.
 218

One application of the CAPM beta is the calculation of the

Treynor Ratio. The Treynor ratio can be used to rank the
desirability of a particular asset in combination with other assets,
where part of the total risk inherent in the standard deviation will
be diversified. The Treynor Ratio is the return in excess of the
risk-free rate divided by the beta.

Definition 4.11: The Treynor ratio TR is the difference

between the annual arithmetic mean return and annual
arithmetic mean risk-free return divided by the annualized Beta
of the fund return, i.e.

rP − rf
TR = .
β(P,MP )

The CAPM can be applied for forecasting the return of a

asset or a portfolio or measuring the deviation α(P,MB) of an
asset or a portfolio to the relation (2), i.e.

r1 = α(C1, MP) + rf + β(C1, MP) (r1 − rM)

and by (3)

rP = α(P, MP) + rf + β(P, MP) (rP − rM).

This motivates the following definition:

Definition 4.12: The CAPM Alpha or Jensen’s Alpha

α(P, MB) is the factor that reconciles actual return to those
predicted by the CAPM defined by

α(P, MP) = rP − rf − β(P, MP) (rP − rM).

In this regression rM is the independent variable and rP is the

dependent variable. If α is greater than zero, the fund has a
 219

return higher than expected by the CAPM. A negative α

indicates that the fund performed worse than predicted given
the market risk taken.

4.5. Relative adjusted measures

Tracking error and down side risk is useful when measuring

the active risk. In this section we present two measures that
relate relative return to relative risk return.

Definition 4.13: The Sortino ratio SR is the ratio

between the annual average difference and the target return T
provided by the benchmark and the annualized downside
deviation, i.e.

SR =
(rP − T ) ⋅ f
1 N

N
∑ (rj − T)2 ⋅ f
k =1
rj ≤ T

where f is the periodicity of the data per year.

Definition 4.14: The Information ratio IR is a measure of

the benchmark relative return gained for taking on benchmark
relative risk

1 N
(
∑ r − r B,k
N k =1 P,k
)
IR =
TE(1)

and

1 N
(
∑ r − rB,k
N k =1 P,k
)
annualized information ratio = f
TE(1)
 220

where f is the periodicity of the data per year.

The information ratio has the same role for relative

considerations as the Sharpe ratio has for absolute return
consideration.
 221

5. INVESTMENT CONTROLLING

5.1. Investment controlling and the investment process

5.1.1. Introduction

Nowadays it is more and more important for an asset

management company to have efficient and appropriate
management information on the performance of their discretionary
managed portfolios. Without decision-oriented information on the
performance or quality of its products and/or asset managers a
specific asset management company will find it more and more
difficult to stand the increasing challenges of the asset
management industry in future. Clients and consultants have
similar needs where these often correspond to the asset manager
one’s some years ago. Investment controlling deals with such
needs and will be explained in detail in the following.

5.1.2. Definitions

Investment controlling is an area of activity that is part of the

overall controlling within the asset management and is an
important component of the recurring investment decision making
process.

Definition 5.1 (general): From an asset management

company point of view, investment controlling is defined as
information management whose task it is to gather, to process, to
check and to distribute information necessary to meet the overall
objectives of the asset management company.

In this respect the investment controlling objective consists in

configuring the infrastructure – particularly within the framework of
 222

the investment decision making process – in such a way that the

processes (e.g. forecasting, decision making and implementation),
the quality and the results (e.g. returns), the risks (e.g. of using
derivatives) and the costs become more transparent and
comprehensible.

Definition 5.1’ (specific) : Considering the client

perspective, in the following investment controlling is in general
defined as independent monitoring of the performance of asset
management products and/or accounts with the aim of ensuring
that the client gets what was promised in the first place with
respect to quality and performance.

The investment decision making process illustrated

schematically in Figure 4.1 may in reality be highly complex due to
the large number of possible investment instruments, of decision
makers involved (such as the research team, asset allocation team
and specialists in the various investment categories) as well as of
not always transparent levels of the decision making. The effect of
this complexity is that the results of the investment process and
its determining factors are not always apparent. In this context,
investment controlling intents to visualise the contributions of the
individual decisions of the investment process, especially with
respect to return and risk, and to allocate them to the responsible
decision makers. The results and conclusions of the different
investment controlling activities are an important feedback and
input into the investment process to enhance the quality or
performance of the specific asset management product.
Investment controlling is an important part of the investment
process. It is the last step of the investment process and analyses
the result of the overall process - the account performance - with
respect to the sources of return as well as of risk. These results
 223

are again input for the investment process and therefore the
starting point for the ongoing investment process. Furthermore as
illustrated in Figure 5.1 investment controlling may analyse all
steps of the investment process starting from the definition of the
investment target up to the re-balancing of the actual portfolio.
As shown in Figure 5.1, depending on its setup, investment
controlling not only encompasses pure controlling activities but also
compliance and risk management tasks as well as focuses not
only on asset managers but is also concerned with the impact of
consultants and may be also of that of the client. In Switzerland,
the most know and influential consultants are Ecofin,
Complementa Investment - Controlling AG and PPCmetrics.
Internationally we mention Russel and Wyson Watt.
If the investment controlling concentrates on performance or
quality related aspects and does not address financial aspects one
could also speak about performance monitoring. In the following
we use performance monitoring and investment controlling
interchangeable.

Risk management Compliance

(non-investment)

Investment
controlling
Performance
Finance monitoring
costs/revenues

Reporting
(internal and external)

Forecasts Processes Behaviour Results

Different facets of investment controlling Figure 5.1

 224

5.1.3. Purpose and objectives

Form a general point of view investment controlling adds to

the visibility, transparency and credibility of any asset management
company. In detail investment controlling helps

• Implementing best practice in performance measurement and

performance presentation, for example by implementing the
GIPS standards
• Producing an independent performance analysis of the asset
management accounts and/or products
• Enabling deep level analysis which is necessary to identify the
real drivers of the account return and account risk and this from
an ex-post as well as from an ex-ante point of view
• Monitoring risk and return of accounts and/or products against
their designated benchmark and objectives, capturing
performance dispersion
• Reducing unnecessary discussions by using more objective and
less subjective information during the performance review
• Creating of or increasing the transparency and comparability of
the asset management products and/or accounts
• Addressing performance issues on an regular basis and not
leaving them running
• Creating a basis not only for ongoing analyses but also for
structural changes in the investment process
• Reducing of unintended business risks through early addressing
of potential performance issues
 225

5.1.4. Investment controlling activities

Following the different aspects of the investment controlling

one can easily imagine that investment controlling is very manifold
and encompasses a lot of different activities like:

• Performance attribution or more precisely return attribution

and/or risk attribution and this ex-post as well as ex-ante
• Market index and benchmark comparisons, composite
dispersion analysis and peer group analysis with respect to the
return and/or risk but also to characteristics like asset class or
sector weights, duration, exposure to specific risk factors and
so on
• Calculation of performance figures and statistics representing
manager skills or the investment style and running style analysis
• Review of the set up of the specific asset management account
with respect to benchmark, investment guidelines, transaction
costs and management fees, etc.
• Product review against client expectations and best practice
• Identifying of actual and potential performance issues and
highlighting the serious ones to senior management
• Suggesting remedial action to solve performance issues
• Risk decomposition and risk budgeting
• Analyzing and identifying all steps of the investment process
• Review of the investment guidelines and benchmarks
• Checking whether risk levels and limits are appropriate
• (Aggregated) performance reporting to senior management
 226

5.1.5. Necessary resources

To do an efficient and meaningful investment controlling a lot

of requirements has to be fulfilled whereas it is very seldom that all
of them are met:

• Investment controllers with both general controlling and asset

management background or know-how (in theory and practice)
• Availability of all needed data in good quality within a reasonable
timeframe
o Daily data (holdings and prices)
o On stock level
o For the account as well as for the benchmark
o On forecasts as well as tactical asset allocations or actual
transactions

• Peer group information

• Availability of appropriate analytical tools to run performance

attribution or simulations as well as tools to maintain composites
or to do the performance reporting

• Asset management agreements or mutual funds prospects and

investment guidelines
Without a critical amount of resources and without the relevant
data an investment controlling can not be efficient and meaningful.
For example, running a risk attribution for a Swiss fixed income
portfolio with a risk analytics tool that does not incorporate a Swiss
credit model may be risk decomposition but is still useless from an
investment controlling point of view because it does not reflect the
investment process.
 227

5.2. Building blocs and set up of investment controlling

5.2.1. Introduction

Following a general controlling setup the investment

controlling process or in other words the performance monitoring
process can be split up as illustrated in following figure 5.2.
Production / reporting ( quantitative aspect) Portfolio analytics / risk control (qualitative aspect)

MonitoringMonitoring
of results of
(exresults
--post) (ex-post)
and of Inputs
snd (ex
of inputs
-ante)(ex-ante)

Performance Performance Performance Performance Performance Portfolio Performance

measurement administration reporting analysis watch list analytics review

Feed
Feed
forward
forward
and
and
feed
feed
back
back

Efficient controlling of forecast and of investment process outcome

The investment controlling or performance monitoring process

Figure 5.2

The first four steps are quantitatively oriented and are used
for calculating, maintaining or storing, visualising and analysing
different performance figures of a specific account or a specific
asset management product or composite. The last three steps are
less production oriented and instead deliver qualitative statements
on the investment process and its results. The performance watch
 228

list determines problematical asset management accounts and/or

products that are subsequently analysed by portfolio analytics and
are discussed in depth in the performance review. In this respect,
working out proposals for improvement and pointing out possible
consequences for the investment process are the primary
objectives of the qualitatively oriented analyses conducted in
steps 5 to 7. In the following we illustrate the investment
controlling or the performance monitoring process whereas for
more details we refer to the other relevant chapter of the course.
The purpose of this section is to get an overview of the different
parts of the performance measurement and analysis and to see
how to incorporate them to one general investment controlling
framework. We proceed by the different steps of the performance
monitoring process

5.2.2. Different steps of the performance monitoring

process

a. Performance measurement
This first step of the performance monitoring process deals
with all aspects of return and risk measurement, i.e. the calculation
of all necessary return and risk measures or figures like gross and
net account returns, time weighted and money weighted rate of
returns, risk figures such as volatility or tracking error and so on.
Performance measurement normally focuses on the total account
level and is a time series analysis. Figure 5.3 shows an example
for a performance measurement report showing all the necessary
input data to calculate a total return gross and net.
 229

b. Performance administration
Performance administration normally covers the
benchmark calculation and more importantly the composite
construction and maintenance.

Sample performance measurement report Figure 5.3

Definition 5.2: A composite is an aggregation of one or

more portfolio managed according to a similar investment
mandate, objective, or strategy.
The composite return is the asset weighted average of the
performance of all portfolios in the composite. Creating
.meaningful composites is essential to the fair presentation,
consistency, and comparability of performance over time and
among firms.
 230

In order to assess the asset management skills of one or more

asset managers or even of a whole firm it is necessary to classify
the different accounts and assign accounts with same
characteristics like same benchmarks, same investment strategies
and/or styles to a composite as illustrated in Figure 5.4.
Constructing and maintaining composites according to the GIPS
standards may also be part of the performance administration.
Being in compliance with the GIPS standards means best practice
with respect to transparency and comparability in the area of
performance presentation and is seen as the basis for an efficient
investment controlling of an asset management company.

Idea of composite construction Figure 5.4

c. Performance reporting
This step of the performance monitoring process includes the
reporting of different performance figures for specific time periods
- normally on a total account level. If an observer is interested in
more detailed information on the sources of return and risk, he
may start reviewing the performance figures on a total account
 231

level and afterwards analyses a performance attribution which will

be explained in the next section.
The following Figures 5.5 to 5.8 illustrate a sample
performance reporting for a specific equity composite where the
layout and the chosen analysis do not have to be composite or
account specific. The report setup is normally client or user
specific and varies from observer to observer. In practice
depending on the size of the asset management company these
kinds of reports are produced by a spreadsheet or an external
software solution. Considering the external software solution it is
often the case that the composite maintenance software and the
reporting software is part of a single software solution.
In setting up a performance report one has to define a lot of
reporting characteristics like:
• Which product, composite or account should be analysed?
• Which time period should be considered?
• Whether gross or net return should be reported?
• Which return and risk measures should be shown?
• Whether rolling and/or annual performance figures should be
presented?

It should be noted that it is crucial for getting the whole picture

to analyse the account or composite performance from different
aspects since performance figures are very sensitive to used
methodology, time periods and input data. This means that varying
the time period for example only for one month forward may lead
to the fact that a very underperforming account transfers to a very
outperforming account. Another example may be the choice of the
risk free rate if calculating the Sharpe Ratio. In continental Europe
it is common to use the relevant total return of the alternative risk
free investment and in other countries it is more common to use
the actual risk free rate. Depending on the chosen type of risk free
 232

rate you may come up with totally different conclusions. Important

is here especially whether you use the performance figure as a
measure to compare past performance or to estimate and rank
future performance or in other words whether you do an ex-post or
a ex-ante performance analysis.

Sample performance report 1 Figure 5.5

 233

Equities World BM MSCI active Mandates direct

Benchmark MSCI World (ri) in CHF Series Type Asset Weighted Gross Return Reporting Currency CHF
No. of A/Cs 5 Inception Date 01 Jan 1997 Market Value (m) End of Period 84.27
Composite Code ZU-COMP250 Reporting Date 31 Dec 2003

Indexed Cumulative Relative Returns

Periodical Returns in %

Composite Benchmark Relative 135

1 Month 1.67 1.69 -0.03 130
3 Months 6.60 7.11 -0.50
6 Months 8.43 9.58 -1.15 125
1 Year 17.98 19.64 -1.66 120

Indexed Returns
2 Years -12.65 -10.46 -2.18
3 Years -13.77 -11.82 -1.96 115
4 Years -13.70 -11.83 -1.87
110
5 Years -2.33 -2.46 0.13
Since Incep. 4.05 3.93 0.12 105
100
Calendar Year Returns in %
95
Composite Benchmark Relative
90
YTD 17.98 19.64 -1.66 Incep Oct 97 Aug 98 Jun 99 Apr 00 Feb 01 Dec 01 Oct 02 Aug 03
2003 17.98 19.64 -1.66
2002 -35.32 -32.99 -2.33
2001 -15.99 -14.47 -1.52
2000 -13.46 -11.84 -1.62 Monthly Relative Returns
1999 60.21 46.07 14.14
1998 21.29 17.51 3.78
1997 22.50 26.25 -3.75 4

3
Annual Risk Figures in % Composite Benchmark
2
Volatility over 1 Year 16.90 15.80
Volatility Since Inception 22.98 20.80 1
Sharpe Ratio over 1 Year 1.06 1.24
in %

Sharpe Ratio Since Inception 0.10 0.10 0

Tracking Error over 1 Year 2.12 N/A
Tracking Error Since Inception 4.69 N/A -1
Information Ratio over 1 Year -0.78 N/A
Information Ratio Since Inception 0.03 N/A -2
Correlation over 1 Year 0.99 N/A
-3
Correlation Since Inception 0.98 N/A
-4
Jan 97 Nov 97 Sep 98 Jul 99 May 00 Mar 01 Jan 02 Nov 02 Sep 03

Sample performance report 2 Figure 5.6

E q u itie s W o rld B M M S C I a c tiv e M a n d a te s d ire c t

B e n c h m a rk M S C I W o rld (ri) in C H F P e rfo rm a n c e T y p e A s s e t W e ig h te d G ro s s R e tu rn B a s e C u rr e n c y CHF
N o . of A /C s <5 In c e p tio n D a te 01 Jan 1997 M a rk e t V a lu e (m ) 8 4 .2 7
C o m p o s ite C o d e Z U -C O M P 2 5 0 D a te 31 D ec 2003

80 30

60 28

26
40
1 Year Total Return

1 Year Volatility

24
20
C o m p o s ite C o m p o s ite
22
0 B e n c h m a rk B e n c h m a rk
20
-20
18

-40 16

-60 14
97

02

97

02

2
8

99

3
r0

r0
v9

l9

y0

v0

v9

y0

v0
p9

p0

p9

p0
Jul
Jan

Jan

Jan

Jan
Ju

Ma

Ma
No

No

No

No
Ma

Ma
Se

Se

Se

Se

20 8 6
5
7
15 4
6
1 Year Information Ratio

3
1 Year Excess Return

1 Year Tracking Error

10
5 2

5 4 1
0
3
0 -1
2 -2
-5
1 -3
-4
-10 0
7

2
8

99

02
97

3
v9

y0

v0
p9

r0

p0
7

02

97

02

2
97

Ja n
Jan

Jul

Ma
v9

v0

v9

r0

v0

No

No
p9

l9

y0

r0

p0

p9

l9

y0

p0

Ma
Se

Se
Jan

Jan

Jan

Jan
Ju

Ju
Ma

Ma
No

No

No

No
Ma

Ma
Se

Se

Se

Se

Sample performance report 3 Figure 5.7

 234

Equities World BM MSCI active Mandates direct

Benchmark MSCI World (ri) in CHF Performance Type Asset Weighted Gross Return Base Currency CHF
No. of A/Cs <5 Inception Date 01 Jan 1997 Market Value (m) 84.27
Composite Code ZU-COMP250 Date 31 Dec 2003

30 30

20 25
Total Return (ann. if > 1 year)

10
20

Volatility (ann.)
0 Composite Composite
Benchmark 15
Benchmark
-10
10
-20

5
-30

-40 0
1 Month 3 Months 6 Months 1 Year 3 Years 2003 2002 1 Year 2 Years 3 Years 2003 2002 2001

0 6 0

-0.1
-0.5 5
Excess Return (ann. if > 1 year)

-0.2

Information Ratio (ann.)

Tracking Error (ann.)

4 -0.3
-1
-0.4
3
-1.5 -0.5

2 -0.6
-2 -0.7
1
-0.8
-2.5
1 Month 3 6 1 Year 3 Years 2003 2002 0 -0.9
Months Months 1 Year 2 Years 3 Years 2003 2002 2001 1 Year 2 Years 3 Years 2003 2002 2001

Sample performance report 4 Figure 5.8

d. Performance contribution and attribution

Performance contribution and attribution is introduced and
explained by model example in Chapter 2 and 3. Is a central
component of the performance monitoring process and it is
defined as a process that determines the return and risk
contributions of the individual decision making steps within an
investment process. Thus, performance attribution is concerned
not only with the past but also with the future, and determines
which return and risk contributions are due to which decisions
(regarding investment category and instruments) and to which
decision makers on an ex-post as well as ex-ante basis. Figure
5.9 illustrates the various levels of analysis of performance
attribution as well as possible allocation criteria of return and risk
contributions. It is evident that performance attribution can be
 235

carried out in a variety of different ways. On the one hand, return

and risk contributions may be calculated on an absolute basis, i.e.
isolated for an asset management account or for a specific
benchmark, or on a relative basis, i.e. for an asset management
account in comparison to a benchmark. On the other hand, the
performance attribution may be focused on the past (ex post) or
the future (ex ante). In summary, performance attribution is
defined as the decomposition of historical or expected absolute or
relative return and/or historical and expected absolute or relative
risk.

Levels of analysis and allocation criteria of performance analysis

Figure 5.9

As seen in Figure 5.9, in general performance attribution can be

used to measure the return and risk contributions of categories,
sectors and instruments (e.g. asset categories, countries,
currencies or securities), of decision makers such as the client
 236

himself, portfolio managers or consultants, and finally of asset

management activities like the definition of the benchmark, the
strategic or tactical asset allocation or the stock picking.
Performance attribution has been developed since 1980’s. In
the early days, the return and risk contributions were at first
considered at an aggregated level – following a top-down
approach. Particularly in the recent years, more detailed and more
profound analyses were developed. It is increasingly possible to
conduct performance attribution on a stock level and on a daily
basis. In parallel with this development, the focus of performance
attribution has shifted away from returns towards risk and risk-
adjusted returns. However, depending on the investment category
considered and the investment instruments used, risk analysis and
risk-adjusted return analysis still require considerable further
development. The development of the performance measurement
and attribution is illustrated in Figure 5.10.
In the following we present a sample performance attribution
report which is split up into 6 parts (Figures 11 to 16):
• Overview page covering aggregated return and risk information
• Ex-post return decomposition according to sectors
• Ex-post management effects according to sectors
• Ex-post return decomposition according to countries
• Ex-post management effects according to countries
• Ex-ante risk decomposition
 237

Status quo /
best practice
time

Risk ???

Benchmark
Derivatives
"correct" return Stocks
methodology From monthly
Return From Sectors to daily to real
quarterly to time
Absolute monthly Asset classes
profit / countries
from ex post to ex ante

Detailed analysis
Portfolio view
(Return and risk decomposition)

Performance measurement Performance attribution

Status quo of performance attribution Figure 5.10

Attribution by MSCI Attribution by 5 World
Portfolio Return Attribution Effects
Sector Regions

NAME (ID) Composite World MSCI, active, mandates Currency CHF Asset Allocation -0.96% -1.01%
PM AMPE Return Portfolio 18.28% Stock Selection 0.13% -0.59%
BENCHMARK MSCI World in CHF Return Benchmark 19.60% Interaction -0.49% 0.28%
PERIOD 31.12.2002 - 31.12.2003 Return Relative -1.32% Total -1.32% -1.32%

(Average over-/underweight) Attribution Analysis - by MSCI Sector

Asset Allocation Stock Selection
1.5% 1.5%
1.0% 1.0%
0.5% 0.5%
0.0% 0.0%
-0.5% -0.5%
-1.0% -1.0%
-1.5% -1.5%
y

gy

es
s

gy

ls
e

h
ls

gy
y
ar

le

ls

es
s

s
ls

l
y
al

as
itie

h
ar

ta
ria
ia

r
lo

le

ti e
ic

ar
er

as
ria
na

ia

ia
ap
n

ri

ic
C
nc

To
er
rv

ap
no
io

til
e

ol
st

er
nc
En

tili
rv

C
st
io
St

En
at
lth

Se
na

n
et

U
du

St

at
lt h
ch

U
na

du

Se
et

ch
M
cr

er

ea

M
Fi

cr
In

ea
Te

er

Fi
m

In

Te
is

m
m

is
H

m
m

H
D

m
su

D
n

n
su
co
io

co
er

io
er
on

on
at

le

at
m

le
m
C

Te

rm
C
su

Te
su
or

fo
on

on
f
In

In
C

Risk Analysis (end period) Portfolio Benchmark Attribution Analysis - by 5 World Regions
Number of Securities 67 1'550 Asset Allocation Stock Selection
Number of Currencies 8 0 0.6%

Portfolio Value 84'334'091 0.4%

Total Risk (ex-ante) 18.81% 18.21% 0.2%

- Factor Specific Risk 18.66% 18.18% 0.0%

-0.2%
- Stock Specific Risk 2.39% 1.02%
-0.4%
Tracking Error (ex-ante) 2.57%
-0.6%
Value at Risk (at 95%) 3'570'469
-0.8%
R-squared 0.98
-1.0%
Beta-adjusted Risk 18.64% 18.21%
-1.2%
Predicted Beta 1.02 Asia ex Japan Europe Japan North Am erica Cash Total

Predicted Dividend Yield 1.86 2.01

P/E Ratio (E: 12 months) 28.39 26.00 Important Remark: Differences between attribution returns and the returns of the official performance measurement tool are usual. They can be
P/B Ratio (B: year-end) 2.58 2.56 explained by the two systems using two different methodologies and by intraday trading gains or losses. Above figures are subject to future changes.

Sample performance attribution report part 1 Figure 5.11

 238
Average Average Relative Absolute Absolute Difference Difference
by MSCI Sector
PF Weight BM Weight Weight PF Return BM Return PF-BM BM-BM Tot
Consumer Discretionary 12.03% 12.18% -0.15% 15.58% 23.07% -7.49% 3.47%
Consumer Staples 8.83% 9.26% -0.43% 7.74% 5.09% 2.65% -14.51%
Energy 8.29% 7.41% 0.88% 10.51% 13.30% -2.79% -6.30%
Financials 21.70% 22.94% -1.24% 23.90% 24.94% -1.04% 5.34%
Health Care 11.73% 12.53% -0.80% 9.60% 7.31% 2.29% -12.30%
Industrials 10.45% 9.71% 0.74% 34.93% 24.08% 10.85% 4.48%
Information Technology 12.33% 12.49% -0.16% 25.06% 32.79% -7.73% 13.19%
Materials 4.84% 4.53% 0.31% 23.11% 30.33% -7.22% 10.73%
Telecomm Services 6.15% 5.24% 0.91% 15.02% 12.64% 2.38% -6.96%
Utilities 2.44% 3.72% -1.28% 34.09% 15.45% 18.64% -4.15%
Cash 1.21% 0.00% 1.21% 8.94% 0.00% 8.94% -19.60%
Total 100.00% 100.00% 0.00% 18.28% 19.60% -1.32% 0.00%

Import remarks:
- The weight numbers gives an overview of the average weight invested in the different groups (e.g. sectors) with daily weights averaged over the chosen period.
- The absolute return numbers give an overview of the group's absolute performance (e.g. sector) within thc chosen time period (portfolio and bechmark)
- The difference PF-BM compares the group's performance of the portfolio to the group's performance in the benchmark within the chosen time period.
- The difference BM-BM Tot compares the group's performance in the benchmark with the performance of the total benchmark within the chosen time period.
Please note that cash is included in the relevant country or regional groups and shown as "other assets" else (e.g. in sectors). Concerning derivatives, only futures are split up at the
moment; derivatives (call or put options) are not yet included. Moreover, illiquid securities as private placements are not yet taken into account.

Sample performance attribution report part 2 Figure 5.12

Asset S to c k
b y M S C I S e c to r In te ra c t io n T o ta l
A llo c a t io n S e le c t io n
C o n s u m e r D is c re tio n a ry 0 .0 0 % -0 .8 6 % -0 .0 3 % -0 .8 9 %
C o n s u m e r S ta p le s 0 .0 5 % 0 .2 8 % 0 .0 3 % 0 .3 6 %
E n e rg y 0 .0 0 % -0 .1 9 % -0 .0 2 % -0 .2 1 %
F in a n c ia ls -0 .1 0 % -0 .2 1 % 0 .0 1 % -0 .3 0 %
H e a lth C a re 0 .2 0 % 0 .3 5 % -0 .0 3 % 0 .5 2 %
In d u s tria ls -0 .0 4 % 0 .9 9 % 0 .1 2 % 1 .0 7 %
In fo rm a tio n T e c h n o lo g y -0 .2 1 % -0 .9 1 % 0 .0 1 % -1 .1 1 %
M a te ria ls 0 .0 1 % -0 .2 8 % -0 .0 8 % -0 .3 5 %
T e le c o m m S e rv ic e s -0 .1 1 % 0 .2 1 % -0 .0 8 % 0 .0 2 %
U tilitie s 0 .0 2 % 0 .7 4 % -0 .4 1 % 0 .3 5 %
C ash -0 .7 8 % 0 .0 0 % 0 .0 0 % -0 .7 8 %
T o ta l -0 .9 6 % 0 .1 3 % -0 .4 9 % -1 .3 2 %

Im p o r t re m a r k s :
A s s e t A llo c a t io n E ff e c t is th e p o rtio n o f p o rtfo lio e x c e s s re tu r n th a t is a ttrib u ta b le to ta k in g d iffe re n t g ro u p b e ts fro m th e b e n c h m a rk . A n o ve rw e ig h t o f a g ro u p
( e .g . S P I s e c to r "C h e m ic a ls " ) th a t o u tp e rfo rm s th e w h o le b e n c h m a rk ( e .g . S P I) w ill g e n e ra te a p o s itiv e a s s e t a llo c a tio n e ffe c t.
S e c u rity S e le c tio n E ffe c t is th e p o rtio n o f p o rtfo lio e x c e s s re tu r n a ttrib u ta b le to c h o o s in g d iffe re n t s e c u r itie s w ith in g ro u p s fro m th e b e n c h m a rk . A n o v e rw e ig h t
o f a w e ll-p e rfo rm in g s e c u rity ( e .g . N o v a rtis ) in c o m p a ris o n to its g r o u p b e n c h m a rk ( e .g . S P I s e c to r " C h e m ic a ls " ) w ill g e n e ra te a p o s itiv e s to c k s e le c tio n e ffe c t.
In t e ra c tio n E ffe c t is th e p o rtio n o f th e p o rtfo lio e x c e s s r e tu rn w h ic h is n o t a ttrib u ta b le to a s s e t a llo c a tio n n o t s to c k s e le c tio n .

Sample performance attribution report part 3 Figure 5.13

 239
Average Average Relative Absolute Absolute Difference Difference
by 5 World Regions
PF Weight BM Weight Weight PF Return BM Return PF-BM BM-BM Tot
Asia ex Japan 1.86% 3.16% -1.30% 12.52% 31.51% -18.99% 11.91%
Europe 27.62% 28.92% -1.30% 25.40% 24.36% 1.04% 4.76%
Japan 9.83% 8.74% 1.09% 20.16% 21.75% -1.59% 2.15%
North America 59.48% 59.17% 0.31% 15.99% 16.38% -0.39% -3.23%
Cash 1.21% 0.00% 1.21% 8.94% 0.00% 8.94% -19.60%
Total 100.00% 100.00% 0.00% 18.28% 19.60% -1.32% 0.00%

Import remarks:
- The weight numbers gives an overview of the average weight invested in the different groups (e.g. sectors) with daily weights averaged over the chosen period.
- The absolute return numbers give an overview of the group's absolute performance (e.g. sector) within the chosen time period (portfolio and bechmark)
- The difference PF-BM compares the group's performance of the portfolio to the group's performance in the benchmark within the chosen time period.
- The difference BM-BM Tot compares the group's performance in the benchmark with the performance of the total benchmark within the chosen time period.
Please note that cash is included in the relevant country or regional groups and shown as "other assets" else (e.g. in sectors). Concerning derivatives, only futures are split up at the
moment; derivatives (call or put options) are not yet included. Moreover, illiquid securities as private placements are not yet taken into account.

Sample performance attribution report part 4 Figure 5.14

A ss e t S to c k
b y 5 W o rld R e g io n s In te ra c tio n T o ta l
A llo c a tio n S e le c tio n
A sia e x Jap an -0 .1 0 % -0 .5 6 % 0 .3 0 % -0 .3 6 %
E u rop e -0 .0 4 % 0 .3 4 % -0 .0 7 % 0 .2 3 %
Jap an -0 .0 9 % -0 .1 6 % -0 .0 8 % -0 .3 3 %
N o rth A m e ric a 0 .0 1 % -0 .2 0 % 0 .1 4 % -0 .0 5 %
C a sh -0 .7 8 % 0 .0 0 % 0 .0 0 % -0 .7 8 %
T o ta l -1 .0 1 % -0 .5 9 % 0 .2 8 % -1 .3 2 %

Im po rt re m arks:
A s s e t A llo c atio n E ffe c t is th e p ortio n of p ortfo lio e xc ess re turn th at is attribu tab le to takin g d iffe re nt g rou p b e ts from th e be n c hm ark. A n ove rw eig h t of a g roup
(e .g . S P I se c tor "C he m ic als" ) th at ou tp e rform s th e w h ole be n ch m ark (e.g . S P I) w ill g en erate a p ositive asse t allocation e ffe c t.
S e c urity S e lec tio n E ffe c t is th e p ortion of p ortfolio exc e ss re tu rn a ttribu tab le to c h oosin g d iffe ren t se cu rities w ith in g rou p s from the b en ch m ark. A n overw eig ht
of a w ell-p erform ing se curity (e .g . N ovartis) in com p arison to its g rou p b ench m ark (e .g. S P I sector "C he m icals") w ill generate a p ositive stock selec tion e ffe c t.
In tera c tio n E ffec t is th e p ortion of th e p ortfolio exc ess re tu rn w h ich is n ot attrib u tab le to asset alloc ation n ot stoc k se le c tion .

Sample performance attribution report part 5 Figure 5.15

 240

Risk Model: Global Portfolio Benchmark Risk Model: Global Portfolio Tracking Error

Number of Securities 67 1'550 Total Risk (ex-ante) 18.81% 2.57%

Number of Currencies 8 0 Factor Specific Risk 18.66% 1.50%

Portfolio Value 84'334'091 - Region 11.50% 0.18%

Total Risk (ex-ante) 18.81% 18.21% - Country 6.98% 0.83%

- Factor Specific Risk 18.66% 18.18% - Industry 2.64% 0.77%

- Stock Specific Risk 2.39% 1.02% - Fundamental 1.44% 0.78%

Tracking Error (ex-ante) 2.57% - Currency 8.42% 0.27%

Relative Value at Risk 3'570'469 - Covariance (+/-) 9.35% 0.52%

R-squared 0.98 Stock Specific Risk 2.39% 2.08%

Beta-adjusted Risk 18.64% 18.21%

Explication of risk model: Factor risk is a standard deviation that is
Predicted Beta 1.02 measured by multiplying the 5-year exposure of the components of a portfolio
to each risk factor and by multiplying these figures by the externally determined
risk of each factor. The Tracking Error is measured similarly except that it is the
Predicted Dividend Yield 1.86 2.01
difference between portfolio and benchmark exposure that is multiplied.
Specific Risk is the standard deviation that measures the volatility of the risk not
P/E Ratio (E: 12 months) 28.39 26.00
captured by the factor model. The model consists of 3 regional, 21 country, 38
industry and 8 fundamental factors (market cap, 4-year E/P growth, E/P, B/P, 5-
P/B Ratio (B: year-end) 2.58 2.56
year yield, long term debt, 5-year ROE variablity and 5-year earnings variability).

Sample performance attribution report part 6 Figure 5.16

e. Performance watch list

As the name says the performance watch list consists of
accounts or composites that should be watched and may be
reviewed. The reason why accounts or composites are put on the
watch list may be manifold and may be of quantitative but also of
qualitative nature such as a underperformance versus a benchmark
or a peer group, too much or too less risk, client complaints and so
on.
The determination of the watch list follows the process
illustrated in Figure 5.17. It starts normally with a mechanical filter
focusing on the historical and forward looking characteristics of the
products and of the individual accounts, ex-ante risk limits or risk
budgeting constraints as well as on the clients feedback. The
investment controlling committee decides in the next step on the
accounts or products that seem to be problematic and that go onto
 241

the performance watch list. Afterwards within the performance

review meeting each watch list account or product is analysed in
detail considering all kind of information from investment guidelines
up to an ex-ante risk break down. As a result of this performance
review corrective steps may be defined and implemented. If the
performance does not improve over longer time horizons serious
performance problems are reported to the senior management via
an escalation process.

Performance watch list process Figure 5.17

f. Portfolio analytics
Portfolio analytics provides deeper insight into the accounts
or products on the watch list by taking all the available quantitative
information produced in one of the preceding steps of the
performance monitoring process. Portfolio analytics is very often
also referred to as performance attribution.
 242

We distinguish between these two steps to highlight that the

performance attribution analysis on its own and without any
qualitative judgement is nothing else as a detailed performance
reporting. Through interpreting the performance figures by
considering all circumstances the performance attribution becomes
a meaningful management information tool. For the definition of
portfolio analytics we refer to the Introduction 1
In comparison performance attribution is the quantitative and
portfolio analytics the qualitative assessment of the quality of an
asset management portfolio or product.
What problems or pitfalls may occur during this step of the
performance monitoring process is explained in a later chapter. At
this point it should only be mentioned that performance attribution
or portfolio analytics bear quite a high risk of misinterpretations and
should be handled carefully.

g. Performance review
The performance review deals with the accounts and the
products which are on the performance watch list and tries to
analyse where the performance problems came from and whether
any corrective action is necessary or would help to bring the
product again on the right track. Within this step of the
performance monitoring process the portfolios or products are
analysed in detail considering all kind of information from
investment guidelines up to an ex-ante risk break down. The
following questions may be addressed within a typical performance
review (The figures in brackets refers to Figure 5.18):
• Where does the return come from and from which decisions do
they originate (1)?
• What risks have been taken (relative/absolute)?
• Is the choice of the benchmark still sensible? Are the general
circumstances still valid or reasonable (6)?
 243

• Are the investment guidelines still reasonable and have they

been respected (4)?
• What was the impact of the costs on the overall return (5)?
• Is the risk profile and risk budget still appropriate ((2) and (3))?
• What was the performance of the competitors?

5.2.3. Example of a performance review

In the following we present an example for a performance

review incorporating most of what was said or explained in the
above chapters. A performance review follows normally the
investment process of the specific portfolio or product and
analyses and reviews all steps of the investment process as
illustrated in Figure 5.18.
In this example the above mentioned composite “equities
world BM MSCI active direct” or as mentioned in the performance
attribution section the “composite world MSCI, active, mandates”
was put on the performance watch list due to it negative excess
return of -1.32% over the last 12 months, as of end of December
2003. The investment controller got the task to run a performance
review and to analyse why this product underperformed quite
substantially.
 244

Actual market data and

forecasts

3
1 Portfolio construction

Asset allocation process

4

Rebalancing
Investment
guidelines

2
Portfolio analytics /
Performance analysis

Investment target, risk profile and benchmark

Aspects of the performance review Figure 5.18

After analysing the different performance reports and

performance analysis the investment controller came up with the
following facts and conclusions (see Figures 5.14 to 5.16):
• The asset manager did not do big sector bets => maximum
overweight 1.28%
• The asset manager did not do big country bets => maximum
overweight 1.30%
• The asset manager did big security specific bets by investing
“only” in 67 out of 1550 securities in the benchmark
• The asset manager took quite a lot of security specific risk =>
stock specific risk of 2.08% versus factor specific risk of
1.50%
• Conclusion 1: the asset manager pursued a stock picking
approach with neutral sector and country bets!
 245

• The biggest return contributions come from

Ö Overweight in cash from March to May => - 0.78%
Ö Stock selection
ƒ Consumer discretionary => - 0.86%
ƒ Information technology => - 0.91%
ƒ Industrials => + 0.99%
ƒ Utilities => + 0.74%
Ö Interaction utilities => - 0.41%
• The biggest relative risk contribution come from the stock
specific risk => 2.08%

• Conclusion 2: Stock picking that did not pay out!

As described, the negative excess return comes mainly from the

cash bet in March until May which explained nearly 100% of the
asset allocation effect. In addition the underweight of utilities - an
outperforming sector - results in a quite high negative interaction
effect. One could conclude that the asset manager had quite bad
luck with his asset allocation decisions over the last 12 month. As
a result of this analysis the investment controller could conclude
that no remedial action is required and could decide to wait for
another period seeing whether the product recovers or not. If the
investment controller decides that remedial action is required his
suggestions could range from a change in the investment strategy
up to the replacement of the asset manager.
 246

2.0%

1.5%

1.0%

0.5%

0.0%

-0.5%

-1.0%

-1.5%
31.01.2003 28.02.2003 31.03.2003 30.04.2003 31.05.2003 30.06.2003 31.07.2003 31.08.2003 30.09.2003 31.10.2003 30.11.2003 31.12.2003
Total Monthly 0.04% 0.53% -0.57% 0.49% -0.99% -0.08% 0.49% -0.06% -0.83% -0.15% -0.14% -0.14%
Asset Allocation Monthly -0.12% 0.02% -0.11% -0.63% -0.12% 0.01% 0.05% -0.15% -0.17% -0.07% -0.02% 0.08%
Stock Picking Monthly 0.47% 0.47% -0.51% 1.23% -1.00% -0.05% 0.41% 0.03% -0.67% 0.04% -0.05% -0.18%
Interaction Monthly -0.31% 0.04% 0.05% -0.11% 0.13% -0.04% 0.03% 0.06% 0.01% -0.12% -0.07% -0.04%
Total Cummulated 0.04% 0.53% -0.04% 0.49% -0.17% -0.25% 0.26% 0.21% -0.80% -0.98% -1.12% -1.32%
Asset Allocation Cummulated -0.12% -0.10% -0.21% -0.80% -0.52% -0.54% -0.51% -0.70% -0.90% -1.03% -1.05% -0.96%
Stock Picking Cummulated 0.47% 0.91% 0.40% 1.66% 0.63% 0.62% 1.08% 1.16% 0.32% 0.41% 0.36% 0.13%
Interaction Cummulated -0.31% -0.28% -0.23% -0.37% -0.28% -0.33% -0.31% -0.25% -0.22% -0.36% -0.43% -0.49%

Management effect over time Figure 5.19

25%

20%

15%

10%

5%

0%
Jan 03 Feb 03 Mrz 03 Apr 03 Mai 03 Jun 03 Jul 03 Aug 03 Sep 03 Okt 03 Nov 03 Dez 03

Consumer Discretionary Consumer Staples Energy Financials

Health Care Industrials Information Technology Materials
Telecomm Services Utilities Other Assets

Sector weights over time Figure 5.20

 247

3.0%

2.5%

2.0%

1.5%

1.0%

0.5%

0.0%
Jan 03 Feb 03 Mrz 03 Apr 03 Mai 03 Jun 03 Jul 03 Aug 03 Sep 03 Okt 03 Nov 03 Dez 03

Total Tracking Error Factor Specific a) Region b) Country c) Industry

d) Fundamental e) Currency f) Covariance Stock Specific

Ex-ante risk contributions over time Figure 5.21

 248

5.3. The benefits of GIPS reporting

As mentioned in the previous chapter GIPS composites are

the basis for an efficient investment controlling of an asset
management company. In this chapter we will explain the need for
being in compliance with the GIPS standards (Global Investment
Performance Standards) and show that the value added is not only
limited to the asset manager but is also to the client’s benefit. In
other words we will discuss the performance of an asset manager
from the senior management’s as well as from a client’s point of
view. For details on the GIPS standards or on specific country
versions like SPPS (Swiss Performance Presentation Standards),
please see the relevant chapters beforehand [2, CFA].
The case for the GIPS standards is best explained by looking
at Figure 6.1. Assuming that a client is searching for an asset
manager to manage a specific discretionary account and got just a
performance number back from the different marketing officers. In
the sample below asset manager A and B show the same
historical performance track record for the specific asset
management product. Looking at the two annualized return figures
– which are identical – it is difficult for the client to decide on which
asset manager to take. To come to a final decision a lot of
additional information is needed. Considering just the return figures
the client can not come to a meaningful decision because he does
not know:
• Where do the return figures come from: a model portfolio, a
GIPS composite, a representative account, the best performing
account, the largest account, etc.
• Whether the return figures belong to accounts managed by the
relevant asset management company or belong to a track
record of a specific asset manager generated with one of his
former employer
 249

• Whether the return figures are gross or net returns and how the
different fees are reflected within the calculation; for example in
continental Europe the net return may be net of custody fees
and in North America vice versa
• What is the underlying reporting period: since inception, last 12
months, the best looking period, etc.
• How the return figures where calculated: as a money weighted
rate of return, as a time weighted rate of return and if the latter
what approximation method is used
• What was the investment strategy or the investment objective?
• Which benchmark is representing the investment strategy best?

asset manager A asset manager B

5 % p.a. 5 % p.a.

Two asset managers that manage a comparable account

Who is the better portfolio manager? Figure 5.22

Without an answer to the above questions a client is hardly able to

judge and to compare the different asset management companies
respectively their performance track records. „Cherry Picking“, i.e.
the intentional selection of a specific account or observation
period, the use of sample accounts, model portfolios or
simulations, the transfer of performance track records as well as
the not standardized methodologies for calculation performance
figures lead to the issue of missing comparability of performance
figures of different asset management companies. The issue of
 250

the impossibility to compare performance figures and to assess the

performance of asset managers lead to the necessity for a unique
set of rules regulating the performance presentation and its
calculation. In other words – as illustrated in figure 2 – the issue of
performance presentations can be summarized as the case where
the client and also the senior management have to ask a lot of
unnecessary questions to get an objective and correct picture of
the performance or of the quality of the asset manager. Such
unnecessary questions are for instance
• Which accounts are presented?
• How has the return been calculated?
• Which accounts are behind the track record?
To illustrate the main issue in presenting performance – the
determination of the performance track record – we use a fictitious
case for a sample asset management company XYZ. In our
example a prospective client asked the asset manager to present
his historical performance track record for a specific product ABC.
As shown in figure 6.2 the asset manager XYZ managed 3
accounts in this product category for the period n until n+2.
Account C was terminated at the end of n and account B closed at
the end of n+2. Only account A was managed for the whole period
from n to n+2. Beside the account returns figure 3 also
encompasses the asset under management as of the beginning of
each year. The question is now which performance track record
should be shown to a prospective client.
 251

Year n n+1 n+2

Account A
Return 6.2% -2.0% 4.2%
Assets on the 1st of January 10.0 10.6 10.4
Account B
Return 5.0% -3.3%
Assets on the 1st of January 100.0 105.0
Account C
Return 4.1%
Assets on the 1st of January 500.0

Which track record is the most representative to be shown?

Figure 5.23

In determining a performance track record an asset manager has

several different alternatives to chose from such as:
• Alternative 1: Sample or representative account
• Alternative 2: Account with the longest performance history
• Alternative 3: Equally weighted average of the returns for all
actual accounts
• Alternative 4: Equally weighted average of the returns for all
accounts ever managed
• Alternative 5: Capital weighted average of the returns for all
actual accounts
• Alternative 6: Capital weighted average of the returns for all
accounts ever managed
• Alternative 7: Model portfolio or model strategy
• Alternative 8: Account with the best performance history
 252

In our sample case, after some discussions the asset manager

decided to follow alternative 3 which seems to be reasonable at
first sight. Figure 4 illustrate this performance track record by the
indexed cumulative return which is in our case identical to the
performance track record of account A.

110

Selection of
alternative 3, as
all actual 105

accounts are
reflected
100

95
2000 2001 2002 2003

Is it possible to asses the performance with this track record?

Figure 5.24

The question which arises now is whether a prospective client can

asses the quality of the asset manager by using such a
performance track record. The answer is clearly no because the
performance of an asset manager is normally not identical with the
performance of the actual managed accounts but with the
performance of all ever managed accounts. Neglecting the
terminated accounts in determining the performance track record
results in an effect called “survivor of the best”. The effect “survivor
of the best” means in this context that neglecting the terminated
accounts is normally the same as neglecting the bad performing
accounts because it is normally a bad performing account which is
 253

terminated – and it are the good performing accounts which

normally survive. In our sample case the performance track record
– following alternative 3 – is becoming better over time because
the on average bad performing accounts C and B were terminated
at the end of n and n+1 and the performance track record was
adjusted accordingly. Incorporating the terminated accounts –
means alternative 4 – would result in a cumulative return for the
period n to n+2 of 6.61% in comparison of 8.45% for alternative
3.
Alternative 4 seems to be a good way of determining a
performance track record but is not considering the assets under
management of the different accounts A, B, and C. This means
that the returns of the different accounts are equally weighted
which results in a bias towards the smaller accounts. In our
example in 2001 the performance track record profit from the
equal weighting of the large account C with a relatively bad return
of 4.1%. Considering the assets under management of the
different accounts over time – means alternative 6 – would result
in a performance track record shown in figure 5. The performance
track record of alternative 6 is not so good than that of alternative
4 because the bad performing accounts C and B had quiet a high
weight to the overall performance in the first two years.
Determining a performance track record following alternative 6 is
best practice because it follows the spirit of the GIPS standards
(Global Investment Performance Standards or in Switzerland the
country version of GIPS the Swiss Performance Presentation
Standards – SPPS) which are explained in detail in one of the
former chapters. The question which arises is now whether such a
performance track record is sufficient to asses the quality of an
asset manager. The answer is again clearly no. An observer needs
more information on the asset manager and its product which
ranges from the benchmark return, the number of accounts
 254

managed in such a product, the calculation method used, and so

on. Such a performance presentation is illustrated in figure 6 which
includes the basic information needed to get a good starting
picture in evaluating the quality of an asset manager.

110

Selection of Alternative 3

alternative 6, as
all accounts ever 105

managed are
reflected Alternative 6

100

95
2000 2001 2002 2003

In order to asses the performance do we need further information?

Figure 5.25

Composite Benchmark Number of Composite

return return accounts asset size

n+2 4.2% 4.8% 1 10.4

n+1 -3.2% -1.5% 2 115.6
n 4.3% 3.3% 3 610.0

=> The basis of the performance discussion and assessment

!
should always be a GIPS compliant report

Simple composite performance presentation Figure 5.26

 255

Figure 6 shows a sample performance presentation for a global

equity composite which follows the GIPS standards and includes
the minimum information required by the GIPS standards. This
basic information on the performance for a specific product or
composite enables the prospective client to concentrate on
meaningful questions and to get ride of unnecessary questions.
Assessing the quality of an asset management company or a
specific product or composite is quiet easy if one considers a GIPS
compliant performance presentation – and this is not only true for
prospective clients but also for the senior management of the
asset management company itself.

Sample GIPS compliant performance presentation Figure 5.27

 256

In summary the starting point of a performance evaluation from a

client’s as well as from a senior management’s perspective should
always be a GIPS compliant performance presentation because
• It enhances transparency and especially enhances the
understanding of the used performance measurement methods
and the performance presentation itself,
• It avoids cherry picking of accounts, time periods, and so on,
• It increases the comparability of different products and asset
managers,
• It improves the ability to judge the quality of the asset manager
or a specific product or composite and
• It enables objective discussions on the performance and
therefore enables to focus on the essential issues.

At the end we would like to stress that investment controlling

can only be done in an efficient and effective way with composite
information which preferably was constructed and maintained
according to the GIPS standards. Enabling a meaningful evaluation
of the performance and the quality of an asset manager is the
main benefit for the client and for the asset management company
from GIPS compliant performance presentations.
 257

5.4. Some critical issues in performance attribution

In this chapter we address certain issues in performance

attribution which people usually are not aware of. Performance
attribution is very complex not only to calculate but also to set up.
Setting up the performance attribution in a not suitable or even
wrong way results in misleading information or misinterpretation
and with this leads to wrong feedback into the investment process.
In the following we explain some of these issues but be aware that
there are other issues left.
As said above performance attribution is very sensitive to
misinterpretation because performance

• Is a result from several obvious and less obvious decisions

• Originates from several obvious and less obvious decision
makers
• Can vary over time

Misinterpretation of the historical performance may lead to

erroneous or even wrong decisions as they
• Asses the quality of the decision makers
• Give feedback into the decision making process

Furthermore the investment universe as well as the implementation

of the investment strategy has to be considered by setting up the
performance attribution. As shown in figure 6.12, in this situation
one especially has to define whether certain decisions are stock
picking or asset allocation decisions. For case A) it has to be
defined whether the 20% investments in small and mid caps is a
stock picking or an asset allocation decision? If the performance
analyst sets up the performance attribution wrongly than of course
 258

he get misleading figures for the management effects. In case B)

the performance analyst has to specify whether the investments in
(US) biotech stocks is a stock picking decision versus the US
equity index or whether its is an asset allocation decision versus
the overall benchmark because biotech stocks are not an explicit
part of the overall benchmark.

A) Benchmark SMI
Investment universe SPI

20% Investments in Small&Mid Cap

B) Benchmark MSCI World Stock picking

with 10 sub accounts or
Asset allocation ?
with sub account US-Biotech

Relative return +2.0%

Asset allocation -0.5%
Stock picking +2.7%
Interaction -0.2%

Is each decision in the investment process reflected? Figure 5.28

.

Until now we just addressed issues in setting up a return

attribution but there are similar problems if setting up a risk
attribution. The issues with risk attribution are a bit bigger because
the risk attribution software available is not as flexible as the return
attribution software. Normally the risk attribution software
decomposes the absolute or relative risk following a specific risk
 259

model which may not represent the actual investment process.

Therefore the figures of a risk attribution have be handled with
care because the different risk factors and their contributions to
the overall absolute or relative risk are normally not linked to the
steps of the investment process or the different decisions taken.
Figure 6.14 illustrates this issue by indicating that in an ideal world
the risk attribution have to be linked to the return attribution.

Risk Model: Global Portfolio Tracking Error

Total Risk (ex-ante) 18.81% 2.57%

Factor Specific Risk 18.66% 1.50%

- Region 11.50% 0.18%

- Country 6.98% 0.83%

- Industry 2.64% 0.77%

- Fundamental 1.44% 0.78%

- Currency 8.42% 0.27%

- Covariance (+/-) 9.35% 0.52%

Stock Specific Risk 2.39% 2.08%

Asset Stock
by MSCI Sector Interaction Total
Allocation Selection
Consumer Discretionary 0.00% -0.86% -0.03% -0.89%
Consumer Staples 0.05% 0.28% 0.03% 0.36%
Energy 0.00% -0.19% -0.02% -0.21%
Financials -0.10% -0.21% 0.01% -0.30%
Health Care 0.20% 0.35% -0.03% 0.52%
Industrials -0.04% 0.99% 0.12% 1.07%
Information Technology -0.21% -0.91% 0.01% -1.11%
Materials 0.01% -0.28% -0.08% -0.35%
Telecomm Services -0.11% 0.21% -0.08% 0.02%
Utilities 0.02% 0.74% -0.41% 0.35%
Cash -0.78% 0.00% 0.00% -0.78%
Total -0.96% 0.13% -0.49% -1.32%

Is the Risk contribution consistent with the return attribution?

Figure 5.29

After explaining some of the pitfall with setting up and running a

performance attribution it would be useful to have some rules to
hinder these kind of misleading performance analyses such as:
 260

• Reflect each step of the investment process

• Reflect each decision in the investment process
• Mirror the investment style correctly
• Analyse the impact of the investment guidelines
• Define whether a certain product or an asset manager has to
be assessed
• Analyse the variation of the management effect over time
• Analyse the impact of leverage

As you may be expect there is nothing like a complete list of

rules to follow to prevent such pitfalls. The only thing what helps in
general is transparency and here there is a good example for such
general rule or standard on transparency – the GIPS standards.
We argue that there is also a need for performance attribution
presentation standards which would help in solving this issue.
At the end we conclude with summarizing some of the main
points on investment controlling:
• Investment controlling is in its closer but also in its wider
meaning an absolute must
• Complying with GIPS or using GIPS composites is imperative
for a sound and convincing investment controlling
• The client perspective has to be reflected
• Transparency is the key for a meaningful feedback into the
decision making or investment process
• Investment controlling enables:
o Realistic and unbiased performance discussion
o Controlling of the consistent implementation of investment
strategies
o Reduction of unintended or not obvious investment and
non-investment risks
o Independent performance reviews
 261

o Focus on ex-ante risks and the investment process

o Early awareness of potential performance or process
oriented problems
Index
Annualized Return, 48

Asset class, 170

Attribution, 27

Benchmark, 22

Beta, 212

Compounding, 9

Contribution, 18

Correlation, 129

Covariance, 128

Excess Return, 22

Geometrical Attribution, 38

Money weighted rate of return (MWR), 68

Time weighted rate of return (TWR), 52

Portfolio Analytics, 5

Return, 8

Segment, 20

Standard deviation, 119

Tracking Error, 130

Variance, 119

Volatility, 120
Literature

[1] Bonafede Julia K., Foresti Steven J., Matheos P.,

Fall 2002
A multi-period Linking Algorithm that has stood the test of
time
Journal of Performance Measurement

[2] Brian R. Bruce, 1990

Quantitative International Investing
McGraw-Hill Book Company

[3] CFA Institute, May 2006

Global Investment Performance Standards (GIPS)

[4] Copeland T., Weston F, 1988

Financial Theory and Corporate Policy
Addison – Wesley, 3rd edition

[5] Dalang R., Marty W., Osinski Ch., Winter 2001/2002

Performance of quantitative versus passive investing
Journal of Performance Measurement

[6] Diderich C., Marty W., June 2000, pp. 238-245

The Min-Max Portfolio Optimisation Strategy: An empirical
study on Balanced Portfolios
Lecture Notes in Computer Science
Numerical Analysis and its Applications 2000, pp. 238-245

[7] Dougherty Ch., 2002

Introduction to Econometrics
Oxford university press
[8] Elton, E. J., Gruber M., 2007
Modern portfolio theory and investment analysis
Wiley, 7th Edition

[9] Feibel B. J., 2003

Investment performance measurement
Whiley, Finance

[10] Gabisch G., Lorenz H.-W., 1989

Business Cycle Theory 2nd edition Springer Verlag

[11] Kleeberg, Jochen M., 1995

Der Anlageerfolg des Minimum-Varianz Portfolios.
Schriftenreihe “Portfoliomanagement”. Uhlenbruch Verlag

[12] Kreyszig E., 1978

Introductory Functional Analysis with Applications
John Wiley & Sons

[13] Lorenz H.-W., 1993

Nonlinear Dynamical Economics and Chaotic Motion
Springer-Verlag
Heidelberg, Berlin, New York

[14] Illmer S., Marty W., Summer 2003

Decomposing the Money-Weighted Rate of Return,
Journal of Performance Measurement

[15] Neumaier A., 2001, page 61-90

Introduction to Numerical Analysis
Cambridge University Press
[16] Malkiel G. Burton, 2007
A Random Walk Down Wall Street
W.W. Norton & Company

[17] Marty W., 6.10.2006

Stability is the Enemy of Optimality unpublished
Translated from NZZ

[18] Miller and Modigliani

Dividend Policy, Growth, and the Valuation of Shares
Journal of Business 34 (Oct. 1961) pp. 411 - 433

[19] Rustem B., Marty W., Becker R., 2000, pp. 1591 – 1621
Robust min max portfolio strategies for rival forecast and risk
scenarios
Journal of Economic Dynamic and Control Vol. 24 (11-12)

[20] Ross S.A. March 1977, page 177 − 184

The Capital Asset Pricing Model Short sales restriction and
related issues Journal of Financial March Ross, S.A.

[21] Sharpe W.F., 1995

Investments
Prentice Hall International Editions, 5rd edition

[22] Shestopaloff Y, 2009

Science of Inexact Mathematics
Akvy Press, 1st edition

[23] Taleb N. N., 2007

The Black Swan The impact of the highly improbable
Random House
[24] Varga R., 2000, page 22 - 53
Matrix Iterative Analysis
Springer Verlag, 2nd edition

[25] Vörös J., 1987 page 302 - 310

The explicit derivation of the efficient portfolio frontier in the
case of degeneracy and general singularity
Eur. J. Operat. Res. 32
The author

Dr. Wolfgang Marty is vice president at Credit Suisse. He

joined Credit Suisse Asset Management in 1998 as Head Product
Engineering. He specializes in Performance Attribution, Portfolio
Optimisation and Fixed Income in general.
Prior to joining Credit Suisse Asset Management, Marty
worked for UBS AG in London, Chicago and Zurich. He started
his career as assistant for applied mathematics at the Swiss
Federal Institute of Technology.
Marty holds a university degree in Mathematics from the Swiss
Federal Institute of Technology in Zurich and a doctorate from the
University of Zurich. He chairs the method and measure sub
committee of the European Bond Commission (EBC) and he is
president of the Swiss Bond Commission (OKS). Furthermore he
is a member of the Fixed Income Index Commission at the Swiss
Stock Exchange and a member of Index team that monitors the
Liquid Swiss Index (LSI).