Vous êtes sur la page 1sur 102

Portfolio

construction using alternative benchmark indices


- An empirical investigation of alternatively weighted versions of the cap-weighted MSCI AC
World index and their influence on the performance of Jyske Invests portfolio Global Equities



Authors
Christian Curtz Henriksen (288780)
Camilla Lindved Hansen (300305)

Supervisor
Christian Schmaltz
Department of Economics and Business





September 2013
MSc. Finance
Aarhus University
Business and Social Sciences

Abstract
Today, cap-weighted (CW) indices have become an integral part of the construction strategy of
actively managed equity portfolios throughout the world. The primarily goal of active portfolio
managers is to outperform these CW benchmark indices and therefore they have a critical im-
pact on both asset allocation and performance measurement. Consequently, a lot of attention
has been paid to the essential question of how to construct an adequate benchmark index. De-
spite a growing body of literature within the field of index construction, only limited, if any, con-
sensus has yet been found. Additionally, no studies that specifically investigate how the choice
of benchmark index and tracking error constraint impacts actively managed portfolios are evi-
dent in the academic literature.
Inspired by the circumstances above, this thesis sets out to investigate whether alternatively
weighted MSCI AC World indices are superior to their CW counterpart and how they affect the
performance of Jyske Invests actively managed portfolio Global Equities (GE). Coherent with the
literature on index construction and evaluation of portfolio construction strategies, this entire
study is set up in a controlled backtest environment and applies a sample period from 2006-
2013.
The evaluation of the alternative benchmark indices are conducted on the basis of a customised
framework consisting of conventional performance measures as well as quality criteria (turno-
ver, concentration, representativeness, liquidity and transparency). A key finding is that the
equal-weighted, minimum-volatility-weighted and maximum-Sharpe ratio-weighted indices
show robust outperformance of the CW MSCI AC World index while preserving a sufficient level
of quality.
In terms of GE, this thesis finds that choosing the more efficient indices as benchmark has a
large performance enhancing effect on GE when the tracking error is minimised. Conversely,
loosening the tracking error constraint is only performance enhancing for the less efficient cap-
weighted and fundamental-weighted indices. These empirical findings suggest that Jyske Invest
should reappraise their current portfolio construction strategy by either (1) maintaining the CW
MSCI AC World index as benchmark and allow for higher tracking error or (2) choosing a more
efficient benchmark index and maintain their strict tracking error constraint.

List of contents
1. Introduction ............................................................................................................................................... 1
1.1 Problem statement ............................................................................................................................................... 2
1.2 Contribution ............................................................................................................................................................ 3
1.3 Delimitations ........................................................................................................................................................... 4
1.4 Preliminary evaluation of sources ................................................................................................................. 4
1.5 Thesis structure ..................................................................................................................................................... 5
1.6 Jyske Invests Global Equities at a glance .................................................................................................... 6
1.6.1 Stock picking ................................................................................................................................................... 6
1.6.2 Portfolio construction ................................................................................................................................. 7
2. Theory and literature review ............................................................................................................... 9
2.1 Portfolio construction ......................................................................................................................................... 9
2.1.1 Expected returns ......................................................................................................................................... 10
2.1.2 Covariance matrix ....................................................................................................................................... 13
2.2 Portfolio performance measures .................................................................................................................. 14
2.2.1 Risk-adjusted ratios ................................................................................................................................... 14
2.2.2 Extreme risk (95% Value at Risk) ....................................................................................................... 16
2.2.3 Factor exposures ......................................................................................................................................... 16
2.3 Equity indices as benchmarks ....................................................................................................................... 17
2.3.1 The evolution of benchmark indices .................................................................................................. 18
2.3.2 Evaluating benchmark indices .............................................................................................................. 19
2.3.3 Criticism of efficiency of cap-weighted benchmark indices ..................................................... 22
2.3.4 Alternatives to cap-weighted benchmark indices ........................................................................ 24
2.3.5 Why still cap-weighted benchmark indices? .................................................................................. 27
2.4 Tracking error constraints .............................................................................................................................. 28
2.4.1 Criticism of tracking error constraints .............................................................................................. 29
2.4.2 Tracking error from different perspectives .................................................................................... 30
3. Methodology ............................................................................................................................................ 32
3.1 Introduction to the backtesting methodology ........................................................................................ 32
3.2 Constructing alternatively weighted MSCI AC World indices (SRQ 1) ......................................... 33
3.2.1 Equal-weighted MSCI AC World index .............................................................................................. 34
3.2.2 Fundamental-weighted MSCI AC World index ............................................................................... 35
3.2.3 Minimum-volatility-weighted MSCI AC World index .................................................................. 37
3.2.4 Maximum-Sharpe ratio-weighted MSCI AC World indices ....................................................... 40
3.3 Impact of alternative benchmark indices on Global Equities (SRQ 2) ......................................... 42
3.4 Impact of loosening tracking error constraint on Global Equities (SRQ 3) ............................... 44
4. Data ............................................................................................................................................................. 47
5. Empirical findings ................................................................................................................................. 49
5.1 Preliminary analysis of Global Equities ..................................................................................................... 49
5.1.1 Performance compared to the CW MSCI AC World index ......................................................... 49
5.1.2 Performance compared to peer group of other mutual funds ................................................ 52
5.1.3 Performance had Global Equities been equal-weighted ............................................................ 52
5.2 Evaluation of alternatively weighted MSCI AC World indices (SRQ 1) ........................................ 54
5.2.1 Performance measures ............................................................................................................................. 54
5.2.2 Turnover ......................................................................................................................................................... 60
5.2.3 Concentration ............................................................................................................................................... 61
5.2.4 Representativeness .................................................................................................................................... 62
5.2.5 Liquidity .......................................................................................................................................................... 64
5.2.6 Transparency ................................................................................................................................................ 65
5.2.7 Robustness tests .......................................................................................................................................... 66

II

5.2.8 Qualitative scores summarising the findings for each MSCI AC World index .................. 70
5.3 Impact of alternatives benchmark indices on Global Equities (SRQ 2) ....................................... 71
5.3.1 Return and risk ............................................................................................................................................ 72
5.3.2 Risk-adjusted ratios ................................................................................................................................... 73
5.3.3 Significance tests ......................................................................................................................................... 73
5.4 Impact of loosening tracking error constraint on Global Equities (SRQ 3) ............................... 75
5.4.1 Return and risk ............................................................................................................................................ 75
5.4.2 Risk-adjusted ratios ................................................................................................................................... 77
5.4.3 Significance tests ......................................................................................................................................... 78
6. Evaluation of the findings ................................................................................................................... 81
6.1 Data ............................................................................................................................................................................ 81
6.2 Methodology .......................................................................................................................................................... 82
6.3 Further research .................................................................................................................................................. 83
6.4 Practical implications ........................................................................................................................................ 84
7. Conclusion ................................................................................................................................................ 86
Bibliography ................................................................................................................................................. 89
Appendix ....................................................................................................................................................... 97

III

List of figures
Figure 1.1 - Thesis structure ......................................................................................................................................... 5
Figure 1.2 - Intuition behind Jyske Invest's investment philosophy (VAMOS) ...................................... 7
Figure 2.1 - Benchmark index evaluation framework ..................................................................................... 20
Figure 2.2 - Top 10 constituents in the CW MSCI AC World index ............................................................. 23
Figure 2.3 - Proportion of financial stocks in the CW and EQW MSCI AC World index .................... 24
Figure 5.1 - Region weights ......................................................................................................................................... 50
Figure 5.2 - Sector characteristics ............................................................................................................................ 51
Figure 5.3 - Indexed cumulative return ................................................................................................................. 51
Figure 5.4 - Comparison to other European mutual funds ............................................................................ 52
Figure 5.5 - Region characteristics ........................................................................................................................... 62
Figure 5.6 - GICS sector characteristics .................................................................................................................. 63
Figure 5.7 - % Sharpe ratio change for GE when tracking alternative indices ..................................... 74
Figure 5.8 - Return and standard deviation when loosening tracking error constraint .................. 76
Figure 5.9 - Sharpe ratio and information ratio when loosening tracking error constraint ........... 77
Figure 5.10 - % Sharpe ratio change for GE when loosening tracking error constraint ................... 79

List of tables
Table 1.1 - Jyske Invest's weight limits .................................................................................................................... 7
Table 3.1 - Description, directional impact and weight of Jyske Invest's VAMOS score ................... 42
Table 5.1 - Size and weight characteristics of GE and the CW MSCI AC World index ........................ 50
Table 5.2 - Performance measures of GE using current and equal weighting scheme ..................... 53
Table 5.3 - Sign. tests of GE differences between current and equal weighting scheme .................. 53
Table 5.4 - Performance measures of alternative MSCI AC World indices ............................................. 54
Table 5.5 - Sign. tests of differences between CW and alternative MSCI AC World indices ............ 56
Table 5.6 - Carhart (1997) 4-factor model return decomposition ............................................................. 58
Table 5.7 - Turnover characteristics ....................................................................................................................... 60
Table 5.8 - Concentration characteristics ............................................................................................................. 61
Table 5.9 - Liquidity characteristics ........................................................................................................................ 64
Table 5.10 - Performance measures for subsamples ....................................................................................... 67
Table 5.11 - Performance measures and turnover for different rebalancing frequencies .............. 68
Table 5.12 - Performance measures, concentration and liquidity for different levels of ............. 68
Table 5.13 - Performance measures for different covariance matrices ................................................... 70
Table 5.14 - Qualitative scores summarising the findings of the alternative indices ........................ 71
Table 5.15 - GE performance measures applying alternative benchmark indices .............................. 72
Table 5.16 - Sign. tests of differences between actual GE and GE applying alternative indices .... 74
Table 5.17 - Sign. tests of differences between GE with loosest and strictest TE constraint ......... 78

IV

List of appendices
Appendix 1 - Review of studies on cap-weighted indices .............................................................................. 98
Appendix 2 - Review of empirical tests of the CAPM ....................................................................................... 99
Appendix 3 - Example of a backtest using Wilshire Atlas (MSRW (VAMOS) index) ....................... 101
Appendix 4 - Performance comparison between indices of this thesis and MSCI ........................... 103
Appendix 5 - Example of constructing a fundamental-weighted index ................................................ 105
Appendix 6 - Example of estimating cov. matrix from exposures and factor returns .................... 106
Appendix 7 - Example of constructing a minimum-volatility-weighted index .................................. 107
Appendix 8 - Example of constructing a maximum-Sharpe ratio-weighted index .......................... 108
Appendix 9 - Example of a minimising tracking error portfolio .............................................................. 109
Appendix 10 - Example of loosening the tracking error constraint ....................................................... 110
Appendix 11 - Cumulative return of alternative indices ............................................................................. 112
Appendix 12 - Fama & French (1993) 3-factor model ................................................................................... 113
Appendix 13 - Development of exposures to sectors and regions of alternative indices ............. 114
Appendix 14 - Data CD ................................................................................................................................................ 119

1. Introduction

1. Introduction
Matching the market is an inefficient investment strategy even in an informationally effi-
cient market

- Robert A. Haugen and Nardin L. Baker (1991)

Despite early criticism of cap-weighted (CW) indices by Haugen & Baker (1991) and Grinold
(1992), they are still to this day by far the most common point of reference for both passive and
active investment funds. Many passive investment funds invest directly in broad global equity
indices, such as the CW MSCI AC World index1, FTSE All-World Index and Russell World Cap
Index. On the other hand, active portfolio managers are evaluated based on their relative per-
formance to these indices the so-called benchmarks. Consequently, a lot of attention has been
paid to the essential question of how to choose and construct an adequate benchmark index as
it plays a crucial role for investment institutions and ultimately also the individual investor. The
attention has especially escalated over the last few years, where numerous alternatives to the
criticised CW indices have been proposed. The main point of criticism of CW indices is their
inefficient risk/return properties primarily caused by (1) reliance on the highly unrealistic as-
sumptions behind CAPM, (2) concentration in a few large stocks and (3) a trend-following fea-
ture. In response to the criticism of cap weighting, several alternative weighting schemes have
been proposed such as equal weighting, fundamental weighting, risk factor weighting, maxi-
mum diversification weighting, minimum-volatility weighting, maximum-Sharpe Ratio
weighting, etc. Proponents of CW indices, however claim that reweighting these traditional indi-
ces does not consistently produce positive excess returns (or alpha) and tend to produce a sys-
tematic beta exposure toward the smaller-cap and value stocks within the targeted index. Fur-
ther, proponents highlight that the best index is not necessarily the one that provides the high-
est return over a given period, but the one that most accurately reflects the mood of the market
(Philips et al., 2011).
Additionally, the choice of benchmark index requires careful consideration as the performance
evaluation of active portfolio managers is highly tied to the tracking error volatility, i.e., the var-
iability of the difference between the managers return and the benchmarks return. Minimisa-
tion of tracking error volatility has become an important criterion for assessing overall manager
performance. In light of this, portfolio managers are hypersensitive to any deviations between
their returns and those of the benchmark as it may ultimately lead to their dismissal (Rappaport

1 All Country World Index
2 As a rule of thumb Jyske Invest sets a $2 billion minimum requirement to the market capitalisation of

stocks included in GE.


3 Throughout this thesis returns are in USD, adjusted for currency effects, include gross dividends but not

1. Introduction
& Mauboussin, 2001). In this environment, a portfolio manager is naturally more concerned
with short-term returns relative to the benchmark, rather than maximising risk-adjusted long-
term returns. This raises a very fundamental question regarding the appropriateness of impos-
ing strict tracking error constraints on active portfolio managers.
Clearly, no simple answer to which index to use as benchmark and how alternative indices and
tracking error constraints affect the performance of actively managed portfolios is evident in
the theoretical literature. The aim of this thesis is to contribute to the understanding of these
key issues and test whether alternative indices and looser tracking error constraints should
represent a new paradigm of equity investing.

1.1 Problem statement


Motivated by the discussion above, this thesis conducts, through a thorough quantitative analy-
sis, a comparison of alternatively weighted versions of the criticised CW MSCI AC World index.
Subsequently, the impact of these alternative benchmark indices and different levels of tracking
error constraint is investigated on a specific actively managed portfolio, namely Jyske Invests
portfolio Global Equities (GE) that applies the CW MSCI AC World index as benchmark. In doing
so, this study is able to specifically address the implications of Jyske Invests current portfolio
construction strategy and critically assess its appropriateness. As the current investment setup
in Jyske Invest is quite universal and commonly used in other active investment funds as well,
our analysis of GE is not only useful to Jyske Invest, but can potentially offer guidance to a wide
range of practitioners. The overall key research question (KRQ) under investigation is broadly
stated as follows:

KRQ: How does alternatively weighted versions of the cap-weighted MSCI AC World index affect
the performance of Jyske Invests actively managed portfolio Global Equities?

To ensure a solid answer and to focus the study, the KRQ is broken down into three sub-
research questions (SRQ):

SRQ 1: Are alternatively weighted MSCI AC World indices superior to the cap-weighted MSCI AC
World index?
SRQ 2: How does the choice of benchmark index affect the performance of Global Equities?
SRQ 3: How does loosening the tracking error constraint on Global Equities affect its perfor-
mance?

1. Introduction
Overall Jyske Invests portfolio construction strategy for GE is constituted by two distinct choic-
es:
1) Which benchmark index to use
2) What tracking error constraint to impose the portfolio managers
SRQ 2 and SRQ 3 specifically address each of these choices. An empirical investigation of alter-
native equity indices and thereby an answer to SRQ 1 on whether or not superior indices exist is
necessary to be able to provide a more thorough, adequate and pragmatic answer to SRQ 2 and
SRQ 3. Moreover, the literature review reveals pronounced dispute on how to construct alterna-
tive indices. This thesis defines superiority and evaluates the indices based on the framework
that is introduced in section 2.3.2, which combines selected performance measures and quality
criteria.

1.2 Contribution
The contribution of this thesis should be evaluated on the basis of the progress already made in
the theoretical, methodological and empirical knowledge about how to construct equity indices
and what benchmark index and tracking error constraint to use for actively managed portfolios.
A considerable amount of studies have been conducted in the area of index construction in the
past decades. Still, no unequivocal answer seems to exist to the fundamental questions of
whether superior alternatives to the CW indices exist. This study adds another piece to the puz-
zle by constructing alternatively weighted versions of the CW MSCI AC World index and evalu-
ate each of them on the basis of a customised framework consisting of performance measures as
well as quality criteria such as turnover, concentration and liquidity. Contrary to related studies
on alternative indices, e.g. Arnott, Hsu & Moore (2005), Amenc et al. (2011), Chow et al. (2011)
and Clare, Motson & Thomas (2013a), this thesis honours the recent call from Amenc, Goltz &
Shuyang (2012) and applies the same stock universe for all constructed indices. Comparing
strategies that involve stock selection with strategies that simply change the weighting within a
given universe is not a correct way to analyse different weighting schemes as the comparison is
based on the different stock universes. As such, the validity and robustness of their conclusions
are challenged. In addition to keeping the stock universe fixed, this thesis also challenges relat-
ed studies by using a more recent data period and different weight constraints and key input
parameters.
Further, the study adds to the existing literature by explicitly evaluating the impact of using
alternative benchmark indices and looser tracking error constraints on a specific portfolio had
this been the code of practice in the past. To the authors knowledge no academic study has yet
taken this extra step, but instead decided to focus on benchmark indices and tracking error con-
straints from a more aggregate point of view. Being granted access by Jyske Invest to infor-

1. Introduction
mation and data on GE regarding their stock picking process, historic constituents, weighting
methodology, portfolio manager constraints, etc. has enabled this study to go one step further.

1.3 Delimitations
There is practically an infinite amount of interesting perspectives and questions related to the
equity investment process in general. In order to achieve an increased focus it is necessary to
make a number of delimitations. First and foremost this thesis limits itself to focus exclusively
on actively managed portfolios even though the performance of passively managed portfolios is
also directly affected by the choice of benchmark and very little by the choice of tracking error
constraint. This limitation is made to ensure a more exhaustive analysis of the implications for
active investment funds, and in particular GE, as the large differences between active and pas-
sive investment funds rule out the possibility of covering both.
Furthermore, the study will not explicitly address the impact on portfolio performance caused
by the chosen stock picking process in Jyske Invest. This varies greatly from institution to insti-
tution. As such, the study explicitly focuses on the portfolio construction strategy in Jyske Invest
constituted by the choice of benchmark index and tracking error constraint, as it is more uni-
versal and thus interesting for a broader range of stakeholders. In practice, this means that the
historical constituent stocks of GE and the CW MSCI AC World index are kept fixed at all times
throughout this thesis and only the various weighting schemes and portfolio construction strat-
egies are being tested.
Several hundred books and articles have been written on portfolio construction strategies, per-
formance measures and benchmark indices. While many of these studies serve as inspiration
and references in this thesis, a complete literature review will not be presented. Instead the
literature review of this thesis is centred around the most cited and recently published studies.
Finally, the reader of this thesis is expected to possess elementary knowledge of modern portfo-
lio theory though a brief introduction is provided in section 2.1. Moreover, the reader should be
familiar with basic statistics and econometrics as it is beyond the scope of this study to provide
an in-depth description of all the applied statistical frameworks.

1.4 Preliminary evaluation of sources


Published articles from respected journals within finance and economics form the literary base
of this thesis. The findings and theories from these papers are considered reliable since they are
thoroughly examined before publication. In particular, the related studies made by Arnott, Hsu
& Moore (2005), Amenc et al. (2010), Amenc et al. (2011), Chow et al (2011), Amenc, Goltz &
Lodh (2012) and Clare, Motson & Thomas (2013a; 2013b) provide a standard of comparison
and are further used to validate the findings of this thesis. Throughout the thesis they are re-

1. Introduction
ferred to as the related studies.
This thesis sample period is limited by the access of Jyske Invest to the historical constituents
of the CW MSCI AC World index, which extends back to January 2006. As such the full sample
period applied in this paper runs from January 2006 to May 2013.
The Wilshire Atlas software package is preferred in this thesis as it provides direct access to the
required proprietary data on the historic constituents of the CW MSCI AC World index and GE.
Further data is retrieved from International Data Corporation (IDC) and Worldscope. The validi-
ty of the data is considered relatively good.

1.5 Thesis structure


The thesis is divided into seven chapters. Figure 1.1 illustrates the structure and shows how
each sub-section relates to the research questions.
Figure 1.1 - Thesis structure
ABC%"B!B("2D%#EB!?3FG
#".
3,2'046)2+0,

!"
&8/0'G.(,4.
A+2/'(26'/.
'/@+/J

!
!"#$%
()*+,-.*/0+%1!23%(2%45,)6%/-6/7+8

!
!"#&
9+-7:;.,<%/;=.7*%5-%>!

!"#

H0'2B0A+0.)0,12'6)2+0,

!"!

H0'2B0A+0.:/'B0'9(,)/.9/(16'/1

!"$

?K6+2G.+,4+)/1.(1.L/,)89('*1

!"%

&'()*+,-./''0'.)0,12'(+,21

$"#

7()*2/12+,-.9/28040A0-G

$"!
$".
F/28040A0-G $"$

3,4/5.)0,12'6)2+0,

!
7/,)89('*.+9:()2.0,.;<

$"%
%".=(2(

>"!

H'/A+9+,('G.(,(AG1+1.0B.;<
?@(A6(2+0,.0B.(A2/',(2+@/.+,4+)/1

>"$

!
7/,)89('*.+9:()2.0,.;<

>"%
C"
?@(A6(2+0,
0B.28/.
B+,4+,-1

!
&'()*+,-./''0'.+9:()2.0,.;<

=(2(
>"#

>".
?9:+'+)(A.
B+,4+,-1

!
!"#'
?,.7</-@%+,,5,%/;=.7*%5-%>!

C"#

!
&'()*+,-./''0'.+9:()2.0,.;<

=(2(

C"!

F/28040A0-G

C"$

I6'28/'.'/1/(')8

C"%

H'()2+)(A.+9:A+)(2+0,1

D".E0,)A61+0,

E0,)A61+0,

Source: Own contribution

Chapter one covers the already clarified research design and introduces Jyske Invests GE port-
folio at a glance. Chapter two establishes a conceptual ground from where the subsequent work
is built and evaluates theoretical literature and empirical findings of related prior research.
Chapter three outlines the methodological foundation of the study. Chapter four presents the

1. Introduction
applied data. In chapter five the empirical findings are presented and discussed. Chapter six
evaluates the study including an assessment of practical implications and finally chapter seven
concludes.

1.6 Jyske Invests Global Equities at a glance


This section on GE is based on meetings with and materials provided by Jyske Invest; see Jyske
Invest (2013a; 2013b; 2013c; 2013d; 2013e). Jyske Invest currently manages 15 stock portfoli-
os, where GE is one of them. The division occupies 10 portfolio managers, who all contribute to
GE. The portfolio was introduced in 1988 and consists of around 100 stocks. The portfolio is
diversified in companies from different regions, countries and industrial sectors but only in
large and mid cap companies2. Typical investors are clients from the personal banking division
with a 3-5 year investment horizon. Other investor profiles include municipalities and corpo-
rate clients.
The objective of the portfolio management is to outperform the CW MSCI AC World benchmark
index. The annualised return3 from January 2006 to May 2013 was 6.9% for GE versus 5.1% for
the CW MSCI AC World index. This shows that before trading and administrative expenses GE
has been able to outperform the benchmark. A more elaborate comparison of the performance
of GE and the CW MSCI AC World index is provided in section 5.1.1. The next two sections intro-
duce the stock picking process and portfolio construction strategy of GE.

1.6.1 Stock picking


To identify and pick attractive stocks, Jyske Invest uses a proprietary factor model that together
with thorough due diligences and risk considerations comprise their overall investment philos-
ophy, which builds on value, momentum and strength (VAMOS) factors. The intuition behind
these three factors is shown in figure 1.2 on the next page.
The global stock universe4 is screened on a weekly basis and based on the factor model each
stock gets a unique VAMOS score from 0 to 100. A full description of each factors impact on the
total score is provided in the methodology section 3.2.4.3. When a given stock not currently
included in GE receives a high score or the score of a stock already included in GE changes sig-
nificantly, a thorough due diligence is set in motion. This includes an analysis of the particular
companys strategy, the competitors, macro factors, triggers and risk factors.

2 As a rule of thumb Jyske Invest sets a $2 billion minimum requirement to the market capitalisation of

stocks included in GE.


3 Throughout this thesis returns are in USD, adjusted for currency effects, include gross dividends but not
adjusted for tax payments and trading and administrative expenses.
4 Consists of approximately 6000-7000 companies and the gross universe is determined as either mem-
ber of MSCI AC World IMI (large, mid and small cap) or a market capitalisation > $2 billion

1. Introduction
Figure 1.2 - Intuition behind Jyske Invest's investment philosophy (VAMOS)

Value
Stocks with an attractive cheap price
outperform expensive stocks

Momentum
Stocks from companies, that show
surprisingly good results outperform
stocks from companies with
disappointing results.

Strength
Stocks from companies with high and
increasing ROIC outperform stocks
from companies with low and
decreasing ROIC

Result: Cheap stocks from quality-companies, that show surprisingly good results, perform well in the long run


Source: Own contribution, Jyske Invest (2013b)

1.6.2 Portfolio construction


Jyske Invests portfolio construction strategy is to a certain extent an integrated part of the in-
vestment process. The goal for the construction strategy is to be exposed to the preferred char-
acteristics (VAMOS factors) and at the same time achieve the best possible diversification.
In order to determine the individual weights of the constituents in GE a range of factors are tak-
en into account, though some are emphasised more than others. These factors include strength
of the investment case, the estimated fair value, investment risks, liquidity, correlation with the
portfolios other stocks and the contribution to the total and relative risk profile. Also, manage-
ment and board of directors have set up some weight limits, which are shown in table 1.1.
Table 1.1 - Jyske Invest's weight limits

Tracking error
Beta

Internal lower limit

Internal upper limit

0.0%
0.9

6.0%
1.1

-15.0%
-10.0%
-10.0%
-2.7%
-2.3%
-7.3%
-5.0%
-10.0%
-5.0%

15.0%
10.0%
10.0%
10.0%
10.0%
10.0%
5.0%
10.0%
5.0%



Region weights relative to benchmark
North America
Europe
Asia

Latin America

Eastern Europe / Africa
Far east
Others

Country weights relative to benchmark

Industrial sector weights relative to benchmark
Source: Jyske Invest (2013e)




However, it remains a fact that instead of specifically addressing each of these factors, the con-
struction is in reality constituted by two distinct choices:
1) Which benchmark index to use
2) What tracking error constraint to impose the portfolio managers

1. Introduction
As mentioned, the chosen benchmark for GE is the CW MSCI AC World index. Since the introduc-
tion of GE the rule of thump for determining the stock weights has always been benchmark
weight + 0.7%. For example, say the portfolio managers have decided to include General Elec-
tric in GE. In the CW MSCI AC World index the weight for General Electric is 0.76% as of 31 May
2013, which means that the weight in GE should be 0.76% + 0.7% = 1.46%. This way of deter-
mining the stock weights is a very simplified and heuristic way to achieve low tracking error.
Also, in cases where a stock that is not part of the benchmark is included in GE, no consistent
procedure is followed in determining the stocks weight.
Despite the rule of thumb, the sum of the individual weights obviously has to be 100%. There-
fore it is necessary to adjust the individual weights in order to obtain a total weight of 100%.
The method for the adjustments is to take one weight constraint (illustrated in the table above)
at the time. The most important constraints are the geographical region and industrial sector
weights. These weights must be within the intervals illustrated in table 1.1 and this is accom-
plished by adjusting the allowed overweight in each stock by +/- 0.2 percentage point from the
0.7% rule of thump. The portfolio is rebalanced every week but only if any stocks need to be
either included or excluded from the portfolio.
In summary, the investment philosophy of Jyske Invest is apparent in their stock picking pro-
cess where the VAMOS factors are pivotal. However, in the construction process the main focus
is on the benchmark weights, hence, the investment strategy is not evident throughout the in-
vestment process and the consistency seems wavering. For this reason, this paper focuses ex-
clusively on the construction part of the investment process in Jyske Invests GE.

2. Theory and literature review

2. Theory and literature review


The purpose of this chapter is to introduce the main theoretical concepts used throughout this
thesis in order to establish a conceptual ground from where the subsequent work is build. It also
aims to give some motivation for the use of the methods applied and expose the relations between
them.
The chapter first includes a brief review of the theory behind portfolio construction and the re-
quired inputs. Second, a presentation of the performance measures relevant to this thesis is provid-
ed including a discussion of pros and cons. Third, equity indices used as investment benchmarks
are reviewed covering quality criteria and criticism of CW indices followed by an introduction of
alternatively weighted indices. Finally, the use of tracking error constraints is discussed and ex-
pounded from different perspectives.

2.1 Portfolio construction


As mentioned, this paper focuses only on the construction part of an investment process, and
thus understanding the theory behind portfolio construction is a prerequisite. In particular, the
vast majority of the alternative equity indices and GE portfolios constructed in this thesis build
on the portfolio theory introduced in the seminal and heavily cited paper by Markowitz (1952).
A very brief introduction to his portfolio optimisation theory is therefore provided with a spe-
cific focus on the two required inputs; expected returns and covariance matrix.
In accordance with Markowitz (1952) and most portfolio theory this thesis assumes that in
comparing two stocks with the same return, a rational investor prefers the one with the smaller
variance. Similarly, in comparing two stocks with the same variance (risk), the one with the
larger return is preferable to a rational investor. When the comparison is not so simple, as for
example when one stock has both higher return and variance than another stock, the choice
depends on the investor's level of risk tolerance. The investor must judge whether the addition-
al return is worth the additional risk.
Comparing portfolios of stocks is more complicated but Markowitz (1952) showed how portfo-
lios of stocks could be compared and constructed such that:

No other portfolio has higher return for the same risk

No other portfolio has lower risk for the same return

He called these portfolios efficient and developed computer algorithms to find all efficient port-
folios from a given set of stocks, i.e. the efficient frontier. All the identified feasible portfolios
minimise risk for a given level of expected return and maximise expected return for a given lev-
el of risk.

2. Theory and literature review


As with every other model, Markowitz (1952)s mean-variance theory is based on a set of as-
sumptions, which are:
1. All investors seek to maximise the expected return of total wealth
2. All investors have the same expected single period investment horizon
3. All investors are risk-adverse, that is they will only accept greater risk if they are com-
pensated with a higher expected return
4. All investors base their investment decisions on the expected return and risk
5. All markets are perfectly efficient (e.g. no taxes or transaction costs)
Much of the work that has followed Markowitz (1952) has addressed practical issues facing the
portfolio manager when constructing mean-variance efficient portfolios. Demowitz, Glen &
Madhaven (2001) and Elton et al. (1993) review the impact of transaction costs while Sengupta
(2003) tests the impact of differences in investment horizon. However, more important is the
theorys sensitivity to the two required inputs; expected returns and covariance matrix. De-
pending on the applied method, estimates of these inputs will suffer from estimation error
and/or specification error, both of which Chopra & Ziemba (1993) conclude will effect the port-
folio optimisation in such a way that the resulting optimal portfolio is not the true optimal port-
folio. Hence, it is of great interest to make the estimates as good as possible in order to reduce
this uncertainty. Thus, the next section reviews methods applied in this thesis for estimating
expected returns and the covariance matrix.

2.1.1 Expected returns


Future returns depend heavily on changes in expectations, and since these are unobservable
and investors have diverse opinions, it is difficult to forecast the changes in aggregate expecta-
tions. For that reason, literature has suggested an infinite number of methods to estimate ex-
pected returns, see for example Merton (1980), Campbell & Shiller (1988), Fama & French
(1988), Lamont (1998), Lettau & Ludvigson (2001) and Thompson & Polk (2006). Still, no single
model has been universally accepted and thus this study uses three different methods. Based on
a thorough review, the three methods have chosen to be the:
1. Historical average of actual returns
2. Historical average of Wilshires GR6 multi-factor model returns
3. Jyske Invests VAMOS score
This section serves as an introduction to the theoretical intuition behind each method and dis-
cusses pros and cons, whereas the technicalities of the methods are covered in chapter 3 on
methodology.

10

2. Theory and literature review


2.1.1.1 Historical average of actual returns
The simplest way to forecast future returns is to use some average of actually realised returns.
Many investors and analysts favour this method, since it does not create a lot of noise and is
easy to implement in opposite to more sophisticated financial models that require numerous
assumptions about the underlying asset. Bodie, Kane & Marcus (2010) argue that if the focus is
on future performance, the historical average is the statistic of interest as it is an unbiased esti-
mate of an asset's future returns. Campbell (2001) explains more detailed that when returns are
serially uncorrelated that is when one years return is unrelated to the next years return the
historical average represents the best forecast of future returns. However, Campbell & Thomp-
son (2008) show in their study that theoretically restricted valuation models often outperform
return forecasts based on historical averages of stock returns.
The average returns are in this paper based on five years of monthly returns, which is in ac-
cordance with the investment horizon in Jyske Invest and the suggestions by Peare & Bartholdy
(2005) and Welch & Goyal (2008).
2.1.1.2 Historical average of Wilshires GR6 multi-factor model returns
Multi-factor linear risk models attribute risk and return to a series of factors that are believed to
drive market behaviour. The return of each security contained within a portfolio is sensitive to
these factors and factor models attempt to quantify this sensitivity and thus relate returns to
movements in the market. Specifically, it is proposed that the contribution to the excess return
of a given security attributable to a particular factor is the product of a securitys sensitivity to
that factor, known as the exposure, and some unknown coefficient referred to as the factor re-
turn. The total excess return of the security is the sum of these contributions plus an error term
of expected value zero. That is:

!"#$ ! !%$ = #&"#' #$ %' #$ +""#$

(2.1)

'

where ri,t is the return of the ith security at time t, rft is the risk free return at time t, ei,k,t is the
exposure of the ith security to the kth factor at time t, fk,t is the factor return of the kth factor at
time t and i,t is the residual return that accounts for any contribution to return not captured by
the factors.
Linear factor models have found increasing favour in the financial world and now possess a rich
history amongst financial practitioners and academics alike. The development of the single fac-
tor CAPM in 1964, in which the expected return of a given security was related to the return of
the market, formed the basis. However, an almost infinite number of studies such as the ones
provided by DeBondt & Thaler (1985) and Jegadeesh & Titman (1993) criticise the predictive

11

2. Theory and literature review


power of the CAPM and according to a study by Peare & Bartholdy (2005) the CAPM on average
only explains three percent of the differences in returns.
Additional work by Merton (1973), Ross (1976), Breeden (1979), Fama & French (1993) and
Carhart (1997) generalised the CAPM to frameworks with multiple factors. Namely the Inter-
temporal CAPM (ICAPM), Arbitrage Pricing Theory, Consumption CAPM (CCAPM), Fama-
Frenchs three-factor model and Carharts four-factor model. Rich academic studies have shown
that average return on common stock are related to firm characteristics like size, earn-
ings/price, cash flow/price, book-to-market equity, past sales growth, long-term past return
and short-term past return, etc. These factors are also referred to as anomalies and are patterns
in the return, which are not captured by CAPM5.
This paper uses Wilshires Global Region 6 multi-factor model (GR6) to forecast future returns
and estimate covariance matrices. It is built around a set of factors that are recognised as being
both intuitive and statistically significant in the analysis of risk and return. The model compris-
es six regional sub-models, each of which describes the behaviour of return in one of the follow-
ing regions: North America, Europe, Asia, Mediterranean, Pacific and Latin America. The region-
al sub-models share a common factor structure and possess a group of country of risk factors,
industry or sector factors and a set of fundamental factors (Wilshire Associates, 2006). A more
technical description of the model and how it is estimated is provided in section 3.2.3 and 3.2.4
when it is first used for mean-variance optimisation purposes.
2.1.1.3 Jyske Invests VAMOS score
Based on the convincing studies of market anomalies presented in the previous section, it seems
relevant to use Jyske Invests earlier introduced VAMOS score as a proxy for expected return as
well. The VAMOS score builds primarily on the value and momentum anomalies. It also includes
strength factors such as the cash flow return on investment, which is a measure of a companys
historical ability to create or destroy wealth over time. Early studies made by Fama (1992) and
Lakonishok, Shleifer & Vishny (1994) have demonstrated that value strategies can be used to
predict future returns. Conflicting explanations have been presented for the success of these
strategies but most empirical evidence remains strong.
Researchers have also convincingly demonstrated that momentum strategies have the power to
predict future stock returns. Jegadeesh & Titman (1993) showed that strategies buying winners
and selling losers based on returns over the previous 6 to 12 month generate excess returns.

5 Besides specifying multi-factor models on the basis of observable factors, statistical factor models such

as principal components models can for example also be used to derive a set of unobservable factors that
explain stock returns.

12

2. Theory and literature review


Empirical evidence in favour of the momentum strategy is quite convincing, but as with value
strategies the explanation of why it works is incomplete.
Interestingly, value and momentum measures are often negatively correlated and as such it
seems difficult to pursue both strategies. Following a value strategy entails, to some extent, buy-
ing firms with poor momentum and buying firms with good momentum entails to some extent
pursuing a poor value strategy. However, according to Asness (1997) holding momentum con-
stant leads to a more effective value strategy in most cases. That is, the value strategy works
best when not forced to short the effective momentum strategy and similarly, holding value
constant leads to a generally superior momentum strategy. The relation of value and momen-
tum to future returns are not simply stronger holding other variables constant; they are condi-
tional upon each other. Value works in general but largely fails for firms with strong momen-
tum. Momentum also works in general, but is particular strong for expensive firms. Finally, it
should be emphasised that the VAMOS score differs from the other two methods as it combines
backward looking and forward-looking measures.

2.1.2 Covariance matrix


Chopra & Ziemba (1993) argue that estimating the covariance matrix is less important than
estimating the expected returns with the argument that portfolio weights are more sensitive to
changes in the expected return vector than changes in the covariance matrix. However, the diffi-
culty of estimating expected returns compared to estimating the covariance matrix implies that
most improvements can actually be achieved in the covariance matrix estimation (Bengtsson &
Holst, 2002).
The literature has proposed several ways of estimating the covariance matrices including a fac-
tor-based approach, a constant correlation approach and a statistical shrinkage approach
(Amenc et al., 2010). In this paper the factor-based approach is applied using the GR6 model
briefly introduced in the previous section. Among other things, the use of a multi-factor model
allows estimation of the covariance matrix based on knowledge of the factor exposures and fac-
tor returns without needing to store the variance of returns of all securities. Moreover, the GR6
model builds on observable and economically intuitive factors in opposite to strictly statistical
factor models such as principal component analysis.
Studies by Ledoit & Wolf (2004b) and Kuberek & Matheos (2005) show that the estimation of
the covariance matrix often leads to bias, which, in turn, translates into error in the calculation
of risk. Kuberek & Matheos (2005) find that risk estimates from nave estimators may be off by
40% or more. To deal with this issue, they construct all matrices using the so-called SHaPTSE
estimator (Structured Hadamard Product Target Shrinkage Estimator), which explicitly treats

13

2. Theory and literature review


the problem of bias and can reduce errors in risk measurement by approximately 60-70%. The
SHaPTSE estimator is also used in the GR6 model to correct for estimation bias.

2.2 Portfolio performance measures


Throughout this thesis the performance of a number of alternatively weighted benchmark indi-
ces and GE portfolios have to be measured and analysed. To ensure a uniform and fair evalua-
tion of these portfolios, a common set of performance measures have been chosen and will be
presented in this section.
The literature has suggested and discussed a great number of different performance measures.
Obviously, not all measures can be included or are relevant for this thesis. Instead the selection
of performance measures is inspired by the measures used in the related studies on alternative
benchmark indices to ensure comparability6. Consequently, this paper divides the performance
measures into three categories:
1. Return and risk (average return and standard deviation)
2. Risk-adjusted ratios (Sharpe ratio, Sortino ratio and information ratio)
3. Extreme risk (95% Value at Risk)
The exhaustive literature review of performance measures by Le Sourd (2007) is used as refer-
ence for all definitions presented in the succeeding sections. Note that a presentation of the
simple return and risk measure will not be provided. In the analysis of alternatively weighted
benchmark indices, a further decomposition of returns into factor exposures is conducted using
the Carhart (1997) 4-factor model, which is introduced in section 2.2.3.

2.2.1 Risk-adjusted ratios


As return and volatility considered in isolation not sufficiently evaluate the attractiveness of a
portfolio, the risk-adjusted ratios add valuable perspectives to the overall performance of a
portfolio. Some of the ratios require a reference index (information ratio) while others can be
calculated without any references (Sharpe ratio and Sortino ratio).
2.2.1.1 Sharpe ratio
A simple and well-known measure of risk-adjusted performance is the Sharpe ratio. The meas-
ure plays a central part in this thesis and is used to characterise the risk/reward efficiency of all
indices and portfolios considered throughout the study.



6 For a comprehensive overview of most performance measures refer to the literature review by Le Sourd

(2007).

14

2. Theory and literature review


Mathematically, the ratio can be expressed as:

!"#$%& $#'()% =

$% ! $*
"%

(2.2)

where rp is the portfolio return, rf is the risk free rate and p is the portfolio standard deviation.
The Sharpe ratio tells an investor how much excess return is generated for each unit of total
risk. The risk/reward ratio provides a meaningful basis to rank the desirability of portfolios. All
other characteristics being equal, rational investors would prefer a portfolio with a greater
Sharpe ratio than one with a lesser Sharpe ratio.
2.2.1.2 Sortino ratio
The Sortino ratio is an extension of the idea behind the Sharpe ratio, but the focus is here on
investors minimal acceptable return (MAR). The Sortino ratio uses the portfolios semi devia-
tion in the denominator instead of the standard deviation. This is particularly useful when the
return distribution is asymmetric or if a specific return target is of particular importance to the
investor. Mathematically, the ratio can be expressed as:

!"#$%&" #'$%"( =

#( !)*+
,
-

" (# !)*+ )

(2.3)

$=,
#( /$ <)*+

where the numerator is the portfolio return (rp) in excess of the MAR and the denominator is
the portfolio semi-deviation (i.e. the standard deviation of the returns that are below the MAR).
2.2.1.3 Information ratio
The information ratio is a powerful tool for assessing the skill of an active portfolio manager,
since it makes it possible to check that the risk taken by the portfolio manager, in deviating from
the benchmark, it sufficiently rewarded. It is often used for evaluating both security selection
and portfolio construction abilities, since it captures the idea that an active portfolio manager
has to deviate from the benchmark in one way or the other. However, due to the information
ratios great dependence on the applied benchmark, the choice of benchmark can change the
result dramatically and care should therefore be taken when interpreting the information ratio
(Goodwin, 1998). The choice of benchmark index and its impact on information ratio and other
performance measures is analysed thoroughly in section 5.3.
Mathematically, the ratio can be expressed as:

!"#$%&'()$" %'()$* =

15

%* ! %+
" ,-

(2.4)

2. Theory and literature review


where the numerator is the active return defined as the portfolio return (rp) in excess of the
benchmark return (rb) and the denominator is the tracking error defined as the standard devia-
tion of active returns (TE). Tracking error is calculated as:
!

#((
'=(

! !" =

#$ " #% " #$ " #%

))

&

(2.5)

The lower the tracking error, the more closely the portfolio follows its benchmark, and vice ver-
sa. Since tracking error has significant importance for this paper, the measure will be further
clarified in a thorough literature review in section 2.4 including a discussion on the appropri-
ateness of using tracking error as performance measure.

2.2.2 Extreme risk (95% Value at Risk)


When assessing portfolio performance it is also necessary to investigate whether its
risk/reward efficiency comes at the cost of a higher risk of extreme losses. For many institu-
tional investors, and in particular pension funds, the level of extreme risk exposure is crucial as
huge declines in the value of their equity portfolios are very critical. A commonly used statistical
risk measure of potential extreme losses is the Value at Risk (VaR). In a single and easy to un-
derstand number it estimates the worst loss an investor can expect to incur over a given time
horizon with a given level of confidence. This study uses a time horizon of one month and a con-
fidence level of 95%. By convention, the worst loss is expressed as a positive number.
The most general VaR measure makes a normality assumption about the shape of the returns,
but since the existence of non-normality in stock and portfolio returns is well known, a modified
VaR measure is preferable. Based on a Cornish-Fischer expansion the modified VaR applied in
this study accounts for skewness different from zero (i.e. return asymmetry) and kurtosis high-
er than three (i.e. fat tails). Mathematically, it can be expressed as:

" %
- 2
!"#$%$&# '() = *+ ! $ ,+ ,/ !- 0+
, !2, 3 !
/,2 !4, 0/ '( +
/1
2.
# .
&

(2.6)

where S is the portfolio skewness, K is the portfolio excess kurtosis, and z is the number of
standard deviations at the VaR confidence level (Favre & Galeano, 2002; Jorion, 2007).

2.2.3 Factor exposures


Besides the measures already introduced, most investors also have an interest in identifying
where the risk/return properties come from. Even though risk exposures are often implicit re-
sults of portfolio construction (e.g. equal-weighted portfolios are more exposed to small caps
than CW portfolios), the investors want to know how exposed they are to certain factors. The

16

2. Theory and literature review


CAPM betas ability to explain some of the differences in portfolio and stock returns is im-
portant for many practitioners and academics. However, due to the CAPM betas inability to
capture systematic risk stemming from other factors than the market it is often beneficial to
extend the CAPM model, so that it captures other well-known risk factors such as value, mo-
mentum and small-cap exposure.
This paper uses the Carhart (1997) 4-factor model to decompose the returns of the alternatively
weighted indices by regressing the returns of each index in excess of the risk free rate on the
excess returns of the market factor, on the Fama & French (1993) factors for the value and
small-cap premium, and on an additional factor representing a momentum strategy7. The value
factor is a portfolio that is long in high book-to-price stocks (value stocks) and short in low
book-to-price stocks (growth stocks). The small-cap factor is a portfolio that is long in low mar-
ket-cap stocks (small stocks) and short in high market-cap stocks (large stocks). The momen-
tum factor goes long in stocks with high recent returns (winners) and short in stocks with low
recent returns (losers).
Mathematically, the Carhart (1997) 4-factor model can be expressed:

!" ! !# = "$%!&'( !$ ! !# +"!"#$$!%#& '() +" !"#$% &'( +"!"!#$%&! '()

(2.7)

where rb is the benchmark return, rf is the risk free rate, rm is the market return, SMB is the re-
turn of the small-cap portfolio, HML is the return of the value portfolio and WML is the return of
the momentum portfolio. market, small-cap, value and momentum represent the exposure of the
benchmark index to each of the four factors.
Despite the benefits of the factor model are highly acknowledged, the theory has still been heav-
ily scrutinised and criticised by numerous theoreticians. Black (1993a; 1993b) and MacKinlay
(1995) argue that the results of Fama & French (1993) were likely an artefact of data mining
since they chose their explanatory variables based on the results of earlier empirical studies.
Further, Kothari, Shanken & Sloan (1995) call in questions of survivorship bias and beta mis-
measurement. For a comprehensive literature review and summary of the empirical findings
obtained when decomposing the returns of indices and mutual funds refer to Cuthbertson &
Nitzche (2010).

2.3 Equity indices as benchmarks


An equity index is a tool used by investors and portfolio managers to describe the market, and
to compare the returns on specific investments. Indices, such as the CW MSCI AC World, FSTE

7 The

global market, value, small-cap and momentum factor portfolios are retrieved from Kenneth
Frenchs website. See Fama & French (2012) for more on the construction of the global factor portfolios.

17

2. Theory and literature review


All-World and Russell World Cap, are used as broad global equity benchmarks for asset alloca-
tion as they provide a consistent and complete representation of the global equity markets
(Amenc, Goltz & Tang, 2011). Several chief investment officers (CIO)8 and portfolio managers
use the equity indices as benchmark for actively managed portfolios, as it is also the case in
Jyske Invest. The CW MSCI AC World index is the centre of attention in this thesis as it is the
benchmark index used for Jyske Invests GE portfolio. The CW MSCI AC World index is widely
tracked and serves as the basis for over 500 exchanged traded funds throughout the world
(MSCI, 2013d).
This section presents the key moments in the evolution of benchmark indices followed by a de-
scription of the quality criteria that in combination with the performance measures presented
in section 2.2 are applied to evaluate the alternatively weighted MSCI AC World indices. A re-
view of the literatures main points of criticism of CW benchmark indices is then conducted be-
fore alternative benchmark indices are presented. Finally, a discussion of why the CW indices
remain the most common point of reference is included.

2.3.1 The evolution of benchmark indices


The history of indices starts back in the 19th century when the first stock market index was in-
troduced. Ever since the beginning of the industry for indexing, indices have been playing a cen-
tral role in portfolio management and the importance of the industry has grown remarkably.
The first index to see the world was the Dow Jones Industrial Average index (DJIA) introduced
in 1896. It still exists today and is a price-weighted index representing an average of 30 stocks
from leading American industries. In 1923, Standard & Poors introduced the first CW stock
index with the purpose of providing information on the markets mood and direction. Since then
the number equity index products have been steadily increasing and especially over the last
couple of years the interest in equity indexing has exploded. As of June 2011 close to $6 trillion
worth of assets worldwide is passively tracking some variant of index product, up from $1 tril-
lion in 2005 (Olsen, 2011).
2.3.1.1 Index providers
In the view of this development many index providers have emerged worldwide; not only or-
ganisations specialising in index services such as Dow Jones, Standard & Poors (S&P), Morgan
Stanley Capital International Inc. (MSCI), FTSE Group, Markit, Russell or Wilshire, but also stock
exchanges such as New York Stock Exchange, Deutsche Boerse, Euronext and COE as well as
investment banks including Barclays Capital, JP Morgan and so forth (Amenc, Goltz & Tang,
2011). In creating the indices each of the providers face a series of decisions such as choice of

8 Defined in this paper as the ones being responsible of managing the portfolio managers and overseeing

their performance.

18

2. Theory and literature review


weighting constraints on constituents, stock universe and rebalancing frequency. Thus the indi-
ces will vary depending on the choices that the index creators make.
The CW MSCI AC World index used as benchmark for Jyske Invests GE portfolio is rebalanced
quarterly and only includes mid- and large-cap stocks from 24 developed countries and 21
emerging countries (MSCI, 2013d).
2.3.1.2 Index users
Monitoring of active portfolio managers depend upon the choice of benchmark index. Active
managers aim to beat their reference index, but since they are often subject to certain tracking
error constraints, the managers have limited opportunities to deviate much from the chosen
index. In this environment active managers in general hold portfolios that are similar to the
index and the overall performance of the active funds can be argued to be more affected by the
choice of index than the specific choices made by the active portfolio manager (Rappaport &
Mauboussin, 2001).
Bailey (1992) underlines the importance of choosing an index based on the objective of the in-
vestment in question and argues that the suitability of a benchmark index can and should be
evaluated. In particular with an increasing number of options available to the investors and
index users, it gets more important (and complicated) to choose an adequate index.

2.3.2 Evaluating benchmark indices


First and foremost the already introduced performance measures attach high importance in the
overall evaluation of the attractiveness of an index used as benchmark. In addition, a range of
quality criteria must be considered as well to ensure that the index in question is implementa-
ble in practice and has properties that fulfil the objectives of its users.
Inspired by Bailey (1992), Siegel (2003), Arnott, Hsu & West (2008), Kamp (2008), Amenc,
Goltz & Tang (2011) and Christopherson (2012) this thesis summarises the quality criteria in
five main qualities; turnover, concentration, representativeness, liquidity and transparency. In
combination with the performance measures these quality criteria constitute this thesiss
framework for evaluating indices as benchmarks. The related studies of alternative benchmark
indices focus primarily on the performance measures and only briefly address some of the qual-
ity criteria. The evaluation framework applied in this thesis and illustrated in figure 2.1 aims to
correct this.

19

2. Theory and literature review


Figure 2.1 - Benchmark index evaluation framework
!"#$%&'(
6
8$'9&':";2$1
:$"*+'$*

)$"*+'$
!"#$%&#'$#()$*

=#()$*'%*,'$-67

+(%*,%$,',#"-%(-.*
+/%$0#'$%(-.

=-67>%,K)6(#,'$%(-.6

+.$(-*.'$%(-.
1*2.$3%(-.*'$%(-.

.+"/0#(1
2'0#$'0"

45($#3#'$-676

<

I)$*."#$

C.*A#*($%(-.*

=#0$#6#*(%(-"#*#66

L-M)-,-(@

I$%*60%$#*A@

89:';%<)#'%('=-67
!**)%<'.*#>?%@'()$*."#$
1*,-22#$#*A#'<#"#<'.2'($%*6%A(-.*'A.6(6
!"#$%&#'#22#A(-"#'A.*6(-()#*(6
=#<%(-"#',#A.*A#*($%(-.*'?B$B(B'CD'E+C1'!C'D.$<,
F#.0$%/-A%<'$#&-.*'#50.6)$#
1*,)6($-%<'6#A(.$'#50.6)$#
C!G'$%(-.
D#-&/(#,'3.*(/<@'H'($%,-*&'".<)3#
J)%<-(%(-"#'%66#663#*(

Notes: The measures used to assess the quality criteria of the alternative indices are introduced when used in section 5.2.2-5.2.6.
Source: Own contribution

2.3.2.1 Turnover
High index turnover means high transaction costs for the portfolio manager or investor tracking
the index and it is therefore preferred to keep turnover as low as possible. The two sources of
turnover are reconstitution and weight rebalancing and vary a lot from index to index. Reconsti-
tution occurs when new stocks enter and existing stocks fall out of the index, while weight re-
balancing occurs when the weights proposed by a given weighting scheme deviate from the
actual weights of the index. By construction CW indices have very low turnover because virtual-
ly the entire turnover arises from reconstitution only as the weights are automatically adjusted
to reflect market prices. On the other hand, alternatively weighted indices must also trade
stocks to adjust their weights in accordance with the weighting scheme dictated by the index
design, which naturally increases turnover. The increased turnover of alternative indices is im-
portant to factor in when assessing how much of their performance is eroded due to the occur-
rence of transaction costs. Finally, the rebalancing frequency of an index also affects turnover
significantly (Siegel, 2003).
2.3.2.2 Concentration
A high concentration in a few companies means that investors are exposed to a large amount of
firm-specific risk, which is not beneficial if investors want a well-diversified portfolio. According
to Markowitz (1952) and modern portfolio theory, diversification should be preferred as it is
efficiency enhancing. CW indices have been documented to be overly concentrated, which is
further elucidated in section 2.3.3 covering the main points of criticism of the CW benchmark
indices.

20

2. Theory and literature review


2.3.2.3 Representativeness
Representativeness is rarely well defined by index providers or theoreticians proposing alterna-
tive index strategies. As such, it remains an outstanding question how to measure whether an
index is representative or not. Following Amenc, Goltz & Tang (2011) and taking into account
that the global CW MSCI AC World index is considered, this thesis defines the representative-
ness criteria as being adequately exposed to all geographical regions9 and industrial sectors
(GICS sectors). Additionally, Christopherson (2012) argues that an index should reflect assets
available to all investors and adjust constituent weights for restricted shares, cross-owned
shares and shares that are not available to the public. All of which are fulfilled by the constituent
selection of the CW MSCI AC World index and therefore not addressed further.
2.3.2.4 Liquidity
Liquidity ensures that an index is tradable by many investors without any significant price im-
pact (Amenc, Goltz & Tang, 2011). This means that at reconstitution, whether a stock is included
in an index or not, the price should not be affected. However, it has been shown that, on average,
when a stock goes into the S&P 500, its price rises, and when a stock is dropped from the S&P
500, its price falls10. Among the reasons for this price change is a supply-and-demand effect.
Generally, one would expect that when a stock enters a popular index, demand for the stock
increases and the effect will be more disturbing when fewer stocks are included in the index
(Christopherson, 2012).
2.3.2.5 Transparency
Transparency focuses on the availability of documentation of the concepts and methodology
used to construct the index as well as availability to current and historical data on index values
and composition. This is critical for investors as transparency helps investors to fully under-
stand what indices are doing. In Arnott, Hsu & West (2008)s definition, transparency also in-
cludes consistent and systematic rules in the index construction methodology so that replica-
tion is possible. Further, Christopherson (2012) points out that the index should only include
assets for which complete and reliable data can be repeatedly obtained in a timely manner.
Transparency is typically a bigger issue for sophisticated alternative indices such as the ones
relying on advanced portfolio optimisation techniques compared to the CW ones.


9 Defined in accordance with Wilshires GR6 multi-factor model to be (1) North America, (2) Europe, (3)

Pacific, (4) Asia, (5) Latin America and (6) Africa/Middle East
10 For example Yahoo!s stock price rose by 24% on December 7, 1999, the day before it was added to the
S&P 500 index, due to an obvious mismatch between supply and demand for the stock (Christopherson,
2012).

21

2. Theory and literature review

2.3.3 Criticism of efficiency of cap-weighted benchmark indices


Having discussed the quality criteria relevant for benchmark indices, the thesis now highlights
the main points of criticism of the CW indices.
CW indices have been criticised ever since the early studies by Haugen & Baker (1991) and
Grinold (1992). The original intention of the CW indices was to measure where the market is
rather than being a tool for long-term investors. However, accurately reflecting how the market
behaves does not necessarily lead to desirable risk/return properties (i.e. high Sharpe ratios)
for investors (Amenc, Goltz & Tang 2011). Hence, the harshest criticism relates to the efficiency
(or maybe more suitable inefficiency) of CW indices as they, in summary, are claimed to suffer
from (1) reliance on the highly unrealistic assumptions behind CAPM, (2) concentration in a few
large stocks and (3) a trend-following feature (Haugen & Baker, 1991; Fama & French, 2004;
Amenc, Goltz & Le Sourd, 2006; Hsu, 2006; Goltz & Le Sourd, 2010). See appendix 1 for a brief
overview of some CW indices that have been tested in empirical studies and their conclusions
on efficiency.
The criticism has escalated over the last few years and resulted in the launch of many new index
offerings trying to correct the shortcomings of the CW indices (Amenc, Goltz & Lodh, 2012). The
three main points of criticism are reviewed separately in the succeeding sections.
2.3.3.1 Reliance on unrealistic assumptions behind CAPM
Haugen and Baker (1991), Fama & French (2004) and Goltz & Le Sourd (2010) have among
many others reviewed the theoretical literature on cap weighting and they uniformly point out
that a prediction that CW indices have the lowest possible volatility given their expected return
must be based on a set of assumptions for the CAPM formulated by Sharpe (1964), which in-
cludes the following.
1. Investors are rational mean-variance investors
2. Investors wealth is entirely described by their holdings in tradable securities
3. It is possible to borrow unlimited amounts at the risk free rate
4. No restrictions on short selling
5. All investors have identical preferences and identical investment horizons
6. Taxes, transaction costs, and other frictions are insignificant for most investors and as-
sets
Haugen & Baker (1991), Fama & French (2004) and Goltz & Le Sourd (2010) argue that if only
one of these strict assumptions is not fulfilled, there will exist alternatives to CW indices that
have the same expected return but lower volatility and hence that CW indices are not efficient in
the risk/reward sense. From the list of assumptions, it becomes clear that it is unreasonable to

22

2. Theory and literature review


expect that all these assumptions hold in practice. After all, investors are unlikely to have the
same preferences and the same investment horizons. In addition, the existence of taxes and
transaction costs is quite real. Nor is unlimited borrowing obtainable for most investors. Like-
wise, not all assets are tradable, and investors may invest significant portions of their wealth in
non-tradable assets (e.g. human capital) or illiquid assets (e.g. real estate). So it is hardly sur-
prising that the model frequently fails in empirical tests. A review of some empirical tests of the
CAPM is provided in appendix 2.
Besides fulfilling the CAPM assumptions Goltz & Le Sourd (2010) emphasise that the stock mar-
ket indices must also be equal to the market portfolio. In reality stock market indices only re-
flect a fraction of wealth in the economy, ignoring the share of wealth represented by human
capital, social security benefits and illiquid assets. Even if it were possible to build and hold the
market portfolio that includes all these assets, the market portfolio would be efficient only if the
before-mentioned assumptions hold. As such, under more realistic assumptions financial theory
does not conclude that the CW benchmark indices are efficient.
2.3.3.2 Concentration in a few large stocks
Another characteristic of the CW indices that has been strongly criticised is the concentration in
a few large stocks11. Malevergne, Santa-Clara & Sornette (2009) argue that CW indices in gen-
eral are heavily concentrated in a few large firms. By construction, cap weighting gives some
companies large weights which leads to concentrated portfolios. Usually the 10-30 largest
stocks make up the majority of the weighting in the index. For example, as of 31 May 2013 the
10 largest stocks of the CW MSCI AC World index (out of 2428 in total, i.e. 0.41% of all stocks)
account for 8.42% of the total index capitalisation as illustrated in figure 2.2.
Figure 2.2 - Top 10 constituents in the CW MSCI AC World index

104%
1>4%
C4%
H4%
/4%
>4%

0>13%

04%

0>10%

12304
120@4
>2C@4
>2@H4
>2@/4
>2@34
>2@34
>2HN4
>2H@4
>2H?4
CDE?F

0>11%

/00
/>?
0@C
0/0
03C
03?
033
00>
013
01>
?@AB

0>1>%

&'+,%-$(.
5'$;<=
&'+,%-$(.
&'FGAB;9E#A
5'$;<=
L$E#B.%(E;$
&'+,%-$(.
&'+,%-$(.
:,'2%*BE"#$A
:,'2%*BE"#$A

G/"#"/*+"($"H$*"#$%&$'"()*+*,-(*)I$
?&&@J?&%K

0>>N%

)*
)*
)*
)*
)*
)*
)*
)*
*P9BQ$;#E'F
)*

2(4-5$<-+=>*

0>>C%

12.3$)-'*"/

0>>@%

!""#$%&'(
566,'%7,89#%:,
79(;,A,+B%:,
D$'$;E#%5#$(%:,
:.$I;,'%:,
J,.'A,'%K%J,.'A,'
D,,<#$%&'(
&'B$#%MGA9'$AA%7(.'
O$AB#$%*E
R;,(B$;%K%DES8#$%:,
!"*67$

.",(*/0

0>>H%

!"#$%&$'"()*+*,-(*)$

2(4-5$
'6#+*67+)6*+"($
89$:+77+"();

Note: Constituent list is retrieved as of 31 May 2013.


Source: Own analysis, Wilshire Atlas


11 Note that concentration is included as one of the quality criteria in the overall evaluation of the alterna-

tively weighted MSCI AC World indices.

23

2. Theory and literature review


2.3.3.3 Trend following
Besides suffering from concentration in a few large stocks, CW indices also tend to exhibit trend
following in the sense that at certain stages some industrial sectors have a very dominant
weight in the index (Amenc, Goltz & Le Sourd, 2006). This is simply explained by the fact that
CW indices assign weight to stocks only by their market capitalisations, which is the product of
the stock price and the total amount of stocks. If there is no new stock offered, the weight of any
stock depends solely on the stock price. Whenever the stock price goes up, the market capitali-
sation of that company goes up. The opposite thing will happen whenever the stock price goes
down (Amenc, Goltz & Tang, 2011). An example of this is illustrated in figure 2.3 below, which
shows the historical proportion of financial stocks in the CW and equal-weighted (EQW) ver-
sions of the MSCI AC World index12. It clearly shows the trend-following tendency of the CW
MSCI AC World index relative to its EQW counterpart. In particular, during the financial crisis
the proportion of stocks invested in financials for the CW MSCI AC World index dropped from a
peak of 25.9% in January 2007 to a low point of 16.1% in March 2009 corresponding to a 38%
decline. Conversely, the EQW MSCI AC World index only dropped 7% over the same time period.
Figure 2.3 - Proportion of financial stocks in the CW and EQW MSCI AC World index
("#"$%

-=I%J.5G.CK%'""*%L%'&#,$%
EH=I%?.CM:%'"",%L%!+#*$%

'&#"$%

'"#"$%

!&#"$%

EH=I%J.5G.CK%'""*%L%'"#!$%
-=I%?.CM:%'"",%L%!)#!$%

!"#"$%
'"")%

'""*%

'""+%

'"",%

'"!"%

-./01.203.104567809:18;%<-=>%?@-A%B-%=4C2;%05;8D%

'"!!%

'"!'%

'"!(%

EFG.267809:18;%<EH=>%?@-A%B-%=4C2;%05;8D%

Source: Own analysis, Wilshire Atlas

2.3.4 Alternatives to cap-weighted benchmark indices


In response to the criticism of the efficiency of CW indices, numerous alternatively weighted
indices have recently been proposed. Among the most popular are (1) equal weighting, (2) fun-
damental weighting, (3) minimum-volatility weighting and (4) maximum-Sharpe ratio (effi-
cient) weighting. Besides being the centre of attention of theoreticians, these four alternatives


12 The EQW version of the MSCI AC World index is simply constructed by setting each constituent weight

equal to 1/N at year-end, where N is the number of constituents in the index at each rebalancing date. For
more on EQW indices refer to section 2.3.4.1.

24

2. Theory and literature review


have also been adopted by most of the main index providers13 and therefore this study has cho-
sen to focus on these approaches.
In the following a comparison of the concept behind each of these weighting schemes will be
conducted in order to assess how they differ from one another. In particular, the focus will be on
their different objectives and how they use very different types of information to attribute
weights, including risk/return data, accounting data, or even ignoring any data, as in the case of
equal weighting.
2.3.4.1 Equal-weighted (EQW) indices
In an EQW index, the same weight is simply attributed to each of the constituents regardless of
any market information available. This diversification rule is known as nave diversification.
Compared to CW indices, the two main benefits of equal weighting are (1) lower stock concen-
tration and (2) less trend following (as shown in figure 2.3). According to Zeng, Dash & Guarino
(2010), Chow et al. (2011), Amenc et al. (2011) and Clare, Motson & Thomas (2013a) this leads
to more favourable risk/reward properties, despite giving up deriving any form of useful infor-
mation for weighting the constituents. The main criticism of equal weighting is besides the na-
ivety and increased turnover, higher liquidity constraints since all stocks in the index are given
the same weight regardless of market capitalisation (Zeng, Dash & Guarino, 2010). Hence, at
rebalancing the demand for the smallest stocks will be relatively higher than that of the biggest
stocks, which can potentially produce liquidity pressure.
2.3.4.2 Fundamental-weighted (FW) indices
In FW indices the constituents are weighted (and in some cases selected) by non-capitalisation
measures such as gross revenue, total book value, gross sales, gross dividends, cash flow, and
total employment. These measures serve as proxies of fundamental value.
FW indices can differ from CW indices in two respects; (1) constituents are selected based on
their fundamental characteristics rather than capitalisation and (2) the weights are determined
based on their fundamental characteristics rather than capitalisation. In an overview of some
alternative equity indices, Amenc, Goltz and Lodh (2012) show that some index providers, such
as FTSEs RAFI index, change both stock selection and weighting scheme when constructing
their FW indices while others, such as MSCIs AC World Value Weighted index, only change the
weighting scheme. Similarly, some theoreticians also construct FW indices changing both as-
pects, e.g. Arnott, Hsu & Moore (2005) and Chow et al. (2011), while others only change the
weighting scheme, e.g. Clare, Motson & Thomas (2013b).

13 EQW indices (S&P and MSCI), FW indices (FTSE), MVW indices (MSCI) and MSRW indices (FTSE) to

name a few.

25

2. Theory and literature review


Arnott, Hsu & Moore (2005), Chow et al. (2011), Amenc et al. (2011) and Clare, Motson &
Thomas (2013b) demonstrate uniformly that their FW indices deliver superior meanvariance
performance relative to CW indices. However, other studies including Perold (2007) and Goltz,
Amenc & Le Sourd (2009) argue that fundamental weighting is just a value strategy to capture
the well-documented value premium and hence does not add any further insights not already
provided by Fama & French (1993).
2.3.4.3 Minimum-volatility-weighted (MVW) indices
Minimum-volatility weighting is an initial approach to building scientifically, rather than naive-
ly, diversified portfolios and relies on Markowitz (1952)s portfolio optimisation theory. The
focus here is on finding the constituent weights that lead to the lowest possible portfolio volatil-
ity and thus only the covariance matrix is required (Amenc et al., 2011). Chopra & Ziemba
(1993) introduced the method and suggested that portfolio outcome could be improved by as-
suming that all stocks have the same expected returns. Under this seemingly harsh assumption,
the optimal portfolio is the minimum-variance portfolio. Moreover, Chopra and Ziemba (1993)
examined the relative impact of estimation errors in expected returns, variances, and covari-
ances. They find that (1) errors in expected returns are more than ten times as important as
errors in variances and (2) errors in variances are about twice as important as errors in covari-
ances. As a result, the minimum-volatility portfolio, which does not rely on expected return
forecasts, might be the best way to create an efficient portfolio. Superior risk-adjusted returns
to the CW indices can be explained by exposure towards low beta stocks, low market-cap stocks
and often to value stocks as well (Nielsen & Aylursubramanian, 2008).
MVW indices are typically very concentrated in the stocks with the lowest volatility, which
brings along concerns similar to those mentioned for CW indices. The high concentration in
MVW indices is a widely recognised issue; see Karceski & Lakonishok (1999), Clarke, de Silva &
Thorley (2006) and DeMiguel, Garlappi & Uppal (2009). Further considerations relate to the
opaqueness of the methodology and data used to the construct MVW indices (Nielsen & Aylur-
subramanian, 2008).
2.3.4.4 Maximum-Sharpe ratio-weighted (MSRW) indices
Given that all stocks are unlikely to have the same expected returns, a MVW index is theoretical-
ly unlikely to be the index with the maximum ex ante Sharpe ratio. To improve upon MVW indi-
ces, some theoreticians therefore advise to incorporate useful information on future stock re-
turns.
Amenc et al. (2010) argue that as long as robust parameter estimates for expected returns and
covariance matrices are used, the optimisation-based MSRW index produces greater efficiency
than any other index. In addition, they state that for a rational investor the goal is not to have

26

2. Theory and literature review


the most representative portfolio or the portfolio with the lowest risk; it is instead to hold the
portfolio that achieves the highest risk-adjusted performance. As such, they construct a MSRW
index and claim that this index provides higher welfare for investors in the mean-variance space
and could thus be seen as more representative of the equity risk premium accessible in the
stock market.
The main challenge of the MSRW index is naturally to accurately estimate the covariance matrix
and expected returns of all stocks in the index. Martellini (2008) and DeMiguel, Garlappi & Up-
pal (2009) address these issues and argue that the presence of estimation error in input param-
eters almost entirely invalidates the performance of formal optimisation-based portfolios.

2.3.5 Why still cap-weighted benchmark indices?


Despite heavy criticism, the CW indices are still to this day the most common point of reference
for both passive and active investment funds. Reasons for the continued use can be found in
both theoretical and practical advantages, and this section highlights the most important argu-
ments.
2.3.5.1 Theoretical justifications
The most basic theoretical reason for the continued use is that weighting by market capitalisa-
tion is an objective way of measuring the relative economic importance of index constituents.
Shares of companies offered in the market comprise the opportunity set for investors and valu-
ing available shares at market prices clearly measures the markets assessment of the relative
values of the companies (Christopherson, Cario & Ferson, 2009).
Further, proponents of CW indices argue that capitalisation weighting has a property called
macro consistency. According to Siegal (2003) macro consistency is present if all investors hold
CW index funds, and if there were no active investors. In this case all shares would be held with
no left over. With other weighting schemes, not all investors could hold the index due to their
inferior liquidity and investment capacity. A CW index is therefore a better representation of a
typical investors opportunity set than other methods.
Philips et al. (2011) also support CW indices and highlight that the best index is not necessarily
the one that provides the highest return over a given period, but the one that most accurately
reflects the market. Interestingly, their final point illustrates that a large part of the dispute
stems from completely different views on the purpose of equity indices. Some emphasise quali-
ty criteria such as reflecting the mood of the market, low turnover, high liquidity, etc. while oth-
ers emphasise superior performance measures. Applying the benchmark index evaluation
framework illustrated in figure 2.1, this thesis assesses how each of the investigated indices
performs in terms of performance measures as well as quality criteria.

27

2. Theory and literature review


2.3.5.2 Practical justifications
In addition to the theoretical advantages of CW indices, a number of practical advantages for the
investors and portfolio managers also exist. First, CW indices require less rebalancing than oth-
er weighting scheme. Index adjustments due to corporate actions, such as mergers and spin-
offs, and dividends can result in weight changes for any type of weighting scheme. But changes
in prices alone do not result in adding or removing shares in a CW index as it automatically re-
mains cap-weighted after a change in prices. Other weighting schemes require periodic re-
balancing trades to bring the weights back to those required by the scheme. For an investor
attempting to replicate the index, these trades incur transaction costs, which are not required
by the essentially buy-and-hold strategy of a CW index (Christopherson, Cario & Ferson, 2009).
Amenc et al. (2012) add that when moving away from CW indices to enhance diversification and
increase risk-adjusted performance over long horizons, it should be recognised that each alter-
native to the CW scheme will expose an investor to two related types of risk, namely, (1) model
selection risk and (2) relative performance risk.
Model selection risk addresses the fact that different models (e.g. weighting schemes) are fa-
voured by different market conditions. As such there is a risk that the chosen model may not
yield attractive performance in a given period. Relative performance risk to the CW benchmark
indices arises since alternative benchmarks lead to choices of factor exposures that are different
from those of the CW ones. In order to generate outperformance, the alternative strategies need
to deviate from the default CW indices, which will lead to significant levels of relative risk, i.e.
tracking error (Philips et al., 2011).
Lastly, first mover concerns have influence, since, in particularly, the CIO takes on considerable
reputational risk, when deciding to deviate from the CW indices. First, because he no longer has
anyone but him self to blame, if an underperformance occurs. Second, because CW indices rep-
resent a common reference for the peer group of CIOs to whom a relative risk is thus created. So
as a first mover, the CIO will have a hard time comparing the performance under alternative
weighting schemes with the peer group. In fact, Amenc, Goltz & Tang (2011) find that CW indi-
ces, despite the fact that their shortcomings are widely acknowledged, are likely to remain the
most practically relevant reference for equity portfolios for some time to come, due to their
popularity, the extensive availability of track records, and the fact that they represent the aver-
age decision of investors.

2.4 Tracking error constraints


As briefly introduced in section 2.2.1.3, tracking error measures the deviation of the return of a
portfolio from the return of its benchmarks. As such, it tracks how closely the portfolio returns

28

2. Theory and literature review


follow the benchmark returns and is widely used as a key performance measure by both CIOs,
portfolio managers and individual investors. In the following section a brief review of the litera-
tures main points of criticism of tracking error constraints are presented. The thesis then
moves into a discussion of the objective and appropriateness of tracking error for different
portfolio strategies and different stakeholders as large differences exist.

2.4.1 Criticism of tracking error constraints


The effect and use of tracking error has been widely discussed in the literature. One of the first
comprehensive studies of tracking error was made by Roll (1992). He studied the effect of a
tracking error criterion, where portfolio managers are obliged to find the portfolio with mini-
mum tracking error for a given expected return relative to the benchmark. The study showed
that portfolios with tracking error constraints are not total return mean-variance efficient if the
benchmark is not total return mean-variance efficient. Therefore, tracking error managed port-
folios are dominated by other feasible portfolios that have both lower volatility and higher av-
erage return, i.e. higher Sharpe ratios.
Jorion (2003) supports this conclusion and adds further, that when the focus is on excess re-
turns to a benchmark, the active manager ignores the total risk of the portfolio. Clearly, the
measure focuses strictly on relative risk as opposed to focusing on the additional return the
investor might get by taking on more total risk.
In an empirical investigation of 148 actively managed US funds tracking the CW Russell 1000
index, Hope (2008) shows a linear relationship of portfolio returns falling as tracking error de-
clines, which further supports Roll (1992) and Jorion (2003).
Finally, the intentions that tracking error induces for the portfolio managers is also subject to a
lot of criticism. This is closely connected to the agency problem discussed in section 2.4.2.2.
Straying too far from the pre-defined benchmark index may have severe career implications
for portfolio managers. The disappointing performance of actively managed portfolios relative
to market indices are well documented, and the most plausible explanation lies according to
Rappaport & Mauboussin (2001) in the apparent financial incentives of portfolio managers.
Portfolio managers widely suffer from what they call benchmark addiction where portfolio
managers focus on short-term relative performance and are hypersensitive to any deviations
between their returns and those of the benchmark. In such environment a portfolio managers
preoccupation with short-term returns relative to an index, rather than maximising risk-
adjusted long-term returns is not surprising. A study by Cremers & Petajisto (2006) found that
one-third of actively managed US funds are closet indexers and hence suffer from benchmark
addiction. Moreover, it is shown that actively managed funds with low tracking error budgets

29

2. Theory and literature review


exhibit lower alpha, higher beta, and lower average performance compared to funds with high
tracking error budgets (Chen, Noronha & Singal, 2006; Israelsen & Cogswell, 2007).

2.4.2 Tracking error from different perspectives


Despite the drawbacks of using tracking error discussed above, the tracking error measure has
emerged as a primary measurement for evaluating the performance of portfolio managers. As
mentioned previously the objective and appropriateness of tracking error constraints depends
heavily on the chosen portfolio strategy (active or passive) and the stakeholder of a given port-
folio (CIO, portfolio manager or individual investor). As such, this discussion is divided into a
section on portfolio strategy and a section on the agency problem.
2.4.2.1 Passive vs. active portfolio strategies
The objective of a passive portfolio strategy is to construct a portfolio that behaves as much as
possible as its benchmark. It is commonly referred to as indexing and involves minimal expecta-
tional input and instead relies on diversification to match the performance of the benchmark
(Philips, Kinniry Jr. & Schlanger, 2013). Therefore, tracking error is naturally a highly relevant
measure when assessing the performance of a passive portfolio manager where a low tracking
error is viewed positively.
The objective of an active portfolio strategy on the other hand is to outperform the benchmark
in terms of risk-adjusted performance after management fees. To fulfil this, available market
information and various forecasting techniques are used to seek excess returns. For this strate-
gy, the appropriateness of tracking error as a performance measure seems questionable, as an
active manager naturally has to deviate from the benchmark to achieve outperformance. In this
environment tracking error is counterproductive if minimising tracking error becomes a goal
unto itself (Israelsen & Cogswell, 2007).
2.4.2.2 The agency problem
Using tracking error as a performance measure addresses an agency problem, as portfolio man-
agers are often more concerned with the variability of the excess returns above the benchmark
(i.e. tracking error), whereas the investors are more concerned with the variability of the total
returns (i.e. total volatility). According to Wilcox (2000) additional risk will be oriented towards
tracking error hence the portfolio managers aversion to the total risk experienced by the indi-
vidual investor will be disregarded. As a result the portfolio managers have no incentive to pay
a premium for low absolute risk despite an obvious attractiveness seen from the individual in-
vestors perspective. Consider, as an example, a portfolio with the same expected return as the
benchmark, but lower total risk. This portfolio ought to be preferred over the benchmark but in
extreme cases where tracking error is the key performance measure this will not be the case.

30

2. Theory and literature review


The higher tracking error will be penalised more than the risk reduction will be rewarded and
this chain of reasoning does not seem beneficial for the individual investor.
As mentioned, there is something of a logic paradox associated with the use of tracking error
around a benchmark as a risk measure for active management. Individual investors hire active
managers because they are believed to possess special information or skill that will allow for
better risk/return trade-offs. If such special skills exist, then the mean-variance efficient frontier
for that manager will be above the frontier defined by the benchmark portfolio (DiBartolomeo,
2000)
To sum up, despite different perspectives, the literature suggests that a strict constraint on
tracking error (i.e. benchmark addiction) leads to a more inefficient portfolio performance, es-
pecially when the applied benchmark is inefficient. Moreover, the appropriateness of using
tracking error as performance measure seem appropriate for passive portfolio managers
whereas active managers must be allowed an adequate tracking error budget to be able to beat
their benchmark.

31

3. Methodology

3. Methodology
Having laid the theoretical foundation and conducted a literature review within the fields of port-
folio construction, performance measures, benchmark indices and tracking error constraints, this
chapter describes the methodology used to perform the empirical studies of this thesis. The ulti-
mate goal is to provide valid and reliable answers of the established research questions. First, a
general introduction to the backtesting methodology used throughout the thesis is presented. Af-
terwards, the thesis provides a more in-depth description of the methodology applied to answer
each of three sub-research questions on index construction, impact of benchmark indices and im-
pact of tracking error constraint, respectively.

3.1 Introduction to the backtesting methodology


Coherent with the literature on index construction and evaluation of portfolio construction
strategies, this entire study is set up in a controlled backtest environment with full disclosure of
data sources, parameters and estimation methodologies.
Backtesting, also known as investment simulation, is the process of testing an investment strat-
egy over a historical period to assess its efficacy. It is accomplished by reconstructing trades
that would have occurred in the past, using rules defined by a given investment strategy. Based
on the performance of the resulting portfolio, it is possible to gauge the effectiveness of the
strategy and help determining whether or not the current investment strategy should be al-
tered.
To backtest and construct alternatively weighted versions of the MSCI AC World index and GE
portfolios, access to the historical constituents of the CW MSCI AC World index and GE is a pre-
requisite. This proprietary data is provided by Jyske Invest and covers the period back to 2006.
Even though backtesting is the most popular way of evaluating investment strategies by both
theoreticians and practitioners, it has a number of limitations. Most importantly, is the central
and flawed assumption that markets will always behave in a similar manner to the chosen
backtesting period. However, as everyone knows the dynamics of the equity market are con-
stantly changing. Therefore, a strategy that has worked well in the past may start to falter or fail
completely when traded live. Similarly, a strategy that produces poor historical results may
suddenly spring into life when traded live (Quantshare, 2013). To counteract this risk, this the-
sis, in addition to using the entire backtesting period, also tests the performance of the con-
structed indices over three disparate time periods with different market trend characteristics:

January 2006 June 2008 (pre-crisis)

July 2008 December 2010 (financial crisis)

32

3. Methodology

January 2011 May 2013 (sovereign debt crisis)

This follows suggestions made by the related studies on alternative indices and serves as a ro-
bustness test of the performance of each index constructed in this study.
Another limitation of backtesting is its inability to account for the unique and random events
that occurs in markets from time to time. Unexpected or extreme risks are ever present in the
market and are likely to bear little correlation to unexpected events of the past. Events such as
the burst of the dot-com bubble, 9/11, the latest financial crisis and recent sovereign debt crisis
are all examples of this. It is therefore only to be expected that any strategy employed may need
to be changed to accommodate these changing market dynamics (Forex Technical Chartist,
2013).
On a general note, it is of great importance to remember that backtesting can only validate how
a strategy would have performed in the backtesting period. It is never able to make predictions
on whether the strategy will work in current or future markets. However, it remains the most
popular and reliable method for evaluating investment strategies. As such, backtesting is the
applied method in this thesis.
Considering the extensive amount of backtests performed in this study, it has been chosen to
carry them out using the Wilshire Atlas software package provided by Jyske Invest. It is a very
flexible and reliable tool for making backtests and portfolio optimisations and is used in prac-
tice by investment banks all over the world, including Jyske Invest. Also, it has directly access to
the required proprietary data on the historic constituent stocks of the CW MSCI AC World index
and GE. For a stepwise illustration of how a backtest is carried out using Wilshire Atlas refer to
appendix 3.

3.2 Constructing alternatively weighted MSCI AC World indices (SRQ 1)


In this section a description of the methodologies used to construct alternative indices are pro-
vided. As outlined in section 2.3.4 on alternatives to CW indices, the four most popular alterna-
tive weighting schemes are:
1) Equal weighting
2) Fundamental weighting
3) Minimum-volatility weighting
4) Maximum-Sharpe ratio weighting
Consequently, this thesis constructs alternative indices based on each of these different
weighting schemes to answer if superior weighting schemes to the currently used CW indices
exist. The applied construction principles are inspired by but not a replication of the methodol-
ogies introduced in the related studies. This allows for interesting comparisons of the perfor-

33

3. Methodology
mance measures and quality criteria introduced in the literature review between the indices of
this thesis and those presented in the related studies. At the same time, it enables this thesis to
construct indices that are corrected for some of the pitfalls related to the construction princi-
ples used in the related studies.
Common to all the constructed indices in this thesis is the stock universe, which in every case is
constituted by the historical constituent stocks of the CW MSCI AC World index. This ensures
that the changes in index performance and quality are exclusively attributable to the chosen
weighting scheme. Keeping the stock universe fixed, acknowledges the criticism posed by
Amenc, Goltz & Shuyang (2012) of numerous articles comparing alternative indices applying a
flawed methodology. In particular, Arnott (2011) and Chow et al. (2011) are criticised for com-
paring indices that involve stock selection with indices that simply change the weighting
scheme within a given universe. Amenc, Goltz & Shuyang (2012) claim that it is not a correct
way to analyse different indexing schemes, as the comparison is not based on the same uni-
verse. Instead, it must be recognised that different weighting schemes can be applied to any
selection of stocks.
Also, as Jyske Invest utilises the standard CW MSCI AC World index as benchmark, it is naturally
more relevant and fair to apply the same stock universe as this benchmark index, when deter-
mining the impact of alternative benchmark indices on GE compared to its actual performance
(SRQ 2).
Except for the FW index, each weighting scheme is backtested with annual, semi-annual and
quarterly rebalancing to test robustness to different rebalancing frequencies14. Annual re-
balancing is carried out on the basis of end-of-day prices on the last trading day each year. Semi-
annual rebalancing is based on end-of-day prices on the last trading day of June and December,
whereas quarterly rebalancing relies on end-of-day prices on the last trading day of March, June,
September and December. Several other robustness tests are conducted for the two optimisa-
tion-based weighting schemes, minimum-volatility and maximum-Sharpe ratio, which will be
explained in more details in section 3.2.3 and 3.2.4.

3.2.1 Equal-weighted MSCI AC World index


To construct the EQW index, the weight of each constituent in the CW MSCI AC World index is
simply set to 1/N, where N is the number of constituents in the index at each rebalancing date.
As it does not rely on any available market information except which stocks historically consti-

14 As this paper constructs the FW index on the basis of annual accounting measures, testing the impact of

more frequent rebalancing is not relevant. Arnott, Hsu & Moore (2005) applied quarterly accounting
measures and tested the impact of monthly, quarterly and semi-annual rebalancing, but found that index
turnover increased without any appreciable return advantage over annual rebalancing.

34

3. Methodology
tuted the CW MSCI AC World index, it is very straightforward to implement and construct. MSCI
also constructs an EQW version of the standard CW MSCI AC World index (called MSCI AC
World Equal Weighted Index) but uses quarterly rebalancing in opposite to annual rebalancing,
which is the default choice of this thesis (MSCI, 2013a). For a performance comparison between
the indices of this thesis and MSCI refer to appendix 4.

3.2.2 Fundamental-weighted MSCI AC World index


In opposite to the EQW index, applying fundamental weights is more cumbersome as it relies on
a number of non-capitalisation measures of company size such as sales, revenue, cash flow, div-
idends, total employment and total book value of assets. The applied construction principles
draw heavily upon the arguments presented in the highly debated study on fundamental index-
ation by Arnott, Hsu & Moore (2005). However, to construct a FW index that matches the goal of
this thesis, some deviations are required.
The main deviation from Arnott, Hsu & Moore (2005) is actually a question of how fundamental
weighting is defined. Arnott, Hsu & Moore (2005) argue that adopting fundamental indexation
is more than simply changing the basis for weighting the stocks in an index. Therefore they rank
all companies in the U.S. equity universe by each measure and then select the 1,000 largest by
each measure. In other words, they propose to deviate from the stock universe applied by the
indices they intend to compare themselves to. This is not coherent with the aim of this thesis,
which is merely to compare the effect of different weighting schemes and thus the same MSCI
AC World stock universe is kept fixed.
To construct a FW index the point of departure is a decision on which measures of company size
to include. This thesis uses the same four measures as proposed by Arnott, Hsu & Moore (2005)
when constructing their composite index. The four measures are:

The past years total book value (book)

Trailing 5-year average gross sales (sales)

Trailing 5-year average cash flow (cf)

Trailing 5-year average gross dividends (div)15

Total employment is excluded because that information is not always available, whereas reve-
nues are excluded because sales and revenues are very similar concepts.


15 The treatment of dividends as a measure requires some explanation as non-dividend paying companies

is not necessarily a sign of weakness. For example, shareholders in fast growing companies may accept
100 percent earnings retention. Therefore, non-dividend paying companies are treated differently from
low-dividend paying companies. When a company is not paying dividends, this thesis uses the average of
the remaining three size measures instead of the full four size measures.

35

3. Methodology
To calculate the weight for each constituent in the CW MSCI AC World index at each rebalancing
date, the following two-step calculation is performed:
1. For each of the four measures, the constituents of the CW MSCI AC World index are giv-
en a relative weight for that measure. Expressed in mathematical terms using total book
value as an example and with N number of constituents, the relative weight w of constit-
uent i is calculated as:

! "##$ %& =

"##$%&
(

(3.1)

!"##$%&
&='

2. The final composite weight of each constituent is calculated by taking the simple aver-
age of the four relative weighs for each constituent:

! "#$%&'# =

! ())* '# + ! +%&,+'# + ! -" '# + ! .#/ '#


0

(3.2)

The FW index is rebalanced on the last trading day of each year using the most recent annual
reports for each constituent. In most cases, this process means using data that are lagged by at
least two quarters as most annual reports are released during the spring16. Following the sug-
gestions by Arnott, Hsu & Moore (2005) trailing five-year averages are used for gross sales, cash
flow and gross dividends, as substantial volatility in the index weights will result from using
year-to-year data for these measures. In turn this will lead to higher turnover without any im-
provement on performance. An illustrative example of how to construct a FW index containing
10 stocks is shown in appendix 5.
As a final remark, constituents for whom only two out of four measures are obtainable are not
included in the index. Also, three heuristic data validation tests are carried out to remove ex-
treme outliers in the retrieved accounting measures; (1) all constituents with trailing 5-year
average cash flow above trailing 5-year average gross sales are excluded, (2) all constituents
with trailing 5-year average gross dividends above trailing 5-year average gross sales are ex-
cluded and (3) all constituents with past years total book value larger than 50 times the 5-year
average gross sales are excluded. These restrictions lead to an average annual exclusion of 3.8%
of the constituents of MSCI AC World, which is not considered to be significant.
MSCI also constructs a FW version of the standard CW MSCI AC World index (called MSCI AC
World Value Weighted Index) but uses quarterly rebalancing and substitutes dividends with
earnings (MSCI, 2013c). For a performance comparison between the indices of this thesis and
MSCI refer to appendix 4.

16 Using lagged data is a sensible way of preventing look-ahead bias (Chow et al., 2011).

36

3. Methodology

3.2.3 Minimum-volatility-weighted MSCI AC World index


To construct the MVW index, this thesis relies on Markowitzs mean-variance optimisation
framework described in section 2.1 on portfolio construction. However, as finding the constitu-
ent weights that lead to the lowest possible portfolio volatility is the sole objective of this
weighting scheme, only the covariance matrix is required as input parameter.
As mentioned previously the covariance matrix is estimated using Wilshires GR6 multi-factor
model introduced in section 2.1.1.2. To describe the estimation procedure, equation (2.1) is
rewritten to reflect the GR6 model factor structure for each of the six regions:

!"#$ ! !%$ = "&"#' #$ %' #$ +


'

"

#"#"()*+&,#$ %"()*+&,#$ +

"()*+&,

"

#"#,-.($!/ #$$"#$ %,-.($!/ #$ +%"#$

(3.3)

,-.($!/

where ri,t is the return of the ith security at time t, rft is the risk free return at time t, ei,k,t is the
exposure of the ith security to the kth fundamental factor17 at time t, fk,t is the factor return of
the kth fundamental factor at time t, i,ind/sec,t is the exposure of the ith security to its industry or
sector group18 at time t, find/sec,t is the industry or sector factor return at time t, i,country,ti,t 19 is
the exposure of the ith security to its particular country of risk20 at time t, fcountry,t is the country
of risk factor return at time t and i,t is the residual return. Note that the factor exposures
i,ind/sec,t, and i,country,t are dummy variables, that take the value 0 or 1 depending upon whether
or not the security belongs to the industry group or country in question. The exposures to the
fundamental factors are expressed as normalised z-scores of the raw values of each fundamen-
tal factor. For a given factor, the z-score of the ith security is given by:

!"#$%&'( =

) ( ! ()

(3.4)

" ()
"

where Xi is the ith securitys raw fundamental item value, ! is the average of the Xth fundamen-
"

tal item and ! is the standard deviation of the Xth fundamental item.
Having specified the model, the factor returns are estimated through cross sectional least
squares regressions based on the observable security exposures to each factor. That is, each

17 The following five fundamental are incorporated: Market capitalisation, E/P ratio, B/P ratio, total secu-

rity return over the initial eleven months of the trailing year (momentum) and total monthly return vari-
ance over the trailing 24 months.
18 GICS-sector groups are used for Latin America, Mediterranean and Asia, whereas the more granular
GIGS-industry groups are used for North America, Europe and Pacific as more data is available.
19 is the securitys CAPM beta measured against a local market index from the securitys country of
i,t
risk.
20 A securitys country of risk is subtly different from its country of incorporation, although these two
variables often take the same value. Generally, the country of risk of a security is the same as its country
of incorporation, unless the issuer assumes most of its business risk elsewhere. In this case, the country of
risk is simply the country to which the security has most risk.

37

3. Methodology
securitys factor exposure is observed from a cross-section of stocks at some fixed time t. With
exposures in hand, the factor returns themselves are estimated, at the same time t, using a mul-
tivariate regression procedure. By repeating this process for a number of time steps, the covari-
ance matrix of the factor returns is estimated. Throughout this thesis 5-year monthly data is
used as this time horizon is in accordance with most literature and matches Jyske Invests rec-
ommended investment horizon. Further, the equal-weighted covariance matrix is preferred
over the exponential-weighted, so that older factor returns are given the same weight as recent
factor returns in opposite to putting more weight on the most recent factor returns. This choice
has been made to avoid the risks pointed out by Littermann & Winkelmann (1998), who argue
that using exponential-weighted covariance matrices is similar to using a shorter time horizon.
This results in noisier covariance estimators arising from not exploiting enough information to
get well-behaved covariance estimators. In section 5.2.7.4 the robustness of the optimisation-
based indices to these choices is tested by applying covariance matrices estimated using daily
data (500 days) and exponential weighting. The (i,j)th element for each of the two weighting
schemes are calculated as:
+

*
!! "#$%&'() = # ,('- " ,( ,)'- " ,)
+ -=*

)(

+
* +
!! "./'() = # $+"-+* ,('- " ,( ,)'- " ,) ' 0 = # $+"-+*
0 -=*
-=*

)(

(3.5)

where !! "# is the covariance matrix of the ith and jth factor returns, k is the number of time steps
over which the covariance is estimated, fi,t is the factor return of the ith factor at time t, !" is the
mean factor return of the ith factor, fj,t is the factor return of the jth factor at time t, !" is the
mean factor return of the jth factor and a pre-specified weight less than 1 that dampens the
contribution made by older factor returns. As gets smaller more weight is put on recent factor
returns, whereas setting =1 corresponds to using equal weighting.
As a final step in estimating the covariance matrices, the SHaPTSE estimator is applied to reduce
estimation bias but is out of the scope of this paper to cover21.
The covariance matrix of all security returns at time t, written in matrix form where N is the
number of securities and F is the number of factors, is then given by:

! = # "! #$
!
"
" " "

(3.6)


21 For an in-depth theoretical justification of the SHaPTSE estimator refer to Kuberek & Matheos (2005)

38

3. Methodology
where et is an N F matrix containing the exposures of each security i with respect to the corre-
sponding factor at time t (its transpose is denoted !#" ) and !! " is the estimated F F covariance
matrix of the factor returns at time t. An illustrative example of how to calculate the covariance
matrix using security exposures and factor returns is shown in appendix 6.
The constituent weights for the MVW index can now be expressed as the solution to the follow-
ing optimisation problem:

$0
&" $ = /
&
!"# $ %!$
'()*+,- -. % "=/ "
$

&
'1 # $ " # (

(3.7)

! is the estimated covariance matrix of the


where w is the vector of constituent weights and !
constituent returns. The usual constraint that weights have to sum to one is enforced in all op-
timisations to avoid any excess cash not invested in the constituents. Regarding weight con-
straints on the individual constituents, this study applies the approach for determining upper
and lower bounds proposed by Amenc et al. (2010). They suggest imposing weight constraints
that depend on the number of constituents N in their index and calculate the boundaries as:

!=

"
!#

$=

!
#

% ! ""

(3.8)

where is a flexible parameter. Setting =1 corresponds to forcing the optimiser to weight all
constituents equally, whereas higher values of allow the optimiser to determine constituent
weights more freely. These constraints ensure that all MSCI AC World constituents are included
and that no constituents obtain a negative weight that would lead to short sales. This preserves
the comparability between all the indices constructed in this study as the stock universe is kept
intact.
Amenc et al. (2010) sets =2 which resembles equal weighting to a very large extent22. Jaganna-
than & Ma (2003) show in their study that the performance problems associated with mean-
variance optimisation strategies are largely ameliorated when constraints are relaxed. There-
fore, this study by default sets =5 to give the optimiser more flexibility without comprising the
level of concentration too much. To test the impact of this choice, the optimisation-based indices
are also constructed setting =2 and =15. With =5 and N=2500, this study constrain constitu-
ent weights to fluctuate between 0.008% and 0.2%. N is set to 2500 because the number of con-

22 For an index with 1000 constituents, setting =2 would constrain constituent weights to fluctuate be-

tween 0.05% and 0.2%, which is close to the 0.1% weight each constituent would get applying an equal
weighting scheme.

39

3. Methodology
stituents in the CW MSCI AC World index has historically fluctuated around this level. An illus-
trative example of how to construct a MVW index containing 10 stocks is shown in appendix 7.
MSCI also constructs a MVW version of the standard CW MSCI AC World index (called MSCI AC
World Minimum Volatility Index) but uses semi-annual rebalancing and a number of weight
constraints on exposures to GICS sectors, countries, risk factors as well as individual weight
constraints and turnover limits (MSCI, 2013b). For a full description of the weight constraints of
the MSCI AC World Minimum Volatility index and a performance comparison to the MVW index
constructed in this thesis refer to appendix 4. Regarding the individual weight constraints of the
MSCI AC World Minimum Volatility index it is noticeable that no constraint is imposed to ensure
that the stock universe is the same as the CW MSCI AC World index. In fact, as of 31 May 2013
only 279 stocks constitute the MSCI AC World Minimum Volatility index whereas 2428 stocks
are included in the CW MSCI AC World index. Considering that over-concentration in a few large
stocks is one of the main shortcomings of CW indices according to the literature review, it seems
counterintuitive to construct an alternative index containing only 11% of the stocks in its CW
counterpart.

3.2.4 Maximum-Sharpe ratio-weighted MSCI AC World indices


Given that all stocks are unlikely to have the same expected returns, the MVW index is theoreti-
cally unlikely to be the index with the highest ex ante Sharpe ratio. Therefore, this thesis incor-
porates different methods to estimate expected returns in an effort to improve upon the per-
formance of the MVW index. As stated in section 2.1.1 this thesis uses (1) historical average of
actual returns, (2) historical average of Wilshires GR6 multi-factor model returns and (3) Jyske
Invests VAMOS score to forecast future returns. Each method will be presented in the succeed-
ing three sections, but before that a brief description of the general methodology applied to con-
struct MSRW indices (irrespective of the method used to estimate expected returns) is provid-
ed.
Common for all three MSRW indices is the applied covariance matrix and the weight con-
straints, which are identical to the MVW index23. Consequently, the constituent weights for the
MSRW indices can be expressed as the solution to the following optimisation problem:

*1
" $ %& %
,( $ = 0
'' '()*+,- -. + /=0 /
!"# $$
$
& &
,
# $ %!$
-2 ) $ / ) (

(3.9)


23 Refer to the previous section on the minimum-volatility index for a detailed description of the covari-

ance matrix and weight constraints applied in the mean-variance optimisations.

40

3. Methodology
where w is the vector of constituent weights, ! is the vector containing the relevant proxy for

! is the estimated covariance matrix of the constitu-


expected returns for each constituent and !
ent returns. Again, the thesis examines the robustness of the various MSRW indices to changes
in covariance matrix, weight constraints, rebalancing frequency and backtesting period (see
section 5.2.7). An illustrative example of how to construct a MSRW index containing 10 stocks is
shown in appendix 8.
3.2.4.1 Method 1: 5-year average of actual monthly returns
The first and simplest method calculates for each constituent the actual annualised return based
on five years of monthly data. As pointed out in section 2.1.1.1, both Campbell (2001) and Bodie,
Kane & Marcus (2010) claim that the historical average return is the best forecast of future re-
turns as it is unbiased, easy to implement and do not require a lot of questionable assumptions.
The choice of investment horizon matches that of Jyske Invests investors and is also in accord-
ance with the suggestions by Peare & Bartholdy (2005) and Goyal & Welch (2008).
3.2.4.2 Method 2: 5-year average of monthly GR6 model returns
The second and slightly more complicated method calculates for each constituent the GR6-
model annualised return based on five years of monthly data. The rationale behind this method
is that the historical exposure and related returns to the intuitive and thoroughly studied fac-
tors of the GR6 model can potentially be a reasonable proxy for future performance. To account
for occasionally extreme model returns (particularly evident for companies from emerging
countries) a truncation correction is carried out at each rebalancing date so that all normalised
z-scores of the model returns above 3 are set equal to 3. Note, that similarly to the 5-year aver-
age of actual returns this method does not entail any expectations about the future.
3.2.4.3 Method 3: Jyske Invests VAMOS score
Since the two previous methods are exclusively based on backward-looking data, this thesis also
applies Jyske Invests proprietary VAMOS score, which uses a combination of backward and
forward looking measures. It was briefly introduced in the description of Jyske Invests stock
picking procedure (section 1.6.1) and in the theory and literature review (section 2.1.1.3). The

CONFIDENTIAL INFORMATION

VAMOS score combines value, momentum and strength factors and is calculated using 15 differ-
ent factors in total; seven value factors, five momentum factors and three strength factors. Table
3.1 describes all factors, their relative weights and directional impact on the VAMOS score if
they rank highly for a given factor. Most factors are rather self-explanatory and intuitive and
will therefore not be further elaborated on. However, five of the factors rely on proprietary
measures from Credit Suisses HOLT platform, which calls for further explanations.

41

3. Methodology
The factors Upside to HOLT target price and HOLT company discount rate both rely on
HOLTs automated DCF model (Credit Suisse, 2013). This model assigns more than 20,000 com-
panies worldwide a warranted valuation based on each companys performance and by using
empirical research on how thousands of companies with similar characteristics have performed
in the past. The three strength factors are calculated on the basis of the HOLT CFROI (Cash Flow
Return On Investment) metric, which is a measure that converts standard accounting infor-
mation into an internal rate of return that more accurately approximate a companys underlying
economics. For example, by correcting for highly subjective accounting methods regarding de-
preciation and off-balance sheet items. The resulting CFROI measure can be used to assess a
companys historical ability to create or destroy wealth over time. Jyske Invest utilises both the
current CFROI level, the CFROI trend based on the slope of 5-year regressions and the CFROI
volatility based on the annualised standard deviation of five years of monthly data to calculate a
score for a given companys strength.

CONFIDENTIAL INFORMATION

Table 3.1 - Description, directional impact and weight of Jyske Invest's VAMOS score
Category
Value
Value
Value
Value
Value
Value
Value
Momentum
Momentum
Momentum
Momentum
Momentum
Strength
Strength
Strength
Source: Jyske Invest

Factor description
Upside to HOLT target price
HOLT company discount rate
Upside to average analyst target price
Current to historical P/S ratio (5Y average)
Current to historical P/B ratio (5Y average)
12M forward-looking EV/EBITDA
12M forward-looking P/E
Target price momentum 0-6M
EPS momentum 0-3M
EPS momentum 3-6M
Price momentum 0-12M
Price momentum since 12M low
Current HOLT CFROI level
HOLT CFROI trend (based on 5Y regression)
HOLT CFROI volatility (based on 5Y monthly data)

VAMOS impact
of a high rank

Weight
15.0%
5.0%
10.0%
5.0%
5.0%
5.0%
5.0%
10.0%
5.0%
5.0%
7.5%
7.5%
5.0%
5.0%
5.0%

For each factor, all the constituents of Jyske Invests global equity universe (approximately
6000-7000 companies) are ranked and given a score from 0-100. These relative scores are then
weighted by the weight for each factor to arrive at an overall VAMOS score for each constituent.

3.3 Impact of alternative benchmark indices on Global Equities (SRQ 2)


Having constructed a number of alternative indices this section outlines the methodology ap-
plied to assess the impact of these indices on GE had they been used as benchmark in the past
by the portfolio managers of Jyske Invest. In doing so, this study does not only analyse bench-
mark indices on an aggregate level but also investigates how they affect a specific actively man-
aged portfolio.
As described in section 1.6.2 Jyske Invests portfolio construction strategy is constituted by two
distinct choices:

42

3. Methodology
1. Which benchmark index to use
2. Which tracking error constraint to impose the portfolio managers
Therefore, in order to properly test the impact of benchmark indices on GE in isolation (that is
to exclusively change the choice of benchmark index), a backtesting scenario in which the same
tracking error constraint is enforced, must be set up. To track the benchmark, Jyske Invest uses
the very simple and heuristic benchmark weight + 0.7% rule introduced earlier. This rule di-
rectly deploys the weight each GE constituent has in the benchmark to determine its weight in
GE. For that reason, it can be argued that once Jyske Invest have selected their preferred stocks
for GE based on the VAMOS score and thorough due diligences, they weight these stocks in a
way that seeks to track the benchmark as close as possible.
Since Jyske Invests benchmark weight + 0.7% rule in many instances require subjective and
inconsistent adjustments to ensure that the constituent weights of GE add up to 100%, it is not
possible to set up a backtesting scenario that directly replicates this rule. Instead, this thesis
exploits, that the implicit goal of Jyske Invests portfolio construction strategy is to minimise
tracking error and applies the relative mean-variance optimisation setup utilised by Jorion
(2003). In this setup he derives efficient frontiers in relative space rather than absolute space
focusing on tracking error rather than total volatility in his portfolio optimisations. Similar to
Jorion (2003) this thesis considers an actively managed portfolio aiming to outperform a
benchmark index or at least obtain a return that follows the development in the benchmark. For
this task, the portfolio manager must take positions in the stocks within the benchmark index
and, perhaps, other stocks as well. Applying the relative mean-variance optimisation setup, the
portfolio managers of Jyske Invest would carry out the task of minimising tracking error as fol-
lows.
Define the following variables where N is the number of constituents in the benchmark index:

! "# = $%&'()*(+*,%-&./0)1*!%23.'4*56!78
! 9: = $%&'()*(+*;:*!%23.'4*'.0'*/2-2/24%4*')0&12-3*%))()*56!78
! <%$ = ! 9: " ! "# = $%&'()*(+**!%23.'*<%$20'2(-4*,%'!%%-*;:*0-<*,%-&./0)1*56!78
= = %4'2/0'%<*&($0)20-&%*/0')2>*+()*,%-&./0)1*&(-4'2'?%-'*)%'?)-4*?42-3*;@A*/(<%B*56!68
#

Note that the vector of GE weights is an N 1 vector, but naturally contains zeros for all the con-
stituents not included in GE but still part of the benchmark.
The constituent weights for the GE portfolio that minimises tracking error with respect to the
benchmark can now be expressed as the solution to the following optimisation problem:

43

3. Methodology

% *
!"# $ '()
!$ '()
$ %&

$
5
&
"$ %&3" = 4
&
"=4
&
+,-./(01+12+ % 6789 # $ %&3" # :9+;2<+1=(+"+02#,1"1-(#1,+"#0>-'('+"#+?& (3.10)
&
& $ %&3/ = 69+;2<+1=(+/+02#,1"1-(#1,+#21+"#0>-'('+"#+?&
&
'

! is the
where wTE is the vector of GE constituent weights that minimises tracking error and !
estimated covariance matrix of the constituent returns. Note that the weights for the GE con-
stituents are restricted to a minimum of 0.2% and a maximum of 3% as these are the limits ap-
plied by Jyske Invest throughout the backtesting period. For an illustrative example of how this
optimisation problem is carried out in practice refer to appendix 9 in which it is illustrated how
the portfolio weights of a fictive portfolio with four stocks are determined so that the tracking
error of this portfolio is minimised with respect to an EQW benchmark index containing 10
stocks.
To test the impact of benchmark indices on GE, this study solves the above optimisation prob-
lem on a monthly basis throughout the backtesting period applying each of the four alternative-
ly weighted indices. Monthly rebalancing is applied even though GE Stocks is rebalanced every
week since weekly data on GE constituents are not available in the full backtesting period. The
study also backtests the performance of GE applying the traditional CW MSCI AC World index to
the optimisation problem. This backtest is necessary to ensure comparability between the per-
formance of the simulated GE portfolios and the actual GE portfolio. Furthermore, it allows for
an assessment of how well the relative mean-variance optimisation setup mimics the bench-
mark + 0.7% rule actually used by the portfolio managers of Jyske Invest.
Finally, it is worth emphasising that for all simulated portfolios, the historical constituents of GE
are kept fixed, so that only the construction methodology changes.

3.4 Impact of loosening tracking error constraint on Global Equities (SRQ 3)


The thesis now turns into a description of the methodology applied to test the impact of loosen-
ing the tracking error constraint on GE. As described in the previous section, Jyske Invest
weights the GE constituents by means of the benchmark + 0.7% rule that weights each GE
constituent based on its weight in the benchmark. It is a very restrictive and simple rule that
neither leaves much room for the portfolio managers to deviate from the benchmark nor allow
the portfolio managers to account for the interaction effects that influence portfolio properties
when combining the constituents of GE. Consequently, this thesis tests the impact of allowing
the portfolio managers to (1) deviate more from the benchmark while (2) exploiting the de-
pendence structure across stocks.

44

3. Methodology
The first point is facilitated by gradually allowing the optimiser to deviate more and more from
the simulated GE portfolios constructed to minimise tracking error (wTE)24. Another sensible
way of allowing the optimiser to deviate more from the benchmark would be to set specific tar-
get levels of tracking error (e.g. 4%, 5%, 6%, etc.) and then determine the performance of GE for
each of these levels. However, this is not possible using the Wilshire Atlas software package and
therefore the tracking error minimising portfolios from section 3.3 are used instead, which
achieve the exact same purpose as any deviations from the minimum-tracking error portfolios
naturally increases tracking error. The second point of accounting for the dependence structure
of stocks is obtained by maximising the Sharpe ratio of GE using the same logic as for the MSRW
indices.
With the two points above in mind, the constituent weights for the GE portfolio that maximises
Sharpe ratio while gradually allowing for larger and larger deviations from the GE portfolios
constructed to minimise tracking error (wTE) can now be expressed as the solution to the fol-
lowing optimisation problem:

,
5
"
%
.
'
($ %&23 = 4
$ %&(
' )*+,-./0)01) .!"# $
3=4
'
(
$%& $# $ ' !$
.
%&
%& &
./ 4) # 3 * $ '&23 + $ %&23 + 4+ # 3 * $ '&23

(3.11)

))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))2)#6{7892:892;892<89}
where wGE is the vector of GE constituent weights, ! is the vector containing the VAMOS score

! is the estimated covariance matrix of the constituent returns using GR6


for each constituent, !
and wTE is the vector of GE constituent weights that minimises tracking error from which devia-
tions of +/- 20%, 40%, 60% and 80% are allowed25.
The VAMOS score is used as proxy for expected returns as it proves to be the best performing
proxy in the construction of the MSRW indices in section 3.2.4. Moreover, using the VAMOS
score creates consistency between stock picking and portfolio construction in Jyske Invest. For
an illustrative example of how this optimisation problem is carried out refer to appendix 10.
Again, this study keeps the historical constituents of GE fixed and solves the above optimisation
problem on a monthly basis throughout the backtesting period applying each of the five bench-
mark indices considered in this paper. It is necessary to test the impact of loosening tracking

24 For a description of the methodology behind the construction of the GE portfolios that minimises track-

ing error with respect to each of the constructed indices, refer to section 3.3.
25 The maximum deviation of 80% is chosen as this allows the optimiser to assign a given constituent a
weight of 3%(1+80%)=5.4%, which is very close to the upper limit on individual stock weights that the
portfolio managers of Jyske Invest must respect.

45

3. Methodology
error for each index as the tracking error by definition depends on the benchmark index and
investigating differences in performance pattern across benchmarks is therefore relevant.
A backtest that allows deviations from actual historical weights of GE constituents is also per-
formed to test if the same performance pattern is evident for the actual and simulated GE port-
folios.

46

4. Data

4. Data
This chapter presents the data and discusses choices made in the collection process. To ensure va-
lidity and reliability of the dataset, the choices are based on theoretical, methodological and em-
pirical considerations.
The data used to construct and test the performance of the full range of alternative indices and
GE portfolios described in the methodology section is primarily collected from the Wilshire At-
las database. As mentioned, the Wilshire Atlas software package is preferred in this thesis as it
provides direct access to the required proprietary data on the historic constituents of the CW
MSCI AC World index and GE.
This thesis sample period is limited by the access of Jyske Invest to the historical constituents
of MSCI AC World, which starts January 2006. As such the full sample period applied in this pa-
per runs from January 2006 to May 2013 (89 monthly returns). From January 2006 to June
2012 the MSCI AC World and GE constituent lists are available on the last trading day in each
month and from July 2012 to May 2013 they are available on each trading day. To ensure con-
sistency across the sample period and to keep with the related studies on alternative indices,
the thesis only applies data from the last trading day in each month regardless of the availability
of daily data in the most recent year.
All stock prices of this thesis are retrieved from the International Data Corporation (IDC) and
the total returns are calculated on a monthly basis in USD including reinvestments of gross divi-
dends and adjusted for currency effects. In accordance with most related studies no adjust-
ments are made for tax payments and trading and administrative expenses as these are hard to
quantify and very dependent on the type of investor (Arnott, Hsu & Moore, 2005). This way of
calculating returns ensures that a fair comparison to the related studies is obtainable.
All performance measures presented in tables and figures in the following chapters are annual-
ised and based upon the above definition of total returns and of the full sample period unless
otherwise stated. Further, ratios that involve average returns are based on the geometric aver-
age, which reliably reflects multiple holding period returns for investors.
The risk-free rate used to calculate Sharpe ratios, information ratios and factor exposures is the
US one month T-bill rate retrieved from Kenneth R. Frenchs homepage (French, 2013).
The Wilshire Atlas database is connected to a number of available databases. Hence, in con-
structing the FW index data on company-specific annual accounting measures are retrieved
from the Worldscope database. IDC is used to retrieve data on each companys market capitali-
sation needed in the calculation of the CAP ratio introduced in section 5.2.5 on liquidity. IDC

47

4. Data
is also used to retrieve data on the 12-month average monthly trading volume needed in the
calculation of the weighted trading dollar volume (also introduced in section 5.2.5 on liquidity).
Calibration of Wilshires GR6 model is required for all optimisation-based indices and GE port-
folios. As mentioned previously GR6 relies on classifications of companies into different regions,
countries of risk and industry groups. Region and country of risk classifications are retrieved
from IDC, while GICS sector and industry classification is retrieved from MSCI and Standard &
Poors. The five company fundamentals; market capitalisation, E/P ratio, B/P ratio, total securi-
ty return over the initial eleven months of the trailing year (momentum) and total monthly re-
turn variance over the trailing 24 months, are retrieved from Worldscope.
To decompose the returns of the alternative indices into exposures to each of the four Carhart
(1997) factors, Kenneth R. Frenchs homepage is used as source (French, 2013). Fama & French
(2012) constructs global Carhart factor portfolios on the basis of large, mid and small caps from
23 developed countries. Hence, their global stock universe differs from the stock universe of
this thesis, which only comprises large and mid caps and also include stocks from 21 emerging
countries. Nonetheless, Fama & French (2012) state that they would be comfortable using the
global Carhart four-factor model to evaluate the performance of any global portfolio not too
heavily tilted towards micro caps or stocks of a particular region.

48

5. Empirical findings

5. Empirical findings
This chapter presents the empirical findings of this thesis. These findings are based on the previ-
ous three chapters by applying the chosen methodology on the presented data in the selected
theoretical context. The first section contains a preliminary analysis of GE to motivate and under-
line the relevance of investigating the performance and quality criteria of alternative equity indi-
ces and how they affect the portfolio construction strategy for GE. The subsequent three sections
comprising this chapter are in alignment with the three sub-research questions and together pro-
vide new insights on the increasingly discussed alternatively weighted equity indices and their
impact on the actively managed GE portfolio.

5.1 Preliminary analysis of Global Equities


In this preliminary analysis of GE a presentation of its actual performance is provided over the
full sample period. The first section compares the performance of GE to its benchmark, the CW
MSCI AC World index, whereas the second section compares it to other mutual funds. The last
section evaluates the performance of GE had a simple, naive equal weighting scheme been ap-
plied and compares this to its actual performance to highlight potential shortcomings of the
current portfolio construction strategy utilised in Jyske Invest.

5.1.1 Performance compared to the CW MSCI AC World index


As mentioned in section 1.6.2, the portfolio construction in Jyske Invest is highly dependent on
the benchmark, but still the objective is to outperform the chosen benchmark. Effectively, this
means that any outperformance must primarily be a result of the stock picking process in Jyske
Invest.
Before specifically addressing the performance, a presentation of selected size and weight char-
acteristics is shown in table 5.1 on the next page to give a brief overview of GE and the CW MSCI
AC World index. An obvious difference between GE and the benchmark is the number of con-
stituents included, which also explains the weighting differences shown in panel A. The smallest
weight in GE is 0.582% compared to 0.001% in the index.
Panel B shows the ten largest constituents in GE as of 31 May 2013, where the fourth column
presents the weights in GE and the fifth column shows the corresponding weights in the CW
MSCI AC World index. It is worth noticing that, as expected, the top 10 stocks all have large
weights in the benchmark as well. However, the average weight difference turns out to be 1.1%
and thus larger than the allowed +/- 0.2 percentage point deviations from the benchmark +
0.7% rule of thump described earlier. This underlines the subjective and inconsistent nature of
the current portfolio construction strategy in Jyske Invest.

49

5. Empirical findings

Table 5.1 - Size and weight characteristics of GE and the CW MSCI AC World index
A. Size comparisons


GE
CW MSCI AC World
Number of constituents

93


Weight:
Largest
Smallest
Average
Median

B. Top 10 constituents in GE

2428

1.928%
0.582%
1.075%
1.011%


Country
Microsoft Corp
US
Wells Fargo & Co
US
J P Morgan Chase
US
Chevron Corp
US
IBM
US
Nestle Sa
US
Novartis Ag
Switzerland
Sanofi
France
Google Inc
US
Bayer Ag
Germany
Total

Note: Size and weight characteristics are retrieved as of 31 May 2013.
Source: Own analysis, Wilshire Atlas

1.295%
0.001%
0.041%
0.017%

Sector
Info Tech
Financials
Financials
Energy
Info Tech
Con. Staples
Health care
Health care
Info Tech
Health care


GE CW MSCI AC World
weight
weight
1.93%
0.87%
1.83%
0.67%
1.74%
0.66%
1.71%
0.74%
1.68%
0.69%
1.67%
0.67%
1.64%
0.50%
1.60%
0.39%
1.59%
0.73%
1.50%
0.28%
16.88%
6.19%

The close connection between GE and the CW MSCI AC World index is also apparent when com-
paring the composition of the two portfolios. Figure 5.1 and 5.2 show each portfolios weight
allocations to geographical regions and industrial sectors, respectively.
Figure 5.1 - Region weights
A. GE
Asia
9%

Africa,
Middle
East
4%

Europe
28%

Latin
America
4%

B. CW MSCI AC World
Asia
6%

Paci}ic
9%

Africa,
Middle
East
2%

Paci}ic
14%

Europe
27%

North
America
46%

Latin
America
3%

Note: The weights are calculated as the average weight of each region over the full sample period (2006-2013)
Source: Own analysis, Wilshire Atlas

North
America
48%

From figure 5.1 it is seen that the allocation to regions is very alike though GE is a bit more tilted
towards stocks from Asia and Africa/Middle East and a bit less towards stocks from North
America and Pacific compared the CW MSCI AC World index.
Figure 5.2 illustrates the stability of the sector allocation from 2006 to 2013 of GE and the CW
MSCI AC World index, respectively.

50

5. Empirical findings
Figure 5.2 - Sector characteristics
A. GE
100%

Financials
Energy
Consumer Discretionary
Information technologi
Industrials
Health Care
Materials
Consumer Staples
Telecom
Utilities
Not classi}ied

80%
60%
40%
20%
0%
2006

2007

2008

2009

2010

2011

2012

2013

B. CW MSCI AC World
100%

Financials
Energy
Consumer Discretionary
Information technologi
Industrials
Health Care
Materials
Consumer Staples
Telecom
Utilities
Not classi}ied

80%
60%
40%
20%
0%
2006
2007
2008
2009
2010
Note: Industrial sector allocations are based on GICS sectors.
Source: Own analysis, Wilshire Atlas

2011

2012

2013

In terms of the average weight of each sector over the sample period, the two portfolios are
again very alike with average differences between 0-1.4%. The fluctuations of the section alloca-
tion on the other hand are much more pronounced for GE compared to the CW MSCI AC World
index. This is primarily a result of changing stock preferences for the portfolio managers of GE
whereas constituents of the CW MSCI AC World index are much more stable.
To compare the performance of GE with the CW MSCI AC World index, figure 5.3 illustrates the
cumulative return of each portfolio.
Figure 5.3 - Indexed cumulative return
160

Global Equities

577+#&1$"-+$7.1

157

CW MSCI AC World

145

140

!"#$
!""#
!"")
!""*
!"",
!"%"
!"%%
!"%!
!"%$

120
100
80
60
40
2006

2007

2008

2009

2010

2011

Source: Own analysis, Wilshire Atlas

2012

2013

%&'(#& /0123/41
)*+,-,". 5/10'$&6
$%&!'
!%&('
!%&%'
%!&%'
+(,&"'
+(%&,'
$%&)'
$-&('
%)&)'
%$&$'
+$&%'
+#&*'
%#&*'
%#&*'
$$&%'
!(&-'

It is noticed that the performance of GE closely tracks the reference index, yet it has outper-
formed the index in the period from 2006 to 2013. Thus it appears to have fulfilled its overall
goal, but taking trading expenses into account changes the picture (Jyske Invest, 2013a).

51

5. Empirical findings

5.1.2 Performance compared to peer group of other mutual funds


A comparison of GE and 675 other European mutual funds belonging to Morningstars Global
Large-Cap Blend Equity category is presented in figure 5.4 where each funds 10-year returns is
plotted against its standard deviation and tracking error, respectively.
Figure 5.4 - Comparison to other European mutual funds
A. Return / Standard devia[on

B. Return / Tracking error

Return (%)
15

Return (%)
15

10

10

0
25
0
5
10
15
Std. dev. (%)
TE (%)
Mutual Funds Average Global Equi>es
Mutual Funds
Average
Global Equi>es



Notes: Returns, standard deviations and tracking errors are annualised based on 10 years, month-end prices in DKK. All mutual
fund returns are retrieved from Morningstars Global Large-Cap Blend Equity category.
Source: Own analysis, Morningstar
5

10

15

20

Panel A shows that GE has achieved a higher return at the expense of a higher risk compared to
the average fund. Panel B shows that despite GE has incurred a smaller tracking error compared
to the average fund, it has been able to achieve a higher return. Note also that a linear relation-
ship between tracking error and return seems to be evident, a theory Roll (1992), Jorion (2003)
and Hope (2008) also supports under the assumption that the applied benchmark is ineffi-
cient26. Thus there might exist a return-potential in relaxing the tracking error constraint for GE.
This hypothesis is specifically tested in section 5.4 but also briefly touched upon in the following
section.

5.1.3 Performance had Global Equities been equal-weighted


As shown in the two previous sections, the current weighting scheme (benchmark weight +
0.7%) ties the performance of GE very closely to the CW MSCI AC World index. Moreover, it ap-
pears rather inconsistent as no clear construction steps are followed for each constituent, which
translates into weak transparency to the investors of GE. In response to this, the simple and
nave equal weighting scheme is tested on GE (keeping constituents fixed) to test if this bench-
mark independent and easily implemented weighting scheme would have achieved better per-
formance than the actual performance of GE.
Table 5.2 summarises selected performance measures of the two different weighting schemes
using the full sample period and both monthly and quarterly rebalancing.


26 The efficiency of the CW MSCI AC World index is thoroughly tested in section 5.2.

52

5. Empirical findings
Table 5.2 - Performance measures of GE using current and equal weighting scheme
Kolonne2
Monthly rebalancing
Current weighting scheme

Average
return

Standard
deviation

Tracking error
(w.r.t. CW MSCI
AC World)

Sharpe
ratio

Information
ratio

6.6%

20.0%

4.7%

0.25

0.30

Equal weighting scheme

6.9%

21.2%

5.5%

0.26

0.33

Quarterly rebalancing
Current weighting scheme
Equal weighting scheme
Source: Own analysis, Wilshire Atlas


5.9%


20.0%


4.6%


0.22

6.4%

21.1%

5.2%

0.23


0.17
0.24

It is seen that in both cases equal weighting generates higher returns than the current strategy,
but at the same time the standard deviation is also higher for the equal weighting scheme com-
pared to the current weighting scheme. As expected tracking error is also higher when applying
equal weights, since it completely disregards the CW MSCI AC World index as opposed to the
current benchmark + 0.7% weighting scheme.
The fifth column shows that the equal weighting scheme generates slightly higher Sharpe ratios
for both rebalancing frequencies. The information ratio is also higher for the equal weighting
scheme as the increase in returns outweighs the higher incurred tracking error. This indicates
that applying equal weights to the constituents of GE is superior to their current strategy even
though the performance differences are rather small.
5.1.3.1 Significance tests
At first glance the equal weighting scheme appears superior due to the slightly better perfor-
mance measures. However, the performance differences between the two weighting schemes
are very small and based on a relatively limited amount of data. Therefore it is tested whether
this superiority is actually statistically significant. Table 5.3 shows the results of significance
tests of the differences in average return, standard deviation and Sharpe ratio between the cur-
rent weighting scheme and the equal weighting scheme27.
Table 5.3 - Sign. tests of GE differences between current and equal weighting scheme
Average return
Standard deviation
Sharpe ratio

Difference
p-value
Difference
p-value
Difference
p-value
Monthly rebalancing
0.38%
37.2%
1.17%
59.5%
0.0041
89.8%
Quarterly rebalancing
0.47%
31.8%

1.12%
60.9%

0.0107
72.4%
Note: The p-values for differences are computed using a paired t-test for the average returns, a Fischer test for the standard devia-
tion, and a Jobson-Korkie test for the Sharpe ratio28 .
Source: Own analysis, Wilshire Atlas

The results show that none of the differences in the performance measures of GE are statistical-
ly significant. Consequently, the cause of the outperformance could be random effects and it is
not possible to firmly conclude that equal weighting is in fact superior. Despite the insignifi-

27 No reliable and well-adopted significance test of differences in information ratio exists in the academic

literature and is thus disregarded in this thesis (Blatt, 2004).


28 Ledoit & Wolf (2008) criticises the original Jobson-Korkie test proposed by Jobson & Korkie (1981), but
since most academic literature still applies the original test, it is preferred throughout this thesis as well.

53

5. Empirical findings
cance of the differences, a key take-away from this analysis is that the current opaque strategy
in Jyske Invest is not superior to the equal weighting scheme, which most investors would easily
be able to adopt themselves. This gives rise to believe that a more efficient and transparent
portfolio construction strategy exists for GE. A hypothesis that is tested in both section 5.3 and
5.4 that respectively addresses each of the two distinct choices that makes up Jyske Invests
portfolio construction strategy. To answer these questions it is first necessary to investigate if
alternatively weighted versions of the CW MSCI AC World index are superior. The next section
addresses this matter in depth.

5.2 Evaluation of alternatively weighted MSCI AC World indices (SRQ 1)


As mentioned, the thesis now moves into an evaluation of the performance of the alternatively
weighted MSCI AC World indices presented in section 3.2. The analysis builds on the framework
illustrated in figure 2.1 that combines performance measures as well as quality criteria. The first
section addresses performance measures whereas the five subsequent sections deal with each
of the five quality criteria; turnover, concentration, representativeness, liquidity and transpar-
ency. The seventh section tests the robustness of the findings before they are summarised into
qualitative scores in the final section.

5.2.1 Performance measures


Table 5.4 presents the performance measures of each alternative index described in section 3.2
using the full sample period and annual rebalancing29.
Table 5.4 - Performance measures of alternative MSCI AC World indices


CW
EQW
FW
MVW
MSRW (actual)
MSRW (GR6)
MSRW (VAMOS)

Average
return
5.1%
8.5%
4.7%
9.4%
8.4%
8.9%
9.9%

Standard
devia-
tion
18.4%
21.1%
21.8%
14.1%
17.2%
18.5%
18.2%

Semi
devia-
tion
13.3%
14.6%
15.2%
10.1%
12.9%
13.8%
13.3%

Track-
ing
error
-
4.8%
5.1%
6.6%
6.3%
6.9%
4.9%

Sharpe
ratio
0.19
0.33
0.15
0.55
0.40
0.40
0.46

Sortino
ratio
-
0.58
0.31
0.93
0.65
0.65
0.75

Infor-
mation
ratio
-
0.71
-0.08
0.64
0.52
0.55
0.98

CAPM
alpha
0.0%
3.2%
-0.6%
4.8%
3.6%
4.0%
4.8%

CAPM
beta
1.00
1.12
1.16
0.73
0.87
0.93
0.95

Monthly
95%
VaR
6.9%
7.8%
9.0%
4.2%
4.7%
5.0%
5.3%

Notes: As proxy for expected returns MSRW (actual) uses 5-year average of actual returns, MSRW (GR6) uses 5-year average of
GR6 model returns and MSRW (VAMOS) uses Jyske Invests VAMOS score. All measures are annualised except the 95% VaR. A
Cornish-Fisher expansion following Favre & Galeano (2002) and Zorion (2007) is used to compute a Value-at-Risk estimate that
takes into account the mean, volatility, skewness and excess kurtosis of index returns.
Source: Own analysis, Wilshire Atlas

5.2.1.1 Return and risk


The second column of table 5.4 presents the geometric average, annualised returns for each
alternative index. The first point to notice is that except the FW index, all other indices generate
higher returns than the CW MSCI AC World index. The highest returns are achieved by the two
optimisation-based weighting schemes, MV and MSR (VAMOS), which generate annualised re-

29For an overview of the cumulative returns of the alternative weighted indices refer to appendix 11

54

5. Empirical findings
turns of 9.4% and 9.9% respectively. The worst performance is produced by the FW index
(4.7%), which is somewhat surprising when comparing to other studies such as Arnott, Hsu and
Moore (2005), Chow et al. (2011) and Clare, Motson and Thomas (2013b) who all construct
outperforming fundamental indices compared to their CW counterparts. However, given that
the first two studies use a fundamental stock selection approach in addition to fundamental
weighting the deviation is less surprising. In fact, Amenc, Goltz and Lodh (2012) show in a re-
cent study on alternative equity index strategies that fundamental stock selection alone adds
about 50 basis points to annual performance. In addition, they find that fundamental weighting
is the least performance enhancing weighting scheme compared to equal weighting, minimum-
volatility weighting and maximum-Sharpe ratio weighting. Nonetheless, the underperformance
of the FW index constructed in this thesis is noteworthy considering the findings of related
studies and will therefore be addressed further in section 5.2.4 on representativeness and sec-
tion 5.2.7.1 breaking down performance by time period.
The third column in table 5.4 presents the annualised standard deviations of the returns of each
index. The index that generates the highest volatility is the FW index (21.8%) closely followed
by the EQW index (21.1%). As expected the index that produces the lowest volatility is by far
the MVW index (14.1%). The volatility of the remaining indices ranges between 17.2% and
18.5%.
5.2.1.2 Risk-adjusted ratios
Obviously, return and risk seen in isolation is not adequate to fully assess the attractiveness of
each alternative index. Accordingly, the thesis now moves on to an analysis of the three risk-
adjusted ratio presented in section 2.2.1, namely, the Sharpe ratio, Sortino ratio and information
ratio.
The Sharpe ratio reflects the indices risk/reward efficiency by adjusting excess returns over the
risk free rate by the volatility incurred by the indices. Again, all alternative indices outperform
the CW MSCI AC World index with the exception of the FW index. Interestingly, the MSRW indi-
ces do not produce the highest Sharpe ratios despite having it as their primary objective. In-
stead they are beaten by the MVW index. This finding is consistent with the full range of related
studies presented in the literature review and strongly indicates that stock returns are in fact so
unforeseeable that one may as well assume that expected return for all stocks are identical,
which is the underlying assumption of minimum-volatility weighting. Among the MSRW indices,
the index using the VAMOS score as proxy for expected returns achieves the highest Sharpe
ratio implying that this measure most accurately predicts future returns.
The results for the Sortino ratio (which uses downside risk instead of volatility to adjust for
risk) are broadly unchanged and qualitatively similar to the results for the Sharpe ratio.

55

5. Empirical findings
The information ratio reflects the average return difference with the CW MSCI AC World index
when it is adjusted for the incurred tracking error. Except the FW index, all alternative indices
generate rather high information ratios30. The MSRW (VAMOS) index achieves the highest in-
formation ratio (0.98) followed by the EQW index (0.71) and the MVW index (0.64), which indi-
cates that the MSRW (VAMOS) index more efficiently generates excess returns when deviating
from the CW MSCI AC World index. This finding is in accordance with Chow et al. (2011) and
Amenc et al. (2011) who also find their versions of a MSRW index to be the most superior in
terms of information ratio.
5.2.1.3 Extreme risks
In spite of the favourable absolute and relative performance of the EQW, MVW and MSRW indi-
ces, it is also necessary to investigate whether their greater risk/reward efficiency comes at the
cost of a higher risk of extreme losses. The last column in table 5.4 presents the monthly 95%
Value-at-Risk (VaR) statistic for each index. The VaR for the MVW (4.2%) and MSRW (4.7%,
5.0% and 5.3%, respectively) indices are lower than the CW MSCI AC World index (6.9%) sug-
gesting that the superior performance of the optimisation-based indices does not come at the
cost of higher extreme risk. On the other hand, the VaR for the EQW index (7.8%) is bigger than
the CW MSCI AC World index implying that part of its outperformance comes from a larger ex-
posure to extreme risk.
5.2.1.4 Significance tests
Although the alternative indices except the FW index appear attractive overall, it is again inter-
esting to test whether their outperformance is actually statistically significant. This thesis bases
conclusions on a relatively limited amount of data, and any differences could, in principle, be the
results of random effects. Table 5.5 shows the results of significance tests for the average return,
standard deviation and Sharpe ratio. All differences are computed with respect to the CW MSCI
AC World index and results that are significant at the 5% level are indicated in bold.
Table 5.5 - Sign. tests of differences between CW and alternative MSCI AC World indices

Average return

Standard deviation

Sharpe ratio
Difference
p-value
Difference
p-value
Difference
p-value

EQW
3.4%
3.7%
2.7%
20.6%
0.14
8.8%
FW
-0.4%
88.1%
3.4%
11.4%
-0.05
49.8%
MVW
4.2%
18.1%
-4.3%
1.4%
0.36
0.4%
MSRW (actual)
3.3%
21.5%
-1.3%
51.0%
0.20
13.2%
MSRW (GR6)
3.8%
15.8%
0.0%
98.3%
0.20
16.5%
MSRW (VAMOS)
4.8%
1.5%
-0.2%
92.1%
0.26
1.3%
Note: The p-values for differences are computed using a paired t-test for the average returns, a Fischer test for the standard devia-
tion, and a Jobson-Korkie test for the Sharpe ratio.
Source: Own analysis, Wilshire Atlas


30 Grinold & Kahn (2000) asserts that an information ratio above 0.5 is good, above 0.75 is very good,

and above 1.0 is exceptional. While it is not clear whether these breakpoints were determined empiri-
cally they have taken hold as industry standard (Clement, 2009).

56

5. Empirical findings
The results in table 5.5 show that the EQW and MSRW (VAMOS) indices generate significantly
higher average returns than the CW MSCI AC World index. Note that the MVW index does not
produce significantly higher returns despite outperforming the EQW index by 0.8 percentage
points. This is caused by a, compared to the EQW index, relatively higher standard deviation of
the differences in returns between the MVW and CW MSCI AC World index31. Regarding volatili-
ty, the only index with a significantly lower standard deviation is the MVW index. In terms of
risk/reward efficiency, the two optimisation-based indices, MVW and MSRW (VAMOS), generate
significantly higher Sharpe ratios than the CW counterpart. This finding is contrary to similar
tests of the Sharpe ratio differences made by Clare, Motson and Thomas (2013a) who find that
their versions of indices based on minimum-volatility weighting and maximum-Sharpe ratio
weighting do outperform their CW counterpart, but none of them with statistical significance at
the 5% level. However, Amenc et al. (2011) support the findings of this thesis, as their version of
a MSRW index also generate significantly higher Sharpe ratio compared to their CW index.
5.2.1.5 Factor exposures
The above analysis of performance measures provides insight into how the alternatively
weighted indices behave and perform, but does not disclose where the return properties come
from. In column 9 and 10 in table 5.4 the CAPM alpha and beta are computed using the CW MSCI
AC World index as proxy for the market portfolio and show that all alternative indices except
the FW index have positive alphas. Significance tests reveal that the positive alphas are statisti-
cally different from zero at the 5% level for the EQW index (p-value = 4.8%), the MVW index (p-
value = 0.4%) and the MSRW (VAMOS) index (p-value = 0.8%). This implies that these indices
derive their returns from other sources than the CW MSCI AC World index.
To understand where the performance come from the returns of each index are decomposed
using the Carhart (1997) 4-factor model described in section 2.2.332. Table 5.6 presents the re-
sults and shows that when the commonly used equity risk factors are adjusted for, the indices
that show statistical significance are still the EQW (p-value = 0.9%), MVW (p-value = 0.5%) and
MSRW (VAMOS) (p-value = 0.5%) indices. Except the statistical significance of the EQW index,
these findings are in line with Amenc et al. (2011), but opposite to Chow et al. (2011). The latter
finds that only one of 20 alternative indices shows a significant Carhart-alpha and classifies this
unique observation as an outlier and thus concludes that no Carhart 4-factor alpha is present.



31 E.g. the tracking error for the EQW index is 4.8% whereas the MVW index has a tracking error of 6.6%.
32 In appendix 12 the same analysis is carried out using the Fama French (1993) 3-factor model and reach

a similar conclusion.

57

5. Empirical findings
Table 5.6 - Carhart (1997) 4-factor model return decomposition
Small-
Small-
Momen- Momen-
Annual
Alpha Market Market
cap
cap
Value
Value
tum
tum
Index
alpha p-value
Beta p-value
Beta p-value
Beta p-value
Beta p-value
EQW
4.0%
0.9%
1.10
0.0%
0.38
0.0%
-0.18
2.5%
-0.14
0.0%
FW
0.2%
87.5%
1.10
0.0%
-0.05
38.9%
0.24
0.0%
-0.20
0.0%
MVW
4.7%
0.5%
0.74
0.0%
0.17
5.7%
-0.01
93.8%
0.02
51.4%
MSRW (actual)
3.3%
10.4%
0.94
0.0%
0.26
2.6%
-0.32
0.4%
0.10
3.2%
MSRW (GR6)
3.9%
9.9%
0.99
0.0%
0.37
0.6%
-0.39
0.2%
0.09
9.0%
MSRW (VAMOS)
5.0%
0.5%
0.98
0.0%
0.33
0.1%
-0.23
1.5%
0.01
86.4%
Notes: The factor regression results are obtained by regressing the excess returns of each index on the market factor, small-cap
factor, value factor and momentum factor retrieved from Kenneth French's website. The table shows regression coefficients and
the p-values associated with the null hypothesis that the regression coefficients are zero. Coefficients that are significantly different
from 0 at the 5% level are indicated in bold.
Source: Own analysis, Wilshire Atlas and French (2013)

The thesis now turns to an analysis of each of the estimated factor exposures reported in table
5.6.
Market exposure
The results for the market beta confirm the low beta nature of the MVW index, but also show
that the alternative indices are not all uniformly under or overweighted market risk represent-
ed by betas below and above one. Note, however, that all the optimisation-based indices over-
weight low beta stocks. These findings are in line with the alternative indices of Amenc et al.
(2011), Chow et al. (2011) and Clare, Motson & Thomas (2013a; 2013b).
Small-cap exposure
Unsurprisingly the EQW index has the highest bias towards small caps, because equal
weighting, by construction, systematically overweights smaller stocks relative to the CW MSCI
AC World index. However, it is important to note that, as pointed out in chapter 4 on data, the
stock universe of the CW MSCI AC World index used to construct all alternative indices is only
made up of large and mid caps and thus much narrower than the stock universe used by Fama &
French (2012) to construct the Carhart portfolios. This means that the reason for the statistical-
ly significant small-cap exposure of the EQW index is explained by its correlation with the small
cap factor rather than actually containing a larger proportion of small caps as no small caps are
actually contained in the index. The same reasoning applies to the statically significant small-
cap biases of the MSRW indices.
Compared to the related studies, the small cap exposures of the alternative indices in this thesis
are similar to Amenc et al. (2011) and Clare, Motson & Thomas (2013a; 2013b) but differs from
Chow et al. (2011) who find positive small-cap exposure for all indices including their FW index.
Value exposure
Examining the value exposure it must first be noted that the EQW index should, in fact, be free
of any value or growth bias, since no information on valuation affects the determination of its
constituent weights. That the EQW index shows a statistically significant negative value expo-

58

5. Empirical findings
sure (i.e. positive growth exposure) is very counterintuitive and suggests using the value beta
coefficient of the EQW index as reference point for value/growth neutrality. Put differently, in-
dices whose value exposures are not markedly different from that of the EQW index (-0.18) are
not considered to include any value biases. This line of thought follows Amenc et al. (2011) who
also find a value exposure significantly different from zero for their EQW index. The results in
table 5.6 show that only the FW index has a value exposure that is substantially greater than
that of the EQW index. This is consistent with economic intuition as the FW index derives con-
stituent weights from book value, sales, cash flow and dividends, which are all non-
capitalisation measures of company size. As these are typical measures used in value strategies,
it is not surprising to find a substantial value exposure for the FW index. The remaining indices
are not considered to deviate significantly from the EQW indices and hence display no val-
ue/growth bias. For the MSRW (VAMOS) index this is surprising given its partial objective of
favouring stocks that scores high on value factors.
The significant value exposure of the FW index is consistent with the full range of related stud-
ies, but the lack of value exposure of the other indices is not present in the other studies that
generally find their alternative indices to overweight high book-to-market value stocks.
Momentum exposure
Among all the alternative indices, the only two with a statistically significant negative momen-
tum exposure are the EQW and FW indices. Considering that they both, compared to cap
weighting, mechanically rebalances away from stocks that increase in price this is not surpris-
ing and follows the findings of the other related studies. It is surprising though, that the other
underlying partial objective of the MSRW (VAMOS) index of preferring momentum stocks is
likewise not reflected in its momentum exposure that is insignificantly positive.
To sum up, the performance measures of the alternative indices show that the average return and
risk-adjusted ratios of the EQW, MVW and MSRW indices are superior to that of the CW MSCI AC
World index. Statistically speaking only the MVW and MSRW indices generate Sharpe ratios signif-
icantly higher than the CW MSCI AC World index. The only index not showing outperformance is
the FW index.
Attributing the sources of outperformance to the Carhart (1997) 4-factor model, it is shown that
the common equity factors do not capture the entire performance as significantly positive Carhart
alphas are obtained for the EQW, MVW and MSRW (VAMOS) indices. Part of the performance is
attributed to a larger relative small cap exposure whereas no value exposures, except for the FW
index, are evident when using the value beta coefficient of the EQW index as reference point. Posi-
tive momentum exposure is likewise not present.

59

5. Empirical findings

5.2.2 Turnover
Apart from considering performance measures a thorough and adequate evaluation of alterna-
tive indices must also consider various quality criteria (see figure 2.1 presenting the index eval-
uation framework applied in this thesis). The first quality criterion under consideration is turn-
over. Due to the presence of transaction costs, high turnover associated with any of the alterna-
tive indices could potentially erode the outperforming performance measures identified in the
previous section. As mentioned this study does not adjust the performance measures for trans-
action costs, which is consistent with most academic research as it is hard to quantify and trad-
ing costs is very dependent on the type of investor, e.g. investment banks trade cheaper than
private investors (Arnott, Hsu & Moore, 2005)33. Instead this thesis computes the indifference
level of transaction costs to assess the impact of turnover, which reciprocally measures how
large the transaction costs would have to be to completely eliminate the excess returns generat-
ed by each alternative index. Table 5.7 presents the turnover characteristics of each index using
the full sample period and annual rebalancing.
Table 5.7 - Turnover characteristics



Average annual one-
Excess turnover vs.
Excess return vs.
Indifference level of
Index
way turnover
CW MSCI AC World
CW MSCI AC World
transaction costs
CW
4.4%
-
-
-
EQW
17.9%
13.5%
3.4%
25.2%
FW
15.1%
10.8%
-0.4%
n/m
MVW
30.2%
25.9%
4.2%
16.4%
MSRW (actual)
32.3%
27.9%
3.3%
11.7%
MSRW (GR6)
33.9%
29.5%
3.8%
12.8%
MSRW (VAMOS)
43.7%
39.4%
4.8%
12.2%
Notes: The annual one-way turnover is computed as the total amount of new securities purchased or the amount of new securities
sold - whichever is less - over the year, divided by the total net asset value of the index. Turnover is computed at each rebalancing
date and averaged over the entire period. The indifference level of transactions costs is computed as the excess return of the index
versus the CW MSCI AC World index divided by the excess turnover of the index versus the CW MSCI AC World index.
Source: Own analysis, Wilshire Atlas

As expected the CW MSCI AC World index has the lowest turnover even though it is rebalanced
quarterly because virtually the entire turnover in this index arises from reconstitution only. In
contrast, the alternative indices must also adjust the constituent weights to (1) reflect the devia-
tion in the constituent weights from the beginning-of-year policy weights and (2) reflect chang-
es in prices. These changes increase annual turnover from 4.4% for the CW MSCI AC World in-
dex to 15.1% to 43.7% for the alternative indices. Excess turnover is particularly pronounced
for the optimisation-based indices. The indifference level of transaction costs ranges from
11.7% to 25.2% for the alternative indices except the FW index, which has not been computed
as it produces lower returns than the CW MSCI AC World index. These indifference levels are
comparable to similar turnover analyses conducted by Arnott, Hsu and Moore (2005) and

33 Chow et al. (2011) uses a transaction cost model proposed by Keim & Madhaven (1997) in their study

that besides turnover also accounts for stocks with different liquidity characteristics as they generally
incur different transaction costs.

60

5. Empirical findings
Amenc et al. (2010) in their index studies. In practice, it is unlikely that any investor would pay
costs of such magnitude. Moreover, a recent study by Evans & Edelen (2013) on trading costs
and mutual fund performance finds that the average annual expenditures on trading costs is
1.44% for a sample of 1,758 US mutual funds from 1995 to 2006, which is significantly less than
the indifference levels obtained in this study. The improved risk/reward efficiency of the EQW,
MVW and MSRW indices are thus robust to the occurrence of transaction costs.

5.2.3 Concentration
As pointed out in section 2.3.3.2 excessive concentration in a few large stocks with high market
capitalisation is one of the main drawbacks of the CW indices. To shed light on how the concen-
tration of the alternative indices compare to the CW MSCI AC World index, this thesis follows
Strongin, Petsch & Sharenow (2000) in computing the effective number of constituents as the
inverse of the sum of squared constituent weights. Table 5.8 presents the concentration charac-
teristics of each index, again using the full sample period and annual rebalancing.
Table 5.8 - Concentration characteristics


Average effective
Effective constituents to
Relative deconcentration
Index
constituents
nominal constituents
w.r.t. CW MSCI AC World
CW
438
17.1%
-
EQW
2556
100.0%
5.8
FW
412
16.1%
0.9
MVW
606
23.7%
1.4
MSRW (actual)
603
23.6%
1.4
MSRW (GR6)
603
23.6%
1.4
MSRW (VAMOS)
601
23.5%
1.4
Note: The effective number of constituents is computed as the inverse of the sum of squared constituent weights at each rebalanc-
ing date and averaged over the full sample period (2006-2013). The relative deconcentration w.r.t. the CW MSCI AC World index is
calculated as the average effective constituents of each alternative index divided by the average effective constituents of the CW
MSCI AC World index.
Source: Own analysis, Wilshire Atlas

By definition, concentration in the EQW index is as low as possible illustrated by its average
effective constituent being equal to the actual average number of constituents in the CW MSCI
AC World index. In fact, the EQW index is almost six times less concentrated than the CW MSCI
AC World index in terms of the number of effective constituents whereas the relative deconcen-
tration is only 1.4 for the optimisation-based indices34. Interestingly, this suggests that a mere
de-concentration effect by itself does not correct the inefficiency of the CW MSCI AC World in-
dex since the performance measures of the two optimisation-based indices MVW and MSRW
(VAMOS) are superior to the EQW index (see table 5.4), even though they are more concentrat-
ed. A further inference is that the portfolio optimisation methodology applied in this thesis adds
more useful information than the nave equal weighting scheme. Note that the FW index proves

34 The almost identical concentration characteristics of the optimisation-based indices is a direct result of

the applied optimiser that, irrespective of the objective function, tends to assign the preferred stocks the
maximum allowed weight and then let the remaining stocks be weighted by the minimum weight so that
the sum of all stocks equal 100%.

61

5. Empirical findings
to be more concentrated than the CW MSCI AC World index, which is in accordance with most of
the fundamental indices constructed by Arnott, Hsu & Moore (2005).

5.2.4 Representativeness
Another relevant quality criterion for benchmark indices is representativeness defined as the
exposure towards geographical regions and industrial sectors. Representativeness is especially
relevant for global indices, as they are intended to be representative of the global equity market.
Much of the representativeness is obtained through stock selection and in the case of the CW
MSCI AC World index the constituents are selected from 24 developed country indices and 21
emerging country indices. However, the weighting scheme applied to these global constituents
also influence representativeness significantly as shown in figure 5.5 and figure 5.6.
Figure 5.5 - Region characteristics


A. EQW

B. FW

50%
40%
30%
20%
10%
0%

2.0
1.0
0.0
Asia

Europe

Latin America

North America

50%
40%
30%
20%
10%
0%

3.0

Paci}ic

C. MVW

4.0

Africa, Middle
East

Asia

Europe

Latin America

North America

Paci}ic

Africa, Middle
East

4.0 50%
3.0 40%
30%
2.0
20%
1.0 10%
0.0 0%

D. MSRW (actual)

4.0 50%
3.0 40%
30%
2.0
20%
1.0 10%
0.0 0%

2.0
1.0
0.0
Asia

Europe

Latin America

North America

Paci}ic

50%
40%
30%
20%
10%
0%

3.0

Africa, Middle
East

Asia

Europe

Latin America

North America

Paci}ic

Africa, Middle
East

E. MSRW (GR6)

4.0

F. MSRW (VAMOS)

4.0 50%
3.0 40%
30%
2.0
20%
1.0 10%
0.0 0%

4.0
3.0
2.0
1.0
0.0
Asia

Europe

Latin America

North America

Paci}ic

Africa, Middle
East

Asia

Europe

Latin America

North America

Paci}ic

Africa, Middle
East

Note: The weights are calculated as the average weight of each region over the full sample period (2006-2013)
Source: Own analysis, Wilshire Atlas

Figure 5.5 presents region characteristics of each alternative index including the average weight
over the backtesting period for each region and the relative region weights compared to the CW

62

5. Empirical findings
MSCI AC World index. The first thing to notice is that all regions are represented in all indices
and that no region has a lower weight than the 2% lower bound in the CW MSCI AC World index
(Africa, Middle East) as illustrated in figure 5.1. This indicates that the alternative indices are at
least as geographically widespread as the CW MSCI AC World index. However, among the rela-
tive region weights to the CW MSCI AC World index great differences are evident illustrated by
the fluctuating lines in all panels of figure 5.5. Particularly noteworthy is the overweighting of
Africa, Middle East (all indices), Pacific (all indices), Asia (all indices) and Latin America (all
indices except the MVW). This overweighting comes at the cost of the highly developed regions
North America and Europe that are underweighted in all indices except Europe in the FW index.
Figure 5.6 presents GICS sector characteristics of each alternative index including the average
weight over the backtesting period for each sector and the relative sector weights compared to
the CW MSCI AC World index.
Figure 5.6 - GICS sector characteristics


A. EQW

B. FW

40%

3.0 40%

30%

3.0

30%

2.0

20%

2.0

20%
1.0

10%
0%

0.0

0%

30%

3.0
2.0

20%
1.0

10%
0%

0%

0.0
Utilities

Telecom
Services

Materials

Information
Technology

Industrials

Health care

Financials

Energy

Consumer
Staples
Consumer
Discretionary

Utilities

Telecom
Services

Materials

Information
Technology

Industrials

Health care

Financials

Energy

Consumer
Staples
Consumer
Discretionary

40%

1.0

10%

0.0

E. MSRW (GR6)

F. MSRW (VAMOS)

3.0 40%

30%

3.0

30%

2.0

20%

2.0

20%
1.0

10%
0%

1.0

10%

0.0

0%

63

Utilities

Telecom
Services

Materials

Information
Technology

Industrials

Health care

Financials

Energy

Consumer
Staples
Consumer
Discretionary

Utilities

Telecom
Services

Materials

Information
Technology

Industrials

Health care

Financials

Energy

Consumer
Staples
Consumer
Discretionary

0.0

Note: The weights are calculated as the average weight of each GIGS sector over the full sample period (2006-2013)
Source: Own analysis, Wilshire Atlas

Utilities

Telecom
Services

Materials

Information
Technology

Industrials

Health care

Financials

Energy

3.0 40%

30%

2.0

D. MSRW (actual)

40%
20%

0.0
Consumer
Staples
Consumer
Discretionary

Utilities

Telecom
Services

Materials

Information
Technology

Industrials

Health care

Financials

Energy

Consumer
Staples
Consumer
Discretionary

C. MVW

1.0

10%

5. Empirical findings
Again it is noteworthy that all sectors are represented in all indices and that no sector has a
lower weight than the 4% lower bound in the CW MSCI AC World index (Health care). Thus, the
alternative indices are also at least as industrially widespread as the CW MSCI AC World index.
As with region weights, large differences are obvious among the relative sector weights to the
CW MSCI AC World for all indices. Especially noticeable is the MVW indexs consistent over-
weight in the four defensive sectors; Consumer staples, Health care, Telecom services and Utili-
ties. This is in accordance with its underlying construction principle of minimising volatility and
a reason for the significantly lower standard deviation compared to the CW MSCI AC World in-
dex. The MSRW indices are also overweighted in the defensive sectors but not to the same de-
gree as the MVW index. Another interesting observation is the large weight in financial stocks of
the FW index. In light of the occurrence of the financial crisis and the sovereign debt crisis that
make up a large part of the backtesting period compared to other studies, the inferior return of
the FW is less surprising.
Taken all together the representativeness of the alternative indices is similar to the CW MSCI AC
World in terms of an adequate presence in all regions and sectors although significant differ-
ences among the relative region and sector weights appear. Appendix 13 further shows the his-
torical exposures to regions and sectors for each index and reveals pronounced instability for
the optimisation-based indices.

5.2.5 Liquidity
To be able to serve as a usable and applicable index in practice the alternatively weighted indi-
ces must have a certain degree of liquidity and investment capacity such that many investors
can trade them without any significant price impact. There are several useful ways to gauge
liquidity of which this thesis computes a CAP ratio35 and the weighted monthly dollar trading
volume. The two measures are presented in table 5.9 for each index using the full sample period
and annual rebalancing.
Table 5.9 - Liquidity characteristics

Weighted monthly $
Index
CAP ratio
trading volume (millions)
CW
1.00
383
EQW
0.20
76
FW
0.83
294
MVW
0.28
98
MSRW (actual)
0.27
99
MSRW (GR6)
0.22
75
MSRW (VAMOS)
0.30
115
Notes: The CAP ratio is computed as the weighted average market capitalisation of each index divided by the weighted average
market capitalisation of the CW MSCI AC World index. The weighted monthly $ trading volume is computed as the weighted aver-
age of the product of the 12 month average monthly price and 12 month average monthly trading volume for each constituent. All
measures are computed at each rebalancing date and averaged over the full sample period (2006-2013).
Source: Own analysis, Wilshire Atlas


35 Inspired by Arnott, Hsu & Moore (2005).

64

5. Empirical findings
The CAP ratio measures the relative investment capacity of each alternative index by dividing
the weighted average market capitalisation of each index by the weighted average market capi-
talisation of the CW MSCI AC World index. For example a CAP ratio of 0.30 for the MSRW
(VAMOS) index suggests that the weighted average market capitalisation of the constituents in
this index is a little less than one-third of the CW MSCI AC World index. Another inference is that
the aggregate amount of money that can be benchmarked to or invested in the MSRW (VAMOS)
index is approximately one-third the amount that could be benchmarked to or invested in the
CW MSCI AC World index. The CAP ratios for the other optimisation-based indices are a little
lower than the MSRW (VAMOS) index, whereas the CAP ratio of the EQW index is only 0.20. This
is clearly expected since the all constituents in the EQW index are given the same weight regard-
less of market capitalisation. Hence, at rebalancing the demand for the smallest stocks will be
relatively higher than that of the biggest stocks, which can potentially produce liquidity and
price pressure.
The weighted monthly $ trading volume of the indices paints the same picture and indicates
that the alternative indices except the FW index have liquidity between one-third and one-fifth
that of the CW MSCI AC World index. The lower liquidity and investment capacity of the alterna-
tive indices means that they might potentially face relatively higher increases in transaction
costs than the CW MSCI AC World index. Arnott, Hsu & Moore (2005) claim that their alterna-
tive indices with half the liquidity of their CW reference index are not seriously restricted given
that in 2005 more than $1 trillion is passively managed in some variant of CW indices. Follow-
ing their argumentation and taken into account that as of 30 June 2011 close to $6 trillion worth
of assets is passively managed (Olsen, 2011), the inferior liquidity and investment capacity of
the alternative indices do not seem to be a serious constraint.

5.2.6 Transparency
The final quality criterion is transparency defined in section 2.3.2.5 as the availability of infor-
mation on the concept, methodology and data used to construct any given index. In other words,
transparency helps investors understand the objective of each index and how it could be repli-
cated. Naturally, the two most transparent indices are the CW and the EQW as they are very
easy to understand and would be fairly easy to replicate by most investors. Next comes the FW
index that also has a quite intuitive concept, but requires access to financial databases in order
to retrieve the substantial amount of annual report figures of the constituents and is therefore
harder and more cumbersome to replicate.
The most opaque indices are by far the optimisation-based indices relying on numerous param-
eter estimates and decisions with respect to individual weight constraints, covariance-matrix,
proxy for expected returns, etc. As such, the replicability of the optimisation-based indices is

65

5. Empirical findings
very complex as the underlying methodology used to construct these indices is very technical.
By means of chapter 3 and 4 on methodology and data, this thesis provides sufficient infor-
mation on the optimisation-based indices to enable others to reproduce them, but acknowledg-
es the involvement it would require.
Taken together, the analysis of the quality criteria of the alternative indices does not point out any
serious restrictive features that erode their superior performance measures or disallow their usage
in practice. Turnover is increased but still very robust to the occurrence of transaction costs. Ex-
cept the FW index all indices are less concentrated than the CW MSCI AC World and thus reduces
one of the main points of criticism of the traditional reference index. Lower concentration does
translate into lower liquidity and investment capacity, but not to a degree where the alternative
indices would not be implementable in practice. Representativeness is also intact for all indices
despite that the optimisation-based indices show a general tendency to favour stocks from emerg-
ing regions and defensive industrial sectors. Compared to cap-weighting transparency is naturally
reduced when using optimisation-based weighting schemes.

5.2.7 Robustness tests


To test the robustness of the main findings presented in the previous sections, the study will
now examine the sensitivity of these findings to changes in the key parameters that make up the
construction of the alternative indices. As pointed out in section 3.2 the four key parameters are
the choice of (1) backtesting period, (2) rebalancing frequency, (3) weight constraints on con-
stituents and (4) covariance matrix. Note that the proxy for expected returns is clearly also a
key parameter and therefore included throughout the evaluation of alternative indices as well
as in each of the following four sections.
5.2.7.1 Backtesting period
First, the backtesting period is decomposed into three disparate time periods with different
market trend characteristics to test how this influences performance of each of the constructed
indices. The three time periods are:

January 2006 June 2008 (pre-crisis)

July 2008 December 2010 (financial crisis)

January 2011 May 2013 (sovereign debt crisis)

Table 5.10 breaks down the average return, standard deviation and Sharpe ratio of each index
by time period. It is clearly seen that the three time periods lead to different performance levels
of the alternative indices. For example is the EQW index in terms of average return the best-
performing index during the financial crisis but the worst performing index during the sover-
eign debt crisis. The performance fluctuations are also pronounced in similar analyses by Chow

66

5. Empirical findings
et al. (2011) and Clare, Motson & Thomas (2013a; 2013b), which underlines the impact of the
prevailing market conditions on the performance of the various indices.
Table 5.10 - Performance measures for subsamples

January 2006 - June 2008

July 2008 - December 2010

January 2011 - May 2013

Average Standard
Sharpe
Average Standard Sharpe
Average Standard
Index
return deviation
ratio
return deviation
ratio
return deviation
CW
8.2%
12.0%
0.33
-0.2%
26.0%
-0.02
7.6%
14.8%
EQW
12.7%
14.5%
0.59
8.2%
29.7%
0.27
4.6%
16.3%
FW
8.2%
13.3%
0.30
0.8%
30.9%
0.01
5.3%
18.0%
MVW
13.0%
10.9%
0.81
5.8%
19.9%
0.28
9.4%
9.7%
MSRW (actual)
18.5%
15.0%
0.95
-2.0%
23.6%
-0.10
9.7%
10.3%
MSRW (GR6)
20.6%
16.3%
1.01
-1.3%
25.4%
-0.06
8.5%
10.7%
MSRW (VAMOS)
16.7%
15.0%
0.83
2.7%
25.6%
0.09
10.8%
11.5%
Note: The performance measures are annualised and based on monthly data from each of the three subsample periods.
Source: Own analysis, Wilshire Atlas

Sharpe
ratio
0.51
0.28
0.29
0.97
0.94
0.78
0.94

Nonetheless, the MVW and MSRW (VAMOS) indices stand out by consistently outperforming the
CW MSCI AC World index in all three periods in terms of average return and Sharpe ratio. This
indicates that the outperformance of these two indices are in fact robust to changing market
conditions, which supports the finding in section 5.2.1.4 that only these two indices have signifi-
cantly higher Sharpe ratios than the CW MSCI AC World index. Another notable feature is that
for each time period the index with the lowest standard deviation is the MVW index. Lastly, it
should be noted that the FW index performs much worse than the CW MSCI AC World in the
most recent period, which partly explains its lower performance compared to the other related
studies that do not include this time period.
5.2.7.2 Rebalancing frequency
To test the impact of how frequently the alternative indices are rebalanced, this study con-
structs variations of the EQW index and the four optimisation-based indices where the re-
balancing is carried out semi-annually and quarterly36. Table 5.11 on the next page presents
selected performance measures and turnover characteristics for the different rebalancing fre-
quencies.
Overall, it shows that the performance of the alternative indices is reduced when the rebalanc-
ing frequency is increased. In particular, the Sharpe ratio drops consistently for all indices and
the same pattern is almost evident for the average return and information ratio. Additionally,
the turnover naturally increases with more frequent rebalancing, which in combination with the
lower returns, results in indifference levels of transaction costs that are quite close to the 1.44%
mutual fund trading costs estimated by Evans & Edelen (2013). Consequently, this thesis reach-
es the same conclusion as Chow et al. (2011) and finds no benefits from more frequent re-
balancing.

36 As this paper constructs the FW index on the basis of annual accounting measures, testing the impact of

more frequent rebalancing is not relevant.

67

5. Empirical findings
Table 5.11 - Performance measures and turnover for different rebalancing frequencies
Index
CW

Indifference
level of trans-
action costs

Average
return
5.1%

Standard
deviation
18.4%

Sharpe
ratio
0.19

Information
ratio
-

Annual one-
way turnover
4.4%


21.1%


0.33


0.71


17.9%



25.2%

14.1%
17.2%
18.5%
18.2%

0.55
0.40
0.40
0.46

0.64
0.52
0.55
0.98

30.2%
32.3%
33.9%
43.7%

16.4%
11.7%
12.8%
12.2%


21.1%
13.9%
16.7%
19.7%
19.6%


0.29
0.52
0.39
0.39
0.41


0.54
0.55
0.48
0.59
0.97


23.4%
44.5%
48.6%
49.6%
77.4%


13.4%
9.0%
6.5%
9.0%
6.0%


Annual
rebalancing
EQW
MVW
MSRW (actual)
MSRW (GR6)
MSRW (VAMOS)

8.5%
9.4%
8.4%
8.9%
9.9%


Semi-annual
rebalancing
EQW
MVW
MSRW (actual)
MSRW (GR6)
MSRW (VAMOS)

7.7%
8.7%
8.0%
9.2%
9.5%








Quarterly
rebalancing
EQW
7.4%
21.5%
0.27
0.46
30.4%
8.8%
MVW
8.8%
13.9%
0.52
0.56
62.2%
6.3%
MSRW (actual)
7.8%
16.7%
0.37
0.43
70.0%
4.0%
MSRW (GR6)
8.9%
19.4%
0.38
0.56
71.8%
5.7%
MSRW (VAMOS)
8.5%
18.4%
0.38
0.72
113.9%
3.1%
Note: Refer to table 5.7 for a description of how annual one-way turnover and the indifference level of transaction costs are calcu-
lated.
Source: Own analysis, Wilshire Atlas

5.2.7.3 Weight constraints on constituents


Among the numerous related studies of alternative indices presented in the literature review no
consensus on how to constrain the individual constituent weights exists. For this reason it is
particularly important to investigate how the indices respond to both tighter (=2) and looser
(=15) weight constraints.
Table 5.12 - Performance measures, concentration and liquidity for different levels of
Average
Sharpe Information
effective
ratio
ratio constituents
0.19
-
438

Relative
deconcen-
tration w.r.t.
CW MSCI AC
World
-

Average
return
5.1%

Standard
deviation
18.4%

=5 (0.008% < wi < 0.200%)



MVW
9.4%
MSRW (actual)
8.4%
MSRW (GR6)
8.9%
MSRW (VAMOS)
9.9%


14.1%
17.2%
18.5%
18.2%


0.55
0.40
0.40
0.46


0.64
0.52
0.55
0.98


606
603
603
601


1.4
1.4
1.4
1.4

0.28
0.27
0.22
0.30

=2 (0.02% < wi < 0.08%)


MVW
MSRW (actual)
MSRW (GR6)
MSRW (VAMOS)


17.6%
20.8%
19.3%
19.6%


0.46
0.38
0.39
0.38


1.28
0.85
0.88
0.95


1698
1697
1697
1696


3.9
3.9
3.9
3.9

0.23
0.23
0.21
0.24

Index
CW


9.6%
9.3%
9.0%
9.0%

CAP ratio
1.00

=15 (0.003% < wi < 0.600%)








MVW
9.4%
12.1%
0.65
0.44
196
0.4
0.39
MSRW (actual)
7.9%
15.7%
0.41
0.31
187
0.4
0.33
MSRW (GR6)
8.5%
18.9%
0.37
0.37
185
0.4
0.23
MSRW (VAMOS)
9.3%
16.9%
0.46
0.77
183
0.4
0.38
Note: Refer to table 5.8 and 5.9 for a description of how average effective constituents, relative deconcentration w.r.t. the CW MSCI
AC World index and CAP ratio are calculated.
Source: Own analysis, Wilshire Atlas

68

5. Empirical findings
Table 5.12 presents selected performance measures as well as concentration and liquidity char-
acteristics when the flexibility parameter, , introduced in section 3.2.3, is changed for the opti-
misation-based indices. None of the Sharpe ratios of the optimisation-based indices are materi-
ally affected by changing weight constraints, except the MVW index that is able to increase
Sharpe ratio when =15 and vice versa when =2. This changes when looking at the impact on
information ratio, which is substantially reduced when loosing weight constraints as it increas-
es tracking error. The opposite is true when tightening weight constraints. Addressing the im-
pact on concentration and liquidity characteristics, tighter weight constraints (resembling equal
weights), as expected, lead to less concentrated indices but more liquidity pressure relative to
the CW MSCI AC World index. On the contrary, looser weight constraints lead to more concen-
trated indices but less liquidity pressure. Note that for each level of , the optimisation-based
indices once again have very similar concentration characteristics caused by the applied opti-
misers tendency to assign the preferred stocks the maximum allowed weight, and then let the
remaining stocks be weighted by the minimum weight so that the sum of all stocks equal 100%.
This tendency also explains the large differences in concentration characteristics when chang-
es, and clearly demonstrate the optimisation-based indices sensitivity to the applied weighting
constraint37.
Balancing the observations from table 5.12, setting =5 seems appropriate compared to =2 and
=15 as it produces similar performance measures while maintaining an adequate level of de-
concentration and liquidity compared to the CW MSCI AC World index.
5.2.7.4 Covariance matrix
To test the impact of how the covariance matrix of the optimisation-based indices is construct-
ed, this study estimates alternative covariance matrices where the estimation period, data fre-
quency and weighting scheme are changed. Table 5.13 on the next page presents selected per-
formance measures for the different covariance matrices.
It clearly shows that the covariance matrix estimation methodology has a significant impact on
the performance of the optimised indices. Ex ante, investors have little reason to expect one of
the four techniques to consistently offer better performance than the others. Still, it is notewor-
thy that the Sharpe ratios in all cases are at least double that of the CW MSCI AC World index.


37 The nature of the optimiser also explains the instable exposure to geographical regions and industrial

sectors shown in appendix 13.

69

5. Empirical findings

Table 5.13 - Performance measures for different covariance matrices


Average
return
5.1%

Standard
deviation
18.4%

Tracking
error
-

Sharpe
ratio
0.19

Information
ratio
-


9.4%
8.4%
8.9%
9.9%


14.1%
17.2%
18.5%
18.2%


6.6%
6.3%
6.9%
4.9%


0.55
0.40
0.40
0.46


0.64
0.52
0.55
0.98

Monthly, exponential (=0.91)


MVW
MSRW (actual)
MSRW (GR6)
MSRW (VAMOS)


6.9%
11.7%
9.7%
7.8%


14.3%
16.1%
18.5%
18.4%


6.1%
6.2%
6.2%
4.7%


0.38
0.63
0.44
0.34


0.29
1.07
0.73
0.58

Daily, equal
MVW
MSRW (actual)
MSRW (GR6)
MSRW (VAMOS)


12.3%
12.3%
10.9%
11.0%


14.9%
17.3%
19.6%
19.4%


6.3%
6.0%
6.4%
5.1%


0.72
0.62
0.48
0.49


1.15
1.20
0.91
1.16

Daily, exponential (=0.97)






MVW
9.4%
14.1%
6.6%
0.55
MSRW (actual)
8.8%
15.8%
6.2%
0.46
MSRW (GR6)
11.6%
18.8%
6.6%
0.54
MSRW (VAMOS)
11.0%
19.4%
5.1%
0.49
Note: Monthly and daily covariance matrices are based on 60 observations and 500 observations, respectively.
Source: Own analysis, Wilshire Atlas


0.64
0.59
0.98
1.16

Index
CW AC World
Monthly, equal
MVW
MSRW (actual)
MSRW (GR6)
MSRW (VAMOS)

Generally, this thesis finds that changing key construction parameters often lead to different in-
sample performances. Ex-ante, it is therefore difficult to conclude which combination of the key
parameters that would lead to the strongest out-of-sample performance. Nonetheless, all alterna-
tive indices except the FW index have consistently outperformed the CW MSCI AC World index irre-
spective of changes in rebalancing frequency, weight constraints, covariance matrix and expected
returns. On the other hand, decomposing the backtesting period into three disparate time periods
shows some examples of underperformance. However, the MVW and MSRW (VAMOS) indices dis-
play robust outperformance in all three time periods.

5.2.8 Qualitative scores summarising the findings for each MSCI AC World index
To summarise the findings of the alternatively weighted MSCI AC World indices, qualitative
scores from 1 (bad) to 5 (good) are given to all indices for each performance measure category
and quality criterion based on the preceding analyses (see table 5.14). In this respect it is im-
portant to keep in mind that the superiority of any given index depends on the objective of its
user. Some investors prefer indices such as the CW MSCI AC World as it has very low turnover
(i.e. easy to rebalance), high liquidity (i.e. low transaction costs) and great transparency (i.e.
intuitive and easy to replicate). These features translate into a total quality criteria score of 20,
which is the highest quality criteria score of all indices. Other investors prefer optimisation-
based indices due to their superior performance measures. According to the findings of this
thesis, the choice would then be between the MVW index and the MSRW (VAMOS) index de-

70

5. Empirical findings
pending on the desired level of risk as they both receives a total performance measures score of
14, which is the highest score.
Table 5.14 - Qualitative scores summarising the findings of the alternative indices

CW
2

EQW
3

FW
1

MVW
5

MSRW
(actual)
4

MSRW
(GR6)
4

MSRW
(VAMOS)
5

Total PM score

14

11

11

14

Turnover

Concentration

Representativeness

Liquidity

Transparency

20

18

16

13

13

13

13

27

26

19

27

24

24

27


Return and risk

Performance Risk-adjusted ratios


measures
Extreme risks

Quality
criteria

Total QC score
Total score

Source: Own contribution, Wilshire Atlas

Overall, the three indices with the highest total score are the CW MSCI AC World, MVW and MSRW
(VAMOS) indices. Given their extremely different properties, the necessity of carefully assessing
which properties are important in a given investment context is emphasised as there exists no one
index that fulfils all objectives equally well. As pointed out by Amenc et al. (2011), portfolio theory
suggests that when indices are used as investment benchmarks the focus should be on achieving
the highest possible risk-reward ratio. According to this and given the thesis focus on how alterna-
tive equity indices used as benchmarks affect the performance of GE, it should be expected that the
constructed MVW and MSRW (VAMOS) indices have the largest positive impact on the perfor-
mance of GE. Whether this is actually the case is investigated carefully in the next section of this
thesis.

5.3 Impact of alternatives benchmark indices on Global Equities (SRQ 2)


The thesis now concentrates its attention on the GE portfolio and more specifically addresses
the first distinct choice of Jyske Invests portfolio construction strategy concerning which
benchmark index to use. To provide an answer to this, the performance of GE is backtested
(constituents fixed) applying each of the alternatively weighted indices from the previous anal-
ysis as benchmark. To isolate the effect of changing benchmark index it is necessary to accurate-
ly imitate the tracking error constraint imposed on GE historically as this is the other distinct
choice of Jyske Invests portfolio construction strategy38. As mentioned in the introduction to GE
(section 1.6.2) and in the description of the methodology used for this analysis (section 3.3) it
can be argued that the aim of the benchmark + 0.7% rule is to track the benchmark as close as
possible. Moreover, figure 5.3 in the comparative performance analysis of GE and the CW MSCI
AC World index shows that their cumulative returns have followed each other closely despite

38 For an in-depth analysis of the impact of tracking error constraints on GE refer to section 5.4.

71

5. Empirical findings
big differences in their stock universe. Consequently, simulated GE portfolios are constructed by
solving the optimisation problem (equation 3.10) that minimises tracking error to each of the
indices investigated in the previous analysis. For a full description of the methodology used to
carry out this analysis refer to section 3.3. Note that to focus the analysis of GE the MSRW indi-
ces based on actual returns and GR6 model returns are disregarded as they prove to be inferior
to the MSRW index based on the VAMOS score. Finally, a selection of the most relevant perfor-
mance measures introduced in section 2.2 is used as framework to evaluate the impact of using
alternative indices as benchmark.
Table 5.15 presents GE performance measures when tracking error is minimised with respect
to each of the alternative indices.
Table 5.15 - GE performance measures applying alternative benchmark indices
Tracking Tracking
Informa- Informa-
error
error
tion ratio tion ratio
w.r.t. w.r.t. CW
w.r.t. w.r.t. CW
TE minimising
Average Standard
applied
MSCI AC
Sharpe
applied
MSCI AC Monthly
GE portfolios
return deviation
index
World
ratio
index
World 95% VaR
GE (actual)
6.9%
19.4%
4.3%
4.3%
0.28
0.41
6.6%
GE (w.r.t. CW)
6.5%
18.7%
2.8%
2.8%
0.27
0.52
0.52
6.6%
GE (w.r.t. EQW)
8.1%
20.8%
4.1%
4.3%
0.32
-0.09
0.71
7.3%
GE (w.r.t. FW)
6.8%
20.5%
4.8%
3.6%
0.26
0.43
0.46
7.7%
GE (w.r.t. MVW)
7.9%
15.8%
4.9%
5.6%
0.41
-0.29
0.50
4.9%
GE (w.r.t. MSRW (VAMOS))
8.4%
18.2%
3.6%
5.0%
0.38
-0.42
0.65
5.5%
Notes: Tracking error and information ratio have both been calculated with respect to the applied benchmark index as well as the
CW MSCI AC World index. All measures are annualised except the 95% VaR. A Cornish-Fisher expansion was used to compute a
Value-at-Risk estimate that takes into account the median, volatility, skewness and excess kurtosis of GE returns.
Source: Own analysis, Wilshire Atlas

Before analysing the performance measures it should be noticed that except the traditional av-
erage return, standard deviation and tracking error measures, all other performance measures
between the actual GE portfolio and the tracking error minimising GE portfolio with respect to
the CW MSCI AC World index are quite similar. This implies that the simulated GE portfolios in
this analysis are in fact a valid approximation of the benchmark + 0.7% construction strategy.
Nonetheless, it seems like the simulated GE portfolios impose somewhat stricter tracking error
constraints than the actual portfolio.

5.3.1 Return and risk


From table 5.15 it is clear that the choice of benchmark index has a very large impact on GE
when a tracking error minimising construction strategy like the benchmark + 0.7% rule is
pursued. The index with the biggest impact on the average return of GE is the MSRW (VAMOS)
index. Tracking this index increases the average return from 6.9% to 8.4%, whereas tracking
the FW index actually reduces the average return to 6.8%. Applying the EQW and MVW indices
as benchmark also has a positive impact on average return producing returns of 8.1% and 7.9%,
respectively.

72

5. Empirical findings
The greatest reduction in volatility for GE from its actual 19.4% standard deviation is unsurpris-
ingly achieved by tracking the MVW index (15.8%) followed by the MSRW (VAMOS) index
(18.2%). Conversely, had the EQW or FW indices been used as benchmark the standard devia-
tion would have increased to 20.8% and 20.5%, respectively.
The pattern in return and risk described above is very similar to the alternatively weighted in-
dices (see table 5.4), which illustrates that the performance measures of GE is closely connected
to the performance measures of the chosen benchmark when tracking error is kept at a mini-
mum. The only exemption is the GE portfolio tracking the EQW index, which generates higher
return than the GE portfolio tracking the MVW index and higher standard deviation than the GE
portfolio tracking the FW index. These relative GE differences are opposite to the performance
measures of the corresponding indices, where the MVW index beats the EQW index in terms of
average return and the FW index produces a higher standard deviation than the EQW index.

5.3.2 Risk-adjusted ratios


The uniform pattern in return and risk between the tracking error minimising GE portfolios and
their respective benchmark indices naturally translates into a similar pattern in Sharpe ratio. In
fact, the pattern proves to be identical and, as such, suggests that a direct relationship exists
between the Sharpe ratio of the benchmark index (see column 6 in table 5.4) and the Sharpe
ratio achieved by GE (see column 6 in table 5.15), preconditioned that the chosen benchmark is
tracked as closely as possible. In other words, tracking more efficient indices appears to have an
efficiency enhancing effect on GE.
A somewhat similar connection seems to be present for the information ratio calculated with
respect to the CW MSCI AC World index. An index that generates a high information ratio when
the CW MSCI AC World index is the benchmark (see column 8 in table 5.4) also results in a cor-
responding high information ratio for GE (see column 8 in table 5.15). When the information
ratio of GE is calculated with respect to the applied benchmark index it becomes clear that GE
only has positive information ratios (i.e. manages to beat the benchmark) for the CW and FW
indices whereas choosing the EQW or optimisation-based indices as benchmark produces nega-
tive information ratios as the return of the GE portfolios in all cases are inferior to their respec-
tive benchmark index. This finding is quite interesting as it implicitly indicates that active man-
agement is only rewarding when the benchmark is less efficient, while it produces inferior re-
turns to the benchmark when the benchmark is more efficient.

5.3.3 Significance tests


Similar to the analysis of the alternative indices, statistical significance tests are now conducted
to test if firm conclusions can be made on the changes in the performance of GE. For each of the
applied benchmarks, the differences in average return, standard deviation and Sharpe ratio

73

5. Empirical findings
between the simulated tracking error minimising GE portfolio and the actual GE portfolio are
tested for statistical significance. The results are presented in table 5.16.
Table 5.16 - Sign. tests of differences between actual GE and GE applying alternative indices
Average return
Standard deviation
Sharpe ratio
TE minimising GE portfolios
vs. actual GE portfolio
Difference p-value Difference p-value Difference
p-value
GE (w.r.t. CW)
-0.4%
64.3%
-0.7%
74.1%
-0.01
87.3%
GE (w.r.t. EQW)
1.2%
28.0%
1.4%
50.7%
0.04
54.2%
GE (w.r.t. FW)
-0.1%
95.1%
1.1%
60.2%
-0.02
79.7%
GE (w.r.t. MVW)
1.0%
88.2%
-3.6%
5.5%
0.13
16.3%
GE (w.r.t. MSRW(VAMOS))
1.5%
39.0%
-1.3%
53.2%
0.10
15.7%


Note: The p-values for differences are computed using a paired t-test fir the average return, a Fischer test for the standard devia-
tion and a Jobson-Korkie test for the Sharpe ratio.
Source: Own analysis, Wilshire Atlas

The main thing to notice is that none of the statistical tests are significant at the 5% level. In fact,
at the 10% significance level the only simulated GE portfolio that has a significant difference
from the actual GE portfolio is the one that is tracking error minimised with respect to the MVW
index. This portfolio generates significantly lower volatility at the 10% level. As such, from a
strictly statistical point of view it is not possible to conclude any performance differences when
alternatively weighted versions of the CW MSCI AC World index are used as benchmark for GE.
Despite statistical insignificance it is still reasonable to claim that, from an economically point of
view, the observations made this far prove that the choice of benchmark index heavily impacts
the performance of GE. To illustrate this, figure 5.7 shows for each alternative index the per-
centage change between the Sharpe ratio obtained when choosing a different index and the ac-
tual Sharpe ratio obtained with the CW MSCI AC World index as benchmark. The indices on the
x-axis are ranked according to their total performance score (i.e. efficiency) calculated in table
5.14.
Figure 5.7 - % Sharpe ratio change for GE when tracking alternative indices
50%
40%
30%
20%
10%
0%
-10%
GS (w.r.t. FW)

GS (w.r.t. CW)

GS (w.r.t. EQW)

GS (w.r.t. MSRW
(VAMOS))

GS (w.r.t. MVW)

Source: Own analysis, Wilshire Atlas

Figure 5.7 illustrates a clear positive correlation between the efficiency of the benchmark indi-
ces and the Sharpe ratio effect on GE of applying alternative indices as benchmark. Choosing a
more efficient index such as the MVW index improves the Sharpe ratio of GE by up to 46%,
while the effect is negative when the less efficient FW index is used as benchmark. This is in
accordance with the proposition of Amenc et al. (2011) who advocate actively managed portfo-
lios to use the most efficient index available when it serves as investment benchmark. Addition-

74

5. Empirical findings
ally, the empirical findings support the theory of Roll (1992), Jorion (2003) and Hope (2008)
who state that tracking inefficient benchmark indices closely lead to inferior performance. To
the knowledge of the authors no previous studies have specifically tested the impact of alterna-
tive benchmark indices on an actively managed portfolio and thus no comparisons are provided.
Taken together, the actual Sharpe ratio achieved by GE when applying the less efficient CW MSCI
AC World index as benchmark is 0.28 over the full sample period. Had the much more efficient
MVW and MSRW (VAMOS) indices been applied and tracked as closely as possible the Sharpe ratio
would have been 0.41 and 0.38, respectively. This implies that Jyske Invest can increase the
risk/reward properties of GE by altering the first distinct choice of their portfolio construction
strategy and choose a more efficient index than the currently used CW MSCI AC World index. In the
succeeding analysis the other choice of their portfolio construction strategy concerning which
tracking error constraint to impose the portfolio managers of GE is investigated along with an
assessment of how this is affected by the choice of benchmark.

5.4 Impact of loosening tracking error constraint on Global Equities (SRQ 3)


Having evaluated the first distinct choice in the overall portfolio construction strategy for Jyske
Invests GE portfolio, the analysis now move on to the second part; what tracking error con-
straint to impose the portfolio managers. The performance of GE when loosening the tracking
error constraint gradually, is analysed using the full range of indices considered in this thesis as
the choice of index is expected to affect the optimal level of tracking error. It should be empha-
sised that tracking errors and information ratios in this analysis are exclusively calculated with
respect to the applied index. For a full description of the methodology used to carry out this
analysis refer to section 3.4. Note that besides solving the optimisation problem (equation 3.11)
using the constituent weights of the five tracking error minimising portfolios constructed in
section 3.3, the tracking error constraint is also loosened for the actual historical weights of GE.
This simulation is performed to test if the same performance pattern is evident for the actual GE
portfolio (that uses the CW MSCI AC World index as benchmark) and the simulated tracking
error minimising GE portfolio constructed using the CW MSCI AC World index as benchmark.

5.4.1 Return and risk


Figure 5.8 presents the average return and standard deviation when loosening the tracking er-
ror constraint for the actual GE portfolio and the five simulated GE portfolios from the analysis
in section 5.3. Notice, that gradually allowing for greater weight deviations from the benchmark
indices does not always translate into increasing tracking error, which is simply explained by
the fact that allowing greater weight deviations can actually, by chance, prove to track the re-
turns of the benchmark closer. This is the reason for the jumps in panel A and the C-shape in
panel E. Also, the level of tracking error and its sensitivity to larger weight deviations is differ-

75

5. Empirical findings
ent for each index, which is unavoidable as generating the same levels of tracking error is natu-
rally not possible as a target level of tracking error is not necessarily equal to the actually in-
curred tracking error.
Figure 5.8 - Return and standard deviation when loosening tracking error constraint

B. GE (w.r.t CW)

A. GE (actual)

8.5%

20.5% 8.5%

20.5%

8.0%

19.0% 8.0%

19.0%

7.5%

17.5% 7.5%

17.5%

7.0%

16.0% 7.0%

16.0%

6.5%
4.4%

4.6%

4.8%
5.0%
Tracking Error

14.5% 6.5%
2.8%
5.2%

3.2%
3.6%
Tracking Error

4.0%

14.5%
4.4%

D. GE (w.r.t. FW)

C. GE (w.r.t. EQW)

8.5%

20.5% 8.5%

20.5%

8.0%

19.0% 8.0%

19.0%

7.5%

17.5% 7.5%

17.5%

7.0%

16.0% 7.0%

16.0%

6.5%
4.0%

4.5%

4.9%
5.4%
Tracking Error

5.8%

E. GE (w.r.t. MVW)

14.5% 6.5%
6.3%
4.8%

14.5%
5.4%

5.9%
6.5%
Tracking Error

7.0%

F. GE (w.r.t. MSRW (VAMOS))

8.5%

20.5% 8.5%

20.5%

8.0%

19.0% 8.0%

19.0%

7.5%

17.5% 7.5%

17.5%

7.0%

16.0% 7.0%

16.0%

6.5%
4.7%

4.7%

4.8%
4.8%
Tracking Error

14.5% 6.5%
4.9%
3.6%

14.5%
4.0%

4.4%
4.8%
Tracking Error

5.2%

Notes: In panel A the the tracking error constraint is gradually loosened for the actual GE portfolio and in panel B-F it is loosened for
the simulated tracking error minimising GE portfolio constructed using the indices in brackets as benchmark.
Source: Own analysis, Wilshire Atlas

When the CW, FW and EQW MSCI AC World indices are used as benchmarks (panel A-D), the
overall picture is that a higher tracking error gives higher returns and lower volatility, implying
that if the portfolio managers of GE are allowed to deviate more from these indices, they can
generate a higher return for a lower risk (i.e. higher Sharpe ratio). Yet, when applying the EQW
index (panel C), it is noticed that the average return peaks with a tracking error of 4.9% and
that further increases in tracking error actually results in lower returns.
Had the MVW index (panel E) been applied as benchmark, the pattern shows that allowing for
greater weight deviations would not have resulted in increased tracking error. In fact, the
tracking error only fluctuates in the interval from 4.7% to 4.9%, which is very narrow compared
to the effect on the other indices. As such, it is difficult in this case to make any insightful
inferences of the effect of allowing for greater weight deviations since they do not entail higher
tracking error. If anything, it can be seen that irrespective of the allowed weight deviation, the
average return and volatility share the same trend and follow each other very closely.

76

5. Empirical findings
Finally, Panel F shows that for the MSRW (VAMOS) index higher tracking error consistently
leads to lower average return and volatility. This implies that more risk averse investors would
prefer looser tracking error constraints when the MSRW index is utilised. However, part of the
reason is also attributable to the applied optimiser that seems to be favouring low risk stocks
when no increase in Sharpe ratio is possible. As in the case of the MVW index the average return
and volatility track each other very closely.

5.4.2 Risk-adjusted ratios


Moving on to the risk-adjusted ratios, figure 5.9 shows the development of Sharpe ratio and
information ratio when the tracking error constraint is loosened.
Figure 5.9 - Sharpe ratio and information ratio when loosening tracking error constraint

A. GE (actual)

0.41

0.65

B. GE (w.r.t. CW)

0.41

0.65

0.38

0.58

0.38

0.58

0.34

0.50

0.34

0.50

0.31

0.43

0.31

0.43

0.36

0.27
2.8%

0.41

0.00

0.41

0.65

0.38

-0.03

0.38

0.58

0.34

-0.05

0.34

0.50

0.31

-0.08

0.31

0.43

0.27
4.3%

4.5%

4.7%
5.0%
Tracking Error

5.2%

C. GE (w.r.t. EQW)

0.27
4.0%

4.6%

5.1%
5.7%
Tracking Error

3.2%

3.6%
4.0%
Tracking Error

0.36
4.4%

D. GE (w.r.t. FW)

-0.10
6.3%

0.27
4.8%

0.36
5.4%

5.9%
6.5%
Tracking Error

7.0%

F. GE (w.r.t. MSRW(VAMOS))

E. GE (w.r.t. MVW)

0.41

-0.30

0.41

-0.30

0.38

-0.35

0.38

-0.35

0.34

-0.40 0.34

-0.40

0.31

-0.45

0.31

-0.45

0.27
4.7%

-0.50

0.27
3.6%

4.7%

4.8%
4.8%
Tracking Error

4.9%

-0.50
4.0%

4.4%
4.8%
Tracking Error

5.2%

Notes: In panel A the the tracking error constraint is gradually loosened for the actual GE portfolio and in panel B-F it is loosened for
the simulated tracking error minimising GE portfolio constructed using the indices in brackets as benchmark.
Source: Own analysis, Wilshire Atlas

From figure 5.9 it can be seen that the Sharpe ratio in panel A, B, C and D increases when track-
ing error increases. According to the previous analysis of return and risk this was expected for
panel A, B and D, whereas the growing Sharpe ratio when the EQW index is applied as bench-
mark (panel C) reveals that the decline in average return when deviations larger than +/- 40%
are allowed (or tracking error increases more than 4.9%) is smaller than the corresponding
decline in volatility. Conversely, the two optimisation-based indices show a slightly declining
Sharpe ratio when tracking error increases, though inferences for the MVW index should still be

77

5. Empirical findings
made with care. Taken together this indicates that with respect to the two most efficient indices
(found to be the MVW and MSRW (VAMOS) indices in section 5.2) the effect of loosening the
tracking error constraint on the portfolio managers of GE actually has a negative impact on
Sharpe ratio, whereas the effect is positive for the less efficient indices.
A somewhat similar trend is evident for the information ratio, however in panel A, B, C and D
the increase in information slowly diminishes or start declining once a certain tracking error
threshold is reached. From this threshold the higher tracking error is not compensated by a
higher active return, which leads to a decreasing information ratio. Thus for investors to whom
the information ratio has the highest importance, stricter tracking error constraints are pre-
ferred compared to investors aiming for a higher Sharpe ratio.
Note also that irrespective of tracking error constraint the information ratio is still negative
when the EQW and optimisation-based indices are applied.

5.4.3 Significance tests


As in the previous analyses, testing whether or not the changes in the GE performance are sta-
tistical significant is required before any firm conclusions can be made. For each of the applied
benchmarks, the differences in average return, standard deviation and Sharpe ratio between the
loosest tracking error constraint (weight deviations of +/-80% from the simulated tracking er-
ror minimising portfolio) and the strictest tracking error constraint (no weight deviations from
the simulated tracking error minimising portfolio) are tested for statistical significance. The
results are presented in table 5.17.
Table 5.17 - Sign. tests of differences between GE with loosest and strictest TE constraint
Average return
Standard deviation
Sharpe ratio
+/-80% weight deviations
vs. no weight deviations
Difference p-value Difference p-value Difference p-value
GE (actual)
0.8%
76.1%
-1.9%
32.8%
0.08
17.3%
GE (w.r.t. CW)
1.1%
65.2%
-2.4%
20.1%
0.11
7.5%
GE (w.r.t. EQW)
0.1%
77.4%
-2.8%
17.4%
0.06
36.0%
GE (w.r.t. FW)
0.8%
92.9%
-3.1%
12.6%
0.09
13.6%
GE (w.r.t. MVW)
-0.5%
52.2%
-0.9%
60.4%
-0.01
89.3%
GE (w.r.t. MSRW (VAMOS))
-1.0%
32.3%
-2.1%
24.9%
-0.01
87.2%
for the standard devia-
Note: The p-values for differences are computed using a paired t-test fir the a verage return, a Fischer test
tion and a Jobson-Korkie test for the Sharpe ratio.
Source: Own analysis, Wilshire Atlas

Similarly to section 5.3.3 none of the statistical tests are significant at the 5% level. In fact, at the
10% significance level the only GE portfolio that has a significant difference is the one that is
allowed to deviate +/- 80% from the tracking error minimising portfolio with respect to the CW
MSCI AC World index, which generate significantly higher Sharpe ratio. As such, from a strictly
statistical point of view it is not possible to firmly conclude any performance differences among
strictly constrained and loosely constrained GE portfolios.
However, once again from an economically point of view, the observations made this far prove
that the level of tracking error heavily impacts the performance of GE despite statistical insignif-

78

5. Empirical findings
icance. Moreover, the impact is very closely tied to the efficiency of the chosen benchmark in-
dex. To illustrate this, figure 5.10 shows for each index the percentage change between the
Sharpe ratio obtained with the loosest tracking error constraint (weight deviations of +/-80%
from the simulated tracking error minimising portfolio) and the Sharpe ratio obtained with the
strictest tracking error constraint (no weight deviations from the simulated tracking error min-
imising portfolio). The indices on the x-axis are again ranked according to their total perfor-
mance score (i.e. efficiency) calculated in table 5.14.
Figure 5.10 - % Sharpe ratio change for GE when loosening tracking error constraint
45%
35%
25%
15%
5%
-5%
GS (w.r.t. FW)

GS (w.r.t. CW)

GS (w.r.t. EQW)

GS (w.r.t. MSRW
(VAMOS))

GS (w.r.t. MVW)

Source: Own analysis, Wilshire Atlas

It is clearly seen that there exists a negative correlation between the efficiency of the bench-
mark indices and the Sharpe ratio effect on GE of loosening the tracking error constraint. Allow-
ing a high tracking error for the less risk/reward efficient indices improves the Sharpe ratio by
up to 40%, while the effect is negative when the two most risk/reward efficient indices are used
as benchmark for GE. Interestingly, this shows the opposite trend of the corresponding graph in
section 5.3 (figure 5.7) that shows a positive correlation between the percentage Sharpe ratio
impact of choosing another benchmark and the efficiency of the chosen benchmark. Put differ-
ently, tracking a very efficient benchmark closely has a large impact on the Sharpe ratio of GE
whereas large deviations from an efficient benchmark do not. On the other hand, closely track-
ing an inefficient benchmark has a limited effect on the Sharpe ratio of GE whereas allowing for
large deviations from an inefficient benchmark has a big impact. This is intuitive as it, all things
being equal, should be harder to improve performance when deviating from an efficient index
than a less efficient index. To the knowledge of the authors no other related studies have been
made on the specific topic of testing the isolated impact of loosening tracking error constraint
for different indices, and thus no direct comparisons are not provided. The empirical findings
are, however, once again supported by the theory presented by Roll (1992), Jorion (2003) and
Hope (2008) who state that tracking inefficient benchmark indices closely lead to inferior per-
formance, which explains why looser tracking error constraints for the less efficient indices
have the highest impact on the performance of GE.

79

5. Empirical findings
To sum up, the actual Sharpe ratio achieved by GE is 0.28 over the full sample period. Had weight
deviations of +/- 80% of the actual constituent weights been allowed, and thereby higher tracking
error, a Sharpe ratio of 0.35 would have been achieved. On the other hand, had the much more
efficient MVW and MSRW (VAMOS) indices been tracked as closely as possible the Sharpe ratio
would have been 0.41 and 0.38, respectively. This implies that Jyske Invest can potentially increase
the risk/reward properties of GE by making one of two choices to in respect to their current portfo-
lio construction strategy; (1) either they maintain the CW MSCI AC World index and alter their
benchmark + 0.7% rule to allow for higher tracking error or (2) choose a more efficiently con-
structed index such as the MVW index or the MSRW (VAMOS) index, and maintain their strict
tracking error constraint, preferably by means of a more consistent process than the benchmark +
0.7% rule.

80

6. Evaluation of the findings

6. Evaluation of the findings


Some of the presented findings are in line with existing literature, some are contradictory, and
some have not previously been presented. This chapter evaluates the study in terms of its validity,
reliability and implications for further research. It provides a critical discussion of areas that po-
tentially impact the results and therefore should not remain unmentioned for future studies. Addi-
tionally, this chapter also addresses the practical implications of the findings.

6.1 Data
The length of the January 2006 to May 2013 sample period is restricted by the access to the
historical constituents of the CW MSCI AC World index, which are not publicly available and
therefore retrieved through Jyske Invest. Though the sample includes more recent data than
any of the related studies, it is significantly shorter than the studies by Arnott, Hsu & Moore
(2005), Chow et al. (2011) and Clare, Motson and Thomas (2013a; 2013b) covering approxi-
mately 40 years of data but comparable to the study by Amenc et al. (2011) covering less than 9
years of data. This suggests that our findings should only carefully be generalised to index and
portfolio construction in general. Moreover, it explains the limited power of many of the statis-
tical tests performed in this study.
Furthermore, the analyses of the impact of benchmark indices and tracking error constraints
are only conduced on a single actively managed portfolio, GE. This limitation is inevitable as the
analyses rely on proprietary data on the historic constituents of the active portfolio in question,
which is naturally only available for GE. Again this means that one must be cautious when gen-
eralising the empirical findings for GE to other actively managed portfolios though its invest-
ment setup is quite universal.
Regarding the Worldscope database, several outliers and invalid data points are identified in the
accounting measures retrieved from this source, especially for companies from emerging re-
gions. Thus, although some heuristic data validation tests (see section 3.2.2) and double checks
of annual reports are performed, the potential for invalid data is still present. Worldscope is
also the database used by Wilshires GR6 model to collect fundamental data, which might ex-
plain some of the spurious model returns especially evident for companies from emerging re-
gions. Despite applying the truncation technique introduced in section 3.2.4.2 this clearly affects
the reliability of the GR6 model and in particular the performance of the MSRW (GR6) index.
As a final note on the applied data, the Carhart (1997) portfolios are, similarly to Amenc et al.
(2011), retrieved directly from Kenneth R. Frenchs homepage (French, 2013). In light of the
differences between the constituents of the global Carhart portfolios from this source (include

81

6. Evaluation of the findings


small-caps but no emerging countries) and the alternative indices of this thesis (include emerg-
ing countries but no small-caps), it would be interesting to analyse factor exposures by means of
simulated Carhart portfolios sharing the same MSCI AC World constituents as the alternative
indices. Chow et al. (2011) simulate Carhart portfolios that better match the constituents of
their alternative indices and report more intuitive factor exposures of their indices than this
study is able to.

6.2 Methodology
From a methodological perspective, an essential discussion is; to which extent does the ex-post
performance measures of the constructed indices and portfolios of this thesis reflect their ex-
pected ex-ante performance measures. This touches the central assumption of the applied
backtesting methodology that markets will always behave in a similar manner to the chosen
sample period. Similarly to the full range of related studies the fluctuations in performance
measures are significant when decomposing the full sample period into shorter disparate time
periods. When combining this observation with the above-described potential limitations of the
data sources, it is natural for proponents of CW indices to argue that the performance fluctua-
tions of the alternatively weighted indices are too great to conclusively confirm their superiori-
ty.
Another important methodological issue to discuss is the choice of weighting schemes consid-
ered in this thesis. Besides equal weighting, fundamental weighting, minimum-volatility
weighting and maximum-Sharpe ratio weighting, a number of other weighting schemes exist as
well. Examples are; risk-cluster equal weighting, diversity weighting, inverse volatility
weighting, equal risk contribution weighting and maximum diversification weighting. In the
studies by Chow et al. (2011) and Clare, Motson & Thomas (2013a) they are also found to be
more efficient than cap weighting and thus interesting to investigate further.
The application of Wilshires GR6 multi-factor model is also worth discussing as it plays a cen-
tral role in all portfolio optimisations performed in this thesis. The estimation of covariance
matrices is in particular a direct result of the GR6 model. The robustness test of the covariance
matrix estimation methodology (section 5.2.7.4) proves that the performance measures are
largely affected by changes to the key estimation parameters. Chow et al. (2011) test the sensi-
tivity of their choice of covariance matrix by applying two Bayesian shrinkage methods (intro-
duced by Clarke, de Silva & Thorley (2006) and Ledoit & Wolf (2004a), respectively) and a prin-
cipal component analysis and find significant changes in their results too. Thus, the optimisa-
tion-based portfolios are very sensitive to the choice of covariance matrix estimation methodol-
ogy, still no one method is yet considered superior to the others.

82

6. Evaluation of the findings


The optimisation-based MSRW indices and portfolios are additionally very dependent on the
choice of proxy for expected returns. As in the case of covariance matrix estimation, no single
proxy is considered superior to the others. This thesis considers (1) historical average of actual
returns, (2) historical average of GR6 model returns and (3) Jyske Invests VAMOS score. Other
related studies postulate a stocks expected return is directly positively correlated to either total
volatility (Coignard & Choueifaty, 2008) or semi-deviation (Amenc et al., 2010) of its return.
Clare, Motson & Thomas (2013a) reject these economic intuitive relationships empirically but
still uses them as proxies, which emphasise the lack of great alternatives and that every choice
of proxy is very disputable.
The choice of constructing tracking error minimising GE portfolios using the relative mean-
variance optimisation setup proposed by Jorion (2003) also deserves elaboration. Its purpose is
to imitate Jyske Invests benchmark + 0.7% rule so that the isolated effect of changing bench-
mark and subsequently of relaxing tracking error constraints can be analysed. While the rule is
very benchmark dependent and the historical performance of GE have tracked the CW MSCI AC
World index very closely, it is debatable whether it can be classified as a true tracking error
minimising strategy. In fact, it shows that when applying Jorion (2003)s framework on GE with
respect to the CW MSCI AC World index a tracking error of 2.8% is incurred, which is somewhat
lower than the actual tracking error of 4.3% (see table 5.15). Had Jyske Invest exclusively cho-
sen stocks from the CW MSCI AC World index another imitation of the benchmark + 0.7% rule
would have been possible to carry out. This involves taking the sum of the benchmark weights
for all GE constituents and then adding a fixed number to all constituents so that the sum of
constituent weights adds to 100%39. Repeating this procedure at every rebalancing date ap-
pears to be a valid and interesting alternative approximation of the benchmark + 0.7% rule
although not perfect either due to the heuristic and inconsistent nature of the rule.

6.3 Further research


In light of the discussion of the limitations of the chosen data and methodology the authors par-
ticularly encourage further research to (1) apply a longer sample period and (2) gather histori-
cal constituent data for more actively managed portfolios. First, extending the sample period for
the analysis of alternative indices could be facilitated by getting access to the MSCI AC World
constituents over a longer time period. Alternatively, constituents could be chosen based on a
stock selection design that e.g. identifies the 2500 largest global stocks by market cap at each
historical rebalancing date, and then subsequently applies the weighting schemes considered in

39 For example, say GE consists of 100 stocks and the sum of benchmark weights for all GE constituents is

50%. Each constituent would then have 0.5 percentage points added to its benchmark weight and the
sum of all constituent weights adds to 100%.

83

6. Evaluation of the findings


this thesis as well as those from related studies. This would allow for more robust conclusions
and enhance the power of the statistical significance tests performed in this study. Second, to
the authors knowledge no prior research has specifically addressed the implications of choos-
ing benchmark index and tracking error constraint for actively managed portfolios. The meth-
odology introduced in section 3.3 and 3.4 is extendable to other active portfolios if the data on
historical constituents is obtainable. A broader analysis of a big pool of actively managed portfo-
lio over a longer sample period would be interesting to challenge the findings for GE. Also, it
would add additional and more general insights on which benchmark index to choose for active
portfolios and how to adjust the imposed tracking error constraints accordingly.
Finally, the sensitivity of this thesis optimisation-based indices and portfolios to the choice of
covariance matrix and proxy for expected returns highlights yet again the need for more robust
and valid parameter estimations when relying on Markowitz (1952)s traditional portfolio theo-
ry.

6.4 Practical implications


In spite of the pitfalls of the applied data and methodology the empirical findings presented in
this thesis have practical implications for investors and a variety of stakeholder groups. Today,
the primary goal of any actively managed portfolio, including GE, is to outperform its CW
benchmark index. This often compromises the ultimate goal for rational investors, which is to
maximise the return given a preferred level of risk. Constructing indices by means of cap
weighting is consistently found to be an inefficient investment strategy in this thesis and most
related studies on alternative indices. Hence, an actively managed portfolio that beats its ineffi-
cient benchmark is not granted to be the portfolio producing the most efficient risk/reward
properties. In fact, this thesis finds that a direct positive relationship exists between the Sharpe
ratio of the benchmark index and the Sharpe ratio achieved by GE when the chosen benchmark
is tracked very closely. In other words, tracking more efficient indices appears to have an effi-
ciency enhancing effect on GE. Interestingly, the large increase in Sharpe ratio when tracking
more efficient indices does not come at the cost of much larger tracking errors with respect to
the CW MSCI AC World index. As such, the information ratio also increases consistently when
applying more efficient benchmark indices. These findings breaks with current industry prac-
tice and implies that todays default choice of using the inefficient CW benchmark indices is
flawed and leaves investors with inferior risk/reward properties.
Analysing the impact of alternative benchmark indices and tracking error constraints also re-
veals that GE is unable to outperform the most efficient indices irrespective of the imposed
tracking error constraint. Actually, the performance of GE decreases when loosening the track-
ing error constraint with respect to the most efficient indices, whereas it improves when deviat-

84

6. Evaluation of the findings


ing more from the less efficient indices. This implies that the actively managed GE portfolio
would be an inefficient investment if the alternative indices constructed in this thesis were in-
vestable through passive investment funds. A scenario that is considered to be very realistic
given the sufficiently high quality scores of the alternatively weighted indices. This observation
fuels the heavily debated public dispute on active versus passive management and the results of
this thesis clearly favours passive management.
Further, the findings of this thesis open up for the possibility of applying the more efficient indi-
ces internally within an active investment fund while keeping a reference with CW indices ex-
ternally. This portfolio construction strategy is expected to both generate higher Sharpe ratios
and information ratios and ensure that the primary objective of active funds of beating their
benchmark is still fulfilled. Additionally, it acknowledges that CW indices are still likely to be the
default benchmark choice for some years to come due to their popularity, extensive availability
of track records and superior transparency and replicability compared to the proposed optimi-
sation-based indices.

85

7. Conclusion

7. Conclusion
Using a quantitative backtesting approach, this thesis provides an empirical investigation of
alternatively weighted versions of the CW MSCI AC World indices and their influence on the
construction strategy of Jyske Invests portfolio GE from January 2006 to May 2013. The alter-
native MSCI AC World indices are constructed by means of equal weighting, fundamental
weighting, minimum-volatility weighting and maximum-Sharpe ratio weighting.
First, a framework for evaluating index performance is proposed. Based on a thorough litera-
ture review a total of eight categories are chosen to compose the applied evaluation framework.
The framework combines performance measures (risk and return, risk-adjusted ratios and ex-
treme risk) and quality criteria (turnover, concentration, representativeness, liquidity and
transparency) and goes a step beyond related studies by considering measures that assess
whether the proposed indices are implementable in practice rather than predominantly ad-
dressing performance measures.
A key finding is that the EQW, MVW and MSRW versions of the CW MSCI AC World index are
superior to the CW MSCI AC World index in terms of average return and risk-adjusted ratios.
Statistically speaking only the MVW and MSRW (VAMOS) indices generate Sharpe ratios signifi-
cantly higher than the CW MSCI AC World index. The only alternative index not showing outper-
formance is the FW index.
Attributing the sources of outperformance to the Carhart (1997) 4-factor model, this thesis
finds significant positive Carhart alphas for the EQW, MVW and MSRW (VAMOS) indices, which
means that the common equity factors (market, size, value and momentum) do not capture the
entire outperformance.
When exploring the quality criteria of the alternatively weighted indices no restrictive features
that erode the superior performance measures or disallow usage in practice are apparent. This
thesis finds that turnover is increased but still very robust to the occurrence of transaction
costs. Except the FW index all indices are less concentrated than the CW MSCI AC World index.
Moreover, lower concentration does translate into lower liquidity and investment capacity, but
not to a degree where the alternative indices are not considered implementable. Representa-
tiveness is also intact for all alternative indices though the optimisation-based indices show a
general tendency to favour stocks from emerging regions and defensive industrial sectors.
Compared to cap-weighting transparency is reduced when optimisation-based weighting
schemes are used.
To test if the superiority of the EQW, MVW and MSRW indices is robust a number of changes to
the backtesting period, rebalancing frequency, weight constraints, covariance matrix and proxy

86

7. Conclusion
for expected returns are performed. This shows that the alternative indices are quite sensitive
to the key construction parameters. Nonetheless, all alternative indices except the FW index
consistently outperform the CW MSCI AC World index irrespective of changes in rebalancing
frequency, weight constraints, covariance matrix and expected returns. On the other hand, de-
composing the backtesting period into three disparate time periods shows some examples of
underperformance. However, the MVW and MSRW (VAMOS) indices display robust outperfor-
mance in all three time periods.
In terms of GE, this thesis finds that Jyske Invests current weighting scheme (benchmark
weight + 0.7%) ties the performance of GE very closely to the CW MSCI AC World index. Apply-
ing the simple and easily implementable equal weighting scheme reveals an insignificant out-
performance compared to the current weighting scheme. Despite statistical insignificance this
finding imply that the current construction strategy in Jyske Invest has improvement potential,
which is further supported by the findings when changing benchmark index and tracking error
constraint presented below.
Applying the superior EQW, MVW and MSRW (VAMOS) indices as benchmark for GE demon-
strates that, from an economic point of view, the performance is significantly improved com-
pared to the actual GE portfolio, though it cannot be concluded statistically. Loosening tracking
error constraint on GE also proves to be performance enhancing but only for the less efficient
CW and FW indices. Conversely, a negative performance impact is found when loosening track-
ing constraint for the superior alternative indices. Again, it is not possible to conclude that the
performance differences have statistical significance.
Taken together, a positive correlation is found between the efficiency of the alternative indices
and the performance of GE when applying alternative indices as benchmark. On the other hand,
a negative correlation between the efficiency of the alternative indices and the performance of
GE is evident when loosening tracking error constraint. These empirical findings suggest that
Jyske Invest should reappraise their current portfolio construction strategy by either (1) main-
taining the CW MSCI AC World index as benchmark and alter their benchmark + 0.7% rule to
allow for higher tracking error or (2) choosing a more efficient benchmark index and maintain
their strict tracking error constraint.
As a final remark, the findings in regards to the alternatively weighted indices are in general
accordance with previous research that also finds CW indices to be inefficient. By explicitly
evaluating the impact of using these alternative benchmark indices in combination with various
tracking error constraints on a specific portfolio, this thesis go beyond previous comparison
studies and offers additional insight to the field of portfolio construction and alternative

87

7. Conclusion
benchmark indices. As such the results are of relevance to a number of stakeholders including
CIOs, portfolio managers, individual investors, and in particular Jyske Invest.

88

Bibliography
Alexander, G.J. & Baptista, A.M. 2010, "Active portfolio management with benchmarking: A frontier based
on alpha", Journal of Banking and Finance, vol. 34, no. 9, pp. 2185.
Amenc, N., Goltz, F. & Lioui, A. 2011, "Practitioner Portfolio Construction and Performance Manangement:
Evidence from Europe", Financial Analysts Journal, vol. 67, no. 3.
Amenc, N., Goltz, F. & Le Sourd, V. 2006, Assessing the Quality of Stock Market Indices: Requirements for
Asset Allocation and Performance Measurement, EDHEC-RISK Institute.
Amenc, N., Goltz, F. & Lodh, A. 2012, "Choose Your Betas: Benchmarking Alternative Equity Index Strate-
gies", The Journal of Portfolio Management, vol. 39, no. 1.
Amenc, N., Goltz, F., Lodh, A. & Martellini, L. 2012, "Diversifying the Diversifiers and Tracking the Tracking
Error: Outperforming Cap-Weighted Indices with Limited Risk of Underperformance", Journal of In-
vestment Management, vol. 38, no. 3.
Amenc, N., Goltz, F., Martellini, L. & Retkowsky, P. 2010, Efficient Indexation: An Alternative to Cap-
Weighted Indices, EDHEC-RISK Institute.
Amenc, N., Goltz, F. & Shuyang, Y. 2012, "Seeing through the Smoke Screen of Fundamental Indexers: What
are the Issues with Alternative Equity Index Strategies?", EDHEC-RISK Institute, .
Amenc, N., Goltz, F. & Tang, L. 2011, EDHEC-Risk European Index Survey 2011, EDHEC-RISK Institute.
Amenc, N., Martellini, L., Goltz, F. & Ye, S. 2011, Improved Beta? A Comparison of Index-Weighting Schemes,
EDHEC-RISK Institute.
Arbnor, I. & Bjerke, B. 2009, Methodology for creating business knowledge, 3. edition edn, Sage Publica-
tions, Thousand Oaks, Calif.
Arnott, R.D. 2011, "Better Beta Explained: Demystifying Alternative Equity Index Strategies", The Journal of
Index Investing, , no. Summer, pp. 51-58.
Arnott, R.D. & Darnell, M. 2003, "Active versus Passive Management", The Journal of Investing, vol. 12, no.
1, pp. 31.
Arnott, R.D., Hsu, J.C. & West, J.M., 2008, The Fundamental Index: a better way to invest, John Wiley & Sons,
Hoboken, N.J.
Arnott, R.D., Hsu, J. & Moore, P. 2005, "Fundamental Indexation", Financial Analysts Journal, vol. 61, no. 2,
pp. 83.
Asness, C.S. 1997, "The Interaction of Value and Momentum Strategies", Financial Analysts Journal, vol. 53,
no. 2, pp. 29.
Bailey, J.V. 1992, "Evaluating Benchmark Quality", Financial Analysts Journal, vol. 48, no. 3, pp. 33.
Bengtsson, C. & Holst, J. 2002, "On Portfolio Selection: Improved Covariance Matrix Estimation for Swedish
Asset Returns", Lund University Working Paper, .
Bhandari, L.C. 1988, "Debt/Equity Ratio And Expected Common Stock Returns: Empiri", The Journal of Fi-
nance, vol. 43, no. 2, pp. 507.
Black, F., Jensen, M.C. & Scholes, M. 1972, "The capital asset pricing model: Some empirical tests. In Studies
in the theory of capital markets", New York: Praeger, , pp. 79-121.
Black, F. 1993, "Beta and return", Journal of Portfolio Management, vol. 20, no. 1, pp. 8.
Black, F. 1993, "Estimating Expected Return", Financial Analysts Journal, vol. 49, no. 5, pp. 36.

89


Blatt, S.L. 2004, "An In-Depth Look at the Information Ratio", Master Thesis (Worchester Polytechnic Insti-
tute), .
Bodie, Z., Kane, A. & Marcus, A.J. 2010, Essentials of investments, 8. ed., international ed. edn, McGraw-
Hill/Irwin, New York, NY.
Bolognesi, E., Torluccio, G. & Zuccheri, A. 2013, "A comparison between capitalization-weighted and equal-
ly weighted indexes in the European equity market", Journal of Asset Management, vol. 14, no. 1, pp.
14-26.
Bossert, T., Fss, R., Rindler, P. & Schneider, C. 2010, "How "informative" is the information ratio for evalu-
ating mutual fund managers?", The journal of investing, vol. 19, no. 1, pp. 67.
Breeden, D.T. 1979, "An intertemporal asset pricing model with stochastic consumption and investment
opportunities", Journal of Financial Economics, vol. 7, no. 3, pp. 265.
Britten-Jones, M. 1999, "The Sampling Error in Estimates of Mean-Variance Efficient Portfolio Weights",
The Journal of Finance, vol. 54, no. 2, pp. 655.
Brown, K. & Brown, G. 1987, "Does the composition of the market portfolio really matter", Journal of Port-
folio Management, , no. Winter, pp. 26-32.
Campbell, J.Y. 2001, "Forecasting US Equity Returns in the 21st Century", Harvard University, vol. July.
Campbell, J.Y. & Shiller, R.J. 1988, "The Dividend-Price Ratio and Expectations of Future Dividends and Dis-
count Factors", The Review of Financial Studies, vol. 1, no. 3, pp. 195.
Campbell, J.Y. & Thompson, S.B. 2008, "Predicting Excess Stock Returns Out of Sample: Can Anything Beat
the Historical Average?", The Review of Financial Studies, vol. 21, no. 4, pp. 1509.
Carhart, M.M. 1997, "On persistence in mutual fund performance", The Journal of Finance, vol. 52, no. 1, pp.
57.
Chen, H., Noronha, G. & Singal, V. 2006, "Index Changes and Losses to Index Fund Investors", Financial Ana-
lysts Journal, vol. 62, no. 4, pp. 31.
Chopra, V. & Ziemba, W. 1993, "The Effect of Errors in Means, Variances, and Covariances on Optimal Port-
folio Choice", Journal of Portfolio Management, vol. 19, no. 2, pp. 6-11.
Chow, T., Hsu, J., Kalesnik, V. & Little, B. 2011, "A survey of alternative equity index strategies", Financial
Analysts Journal, vol. 67, no. 5, pp. 37.
Christopherson, J. 2012, The Making of A Better Benchmark, Russell Investments.
Christopherson, J.A., Cario, D.R., & Ferson, W.E. 2009, Portfolio Management Measurement and Bench-
marking, McGraw-Hill, New York.
Clare, A., Motson, N. & Thomas, S. 2013, "An evaluation of alternative equity indices (Part 1: Heuristic and
optimised weighting schemes)", Cass Consulting, .
Clare, A., Motson, N. & Thomas, S. 2013, "An evaluation of alternative equity indices (Part 2: Fundamental
weighting schemes)", Cass Consulting, .
Clarke, R.G., de Silva, H. & Thorley, S. 2006, "Minimum-Variance Portfolios in the U.S. Equity Market", The
Journal of Portfolio Management, vol. 33, no. 1, pp. 10.
Clement, C. 2009, "Interpreting the Information Ratio", Cornerstone Research, .
Coignard, Y. & Choueifaty, Y. 2008, "Toward maximum diversification", The journal of portfolio manage-
ment, vol. 35, no. 1, pp. 40.

90


Credit Suisse 2013, HOLT's Discounted Cash Flow Model. Available: https://www.credit-
suisse.com/investment_banking/holt/en/index.jsp [2013, 08/24].
Cremers, M. & Petajisto, A. 2006, "How active is your fund manager? A new measure that predicts perfor-
mance", Yale School of Management working paper, .
Cremers, M., Petajisto, A. & Zitzewitz, E. 2010, "Should Benchmark Indices Have Alpha? Revisiting Perfor-
mance", .
Cuthbertson, K. & Nitzsche, D. 2010, "Mutual fund performance", Financial markets, institutions & instru-
ments, vol. 19, no. 2, pp. 95.
Dalang, R., Osinski, C. & Marty, W. 2002, "Performance of quantitative versus passive investing: A compari-
son in global markets", Journal of Performance Measurement, vol. 6, no. 2, pp. 29-44.
DeBondt, W. & Thaler, R. 1985, "Does the Stock Market Overreact?", The Journal of Finance, vol. 40, no. 3,
pp. 793.
DeMiguel, V., Garlappi, L. & Uppal, R. 2009, "Optimal Versus Naive Diversification: How Inefficient is the 1 N
Portfolio Strategy?", The Review of Financial Studies, vol. 22, no. 5, pp. 1915.
DiBartolomeo, D. 2000, "The Enhanced Index Fund as an Alternative to Indexed Equity Management",
Northfield Information Services, .
Domowitz, I., Glen, J. & Madhavan, A. 2001, "Liquidity, Volatility and Equity Trading Costs Across Countries
and Over Time", International Finance, vol. 4, no. 2, pp. 221.
Dowd, K. 2000, "Adjusting for risk", International review of economics & finance, vol. 9, no. 3, pp. 209.
Dybvig, P.H., Farnsworth, H.K. & Carpenter, J.N. 2010, "Portfolio performance and agency", The Review of
Financial Studies, vol. 23, no. 1, pp. 1.
Elton, E.J., Gruber, M.J., Brown, S.J. & Goetzmann, W.N. 2010, Modern portfolio theory and investment anal-
ysis, 8. ed., International student ed. / Edwin J. Elton ... et al. edn, Wiley, Hoboken, N.J.
Elton, E.J., Gruber, M.J., Das, S. & Hlavka, M. 1993, "Efficiency with Costly Information: A Reinterpretation of
Evidence from Managed Portfolios", The Review of Financial Studies, vol. 6, no. 1, pp. 1.
Engsted, T. 2012, "Aktiv vs. passiv forvaltning, held eller dygtighed og mling af portefljeforvalteres per-
formance", Finans Invest, vol. 3, pp. 4.
Evans, R. & Edelen, R. 2013, "Shedding light on "invisible" costs: Trading costs and mutual fund perfor-
mance", Financial analysts' journal, vol. 69, no. 1, pp. 33.
Fabozzi, F.J., Kolm, P.N., Pachamanova, D.A. & Focardi, S.M. 2007, "Robust Portfolio Optimization", Journal
of Portfolio Management, vol. 33, no. 3, pp. 40.
Fama, E.F. 1992, "The cross-section of expected stock returns", The Journal of Finance, vol. 47, no. 2, pp.
427.
Fama, E.F. & French, K.R. 2012, "Size, Value, and Momentum in International Stock Returns", Journal of
Fincancial Economics, .
Fama, E.F. & French, K.R. 2006, "The Value Premium and the CAPM", The Journal of Finance, vol. 61, no. 5,
pp. 2163.
Fama, E.F. & French, K.R. 2004, "The Capital Asset Pricing Model: Theory and Evidence", The Journal of
Economic Perspectives, vol. 18, no. 3, pp. 25.
Fama, E.F. & French, K.R. 1998, "Value versus growth", The Journal of Finance, vol. 53, no. 6, pp. 1975.

91


Fama, E.F. & French, K.R. 1993, "Common risk factors in the returns on stocks and bonds", Journal of Finan-
cial Economics, vol. 33, no. 1, pp. 3.
Fama, E.F. & French, K.R. 1988, "Dividend yields and expected stock returns", Journal of Financial Econom-
ics, vol. 22, no. 1, pp. 3.
Fama, E.F. & MacBeth, J.D. 1973, "Risk, Return, and Equilibrium: Empirical Tests", The Journal of Political
Economy, vol. 81, no. 3, pp. 607.
Favre, L. & Galeano, J.A. 2002, "Mean-Modified Value-at-Risk optimization with hedge fund", Journal of Al-
ternative Investment, Working Paper, .
Ferson, W.E., Kandel, S. & Stambaugh, R.F. 1987, "Tests of Asset Pricing with Time-Varying Expected Risk
Premiums and Market Betas", The Journal of Finance, vol. 42, no. 2, pp. 201.
Forex Technical Chartist 2013, , Limitations of Backtesting. Available:
http://www.forextechnicalchartist.com/limitations-of-back-testing/ [2013, 08/24].
Frankfurter, G.M. 1976, "The effect of "market indexes" on the ex-post performance of the Sharpe portfolio
selection model", The journal of finance, vol. 31, no. 3, pp. 949.
French, K.R. 2013, , Data library. Available:
http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html [2013, 08/22].
Friend, I. & Blume, M.E. 1973, "A new look at the capital asset pricing model", The journal of finance, vol.
28, no. 1, pp. 19.
Fuller, R.J., Han, B. & Tung, Y. 2010, "Thinking about Indices and "Passive" versus Active Management",
Journal of Portfolio Management, vol. 36, no. 4, pp. 35.
Gibbons, M.R. 1982, "Multivariate tests of financial models: A new approach", Journal of Financial Econom-
ics, vol. 10, no. 1, pp. 3.
Gibbons, M.R., Ross, S.A. & Shanken, J. 1989, "A test of the efficiency of a given portfolio", Econometrica, vol.
57, no. 5, pp. 1121.
Goltz, F., Amenc, N. & Le Sourd, V. 2009, "The performance of characteristics-based indices", European
financial management, vol. 15, no. 2, pp. 241.
Goltz, F. & Le Sourd, V. 2010, Does Finance Theory Make the Case for Capitalisation-Weighted Indexing,
EDHEC-RISK Institute.
Goltz, F. & Tang, L. 2011, Scientific Diversification in Practice: Reactions to EDHEC-RIsk Efficient Indices,
EDHEC-RISK Institute.
Goodwin, T.H. 1998, "The Information Ratio", Financial Analysts Journal, vol. 54, no. 4, pp. 34.
Goyal, A. & Wahal, S. 2008, "The Selection and Termination of Investment Management Firms by Plan Spon-
sors", The Journal of Finance, vol. 63, no. 4, pp. 1805.
Grinold, R.C. 1992, "Are Benchmark Portfolios Efficient?", Journal of Portfolio Management, vol. 19, no. 1,
pp. 34.
Grinold, R.C. & Kahn, R.N. 2000, "Active Portfolio Management", New York: McCraw-Hill, .
Gruber, M.J. 1996, "Another Puzzle: The Growth in Actively Managed Mutual Funds", The Journal of Finance,
vol. 51, no. 3, pp. 783.
Harvey, C.R. 1991, "The world price of covariance risk", The Journal of Finance, vol. 46, no. 1, pp. 111.
Harvey, C.R. & Zhou, G. 1990, "Bayesian inference in asset pricing tests", Journal of Financial Economics,
vol. 26, no. 2, pp. 221.

92


Haugen, R. & Baker, N. 1991, "The Efficient Market Inefficienct of Capitalization-weighted Stock Portfolios",
The Journal of Portfolio Management, .
Hope, T. 2008, "Do tracking error limits also limit excess return and diversification?", Navellier Applied
Research - Working Paper, .
Hsu, J. 2006, "Cap-Weighted Portfolios are Sub-Optimal Portfolios", Journal of Investment Management,
vol. 4, no. 3.
Israelsen, C. 2005, "A refinement to the Sharpe ratio and information ratio", Journal of Asset Management,
vol. 5, no. 6, pp. 423.
Israelsen, C.L. & Cogswell, G.F. 2007, "The error of tracking error", Journal of Asset Management, vol. 7, no.
6, pp. 419.
Jagannathan, R. & Ma, T. 2003, "Risk Reduction in Large Portfolios: Why Imposing the Wrong Constraints
Helps", The Journal of Finance, vol. 58, no. 4, pp. 1651.
Jagannathan, R. & Wang, Z. 1996, "The Conditional CAPM and the Cross-Section of Expected Returns", The
Journal of Finance, vol. 51, no. 1, pp. 3.
Jegadeesh, N. & Titman, S. 1993, "Returns to Buying Winners and Selling Losers: Implications for Stock Mar-
ket Efficiency", The Journal of Finance, vol. 48, no. 1, pp. 65.
Jobson, J.D. & Korkie, B.M. 1981, "Performance Hypothesis Testing with the Sharpe and Treynor Measures",
The Journal of Finance, vol. 36, no. 4, pp. 889.
Jorion, P. 2007, Value at risk : the new benchmark for managing financial risk, 3. edition edn, McGraw-Hill,
London.
Jorion, P. 2003, "Portfolio Optimization with Tracking-Error Constraints", Financial Analysts Journal, vol.
59, no. 5, pp. 70.
Jyske Invest 2013, Fact Sheet - Global Equities.
Jyske Invest 2013, Investeringsprocessen, Internal Presentation.
Jyske Invest 2013, Investor Information - Global Equities.
Jyske Invest 2013, Jyske Invest Favorit Aktier, Internal Presentation.
Jyske Invest 2013, Risk limits, Internal material.
Kamp, R.D. 2008, "Debunking 130/30 Benchmarks Fundamentally Weighted Long-Short Versions of Tradi-
tional Indices Can't Work; It's Also the Wrong Approach", Institutional Investor Journal, vol. 2008, no.
1.
Kan, R. & Zhou, G. 2007, "Optimal Portfolio Choice with Parameter Uncertainty", The Journal of Financial
and Quantitative Analysis, vol. 42, no. 3, pp. 621.
Kandel, S., McCulloch, R. & Stambaugh, R.F. 1995, "Bayesian Inference and Portfolio Efficiency", The Review
of Financial Studies, vol. 8, no. 1, pp. 1.
Kaplan, P.D. 2008, "Why fundamental indexation might - or might not - work", Financial analysts' journal,
vol. 64, no. 1, pp. 32.
Karceski, J. & Lakonishok, J. 1999, "On Portfolio Optimization: Forecasting Covariances and Choosing the
Risk Model", The Review of Financial Studies, vol. 12, no. 5, pp. 937.
Keim, D.B. & Madhavan, A. 1997, "Transactions costs and investment style: an inter-exchange analysis of
institutional equity trades", Journal of Financial Economics, vol. 46, no. 3, pp. 265.

93


Kidd, D. 2011, "The Sharpe Ratio and the Information Ratio", Investment Performance Measurement Fea-
ture Articles, CFA Institute, vol. July, no. 1.
Kothari, S.P., Shanken, J. & Sloan, R.G. 1995, "Another Look at the Cross-Section of Expected Stock Returns",
The Journal of Finance, vol. 50, no. 1, pp. 185.
Kuberek, R. & Matheos, P. 2005, "Taming the Beast - Controlling Risk Underestimation Bias in Covariance
Models", Wilshire Associates Incorporated, .
Lakonishok, J., Shleifer, A. & Vishny, R.W. 1994, "Contrarian Investment, Extrapolation, and Risk", The
Journal of Finance, vol. 49, no. 5, pp. 1541.
Lamont, O. 1998, "Earnings and Expected Returns", The Journal of Finance, vol. 53, no. 5, pp. 1563.
Le Sourd, V. 2007, Performance Measurement for Traditional Investment, EDHEC Risk and Asset Manage-
ment Research Centre.
Ledoit, O. & Wolf, M. 2008, "Robust Performance Hypothesis Testing with the Sharpe Ratio", Working Pa-
per, , no. 320.
Ledoit, O. & Wolf, M. 2004, "Honey, I Shrunk the Sample Covariance Matrix", Journal of Portfolio Manage-
ment, vol. 30, no. 4, pp. 110.
Ledoit, O. & Wolf, M. 2004, "A well-conditioned estimator for large-dimensional covariance matrices", Jour-
nal of Multivariate Analysis, vol. 88, no. 2, pp. 365.
Lettau, M. & Ludvigson, S. 2001, "Consumption, aggregate wealth, and expected stock returns", The Journal
of Finance, vol. 56, no. 3, pp. 815.
Littermann, R. & Winkelmann, K. 1998, "Estimating covariance matrices", Goldman Sachs Research, .
MacKinlay, A.C. 1995, "Multifactor models do not explain deviations from the CAPM", Journal of Financial
Economics, vol. 38, no. 1, pp. 3.
Maillard, S., Roncalli, T. & Teletche, J. 2010, "The Properties of Equally Weighted Risk Contribution Portfo-
lios", Journal of Portfolio Management, vol. 36, no. 4, pp. 60.
Malevergne, Y., Santa-Clara, P. & Sornette, D. 2009, "Professor Zipf Goes to Wall Street", National Bureau of
Economic Research, vol. Working Paper, no. 15295.
Markowitz, H.M. 1987, Mean-variance analysis in portfolio choice and capital markets, B. Blackwell, New
York.
Markowitz, H.M. 1952, "Portfolio Selection", The Journal of Finance, vol. 7, no. 1, pp. 77-91.
Martellini, L. 2012, Inefficient Benchmark in Efficient Markets, EDHEC-RISK Institute.
Martellini, L. 2008, "Toward the Design of Better Equity Benchmarks: Rehabilitating the Tangency Portfolio
from Modern Portfolio Theory", Journal of Portfolio Management, vol. 34, no. 4, pp. 34.
Merton, R.C. 1980, "On estimating the expected return on the market", Journal of Financial Economics, vol.
8, no. 4, pp. 323.
Merton, R.C. 1973, "An Intertemporal Capital Asset Pricing Model", Econometrica, vol. 41, no. 5, pp. 867.
MSCI 2013, MSCI AC World Equal Weighted Index. Available:
http://www.msci.com/products/indices/strategy/risk_premia/equal_weighted/ [2013, 08/23].
MSCI 2013, MSCI AC World Minimum Volatility Index. Available:
http://www.msci.com/products/indices/strategy/risk_premia/minimum_volatility/ [2013, 08/23].

94


MSCI 2013, MSCI AC World Value Weighted Index. Available:
http://www.msci.com/products/indices/strategy/risk_premia/value_weighted/ [2013, 08/23].
MSCI 2013, MSCI Global Equity Indices. Available: http://www.msci.com/products/indices [2013, 08/20].
Neukirch, T. 2008, "Alternative Indexing with the MSCI World Index", .
Nielsen, F. & Aylursubramanian, R. 2008, "Far From the Madding Crowd - Volatility Efficient Indices", MSCI
Barra Research, Working Paper, .
Olsen, K. 2011, "Indexed assets surge 25%", Pension and Investments, .
Peare, P. & Bartholdy, J. 2005, "Estimation of expected return", International review of financial analysis,
vol. 14, no. 4, pp. 407.
Perold, A.F. 2007, "Fundamentally Flawed Indexing", Financial Analysts Journal, vol. 63, no. 6, pp. 31.
Philips, C., Kinniry Jr., F. & Schlanger, T. 2013, The case for index-fund investing, Vanguard Research.
Philips, C., Kinniry Jr., F., Walker, D. & Thomas, C. 2011, A Review of Alternative Approaches to Equity In-
dexing, Vanguard Research.
Quantshare 2013, How to overcome limitations of trading system backtesting. Available:
http://www.quantshare.com/blog-506-how-to-overcome-limitations-of-trading-system-
backtesting [2013, 08/24].
Ranaldo, A. & Hberle, R. 2008, "Wolf in Sheep's Clothing: The Active Investment Strategies behind Index
Performance", European Financial Management, vol. 14, no. 1, pp. 55.
Rappaport, A. & Mauboussin, M. 2001, "Does It Pay to Be an Active Investor?", Extra Chapter to Expecta-
tions Investment, Harvard Business School Press, .
Roll, R. 1992, "A Mean/Variance Analysis of Tracking Error", Journal of Portfolio Management, vol. 18, no.
4, pp. 13.
Roll, R. 1977, "A critique of the asset pricing theory's tests Part I: On past and potential testability of the
theory", Journal of Financial Economics, vol. 4, no. 2, pp. 129.
Ross, S.A. 1976, "The arbitrage theory of capital asset pricing", Journal of Economic Theory, vol. 13, no. 3,
pp. 341.
Rowley, I. & Sharpe, W.F. 1993, "International Value and Growth Stock Returns", Financial Analysts Jour-
nal, vol. 49, no. 1, pp. 27.
Rudd, A. 1987, "Business Riske and Investment Risk", Investment Management Review, , pp. 19-27.
Scowcroft, A. & Sefton, J. 2001, "Do tracking errors reliably estimate portfolio risk?", Journal of Asset Man-
agement, vol. 2, no. 3, pp. 205.
Sefton, J. 2011, "Why is low-risk investing succesful?", UBS Investment Research, .
Sengupta, J.K. 2003, "Efficiency tests for mutual fund portfolios", Applied Financial Economics, vol. 13, no.
12, pp. 849.
Shanken, J. 1987, "Multivariate proxies and asset pricing relations", Journal of Financial Economics, vol. 18,
no. 1, pp. 91.
Shanken, J. 1985, "Multivariate tests of zero beta CAPM", Journal of Financial Economics, vol. 14, no. 3, pp.
327.
Sharpe, W.F. 1994, "The Sharpe Ratio", Journal of Portfolio Management, vol. 21, no. 1, pp. 49.

95


Sharpe, W.F. 1964, "A theory of market equilibrium under conditions of risk", Journal of Finance, .
Siegel, L. 2003, Benchmarks and Investment Management, The Research Foundation of the Association for
Investment Management and Research, United States of America.
Stambaugh, R.F. 1982, "On the exclusion of assets from tests of the two-parameter model", Journal of Finan-
cial Economics, vol. 10, no. 3, pp. 237.
Stein, D. 1999, Introducing Tracking Error, Parametric Portfolio Associates, Inc., Seattle, U.S.
Strongin, S., Petsch, M. & Sharenow, G. 2000, "Beating Benchmarks", The Journal of Portfolio Management,
vol. 26, no. 4, pp. 11.
Tabner, I. 2007, "Benchmark Concentration; Capitalization Weights Versus Equal Weights in the FTSE 100
Index", Multinational Finance Journal, vol. 13, no. 3/4.
Thomas, M. & Collie, B. 2012, Volatility Management, Russell Investments.
Thompson, S. & Polk, C. 2006, "Cross-sectional forecasts of the equity premium", Journal of Financial Eco-
nomics, vol. 81, no. 1, pp. 101.
Treynor, J.L. & Black, F. 1973, "How to Use Security Analysis to Improve Portfolio Selection", Journal of
Business, vol. 46, no. 1, pp. 66.
Vinod, H. & Morey, M. 1999, "A Double Sharpe Ratio", Working Paper, .
Welch, I. & Goyal, A. 2008, "A Comprehensive Look at the Empirical Performance of Equity Premium Predic-
tion", The Review of Financial Studies, vol. 21, no. 4, pp. 1455.
Wilcox, J.W. 2000, "Better Risk Management", The Journal of Portfolio Management, vol. 26, no. 4, pp. 53.
Wilshire Associates 2006, "The Wilshire Atlas GR6 Equity RIsk Model", Wilshire Associates Incorporated
website Resource Library, .
Zangari, P. 1996, "A VaR methodology for portfolios that include options", RiskMetrics Monitor, .
Zeng, L., Dash, S. & Guarino, D. 2010, "Equal Weight Indexing - Seven Years Later", S&P Indices, .
Zhou, G. 2010, "Understanding the Role of Diversification", Q Finance, .
Zhou, G. 1991, "Small sample tests of portfolio efficiency", Journal of Financial Economics, vol. 30, no. 1, pp.
165.

96

Vous aimerez peut-être aussi