Amundi Investment Strategy Collected Research Papers

Investment Strategy
Collected Research Papers

2014 Edition
Editors: Pascal BLANQU & Philippe ITHURBIDE
For professional investors only

Table of Contents
Editors forewords p. 2
- Amundi Research in a nutshell p. 4
Amundi Discussion Papers Series

- Understanding Smart Beta:
beyond diversification and low risk investing p. 9
- SRI and Performance:
Impact of ESG Criteria in Equity and Bond Management Processes p. 51
- Risk-Free Assets:
What Long-Term Normalized Return? p. 93
Amundi Working Papers

- Managing Uncertainty with DAMS.
Asset Segmentation in Response to Macroeconomic Changes p. 109
- Optimal Asset Allocation for Sovereign Wealth Funds:
Theory and Practice p. 159
- Unexpected Returns.
Methodological Considerations on Expected Returns in Uncertainty p. 177
- Low Risk Equity Investments:
Empirical Evidence, Theories, and the Amundi Experience p. 211
- Determining the Maximum Number of Uncorrelated Strategies
in a Global Portfolio p. 241
- Social Responsibility and Mean-Variance Portfolio Selection p. 273
- The Management of Retirement Savings revisited p. 307
- Incorporating Linkers in a Global Government Bond Risk Model p. 327
- Market-Implied Inflation and Growth rates adversely
affected by the Brent p. 341
- Breakeven Inflation Rates and their Puzzling Correlation Relationships p. 353
About the authors and editors p. 371
Research publications p. 381
Amundi Investment Strategy Collected Research Papers 1

Editors foreword
PASCAL BLANQU
Global Chief Investment Officer
of Amundi Group
Dear client,
The new normal, evolution, revolution these are all concepts that have
driven the financial markets since the mid-2000s. Financial crises, recession,
banking crises, debt crises, deflation, and so forth all these have been at the sides
of asset allocators and risk managers, more closely intertwined and requiring
more substantive and independent research. Active management, SMART beta,
risk analysis, diversification and entropy are all themes that one should revisit.
Proprietary research matters more than ever, and it may even be time to give it
a bigger role in assessing risks better.
Amundi has therefore chosen to expand its research teams over the past few
years and to get them implicated in its investment decisions. Our approach has
consisted in expanding both our top-down and bottom-down resources and
in involving them with the management teams and asset allocation decisions
on a regular basis. Research teams also take part in the various investment
committees and help create new investment processes. All this makes portfolio
investment returns a shared objective.
This has also helped enhance Amundis visibility worldwide. It is important to
provide guidelines that explain and publicise Amundis views both externally
and internally. The new line of research products (the cross-asset line) has made
a big contribution to improving this duty to explain. In this way, research work
promotes the groups investment strategies and themes at all times.
We have organised our research to achieve the very important objective of
remaining close to the academic world. Publishing of working papers and
financing research chairs, resulting in conferences and calls for papers, help
achieve this objective.
We have provided our research team with the resources that reflect the role that we
have given it. In addition to proximity with the management teams and clients, the
team offers a broad diversity of profiles and publications. This book shows clearly,
if there was a need, how important research is in Amundis set-up, as well as
the depth and diversity of our research capabilities. It highlights both the role
of research in the investment process and its ability to make lasting analyses that
form the basis of future action.
I wish you all a good reading.
2 Amundi Investment Strategy Collected Research Papers

Editors foreword
PHILIPPE ITHURBIDE
Global Head of Research,
Strategy and Analysis
Dear reader,
The book you have in your hands is emblematic of what Amundi research can
provide, in addition to macroeconomic research, strategy and forecasts, as well
as research on asset classes.
During the past years, Amundi has developed a diversified and decentralised
platform of independent research services supporting both investment teams and
clients. The research team (130 analysts including joint ventures) is organized as a
business line, both top down and bottom up, with different entities across the world
working collectively. This team has several characteristics:
Proximity with portfolio management is key: meetings, portfolio reviews,
sector reviews, investment strategies, internal rating, target price
Proximity with clients is also key: top down research, thematic research,
investment strategies and tailor-made research are crucial to build regular
contacts and meetings with clients, considered as partners.
A research with convictions: to have views and convictions is our DNA:
macroeconomic scenario, financial forecasts, strategies, long term returns,
country and sector allocation, top picks, both top down and bottom up are an
integral part of our duties .
A wide range of research: the team consists of economists, strategists, credit
research, equity research, socially responsible investment research, quantitative
research, real estate research, and contribute on advisory activities, partnerships
with universities, training programmes
Research as an impact player: part of investment committees, part of advisory
teams Amundi research is a key player on portfolio construction, optimization, asset
allocation, new investment processes, relative value trades, risk management processes
Diversity of publications: weekly, monthly, special focus, thematic research,
working papers, discussion papers from top down scenario, convictions and
forecasts to academic publications.
Our thematic research is highly multifaceted: prevailing issues, academic,
methodological or pedagogical papers. This book gives our readers a picture of our
great diversity. It is the first issue of a series. We, at Amundi Research, will edit a
collected papers book every year, including a selection of Working Papers and
Discussion Papers published during the year.
I wish you an enjoyable reading.

Amundi Research in a nutshell
12
Nationalities
130
Research people (including JVs),
of which 40% based in Asia
4500
Meetings with companies each year
over the world (bottom up research,
equity, credit and SRI)
Stockholm
Amsterdam Helsinki
Brussels Frankfurt
Luxembourg
London Warsaw
Montreal Paris Prague
Geneva Zurich
Yerevan Beijing
Milan
New York Madrid Seoul
Durham Tokyo
Athens
Shanghai
Casablanca
Hong Kong
20
Taipei
Abu Dhabi
Bangkok
Mumbai Brunei
Kuala
Lumpur
Spoken languages Singapore
Sydney
Santiago
International Investment Center Office dedicated to networkpartners Office dedicated to institutional

clients and third-party distributors
2000
Meetings with portfolio managers
each year (both bottom up and top
2000+ down research): sector reviews,
portfolio reviews, weekly meetings
Meetings with clients each year
in more than 40 countries

30
Number of Working Papers n
Ma 130
published in academic reviews

y-J
ISS un
150 210 e 20
N
rev euros 1-9304 14
ue-
ban
que
.fr
in the past 4 years (Applied 5
E
Eri
The
A RT
Pa
IC
LE
S
an
ac
ad
em
ic
24 ulso an
Economics, Bankers Markets

c DE
BO
DT n Pl dp
A , Fr
ro fe
Ris PPart d
ric an
:W ss io
Sa k Pr ial LO
BE
ho
em Equi n al
Z and
4411 mi
AT Jun Are
Rai
TA
OU
O
I, Vin ium
lib yao revi
the
of rium
ZH
So sing cen AN
G, ew
cial C t LA Sto M Un W inne
& Investors, Economic Policy,

CO ive
Ph
Pe omp E and ra
o rsit
ble del o
ilip ST y Lill
Ma
rse
pe
rfo an Pie
e No
rd de
rs?
(CF ille BERT rre C f
55 RM , Fra
), Vin nce
RA
ND
rm ies SIX
, Neo m
o the Fra
nce
cen , Ale , Aix Ai an Pr mo Co SKE
Do t LA xis GU -Mars ce o ma
Bus dit nv MA
Evid es Em PO YO eill
e le ine
ss ie en
IN
TE T, Aud Unive wit Sch
ool s ienc
Am en plo , Aix
-M encia
rsit
, IAE hC eY
el BE ce ye ars
eill Nante Aix-en orp
European Financial Management,

LA
NE fr om O e e Un s Sch
ive -Pr ora ie ld
FO S, rsit ool of ovence
69 CU
Un
ive
rsit Fran wne (Ai Ma
x-M nag , Cergam
te
Is
SO y of
Tun ce rshi ars
eill
em
ent
e Sch , Cen and AM
N ... is, pR
Co lamic . ..
.... and ool tre SE,
M
Ma
lek ea of Eco for Fin and
nv Eq SA lly nom anc Kedge
Ab
de en ui IH
I, IHE Bo ics) ial and Bus
ri EL tio
, CN ine
Au
rl
lba ty C Car ost RS Risk ss
na Ind & EH Ma School
Finance, Financial Analysts

Tay ie
lor SANN KHAM l C ic tha
ge, Pe ESS nagem ,
s Bus
ine
AJU LIC oun es Un
ive rfo , Fra
nce
ent
ss Sch ST, Un HI,
terp: Insi
rsit
y of rm
ool ive LER Car an
, Tay rsity SEM,
lor of Sai ENCG arts gght tha
ge ce?
s Un
ive
nt- , Cho an
rsit tienne uaib
y, Ma , Fra Dou
dC
lay
sia nce, and li Un
kka om
U
Hu iversit
par
Journal, Journal of Alternative

ma
m y,
yun El Jad
iso
Ka ida nw
bir
SA
RK
, Mo
roc
co,
ith
AR
,
Investments, Journal of Asset Ass

oci
atio In
n fran partne
ais rshi
e de p wit
na
nce
h
Management, Journal of Banking

and Finance, Journal of Business Ethics, Journal of Finance and Risk
Perspectives, Journal of Fixed Income, Journal of International Money and
Finance, Journal of Investment Strategies, Journal of Portfolio Management,
Research in International Business and Finance, Revue Franaise
dEconomie)
2nd 150
2014 Ranking of Amundi equity research Number of conferences each year
in Extel Thomson survey
Amundi Research Center
Amundi research available on our website: www.research-center.amundi.com

Top-down Monetary Policies Money Markets
ence
OK
Investment Strategies
a confer
a video,
an article,
or a
for
Quant Emerging Markets

Search
h Center
i Researc
Amund
nge
LINES its excha
HEAD modifi es
al bank s will this have?
se centr
The Chine What effect ent structu
ral
rate policy needs
to implem has major
t. This to
that China accoun appear s
aware se its capital
A first
step
long been to liberali
High Yield Sovereign Bonds

ENTS
We have, and wants econom ic policy. ADD TO
MY DOCUM
reformsuences for its
conseq
READ MORE
u
REA D
MOST
DAY
T OF THE
THOUGH
EXPERT
TALK
%
1.5014 -
11.03.2
Thought
of the
ion proje
ECB inflat the ECB publish
Day
ction s new
ed its
11.03.2
Macro-e
014 -
conomi
Expert
e: What
lation of
The accumu2013 confirm
talk
's next in
Real Estat ean mark ets? ng in 2H13
the Europ c indicato rsnews
improvi
good
since the
ead
end
s widespr c climate
conomi
.
Private Equity Real Estate Bottom-up
of summer the macro-e
014 -
Expert
Talk
6th March, ons. ment in
On the conomi c projecti improve
14.03.2 s,
Aben omic d macroe
Japa n d by unex pecte
chall enge a was filled
phen omen
ENTS
in Japan of the MY DOCUM
ywhere the victims i three ADD TO
11, ever y for
On March d sympath ake and tsunam
Asset Allocation
offered
with renewe earthqu r Abe, who of Tokyo,
dented Ministe the centre
unprece Prime ble
years ago. souls at with unbearaoutput MY DOCUM
ENTS
for their al ADD TO
prayersfeel uncomf ortable
While industri
might ic weakne ss.
econom
ENTS
MY DOCUM
ADD TO
Corporate Bonds Forecasts

Equities Fixed Income Foreign Exchange

DP-04
Understanding Smart Beta:

Beyond Diversification
and Low Risk Investing
Alessandro Russo,
Head of Equity Quantitative Research
May 2014
Smart Beta is the asset management industrys answer to some well-

known drawbacks of market capitalization-based equity indices
that typically result in high volatility and massive drawdowns,
potentially compromising the risk-return payoff of these indices.
In this study, we provide a formal description of some popular risk-
based smart beta strategies (the minimum variance portfolio, the
portfolio maximizing the diversification ratio, and the risk parity
portfolio), providing some insights in terms of composition.
Specifically we point out their diversification properties, their low
systematic risk characteristics, as well as their exposures to other
significant factors such as small caps and sector reversal.
Smart beta may become a new equity core if the investors relevant
risk measure is absolute risk, and if the liquidity of these strategies
is consistent with the amount of assets the investor holds.
Finally, we discuss whether smart beta should be considered as
passive or active strategies.

UNDERSTANDING SMART BETA:
BEYOND DIVERSIFICATION AND
LOW RISK INVESTING
Introduction
Equity markets have been very challenging during the last 25 years: international
indices often shifted from extraordinary bull market conditions to prolonged
drawdowns with high realized volatility. During the first decade of the new century
equity investors faced a major and unfavorable change in traditional risk-return
payoffs. Such a background stimulated discussions over traditional market cap
weighted index and growing evidence of their inefficiency had been pointed out.
Market cap weighted indexes rely on stocks prices only and, as markets are
not in equilibrium all the times, market value weights may suffer price noise. In
extreme circumstances where bubbles arise, since market cap weighted indices
mimic a buy and hold strategy (with no auto-corrective mean reverting mechanism
embedded), overvalued stocks as telecom before 2000 or financials before 2008
become over-weighted.
In addition, in market cap weighted index large cap are over represented, and small
cap almost neglected.
Weight of Information Technology and Telecom
40%
35%
30%
25%
20%
15%
10%
5%
29/4/11
31/8/11
28/4/95
31/8/95
30/4/96
30/8/96
30/4/97
29/8/97
30/4/98
31/8/98
30/4/99
31/8/99
28/4/00
31/8/00
30/4/01
31/8/01
30/4/02
30/8/02
30/4/03
29/8/03
29/4/05
31/8/05
28/4/06
31/8/06
30/4/07
31/8/07
30/4/08
29/8/08
30/4/09
31/8/09
30/4/10
31/8/10
30/4/12
31/8/12
30/4/04
31/8/04
30/12/11
29/12/95
31/12/96
31/12/97
31/12/98
31/12/99
29/12/00
31/12/01
31/12/02
31/12/03
30/12/05
29/12/06
31/12/07
31/12/08
31/12/09
31/12/10
31/12/12
30/12/94
31/12/04
TMT bubble RISK PARITY INDEX MKT CAP INDEX (MSCI WORLD) Source: Amundi Research

The asset management industry has been proposing several alternative ways of
building equity indices and portfolios, aiming to mitigate the inefficiency embedded
in price-based index construction rules. These alternative indices or portfolios are
known as smart beta equities and they generally belong to absolute risk-returns
strategies: away from the notion of tracking error or information ratio, they focus on
Sharpe ratio or risk adjusted return, and absolute volatility metrics. They can ideally
be grouped into two categories: fundamental-based and risk-based portfolios. In
the first family, as in the case of the RAFI index, stocks weights are proportional to
some fundamental metrics, as revenues, income, cash flows, or dividends. In the
second family, stocks may be weighted according to some risk metrics such as
volatility, correlation and contribution to volatility, or may maximize some risk-based
utility function (minimize volatility or maximize diversification). Within this category,
risk-based weighting schemes may be applied to a restricted investment universe,
according to the exposure of the stocks to some fundamental, technical, and style
measures (also known as risk factors like value, momentum, volatility, or size).
In the last few years, Amundi has deeply investigated smart beta equities,
developing its own range of solutions aiming to Sharpe ratio improvement. They
are based either on the use of instruments providing favorable asymmetry (options
and other derivatives), or they belong to the risk-based family of alternative beta
portfolios as minimum variance, optimal diversification, and risk parity.
I - Smart Beta Strategies

1. 1 The minimum variance portfolio
Amundi claims several years of experience in minimum variance equity management,

with two Europe portfolios (since 2007 and 2009 respectively) and some more
recent portfolios on world developed markets, Japan, emerging markets, Pacific
ex Japan, and other customized universes.
The efficient frontier and the minimum variance portfolio
The efficient frontier represents the set of portfolios that earn the maximum rate of
return for every given level of risk. The minimum variance portfolio is the one sitting on
the very edge of the efficient frontier. In building such a portfolio, expected returns are
not needed as the only requirement is to minimize volatility, while being fully invested.
The simple objective function is thus:
Min (w Vw) T
Such that e w = 1 T
where w is the vector of the optimal portfolio weights, V is the variance-covariance

matrix, and eT is a vector of ones.

We will show in the next section that the minimization of variance is achieved
though both the selection of low risk stocks (low systematic and low specific risk
stocks), and the selection of those stocks that are exposed to uncorrelated even
negatively correlated factors. In other words, we will prove that the minimum
variance portfolio contains both a low risk story, and a diversification story.
An enhanced process
Although we recognize the advantage of such a process being transparent and
intuitive, we are conscious of some typical drawbacks that may arise from minimum
variance portfolios: as shown in Clarke, de Silva and Thorley (2011), minimum
variance portfolios may be quite concentrated on a few low volatility stocks, may
exhibit rather high turnover, may be exposed to valuerelated factors such as
dividend yield, may be invested in small capitalization stocks (with some relevant
implications on liquidity), and may have some volatile exposure to momentum.
Similarly, Thomas and Shapiro (2007) highlight the risk of the minimum variance
portfolio being excessively concentrated on a few low risk sectors, and the lack
of control for involuntary factor exposures. They also express their preference for
tilting portfolios toward some successful stock ranking criteria.
These are all relevant issues in portfolio construction. In order to take them into
account, the best practice of the industry is to implement an enhanced portfolio
construction process, employing filters to the investment universe, applying
optimisation constraints, and allowing discretionary interventions by the fund
managers. We will briefly describe Amundis investment process in the annex.
However, in the next section of this study, except where it is explicitly mentioned,
we will ignore any aspect that is beyond the pure smart beta portfolio construction,
as we want to focus on the impact that the unconstrained minimum variance
process has on portfolio composition.
1.2 The portfolio maximizing the diversification ratio
Several reasonable diversification measures exist, and maximizing each of them

would lead each time to a different portfolio. One of the most popular measures of
diversification is the so called diversification ratio, which is the ratio () of average
stocks volatility on portfolio volatility, as it was originally introduced by Choueifaty
and Coignard in 2008.
w
=
, ww
Since correlations among any pairs of assets are lower than one, the denominator
is lower than the numerator and the ratio is always higher than one. Maximizing this

ratio is thus equivalent to minimizing the average correlation across all the stocks
in the portfolio.
Better diversification and lower correlations explain why the risk of the portfolio
maximizing the diversification ratio is always lower than the risk of a standard
market index. In addition to that, the optimisation contains a pseudo-minimization
of the denominator that is satisfied via the selection of low systematic risk stocks.
On the other hand, at the numerator, the optimisation results in the selection of high
specific risk stocks since they increase average volatility, while having little impact
on the denominator: specific risk doesnt matter at the denominator as it is easily
diversified away.
As a result, the portfolio maximizing the diversification ratio may show an average
total volatility that is not statistically different from that of a standard market index,
but will necessarily result in below average systematic risk stocks (the denominator
effect), and above average specific risk stocks (the numerator effect).
Very often the portfolio maximizing the diversification ratio is presented as a

portfolio belonging to the efficient frontier, or even being the tangency portfolio
(the portfolio maximizing the Sharpe ratio). Actually, this portfolio corresponds to
the maximum Sharpe ratio portfolio only in the hypothesis that expected returns
are strictly proportional to their total volatility. If this hypothesis does not hold, still
being the portfolio that maximizes our specific definition of diversification (), such
a portfolio is below the efficient frontier and does not correspond to the tangency
portfolio. Neither can we state that the portfolio maximizing the diversification
ratio corresponds to the market portfolio, as we would assume that such a market
portfolio is completely insensitive to expected returns.
Maximizing diversification is an intuitive and transparent process, but as for the

minimum variance process it may contain the typical drawbacks of optimisation-
based portfolios, such as overconcentration, lack of liquidity, (involuntary) style
exposures, turnover, low fundamental quality.
For these reasons, when dealing with diversification-based strategies, we believe

that an enhanced process similar to the minimum variance one may be sound.
However from now on, we will ignore any aspect that is beyond the pure smart
beta portfolio construction, as we want to focus on the impact that the risk-based
process alone has on portfolio composition.
1.3 The risk parity portfolio
Risk parity means that each asset (asset class, equity sector, single stock) has an
equal contribution to the total risk of the portfolio.

In order to come out with full risk parity, the following relationship must hold:
= = =
Where RC i is the risk contribution of the i th asset, and MC i is its marginal
contribution to risk, defined as follows
In other words, the risk contribution should be the same for any asset or asset
class and the weight of each asset or asset class should be proportional to the
inverse of its marginal contribution to risk:
1
~
Actually, marginal contributions to risk are both function of volatilities and

correlations of any asset with the rest of the portfolio, with correlations depending
on portfolio composition itself. In other words, weights are the unknowns and
should be proportional to the inverse of marginal contributions that depend
on weights themselves: the problem is clearly recursive, and the solution is
endogenous.
As Maillard, Roncalli, and Teiletche (2009) have pointed out, full risk parity cannot
be obtained in a closed formula unless some unrealistic hypotheses (such as
equal correlation among all the assets in the investment universe) are made, and
may not be achieved through optimisation either, if the number of assets involved
is very high, and correlations are very heterogeneous. For this reason the asset
management industry proposes several proxies.
By far, the easiest but probably the most nave proxy for risk parity is the equally
weighted portfolio: no estimation is made on volatility and correlation and assets
are equally weighted. It would correspond to the true risk parity portfolio assuming
that all stocks have the same volatility, and all the pairs of stocks have identical
correlation. With no risk estimation, the equally weighted scheme only removes
the risk concentration driven by market capitalization: since sectors, countries,
or whatever groups of stocks (based on some style criteria, for instance) are not
equally populated, equally weighting stocks would result in higher concentration
of risk over those sectors, countries, or styles that are over-represented.
Another proxy for risk parity would be the risk weighted scheme where stocks are
weighted proportionally to the inverse of their volatility. This weighting scheme
removes the risk concentration driven by market cap and adjusts for volatility,
but the resulting portfolio is a true risk parity solution only in the hypothesis of

equal correlation across all pairs of assets. However, when correlations are quite
homogenous, although every stock has a similar risk contribution, we would still
have concentration over those families of stocks that are overrepresented.
In order to smooth the risk concentration over such an overrepresented group
of stocks, a two-step risk weighting scheme may be used: risk-weighted sector
baskets should be created first, and the overall portfolio should be created
afterwards by weighting those baskets for the inverse of their volatility.
We can check for the accuracy of each of these solutions computing the
percentage contribution (PC i ), for any basket of stocks:
In a test over the constituents of the MSCI Emu, we have built risk parity portfolios
according to the three methodologies discussed above, at any quarter-end from
2003 to 2012. In the chart below, we show the average contribution of any GICS
sector, computed over these quarterly observations.
Percentage Risk Contributions by Sectors
TELECOM
INDUSTRIALS ENERGY
20%
10%
UTILITIES
FINANCIALS
0%
INF. TECH
CONS. DISCRET
HEALTH CARE MATERIALS
CONS. STAPLES
MSCI EMU Eq. Weighted Risk Weighted (1 Step) Risk Weighted (2 Steps)
Source: Amundi Research

The MSCI index is extremely concentrated on Financial stocks (black line).
Removing the market cap bias we reduce risk concentration on Financials, but
we introduce the same problem on some other over-represented sectors such
as Consumer Discretionary and Industrials (dotted black line). After correcting for
volatility at stock level only, risk distribution only marginally improves (red line).
For a better solution two steps are needed: risk parity should first be achieved
within each sector, and then at a portfolio level (light brown line). In any case, this
two-step risk weighting scheme still generates some deviations from a 10% target
contribution to total risk.
In order to further improve the precision of our risk parity, we have tested an additional
method where correlations are taken into account at least across the sectors, in the
second step. We account for correlations using the marginal contribution to total
risk of any risk parity sector. We observe marginal contributions of any sector, in the
most neutral portfolio composition: the equally weighted composition. Equal weights
as a starting point have the advantage of not being too far from the (still unknown)
optimal solution. In this way the marginal contributions that we use for target weight
calculation are a very good proxy for the marginal contribution that we will observe
after weight calculation, thus ensuring a well-balanced risk contribution. We then
weight sector baskets proportionally to the inverse of these measures.
Percentage Risk Contributions by Sectors
TELECOM
INDUSTRIALS ENERGY
UTILITIES FINANCIALS
CONS. DISCRET INF. TECH
HEALTH CARE MATERIALS
CONS. STAPLES
Risk Weighted (2 Steps) Risk Weighted (2 Steps with Correlations) 10% Target

Taking into account correlations at least across sectors reduces the dispersion
of risk contributions, and the deviations from a 10% target become negligible.
In any case, whatever the precision of our risk parity (with the exception of the
equally weighted approximation), in order to contribute the same to portfolio risk,
high risk stocks must have lower weight relative to stocks with lower risk. This is
the main reason why risk parity portfolios are generally exposed to the low risk
anomaly, as we will show hereafter.
In addition, risk parity strategies have an embedded mean reverting mechanism,

as stocks and sectors with positive performance and increasing weights will be
reduced in order to be aligned back to a risk parity weight. Reversal at a sector
level is a successful risk control strategy, and a two-step approach accounts for
it more effectively.
II - Low Risk Anomaly and Diversification

2.1 The Low Risk Anomaly
Financial theory assumes that higher risk is remunerated on average by higher

returns. However, the outperformance of low volatility stocks during the last 50
years has been among the most puzzling anomalies in equity markets. At the
same time, low risk investing has recently gained a remarkable interest, due to its
documented performance coupled with the unprecedented volatility experienced
during the last two global financial crises.
In our previous work, we showed how researchers have been documenting

such anomaly since the early nineties: Fama and French (1992) show a rather
negative relationship between risk and returns, and Baker and Haugen (1991)
find significant reduction in volatility with no reduction in returns, for US minimum
variance portfolios. We find that most of the relevant empirical studies focus on
systematic risk; some of them state that the low risk anomaly holds regardless of
which dimension of risk systematic or total is used for stock selection. Only few
exceptions instead (Ang et al, 2006) rather refer to idiosyncratic volatility.
In this section we show with a practical example that all of the three smart beta
strategies discussed so far are exposed to the low risk anomaly.
We build three portfolios (in Barra One, at the model date of 12/31/2012),
restricting the investment universe to the constituents of the MSCI World Index.
We impose that no stock can exceed a 5% weight. We then group stocks into
three equally populated families, according to their risk: low risk, average risk
and high risk stocks. Finally we observe the percentage allocated to each family
of stocks, for each of the three portfolios as well as for the MSCI World.

Weight Distribution: Total Risk
100%
90%
80%
70%
60% High Risk

50%
Average Risk
40%
Low Risk
30%
20%
10%
0%
Diversification Minvar Risk Parity Msci
In the chart above we see that while the minimum variance portfolio is exclusively
invested in stocks with below average risk, the risk parity portfolio has only a slight tilt
toward low risk stocks, compared to the standard index. The portfolio maximizing the
diversification ratio is apparently well balanced in absolute terms toward low or high risk
stocks, while it clearly underweights average risk stocks. As a conclusion, using total
risk as a grouping criterion, we see a clear and intuitive exposure to low risk anomaly
for the minimum variance, a slight but intuitive exposure for the risk parity portfolio,
and no exposure at all but rather a barbell allocation for the portfolio maximizing the
diversification ratio.
However we traditionally distinguish two components of risk: the systematic
component (or common factor component according to Barra One terminology) and
the specific component. This distinction is needed because, as we have documented,
the systematic risk is the most relevant measure when addressing the low risk anomaly
and, if we restrict our analysis to this component only, the picture changes.
Weight Distribution: Common Factor (Systematic) Risk

100%
90%
80%
70%
60% High Common Factor Risk

50%
Average Common Factor Risk
40%
Low Common Factor Risk
30%
20%
10%
0%
The portfolio maximizing the diversification ratio now exhibits a much more
significant percentage invested in low risk stocks. The low risk feature of the risk

parity portfolio is somehow more significant as well, while unsurprisingly, the
minimum variance portfolio is still exclusively invested in low risk stocks.
Interestingly we observe some different effects while investigating specific risk.
Weight Distribution: Specific Risk

100%
90%
80%
70%
60%
High Specific Risk
50%
Average Specific Risk
40%
Low Specific Risk
30%
20%
10%
0%
This time, the portfolio maximizing the diversification ratio exhibits almost 50% of
the weight invested in high specific risk stocks, and only marginal weight in low
specific risk stocks.
We have explained in section 1 that the maximization of the diversification ratio
contains a pseudo-minimization of the denominator that is satisfied via the
selection of low systematic risk stocks. On the other hand, at the numerator, the
optimisation results in the selection of high specific risk stocks since the latter
increase the numerator, while having little impact on the denominator: specific risk
doesnt matter at the denominator as it is easily diversified away.
As a result, the portfolio maximizing the diversification ratio may show an average
total volatility that is not statistically different from that of a standard market index,
but will necessarily result in below average systematic risk stocks (the denominator
effect), and above average specific risk stocks (the numerator effect).
2.2 Diversification
Diversification according to the risk model

We now move to investigate how the three investment processes behave in terms
of diversification. In addition to the portfolio maximizing the diversification ratio, we
expect the minimum variance optimisation to exploit uncorrelated stocks as well
as low risk stocks; in the same way, we have seen that the two-step risk parity
process also somewhat relies on low correlations (at least across sectors) and
volatilities. In other words we are supposed to find some diversification evidence in
the minimum variance and in the risk parity portfolios as well.
In order to check for diversification we compute the diversification ratio first.

Diversification Ratio
3
2,8
2,6
2,4
2,2 Diversification
2
Minvar
1,8
1,6 Risk Parity

1,4
Msci
1,2
1
Diversification Ratio Total Risk Diversification Ratio Common
Factor Risk Source: Amundi Research
Unsurprisingly, we find that the minimum variance portfolio is well diversified indeed,
while the risk parity portfolio also provides some diversification improvement,
relative to the standard market index.
We compute the same measure excluding the specific risk component both at
the numerator and at the denominator and, while finding the same hierarchy,
we confirm that the specific risk inflates diversification measures, and better
explains why a process maximizing diversification is tilted toward high specific
risk stocks.
We than compute the average correlation of stocks, according to the CBOE
methodology:
N n
- wi w jm im j lij < - wi2m i2

l avg i , j 1 N <1 N
i 1
2-- wi w jm im j
i 1 j i
Average Correlation
70%
60%
50%
Diversification
40%
Minvar
30%
Risk Parity
20%
Msci
10%
0%
Average Correlation Total Risk Average Correlation Common Factors

Average correlations do not change the picture: the portfolio maximizing the
diversification ratio is the best diversified across risk factors, but once again we
find some evidence of diversification in the minimum variance and in the risk parity
portfolios. Again, the specific risk component reduces measured correlation.
Capital diversification
Risk-based measures of diversification like diversification ratio and average
correlation show that smart beta are better diversified than a standard index,
while within smart beta, optimized portfolios are better diversified than risk parity
portfolios. This is because while the optimisation mainly selects a limited number of
highly uncorrelated stocks, a risk weighting scheme still invests in all the stocks in the
investment universe, regardless of their true diversifying properties. Optimisation-
based portfolios are thus quite concentrated on a few low-risk, low-correlation
stocks and investors are comfortable with such a portfolio when the confidence in
the risk model is very high. In contrast, investors may be concerned by the effect
of using a risk model that is not properly specified, where a relevant risk factor is
neglected, or where the optimisation relies on incorrectly estimated correlations.
In these cases, investors may correctly believe that the ultimate insurance against
unexpected risks is capital diversification.
In order to address the capital diversification of the three portfolios, we employ the
entropy measure on the weights of their constituents. The entropy of a portfolio may be
read as the equivalent number of assets held, if those assets were equally weighted.
As shown in the chart below, optimisation-based portfolios that typically invest in
70-80 assets have an entropy measure of roughly 40-45, meaning that they have
a capital diversification equivalent to an equally weighted portfolio of 40-45 assets.
The risk parity portfolio is obviously much better diversified in terms of capital
allocation, with an entropy measure of roughly 1300, out of a maximum possible
of 1600 (the number of investment universe constituents, if equally weighted). Also
the risk parity portfolio has almost double the entropy of the standard market index,
even investing in the same number of stocks.
Entropy
10000
1000
LOG Scale
100
1 276 740
10
47
39
1

We believe that risk parity is more suitable for investors that are not completely
confident about the estimation of the full variance covariance matrix, thus favoring
capital diversification over risk-model diversification.
However, in order to improve capital diversification of the optimisation-based
portfolios, some more prudent constraints may be used on the maximum weight of
any holdings (compared to the 5% that we use in this example).
III - Smart Beta in asset allocation

Since their introduction into the industries, many questions have been raised about
the use of smart beta in asset allocation: investors wonder about the implication of
introducing smart beta equities in traditional equity-bond allocation.
Another point of growing interest is whether smart beta should replace traditional
equity as an alternative equity core, or whether they should constitute a new
satellite. A similar issue is whether smart beta equities should be used to improve
active returns relative to a traditional benchmark, or whether they should rather
be used by investors seeking absolute returns, and thus replace the traditional
benchmark.
Crucial to all these questions is the detection of the drivers behind the risk-return
profiles of smart beta equities, as investors must be comfortable with them before
introducing them into a strategic asset allocation (will these drivers keep on
delivering low risk outperformance in the long run?).
Also, we need to investigate if smart beta equities exhibit some evidence of different
and hopefully more favorable correlations with bonds.
Finally, investors should monitor liquidity as any equity strategy deviating from
free float adjusted market cap is by definition less liquid than the latter. Is
liquidity enough to allow for such a radical switch from traditional equities to
smart beta?
3.1 Performance drivers

Smart beta strategies have proven to be more efficient than market cap indices from
a risk-return standpoint. Their returns over the last decade are at least equal to and
very often higher than those of standard indices, while volatility and drawdowns are
systematically lower.
We try to explain the sources of these favorable deviations from market cap indices,
for some well-known global equity smart beta benchmarks, as well as for some
Amundi smart investment processes.
As for the diversification family, we have analyzed two well-known indices the
FTSE Tobam Maximum Diversification, and the FTSE Edhec Risk Efficient

together with an Amundi process aiming to enhance diversification by minimizing
average correlations (it should be noted that the Amundi process is applied to a
restricted list of high dividend stocks in the global developed markets). As for the
risk parity family, we investigate the MSCI World Risk Weighted together with an
Amundi risk parity process, as explained in section 1 (the biggest difference with
MSCI being the two-step sector-company approach for the Amundi process).
In the minimum variance family we study the MSCI Minimum Volatility, and two
Amundi processes: the first is a minimum variance with some liquidity constraints,
while the second is a very similar process applied to a restricted list of high
quality stocks according to the Piotroski score.
In the Table below we show the correlation matrix of active returns relative to the
corresponding benchmark for each strategy.
FTSE FTSE MSCI Amundi MSCI Amundi

Amundi Amundi
CORRELATION EDHEC- TOBAM World Risk World MinVar -
Diversif. MinVar
R.E. M.D. RW Parity MinVol Piot
FTSE EDHEC-R.E. 14% 29% 61% 35% 20% 14% 10%
Amundi Diversification 14% 78% 62% 61% 83% 84% 86%
FTSE TOBAM M.D. 29% 78% 66% 64% 82% 85% 84%
MSCI World RW 61% 62% 66% 79% 65% 59% 54%
Amundi Risk Parity 35% 61% 64% 79% 48% 48% 48%
MSCI World MinVol 20% 83% 82% 65% 48% 91.4% 91.2%
Amundi MinVar 14% 84% 85% 59% 48% 91.4% 95.4%
Amundi MinVar - Piot 10% 86% 84% 54% 48% 91.2% 95.4%
We can easily recognize the three family blocs with the FTSE Edhec Risk Efficient
somehow being an outlier among its family as well as among the full sample of
strategies. This is due to the specific constraints that affect holdings on each stock:
any constituent cannot be weighted less than one-third of an equal weighting
schemes, neither more than 3 times such a quantity. Though these constraints are
sound, they make this index half way between a market weighted and an equally
weighted portfolio, and not that close to an unconstrained portfolio maximizing
diversification. Not surprisingly, this index is well correlated to the MSCI World Risk
Weighted index that applies similar constraints. Interestingly we notice that the
diversification bloc is highly correlated with the minimum variance block, while the
risk parity block stays somewhere in the middle.
In any case, the correlation matrix suggests that there is some common behavior
behind the active returns of each strategy and this intuition is confirmed by the principal
component analysis (always on active returns), summarized in the chart below.

Explained Variance
100% 4.0%
95% 3.5%
90%
3.0%
85%
80% 2.5%
75% 2.0%
70% 1.5%
65%
1.0%
60%
0.5%
55%
50% 0.0%
PC 1 PC 2 PC 3 PC 4 PC 5 PC 6 PC 7 PC 8
Cumulative Percentage Variance (LS) Variance of Each Component Source: Amundi Research
The common behavior is confirmed by the 85% variance explained by the first factor,
and by the 91% variance explained by the first two factors alone. One can argue
that we have such a high percentage explained as we use redundant information,
since many strategies in our analysis (almost all the strategies within each family
bloc) are very similar to each other, thus resulting in overlapping behaviors. For this
reason we run a simplified PCA on a restricted sample of one strategy per family
(FTSE Tobam Maximum Diversification, Amundi Risk Parity, Amundi Minimum
Variance).
Explained Variance
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
PC 1 PC 2 PC 3
Cumulative Variance By Strategy Cumulative Variance By PC Source: Amundi Research
With no common factor in place (that is, with perfectly uncorrelated strategies) any
principal component would coincide with a stand-alone strategy, while we can see
that the first component of our simplified sample explains as much variance as the
two most volatile strategies. There is definitely some common behavior underlying
the active returns of smart beta strategies and the true challenge is to identify such
a common performance drivers.
In order to detect those drivers, we regress the first two principal components of
the complete sample of eight strategies, over the explanatory variables listed below.

Variables Description Note
Equity Market The standard market index
Long-short of equally weighted
Sector Reversal basket of GICS sectors versus the
MSCI World Residual returns of linear regression
Momentum MSCI World Momentum on MSCI World
Small Cap MSCI World Equally Weighted
Value MSCI World Value
Residual returns of multiple linear
Dividend Msci World High Dividend regression on MSCI World and the
Value factor
Beta-neutral long-short of low Residual returns of multiple linear
Low (Systematic) Risk
systematic risk stocks (L) versus high regression over all the other
Anomaly
systematic risk stocks (S) explanatory variables
All variables are adjusted for market beta in order to avoid double counting for
the market beta effect and to limit multicollinearity. The dividend yield factor
has been simultaneously regressed over the market index and the value factor,
to delete positive correlation between value and dividend. As for the low risk
anomaly, we have built a long basket of stocks belonging to the lowest quintile
according to systematic risk (cf. common factor risk, estimated by Barra One),
and a short basket with stocks belonging to the highest quintile; baskets are then
weighted inversely proportional to their ex ante Beta in order make the long-short
beta-neutral, and residual (ex-post) market exposures as well as any involuntary
exposure to other factors are canceled out via a multiple regression over all
explanatory variables.
Cumulative Factors Performance
2 2,5
1,8 2,2
1,6 1,9
1,4 1,6
1,2 1,3
1 1
0,8 0,7
01/06/03 01/04/04 01/02/05 01/12/05 01/10/06 01/08/07 01/06/08 01/04/09 01/02/10 01/12/10 01/10/11 01/08/12 01/06/13
Low Syst. Risk Sector Reversal Moment. Small Cap Value Dividend Msci World
The sample period has been characterized by strong equity markets despite
the massive drawdown of 2008, a strong low risk anomaly effect (except during
the rebound of 2009), positive momentum, positive sector reversal (the latter is
interesting as it exhibits very low volatility), and small caps. Value and dividend yield
have been flat.

Low Syst. Sector
CORRELATIONS Mkt Beta Moment. Small Cap Value Dividend
Risk Reversal
Mkt Beta 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
Low Syst. Risk 0.0% 0.0% 0.0% 0.0% 0.0% 0.0%
Sector Reversal 0.0% 0.0% 39.5% -1.4% -18.0% 32.9%
Moment. 0.0% 0.0% 39.5% -5.1% -36.8% 8.8%
Small Cap 0.0% 0.0% -1.4% -5.1% 9.1% -19.0%
Value 0.0% 0.0% -18.0% -36.8% 9.1% 0.0%
Dividend 0.0% 0.0% 32.9% 8.8% -19.0% 0.0%
Correlation of all the explanatory variables with the market factor and the low
volatility factor (as well as between value and dividend yield) are equal to zero by
construction, while other correlations are sufficiently low to exclude muticollinearity
problems.
As mentioned, we regress the first two principal components of smart beta
strategies, over the full set of explanatory variables, and we analyze their exposures
and their explained variance.
Exposures Variance Explained (Log Scale)

MKT (negative) MKT (negative)
2,0
50,0%
1,5 12,5%
Dividend LMHbeta Dividend 3,1% LMHbeta
1,0 0,8%
0,2%
0,5 0,0%
0,0%
- 0,0%
Value Sector Reversal Value Sector Reversal
Small Cap Momentum Small Cap Momentum
PC 1 PC 2 PC 1 PC 2 Source: Amundi Research
The first principal component has a very significant negative market beta, and
significant exposures to all the other explanatory variables with the exception of
momentum (positive but not significant). The variance explained is 60% for market
beta, 15% for the low risk anomaly, 5% for the dividend factor, and about 1% for
value, small caps and sector reversal.
The second component has small cap and sector reversal exposure, both of them
significant, but with small caps only explaining a non-negligible portion of variance (5%).
Overall we would argue that the active performance of smart beta strategies is
finally due to low market beta, low risk anomaly, small caps, and sector reversal.
However we recognize that each strategy may have different exposure to these

explanatory variables, and we need to estimate them separately. We thus run seven
multiple linear regressions, and once regression parameters are estimated, we run
performance attribution in order to quantify the impact that any of these drivers have on
the cumulative active return of the eight strategies. We show cumulative effect over the
period from the end of June 2003 to the end of December 2013 in the following Chart.
Cumulative Active Returns vs. Standard Index: 2003 - 2013

250%
200%
150%
100%
50%
0%
-50%
-100%
-150%
FTSE EDHEC-R.E. Amundi Diversif. FTSE TOBAM MD MSCI World RW Amundi Risk Parity MSCI World MinVol Amundi MinVar Amundi MinVar -
Piot
Unexpl. + Interact. MKT Low Syst. Risk Sector Reversal Momentum Small Cap value Dividend Total
The following chart instead shows the contribution to ex-post tracking error
(computed as the percentage explained variance times the realized tracking error)
for each of them.
Explained Variability of Active Returns by Components

12.00%
10.00%
8.00%
6.00%
4.00%
2.00%
0.00%
FTSE EDHEC-R.E. Amundi Diversif. FTSE TOBAM MD MSCI World RW Amundi Risk Parity MSCI World MinVol Amundi MinVar Amundi MinVar -
Piot
Unexpl. + Interact. MKT Low Syst. Risk Sector Reversal Momentum Small Cap value Dividend
With the exception of the FTSE Edhec Risk Efficient, the regression model explains
80% to 90% of the variance of active returns and its F-test is significant for all the
strategies investigated. The model is thus overall well specified.
Intuitively the low market beta has a negative effect during upward markets, and
it explains a big percentage of the variance of active returns. Interestingly, those
strategies exhibiting the lowest market beta offset much of this negative effect with
a positive contribution by the low risk anomaly.

All strategies benefit from sector reversal, with the Amundi Risk Parity benefiting
the most. In this case, the variance explained is particularly high as the construction
process of this portfolio is based on a systematic sector rebalancing (cf. section 1
about the two-step company-sector methodology).
Small cap effect explains both performance and variance for diversification-based
portfolios and risk parity portfolios, while it is basically absent on minimum variance
portfolios, because small caps bring some additional volatility, and because
Amundi portfolios apply some liquidity filters as well.
The dividend factor explains some variance, but has little impact on returns, as it
is quite flat over the period.
In the same way, the value factor is basically absent in the performance chart and
is also negligible in terms of explained variability.
Unexplained component of returns is positive in the case of Amundi minimum
variance, and it is even higher in the portfolio with a quality (Piotroski) filter: we
can argue that there is some more room for investigation about minimum variance
drivers, especially when the construction process is less constrained than the
MSCI World Minimum Volatility. The quality filter delivers additional value.
Finally, we confirm the outlier behavior for the FTSE Edhec Risk Efficient Index.
We have said about the constraints applied in its construction process and,
unsurprisingly, its deviations from a market weighted index are quite low in terms
of cumulative active returns, and realized tracking error. The only visible source of
active return is the small cap exposure.
3.2 Smart Beta for active or absolute returns, a new equity core?
The choice whether smart beta should be used in an absolute or in an active risk-
return framework, depends on the utility function of the investor (or the mandate of
the fund manager in the case of delegated asset management), and the governance
of the investment process.
While a fund manager with the objective of maximizing information ratio -under
a limited tracking error constraint- may find it difficult to massively move toward
smart beta equities, an institutional investor aiming to maximize wealth under
some absolute risk constraint, could use smart beta equity to make up the bulk
(or the new equity core) of its equity investments. In an investment process that
is based on top-down strategic asset allocation by the investment board, and
equity allocation by the equity department thereafter, if the board allocates wealth
based on traditional benchmarks allowing limited tracking error deviations, the
equity department is likely to exploit the enhanced risk-return profile of smart beta
equities only in some satellites of the global equity allocation, since smart beta
equities bring high tracking error relative to a standard market index. On the other

hand, if the investment board accepts to change its strategic benchmark into a
smart beta benchmark, smart beta equities can effectively become the new equity
core. However such a radical choice implies that several conditions are met.
First, the investor must be confident that the performance drivers identified above
are going to deliver positive performance, just as we have seen in the recent past.
The lower this confidence, the lower the likelihood that the investor will be willing
to allocate a relevant portion of its equity to low beta stocks. As we have seen in
performance attribution, low beta itself penalizes profitability as long as it is not
offset by some other positive effects. With no positive contribution from low risk
anomaly, size, sector reversal and so on, investing in low beta stocks is not efficient
in a classical Markowitz framework either (cf. next section).
Second, the investor should be sure that smart beta strategies provide sufficient
liquidity. If the market can absorb the volumes needed for monthly or quarterly
rebalancing, but cannot quickly absorb the program trades resulting from strategic
asset allocation or tactical asset allocation decisions, the investor should rather
prefer to allocate smart beta equities to the satellite bucket of the portfolio. On the
other hand, if liquidity is not an issue, investors may switch their core equity allocation
to smart beta, but should be ready to change the equity benchmark, since smart
beta equities bring high tracking error relative to a standard market index.
Lets consider the case of a big sovereign investor that is going to implement a big
change in its strategic asset allocation, buying (selling) a relevant amount of global
equity. Lets assume the investor wants to complete the program trade in 10 days,
using up to 20% of the daily average volumes, each day (the daily average volumes
are estimated over the last three months as of end of December 2013). We test
three program trades of USD 10, 25 and 50 billion respectively, times four equity
index hypotheses: the MSCI World, the MSCI World Minimum Volatility, the MSCI
World Risk Weighted, and a risk weighted allocation of the latter two indices (43%
MSCI World Risk Weighted and 57% MSCI World Minimum Volatility, according to
a long-term estimation of volatility).
Program Trade - MSCI World Program Trade - MSCI Risk Weighted
100% 100%
80% 80%
60% 60%
40% 40%
20% 20%
0% 0%
d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20 d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20
10 Bln Usd Prog. Trade 25 Bln Usd Prog. Trade 50 Bln Usd Prog. Trade

Program Trade - MSCI Minimum Vol Program Trade - RW(43)MV(57)
100% 100%
80% 80%
60% 60%
40% 40%
20% 20%
0% 0%
d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20 d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20
10 Bln Usd Prog. Trade 25 Bln Usd Prog. Trade 50 Bln Usd Prog. Trade
The most liquid index is unsurprisingly the maker weighted index: a huge program
trade of 50 billion may be completed in five days. The Minimum Volatility index is
the least liquid, not really because it is more exposed to small caps, but rather
because it is concentrated over a lower number of stocks (248), than the Risk
Weighted Index (1600).
As for smart beta in general, only a USD 10 billion program trade allows a relevant,
though not exhaustive, completion after 10 days. In detail, this is the percentage
completion after 10 days.
Percentage Completion of Program Trade after 10 days
100%
95%
90%
85%
80%
75%
70%
65%
60%
55%
50%
Risk Weighted 43%R.W. Minimum Risk Weighted 43%R.W. Minimum Risk Weighted 43%R.W. Minimum
57%M.V. Volatility 57%M.V. Volatility 57%M.V. Volatility
10 BLN USD 25 BLN USD 50 BLN USD
In order to effectively complete the program trades in 10 days, the investor cannot
hold 100% of equity in smart beta and should dilute his holding with traditional
and more liquid equity investments. In the table below, we show the maximum
allocation in smart beta that the investor can afford, in order to complete each
program trade in 10 days.

Maximum Allocation in Smart Beta
100%
95%
90%
85%
80%
75%
70%
65%
60%
55%
50%
Smart Beta weight MSCI World weight Source: Amundi Research
If the investor is not likely to incur program trades bigger than USD 10 billion, risk
weighted, minimum volatility (to a lesser extent), and a mix of the two indices may
all become a new equity core, as the investor can hold up to 100% of total equity
in smart beta. For higher sizes of program trades, smart beta allocation should be
kept residual with respect to market weighted equity, thus smart beta would be
more suited to being a satellite bucket of the portfolio.
However, if comfortable with the USD 10 billion hypothesis, the investor that
goes for smart beta as a new equity core, should seriously consider changing its
strategic benchmark.
In the next chart, we show the tracking error relative to the MSCI World Index
of all equity allocations from the example above, and the tracking error of an
allocation with 40% in the smart beta above and 60% in global bonds, relative
to a classic balanced benchmark (40% MSCI World Index, and 60% JPM Global
Bond Index).
Tracking Error Relative to Standard BenchParks

5.0%
4.5%
4.0%
3.5%
3.0%
2.5%
2.0%
1.5%
1.0%
0.5%
0%
tracking error vs CW Equity tracking error vs 60-40 Source: Amundi Research
The lower the impact of liquidity issues, the easier the move toward smart beta
equities as a new equity core. But massive investments in smart beta equities

bring high relative risk and it is very unlikely that the investment committee and
fund managers are comfortable with tracking error as high as 2% relative to a
traditional bond-equity composite benchmark, and 5% relative to a traditional
equity benchmark. As a consequence, such a big move toward smart beta
equities increases the likelihood that these traditional benchmarks are replaced by
opportune and maybe customized smart beta indices.
3.3 Bond-Equity Allocation
As historical returns may suggest, as far as the risk is lower while returns are
higher, the risk-return profile of some traditional bond-equity allocation is improved
by simply switching from market weighted equities to smart beta. In the chart
below, we trace two simplified efficient frontiers using the JP Morgan Global Bond
index for fixed income, and the MSCI World Index or the MSCI World Minimum
Volatility for equities. The chart is based on historical data only (returns, variance
and covariance).
Historical Historical Correlations

Returns Volatility with bonds
MSCI World 9.48% 15.72% 20.09%

MSCI World MinVol 9.58% 11.41% 31.90%
JPM GBI 4.75% 6.86%
Despite a slightly higher correlation with bonds, and thanks to the far better risk
return profile of the MSCI Minimum Volatility Index, the improvement in the efficient
frontier is straightforward:
Efficient Frontiers - Historical Data
12%
11%
10%
9%
8%
7%
6%
5%
4%
4% 6% 8% 10% 12% 14% 16% 18%
------- MSCI Min Vol: historical data ------- MSCI Min Vol: historical data
We can state that, for the same level of risk of a traditional bond-equity allocation,
we can increase the relative weight of smart beta equities in the allocation (as
smart beta equities are more conservative than traditional equities), thus improving
performance, via both the higher percentage of equity and the higher return of

smart beta. In the same way, introducing smart beta while keeping the weight of
equity unchanged, we improve the expected return of the portfolio, while reducing
its risk.
These statements are formally correct, have proven to hold up over the last decade,
and may be confirmed in the future. However they imply a precise hypothesis: the
performance drivers discussed above will deliver performance in line with those
we have seen in the recent past. On the contrary, if we apply an alternative and
less favorable scenario where sector reversal, low risk anomaly, small caps, and
all other residual factors are not going to deliver any additional return, investing
in smart beta would be simply equivalent to investing in low beta equity. This
is the reason why we believe that clearly identifying performance drivers (even
demystifying some beliefs about these strategies) and being confident with them is
a required condition for investors to buy smart beta.
To formalize this unfavorable scenario into our simplified efficient frontier, we may
set the expected return of the MSCI Minimum Volatility Index to a level that equalizes
the risk-adjusted return of the MSCI World Index.
Historical Historical Risk Adjusted

Returns Volatility returns
MSCI World 9.48% 15.72% 0.60

MSCI World MinVol 6.88% 11.41% 0.60
JPM GBI 4.75% 6.86% 0.69
This is equivalent to the hypothesis that the MSCI Minimum Volatility Index is simply
a low beta index, as it was a combination of cash and a traditional index. Thus,
investing in low beta equity with the same risk adjusted return of traditional equity
would be equivalent to imposing the constraint of a minimum holding in cash, thus
making the frontier less attractive than using unconstrained equity.
Efficient Frontiers - Historical Data and Risk Adjusted Returns
12%
11%
10%
9%
8%
7%
6%
5%
4%
4% 6% 8% 10% 12% 14% 16% 18%
- - - - MSCI Min Vol: same risk adjusted return and correlation of MSCI World Source: Amundi Research

Actually, in a balanced asset allocation with smart beta, we still have to investigate
if some additional benefit could come from lower correlation with bonds: if so,
smart beta equities would be more diversifying. Unfortunately this is not the case
as bonds correlation with the MSCI Minimum Volatility Index is 32%, while its only
20% with the MSCI Index (cf. previous table).
Taking true correlation into account we have an even less interesting profile:
Efficient Frontiers - Historical Data and Risk Adjusted Returns
12%
11%
10%
9%
8%
7%
6%
5%
4%
4% 6% 8% 10% 12% 14% 16% 18%
- - - - MSCI Min Vol: same risk adjusted return and correlation of MSCI World
- - - - MSCI Min Vol: same risk adjusted return and correlation of MSCI World historical correlation Source: Amundi Research
We may wonder if this evidence is limited to the MSCI World Minimum Volatility
Index, or correlation with bonds is higher for any smart beta. Actually this is rather
generalized evidence, with the exception of the FTSE EDHC Risk Efficient Index
that is close to traditional equity, even from a correlation standpoint.
Correlations with Bonds

100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
FTSE EDHEC Amundi FTSE MSCI Amundi Risk MSCI World Amundi Amundi
-Risk Efficient Diversif. TOBAM MD World RW Parity MinVol MinVar MinVar - Piot
Smart Beta MSCI World Source: Amundi Research
3.4 Diversifying and timing smart beta strategies

As we have seen in previous sections, there are several factors behind the
performance of smart beta equities. Though we may identify some significant
common behavior across them, the relevant factors explaining such performance
deviations from a standard index are not the same for every smart beta strategy.

Furthermore we observe that smart beta strategies perform differently according
to the conditions of the equity market as a whole (bear or bull market, trading
range, high or low volatility, high or low average correlation).
These are all strong arguments for diversifying across smart beta strategies, and
not holding just one of them.
3.4.1 Diversifying across smart beta strategies
There are several ways to diversify across smart beta strategies, and diversification
is possible even if the number of strategies involved is very limited. Diversifying
across smart beta may also be useful in addressing the issue of building a
reasonable multi-strategy smart beta benchmark.
We provide several cases of multi smart beta allocation and for illustrative purposes
we stay within the MSCI family (MSCI World Minimum Volatility, and MSCI Risk
Weighted). We first consider the case of an investor within an absolute risk-return
framework. If the investor wants to achieve diversification by equalizing the two
indices contribution to absolute risk, according to a long term (10 years) covariance
matrix, he would allocate 57% of his assets to the minimum variance and 43% to
the risk weighted indices respectively.
Equal Active Risk Contribution Ey Strategy
100%
80%
57% 50,0%
60% 77%
40%
43% 50,0%
20%
23%
0%
Weights Total Risk Contribution Active Risk Contribution
MSCI World RW MSCI World MinVol

As requested, each strategy has an equal contribution to absolute risk but it is

interesting to notice that risk relative to a standard index is concentrated on the
minimum variance strategy.
We can translate the analysis on performance drivers, using the estimated
parameters of linear regressions and their covariance matrix, from section 3.1.
From an absolute risk perspective, the market factor explains roughly 90% of the
absolute variance, with no surprise as we are dealing with equity portfolios. From
an active risk perspective, risk is rather concentrated over low beta (obvious as
minimum variance has very low beta), low risk anomaly, and dividend yield; 10% of
active variance is unexplained.

Contributions Wo Absolute Risk Contributions Wo Active Risk
100% 100%
Unexplained 10,06%
80% Momentum 80%
12,96%
Sector Reversal
60% 60% 15,43%
Value
92,08%
40% Dividend 40%
Low Syst. Risk
48,67%
20% 20%
Small Cap
0% MKT Beta 0%
Factor Contributions Factor Contributions
In a second example, we assume that another investor prefers to diversify from

an active risk perspective. He still diversifies on the two indices, and does not
yet control for factor exposures directly. In this case he would rather invest 73%
in the risk-weighted index and 27% in the minimum variance index respectively.
Equal Active Risk Contribution Contributions

Ey Strategy Wo Active Risk
100% 100%
8.13% Unexplained
80%
33% 26.7% 50% 5.61% Momentum
80% 12.37%
Sector Reversal
60% 73.3% 16.69%
67% 60% Value
40%
50% 17.19% Dividend
40%
20%
Low Syst. Risk
0% 20% 37.47% Small Cap

Weights Total Risk Active Risk
Contribution Contribution MKT Beta
0%
Factor Contributions
By doing so, active risk would be balanced across the two strategies and, as a
side effect, diversification across the performance drivers would improve as well,
while the percentage of active risk with an unknown source would be reduced.
However, the investor might be much more sensitive to diversification across
the factors than across the two indices themselves. In this third example, we
assume that the investor is willing to maximize the diversification of the sources
of active risk. We thus maximize the measure of entropy as defined in section 2.2,
computed over the active risk contributions by factors, or performance drivers.

Maximum Entropy Over Contributions
Factor Contributions to Active Risk to Active Risk
100% 100%
7,36% Unexplained
21% 16.1%
80% 32% 6% Momentum
80% 11%
60% Sector Reversal
17%
60% Value
40% 79% 83.9%
68% 27% Dividend
40%
20%
Low Syst. Risk
0% 20% Small Cap

29%
0%
In this case, in order to reduce the still dominant contribution of low market beta,
the allocation in the risk-weighted index would increase. Contrarily, small cap,
dividend yield, and low risk anomaly contributions are increased. The change in
the latter, however, is mainly due to a base effect: portfolio exposure to the low
risk factor decreases, but as the active risk decreases as well (risk weighted index
has a much lower tracking error than minimum volatility, relative to the standard
index), its risk contribution as a percentage marginally increases. As a side effect,
unexplained active variance is further reduced.
As a fourth and last case, we now consider an investor that is comfortable with an
objective of diversification across factors in an active management framework, but
that is willing to introduce into his allocation some exposure to the sector reversal
factor, because of its regular and low-volatility historical contribution to performance.
The exposure to this factor is negligible in the three previous allocations.
The smart beta portfolio that is most exposed to sector reversal is the Amundi Risk
Parity, because of the two-step stock-sector construction process. We thus repeat
the last case study, adding Amundi Risk Parity to the set of available strategies.
Maximum Entropy Over Contributions

Factor Contributions to Active Risk to Active Risk
100% 100% 5% Unexplained
21% 16,3% 32% 7%
80% 4% Momentum
42,9% 80% 7%
60%
41% 16% Sector Reversal
34% 60% Value
40%
40,9% 28% Dividend
37% 40%
20% 34% Low Syst. Risk
20%
0% 32% Small Cap
0%
Amundi Risk Parity MSCI World RW MSCI World MinVol

The overall allocation to the risk parity strategies is basically unchanged, but split
into the MSCI Index and the Amundi process. This latter now accounts for 37%
of the assets, 41% of absolute risk, and 34% of active risk. Sector reversal would
be introduced as a source of active risk with a 7% contribution, while the entropy
measure on the risk factor would increase to 5.02 from 4.89. Unexplained risk
would be further reduced to 5%.
3.4.2 Timing smart beta strategies
As several factors drive smart beta performance, the most straightforward way to
implement some timing over the different indices or strategies, should be to time
the underlying factors and consistently allocate strategies.
Investors may develop a reliable style rotation model, and may apply some allocation
where risk contributions match return expectations, rather than maximizing some
diversification measure as we have done in previous case studies.
Another way to time smart beta strategies might be to investigate their behavior
according to different market conditions. The following chart exhibits the 12-month
cumulative outperformance of each of the three Amundi strategies relative to the
standard index, with a quarterly frequency, from June 2003 to December 2013.
Results are interesting and often intuitive as well.
Rolling 12-Month Outperformance (LS) Relative to MSCI World (RS)
0.25 3.5
0.2 2
0.15 3 3.0
1
0.1
0.05 2.5
0
-0.05 4 5 2.0
-0.1
-0.15 1.5
-0.2
-0.25 1.0
2011 03
2011 06
2011 09
2011 12
2003 06
2003 09
2003 12
2004 03
2004 06
2004 09
2004 12
2005 03
2005 06
2005 09
2005 12
2006 03
2006 06
2006 09
2006 12
2007 03
2007 06
2007 09
2007 12
2008 03
2008 06
2008 09
2008 12
2009 03
2009 06
2009 09
2009 12
2010 03
2010 06
2010 09
2010 12
2012 03
2012 06
2012 09
2012 12
2013 03
2013 06
2013 09
2013 12
Amundi Minvar Amundi Diversification Amundi Risk Parity MSCI World

In long and steady bull markets as was the case from 2003 to mid-2007, the
risk parity portfolio often exhibits the best returns, while during market crashes
minimum variance is by far the winning strategy.
When the impressive rebound of March 2009 starts, minimum variance starts
lagging the two other strategies, while diversification and especially risk parity react
well since they keep on delivering some positive outperformance.
When the market is suffering some higher volatility without exhibiting a clear trend
as in the period between mid-2011 and mid-2012, minimum variance is the winning
strategy with some nice resistance by the diversification strategy as well.

During the recent low volatility bull market period, smart beta strategies are slightly
lagging overall (with better risk adjusted returns than the market index, however),
but the risk parity strategy still captures the trend fairly well.
We are conscious that forecasting the market conditions of the future is not an
easy task, and actually this is not our goal. On the other hand, we recognize that
volatility, correlation and turbulence, may be behind each of the five states of
the world described above. In this section, we describe the dynamic allocation
model that Amundi implements on a real money multi-smart beta fund on Eurozone
equities. The model is based on three stand-alone dynamic strategies: each of
them is based on a market signal, has an equally weighted target allocation on the
three smart beta, and assigns an overweight and an underweight of 5% to the most
and least profitable strategy, according to the market signal.
The first model is based on market implied volatility, and works as a typical risk
off risk on model. When the average level of the VIX Index, computed over the
last 10 days, is higher than the average computed over the last 25 days (increasing
VIX), we overweight the minimum variance portfolio and we underweight risk parity,
according to the conditional next-month average returns that we have computed
historically. When the 10-day average is lower than the 25-day average (decreasing
VIX) we overweight risk parity and we underweight minimum variance. According to
the VIX model, the weight of the diversification-based smart beta is always neutral.
When the relative difference between the two averages is less than 5% we do not
apply any under/over weights as we allow the model to be in a neutral position,
in order to reduce turnover and avoid false signals. The chart below exhibits the
next-month average annualized returns, conditional to the VIX configuration, as
estimated from beginning 2003 to mid-2012 (our sample period).
Average Returns Increasing Vix Average Returns Decreasing Vix
0.0%
25.0%
-5.0% 20.0%
15.0%
-10.0%
10.0%
-15.0%
5.0%
-20.0%
0.0%
Risk Parity Diversif. Minvar MSCI Risk Parity Diversif. Minvar MSCI
The second model is based on average market correlation. We have computed

average correlation according to the CBOE methodology, on single country indices,
as well as on the GICS industry group indices of the MSCI World.

As for the case of the VIX, in the charts below we report the average annualized
returns computed over months following an average two-week correlation higher
than the 104-week average (low correlation), and months following an average two-
week correlation higher than the 104-week average (high correlation). Again, the
5% threshold for neutral signals applies, and the sample period ends in mid-2012.
Average Returns High Correlation Average Returns Low Correlation
15.0% 10.0%
5.0%
10.0%
0.0%
5.0%
-5.0%
0.0% -10.0%
The intuition behind this is that when correlation is very high (left chart), there is
less benefit in searching for diversification over risk factors, while searching for
diversification across assets directly is probably more effective. When average
correlation is lower (right chart), strategies based on diversification across risk
factors benefit more than those diversified on stocks directly, since the former
favor those stocks exposed to uncorrelated factors. Overall, returns on the right-
hand chart are lower, as the low correlation across sectors and countries includes
a typical pre-crisis situation (October 2008 and July 2011).
Average Correlation and Returns in the Eurozone
Estimation Period: 01/2003 - 06/2012
0,2
0,15
0,1
0,05
0
-0,05
-0,1
-0,15
-0,2
02/01/03
05/01/03
08/01/03
11/01/03
02/01/04
05/01/04
08/01/04
11/01/04
02/01/05
05/01/05
08/01/05
11/01/05
02/01/06
05/01/06
08/01/06
11/01/06
02/01/07
05/01/07
08/01/07
11/01/07
02/01/08
05/01/08
08/01/08
11/01/08
02/01/09
05/01/09
08/01/09
11/01/09
02/01/10
05/01/10
08/01/10
11/01/10
02/01/11
05/01/11
08/01/11
11/01/11
02/01/12
05/01/12
08/01/12
High Corr Low Corr Neutral Delta 2W - 2Y Av Corr Source: Amundi Research
Our third indicator is a turbulence index. We define market turbulence as the cross-
section dispersion of returns, computed over the GICS industry group indices. As this
indicator is closely correlated with the VIX index, we normalize by the VIX itself, as we want
to capture the turbulence that is not already explained by the market implied volatility.

As for the correlation indicator, we compute rolling averages on two weeks and 104
weeks, we apply the neutrality threshold at 5%, and we finally compute next-month
annualized average returns.
Average Returns Average Returns

High Turbulence Low Turbulence
15.0% 8.0%
6.0%
10.0%
4.0%
5.0%
2.0%
0.0% 0.0%
Results are less intuitive than in the two previous cases, but the turbulence
indicator is uncorrelated with the implied volatility and average correlation signals.
As mentioned, our final dynamic strategy consists in the portfolio that averages
the three model portfolios based on the three market signals above. Each model
portfolio overweights the best performing strategy according to the observed
signal at the end of the previous month.
The chart below shows the gross total return performance of the market-weighted
index, an equally weighted basket of the three smart beta strategies, and the
dynamic strategy described so far. Starting mid-September 2012 data are out-of
sample, while starting from June 2013 the strategy feeds a real money portfolio.
Turnover 72.06% IR 1.27

Mkt Wght Constant Mix Dynamic Strategy
Sharpe Ratio 0.29 0.64 0.67
Return / Conditional Var (5%) 0.53 1.01 1.08
Return 6.1% 9.4% 9.9%
Conditional Var (5%) -12.6% -9.3% -9.1%
Drawdowns -56.2% -50.1% -49.3%
Volatility 17.5% 13.2% 13.2%
Compared to an equally weighted composite of smart beta strategies, the

dynamic model outperforms by more than 40 basis points per year. There is no
significant impact on the absolute volatility, even though the drawdown of 2008
is reduced. The Sharpe ratio rises to 0.67 from 0.64 , and the ratio of returns to
Conditional Var improves too (1.08 versus 1.01).

Performance
3.50
1.05
3.00
1.04
Active Returns
2.50
Returns
1.03
2.00
1.02
1.50 1.01
1.00 1.00
0.50 0.99
2014
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Strategy with Timing (LS) Const. Mix (LS)
Mkt W Index (LS) Cum. Active (RS) Source: Amundi Research
IV - Are smart beta passive or active strategies?

Defining an investment strategy as active or passive is usually not an easy task.
Of course there are some extreme circumstances where such a definition is
obvious: lets consider an ETF benchmarked to the S&P 500 and the investment
fund Berkshire Hathaway run by Warren Buffett: it would be hard not to classify
them as passive and active respectively.
However, there are some frequent intermediate situations where we can distinguish
different levels or intensities in being active or passive.
A classic condition for a passive strategy is that we clearly identify the systematic
replication of a widely recognized, transparent and investable benchmark.
According to our definition, we must recognize both the elements in order to match
the passive condition:
1. Benchmark characteristics: widely recognized (generally accepted),
transparent and investable;
2. Systematic replication.
On the other hand, we define as active whatever strategy exhibits pronounced
deviations (both in terms of holdings and in terms of returns) from a benchmark
index.
Widely recognized and transparent Benchmark
Prior to this point, we need to clarify whether the market cap index should be
considered as the benchmark of a smart beta strategy. We believe this should not
be the case, as the main argument for smart beta investing is the well-documented
inefficiency of market capitalization benchmarks. As a consequence, smart beta
strategies should not be benchmarked to market weighted indices, nor is their
impressive tracking error a valuable argument for defining them as active.

However, market weighted indices offer an interesting reference point for
comparison to our discussion. Together with their composition, index providers
disclose the rules applied for companies inclusion or exclusion from the index
(typically: geographical belonging, sector classification, free float, and size), as
well as the rules for index weighting. All of them rely on continuously and publicly
available information as market capitalization. To some extent, market cap weighted
benchmarks may be replicated even without knowing their composition directly: in
order to derive it, investors could combine construction rules with publicly available
information, at the cost of a negligible margin of error.
The case for smart beta benchmark is different. Lets have an example focusing on
some minimum variance benchmarks (such as the MSCI Minimum Volatility index),
or on some diversification benchmarks (such as the FTSE EDHEC Risk Efficient
Index, or the FTSE TOBAM Maximum Diversification).
These indices do not rely on some objective and easily measurable metrics, as in
the case for market capitalization: rather, all of them depend on:
1. Some risk measure estimation (the variance covariance matrix of stocks)
2. An optimizer, and its numerical algorithms
3. The objective function that is maximized: portfolio variance is minimized
(or some diversification measure maximized), subject to some constraints
(minimum stocks threshold, stocks upper bounds, sector concentration
constraints, etc.)
All the points mentioned above contain provider-specific features (risk model,
optimizer), while some of them (the set of constraints) are also very discretionary
and better shaped to design an investment strategy than to build a traditional
investment benchmark.
Smart beta benchmarks require the estimation of a variance covariance matrix. A

simple historical data approach is almost impossible: a statistical and parsimonious
method (such as principal component analysis, or fundamental factor models) is
usually needed, as a covariance matrix may ideally contain millions of parameters.
Such benchmarks are thus dependent on some necessarily complex risk model.
Furthermore, smart beta composition is definitely influenced by the numerical
algorithm (often very complex too) of the optimizer. This point is rather critical
as we often observe some better consistency among risk forecasts provided by
different risk models, than among portfolios optimized by different optimizers: little
differences in risk and correlation estimates may determine huge differences in
optimal stock weights, and differences in numerical algorithms across optimizers
magnify those discrepancies in portfolio composition.
Furthermore, in order to prevent some typical drawbacks of some optimisation-

based smart benchmarks (as excessive concentration on a few sectors or a few

and sometimes illiquid- stocks, involuntary exposure toward styles such as small
caps or momentum) some prudent constraints may be needed:
minimum holding thresholds;
general upper bound on each stock (in absolute terms or as a function of the
daily liquidity);
sector and country holdings may be constrained in a range around their
weights in a traditional index;
the index may be prevented to have major exposure to some style factors.
Although reasonable, realistic, and prudent, these rules are discretionary and lead
to a benchmark that is provider-specific, rather than universally recognized, as it
would be required instead.
As a consequence, for instance, the minimum variance indices so far available in

the market exhibit reciprocal (historical) tracking errors ranging from 4% to 6%.
Definitely more than the tracking errors among market weighted indices, which are
usually lower than 1%.
In conclusion, while the disclosure of parameters and models may satisfy the
transparency condition, the dependency on different risk models and optimizers,
and the common practice of applying various and sometimes heterogeneous
constraints prevent the benchmark to be easily recognized and universally
representative.
We might argue that, as far as there is no universally representative smart beta

benchmark for any of the three categories, passive management is basically
precluded for smart beta: many alternative benchmarks exist and all are very
different from each other.
We think this conclusion is too radical, it is highly influenced by a traditional definition

of benchmark, and it does not recognize some significant trends in the passive asset
management industry. Nowadays passive management is experiencing spectacular
growth thanks to a highly comprehensive product offer. These products are not
limited to those asset classes with a universally recognized benchmark available. In
contrast, every asset class or strategy, even the more exotic or customized, may be
packaged in an ETF. The only requirements are the mere existence of a benchmark
(and very often we would better say the existence of an underlying asset), and the
concrete possibility for the fund manager to replicate its payoff.
We believe that the universal recognition of the benchmark should rather be

replaced by some less stringent requirement such as the mere existence of the
benchmark itself, its transparency and its investability. If several and different
minimum variance, or risk parity, or diversification benchmarks exist, we may
potentially have several passive mandates replicating them, each with a very
different payoff, but all mimicking the payoff of their own benchmark.

Of course, a smart beta fund manager that is formally benchmarked to a traditional
market cap index, would concretely be a passive manager if he actually tracks an
existing smart beta benchmark, regardless of the decision of declaring the true
benchmark.
On the other hand, a highly innovative or customized smart beta solution could
be packaged in a passive product by asking an index provider to produce a tailor-
made benchmark. We think that disclosing a benchmark is a not a pure formality,
neither is it a trivial decision by the fund manager. A formal benchmark implies
the involvement of an index provider that provides objective calculation and
transparency, and also implies a formal commitment by the fund manager of never
deviating from the benchmark itself. Furthermore having an official benchmark
would be a costly decision as smart beta benchmarks are far more expensive than
traditional ones.
As a conclusion, once we have verified that a pertinent benchmark exists, is
transparent, investable and computed by a third party (the index provider), the
only criterion that we need to apply in order to categorize a smart beta product
as active or passive, is whether or not the fund manager implements a systematic
replication of it.
Systematic replication
Once the benchmark is built and made available, the goal of a passive fund
manager is to systematically apply and comply with it, rebalancing at a pre-
specified frequency with no room left for incorporating (time varying) views on
market and stocks. In other words, stock picking is not contemplated, and tracking
error relative to the benchmark or relative to the reference strategy must be ideally
equal to zero.
On the other hand, for a strategy to be defined as active, deviation from
the benchmark must be relevant, and the main drivers of such deviations are
investment decisions. The latter depend on time-varying forecasts of the future
profitability of stocks.
According to this criterion, smart beta investments may fall in both the active
and the passive category. The easiest example of an active strategy is some
optimisation-based smart beta strategy (minimum variance or diversification),
where the process is applied to a restricted list of stocks, based on some
qualitative and judgmental criteria (the best investment ideas of the buy-side
analysts, for instance). Similarly, a smart beta portfolio is actively managed where
the investment universe is filtered by some quantitative or systematic criteria, if
this filtering is specific to the fund manager, and generates some non-negligible
tracking error relative to any of the existing benchmarks available. In the same
way, portfolios where risk factors exposures are managed according to market
views are active portfolios. Finally, any risk parity portfolio that systematically

applies some unique weighting schemes (as the two-step approach discussed in
previous sections) is actively managed, if those weighting rules differ from those
applied by the available benchmarks.
As an opposite example, any replication of an index is passive, even if the index

is built by maximizing a highly innovative utility function, through a sophisticated
numerical algorithm, employing highly specific sector and country constraints, and
investment universe restrictions. The requirements are that the index is produced
by an independent index provider that discloses calculation methodology, and that
the fund manager assures very low tracking error.
Conclusion
Smart Beta equities are the asset management industrys answer to some well-
known drawbacks of market capitalization-based equity indices such as price
noise, overrepresentation of large caps, absence of an auto-corrective mean-
reversion mechanism. Some of these features may result in high volatility and
massive drawdowns, thus potentially compromising the risk-return payoff of
traditional equities, at least when the investment horizon is shorter than 8-10 years.
In this study we provide a formal description of three popular risk-based smart beta
strategies minimum variance, diversification, and risk parity.
We show that low market beta and the low risk anomaly explain a relevant
portion of the variability of the active returns of the minimum variance strategies,
with some variance explained by sector reversal and dividend yield. Yet the
unexplained variability corresponds to some non-negligible positive contribution to
performance (thus further investigation is needed), while filtering the universe for
some quality criteria proves to provide additional value.
As for the diversification-based strategies (portfolio maximizing the diversification

ratio, risk efficient portfolio, etc.), low market beta and low risk anomaly are still
the most significant factors, with the addition of small cap and sector reversal.
Performance drivers behind the risk parity strategies are basically the same, but
we notice that the low beta and low risk anomaly are less explanatory than
small cap and sector reversal. Sector reversal as a source of outperformance
is more relevant for risk parity than for any other smart beta, especially where (as
is the case in Amundis process) risk parity is achieved through a two-step stock-
sector construction process.
Smart beta may become a new equity core if the investors relevant risk measure
is absolute risk. In this case, however, the liquidity of those strategies must be
consistent with the amount of assets the investor holds. If the investors relevant
risk measure is relative risk, smart beta might still become a new equity core,

but some more pertinent benchmarks should be designed, because smart beta
investing generates high tracking error relative to standard indices.
A clever benchmark, as well as a clever multi-smart allocation should exploit the
circumstance that the exposures to performance drivers are not identical for all
smart beta equities, and thus there is room for diversification. Another argument
for diversifying across smart beta is the different behavior they exhibit in some
typical market conditions. In addition to diversification, investors can translate
these different behaviors into some profitable timing strategies.
Finally, we discuss whether smart beta should be considered as passive or rather
active strategies. According to the pure replication criterion, most smart beta
strategies should fall into the passive category, with the exception of those (such
as for the Amundi minimum variance) where the portfolio construction processes
are combined with discretionary and judgmental investment decisions by the fund
manager. The second traditional criterion of transparency and wide recognition of
the benchmark should be replaced by the less stringent requirement of the mere
existence of a benchmark. If the fund manager replicates an existing smart beta
benchmark, the answer is obvious if he designs his own smart beta process,
then the asset manager himself determines the active or the passive nature of his
product by requiring or not that a tailor-made benchmark is created and maintained
by an index provider.

Appendix
Optimisation-based smart beta portfolios at Amundi
Quality Stocks
We believe that fundamental equity selection can provide some valuable
enhancement in the risk return profile of equity portfolios, at least in the long run.
At the same time we do not want to renounce an optimisation process which is
completely independent from expected returns. Expected returns are very noisy in
forecast and thus responsible for well known error maximization problems. For this
reason, we apply a qualitative filter to our investment universe, excluding the lowest
quality stocks from the optimisation. Basically, each quarter we rank the constituents
of the MSCI World Developed Markets according to a Piotroski (2000) score and
we exclude the two bottom quintiles. Keeping 60% of constituents available for
investments, the optimizer is left with a high degree of freedom and it tilts the optimal
portfolio toward good quality stocks, without using explicit expected returns.
Turnover and liquidity
High turnover is a critical issue in many systematic investment strategies like
Minimum Variance and other optimisation-based strategies. In our case, turnover
in the investment universe is limited as the Piotroski score is based on balance
sheet data that varies very little during one quarter. Furthermore we also rebalance
our portfolio quarterly, as suggested by Baker and Haugen (1991).
Nevertheless, more than turnover itself, our concern is indeed liquidity: we aim to
avoid small illiquid companies as we want to be able to liquidate our portfolio in a
reasonable time lag, without incurring significant market impact costs.
To address this requirement, we limited the amount held in any stock to the
following percentage:
where UB i is the upper bound on the ith stock, D is the number of days that we
accept to liquidate the fund, ADV i is the average daily volume over the last quarter,
and NOT is a notional amount of assets under management of USD 1 billion: quite
conservative as it is still far above the current size of our fund.
Sector, country, and stock concentration
As mentioned above, Minimum Variance and Diversification portfolios provide
excellent diversification across risk factors, but may tend to be poorly diversified
across sectors, countries or single stocks. We have thus applied some constraints
at these levels, without preventing the optimizer from choosing solutions that are
far enough from a market index.

On countries and sectors we accept deviations from the market index of 5% to 10%,
while for single stocks we apply a general upper bound (GUB), thus modifying the actual
upper bound as follows:
Management of asymmetries in factor returns

Furthermore, we are conscious that optimisation-based smart beta portfolios may
be systematically or incidentally exposed to fundamental factors such as size,
value or momentum.
We observe that much of our size exposure is corrected away by the liquidity
constraints. As for other factor exposures, we have decided not to manage them
systematically as again we do not want to excessively restrict the optimisation
process.
On the other hand, we regularly monitor the behavior of all the risk factors of
the BARRA model (size, value, growth, momentum, leverage). The goal of this
monitoring is to detect bubbles or suspicious asymmetries like excessive positive
skewness in recent performance: in the case of significant alerts, from time to
time the fund manager hedges the risk of an exploding bubble, imposing a neutral
exposure to the suspected factor.

Acknowledgements
I would like to thank Sylvie de Laguiche for discussions, suggestions, and

comments that improved the quality of the manuscript.
Bibliography
Ang A., Hodrick R.J., Xing, Y., Zhang, X. 2006. The Cross Section of Volatility and Expected
Returns. The Journal of Finance. Vol. LXI, No. 1 (February)
Baker M., Bradley B., Wurgler J. 2009. A Behavioral Finance Explanation for the Success
of Low Volatility Portfolio. New York University, Working Paper
Baker M., Bradley B., Wurgler J. 2011. Benchmarks as Limits to Arbitrage: Understanding
the Low Volatility Anomaly. Financial Analysts Journal, Vol. 67, No. 1, CFA Institute
Baker N. L., Haugen R. A. 1991. The Efficient Market Inefficiency of Capitalization-Weighted
Stock Portfolios. The Journal of Portfolio Management, Vol. 17, No. 3 (March)
Clarke R., De Silva H., CFA, Thorely S., CFA. 2011. Minimum Variance Portfolio Composition.
The Journal of Portfolio Management, Vol. 37, No. 2 (Winter)
Clarke R., De Silva H., CFA, Thorely S., CFA. 1991. Minimum Variance Portfolios in the US
Equity Market. The Journal of Portfolio Management, Vol. 33, no. 1 (Fall)
Clarke, R., De Silva, H., Thorley, S. 2012. Risk Parity, Maximum Diversification, and
Minimum Variance: An Analytic Perspective. Journal of Portfolio Management, Vol. 39, No.
3,(Spring 2013)
Carvalho R.L., Lu X., Moulin P. 2011. Demystifying Equity Risk-Based Strategies: A Simple
Alpha Plus Beta Description. The Journal of Portfolio Management, Vol. 38, no. 3 (Spring)
Choueifaty, Y., Coignard, Y. 2008. Towards Maximum Diversification. The Journal of
Portfolio Management, Vol.35, No. 1 (Fall)
Markowitz H. 1952. Portfolio Selection. The Journal of Finance, Vol. VII, No.1 (March)
Maillard S., Roncalli T., Teiletche J. 2009. On the Properties of Equally-Weighted Risk
Contributions Portfolios. Social Science Research Network, Working Papers Series
(September)
Piotroski J. D. 2000. Value Investing: The Use of Historical Financial Statement Information
to Separate Winners from Losers. Journal of Accounting Research, Vol. 38, Supplement
2000
Russo A. 2013. Low Risk Equity Investments: Empirical Evidence, Theories, and the Amundi
Experience. Amundi Working Papers, WP-033-2013 (March)
Sharpe W.F. 1964. Capital Asset Prices: A Theory of Market Equilibrium under Conditions of
Risk. Journal of Finance, Vol. XIX, No. 3 (September)
Shefrin H., Statman M. 2000. Behavioral Portfolio Theory. Journal of Financial and
Quantitative Finance, Vol. 35, No. 2 (June)
Thomas R., CFA, Shapiro R., CFA. 2007. Managed Volatility: A New Approach to Equity
Investing. State Street Global Advisors

DP-03
SRI and Performance:

Impact of ESG Criteria in
Equity and Bond Management
Processes
Florian BERG, Sylvie de LAGUICHE,
Tegwen LE BERTHE, Alessandro RUSSO, Quantitative Research
Antoine SORANGE, Extra-financial Analysis
March 2014
In this study, we examine whether an ESG signal adds performance

when incorporated into a portfolio management strategy. Amundi
uses a best-in-class approach on a global investment universe.
What makes Amundis extra-financial rating system unique is that
it complements the agency consensus with a subsequent internal
analysis. SRI portfolio management is based partly on exclusion
criteria for the worst-rated securities and partly on the portfolios
overall ESG rating, in absolute and relative terms compared to its
benchmark.
The study revealed that after correcting for geographical and style
bias, there is no significant underperformance or outperformance
over the assessed period. In addition from an incompressible risk
standpoint, there is no significant cost on either the European or
global investment universes over the observed period. In other
words, it can reasonably be said that SRI management currently
offers investors a relatively cost-free way to benefit from the results
anticipated from increasing awareness of ESG criteria among
companies and investors, as well as non-monetary advantages in
terms of reputation and responsibilities to future generations.

SRI AND PERFORMANCE:
IMPACT OF ESG CRITERIA IN EQUITY
AND BOND MANAGEMENT PROCESSES
Introduction
and Previous Literature
The relationship between firms environmental, governance and social practices
and their financial performance has attracted much debate in recent years.
This controversy has been fuelled by arguments from economics, management
and finance. As reminded by Kacperczyk (2009), the two main theses in play
could be described as the shareholder theory and the stakeholder theory. A
stakeholder as defined by Freeman (1984) is any group or individual who can
affect or is affected by the achievement of an organizations purpose. Both
theories defend different views on the role CSR (Corporate Social Responsibility)
should play in the definition of a firms objectives.
According to the shareholder theory corporate managers should focus solely
on increasing their shareholders wealth. The responsibility towards shareholders
should always be considered as more important than the responsibility towards
non-shareholding stakeholders such as employees, customers, natural
environment or communities. This thesis is notably upheld by Friedman (1970),
Jensen and Meckling (1976) and Fama and Jensen (1983).
The stakeholder theory (Freeman (1984) and Freeman et al. (2007)) states that
corporations should consider the interests of each stakeholder in their decision
making. In a modern pluralistic society, a firm cannot simply maximize one
objective function in order to deal with all potential contingencies. Furthermore,
no stakeholder should have a prima facie obligation over another (Kacperczyk
2008). According to Freeman (2004), the stakeholder theory asks for the purpose
of a firm and shared values with all stakeholders. Efficient corporate governance
for instance may foster financial performance and facilitate debt financing. Having
close relationships with suppliers and being attentive to customers needs might
establish a form of loyalty that helps to reduce uncertainty and strengthen a

company during hard times. Good CSR could also be seen as a risk mitigating
policy by preventing the risk of extreme negative events. A sound environmental
profile, for instance, indicates that a firm is less likely to cause an environmental
disaster that may induce unexpected expenses being financed by issuing debt.
The subsequent deterioration of the firms balance sheet may induce a higher
cost of debt and raise its default risk. Inadequate corporate governance, for
instance, may also affect the firms financial performance and thereby increase
its default risk.
Literature surveys by Orlitzky et al. (2003) and Margolis et al. (2007) report evidence
of a positive correlation between corporate social performance (henceforth CSP)
and the firms financial performance as measured by stock market capitalization
or accounting measures. To our knowledge, most studies on the link between CSP
and financial securities focus on the stock market. For instance, in a recent paper
Krger (2009) assesses the impact of ESG (Environment, Social, Governance)
news on stock returns. He shows that a significant negative abnormal return is
observed after the release of a negative event. A positive event has no significant
effect.
However, empirical results from the stock market cannot be applied directly on
the corporate bond market. Stocks and bonds are not affected by news through
the same channel. Bondholders and shareholders do not have the same loss
functions. Bondholders are fixed claimants and have an asymmetric exposure to
the downside risk of their securities relative to the upside potential. In the position
of residual claimants, shareholders are sensitive to upside and downside potential.
Sharfman and Fernando (2008), Menz (2010) and Chen et al. (2007) study the
relationship between different aspects of corporate social responsibility and the
cost of debt financing. On the whole, they do not find evidence that a high CSP
reduces the cost of debt, which contradicts their theoretical analysis. On the
contrary, Goss and Roberts (2011) show firms with high CSP benefit from slightly
lower interest rates on the bank loan market. Moreover, Bauer and Hann (2010)
find a negative relationship between the strength of the environmental profile of US
public firms and their credit spread.
Although the general picture seems to show a positive link between corporate
social performance and corporate financial performance, some individual studies
at the portfolio level, for instance, conducted by Renneboog et al. (2008) and
Amenc et al. (2008) report statistically neutral and sometimes even slightly negative
results, albeit not significant.
These studies tend to confirm the stakeholder theory. Taking into account
environmental, social and governance factors might have a positive impact on the
firms financial performance. However, socially-compliant mutual funds exhibit a
statistical neutral outperformance according to the literature.

Even though mutual funds are neutral in terms of performance, an investor might
have a preference for integrating environmental, social and governance factors
into his portfolio since his utility function might comprise a socially-responsible
component in addition to its financial component.
I - Amundis extra-financial
analysis process
1.1 Philosophy
Responsible Investment is the financial translation of sustainable development. The

purpose of the extra-financial analysis on which it is based is to make businesses
aware of the notion and encourage them to take a sustainable development
approach by assigning them an ESG (Environment, Social, Governance) rating
based on a set of criteria.
t Environmental Dimension:
Energy consumption and CO2 emissions, protection of biodiversity, water,
etc.
t Social Dimension:
Human rights, human resources, link with local communities, etc.
t Governance Dimension:
Independence of the Board of Directors, shareholders rights, anti-corruption
actions, etc.
This analysis incorporates the intangible risks tied to the business activity.
Extra-financial analysis is a so-called best-in-class analysis, as it compares
the securities of a single sector among themselves. Therefore, it promotes the
selection of those companies that are in the best position to manage the risks
and opportunities of sustainable-development issues within homogeneous
industries.
The calculations for weighting ESG criteria (cf. pre-analysis) and ESG ratings (cf.
analysis) are based on the analysis of links among ESG criteria, and the value of
the business.
1.2 Analysis Process
Analysts score businesses on a seven-step scale from A to G. The assessment

of securities from the extra-financial angle includes three phases: a pre-analysis
phase, an analysis phase and a post-analysis phase:

Analysis process
PRE-ANALYSIS ANALYSIS AND CALCULATION POST-ANALYSIS
xESG Criteria Data processing

xWeightings
xProviders selection Rating Calculation tool Quantitative research
and backtesting
Brokers
&
Financial
analysis !
Extra-financial
data providers
Automatic Interaction
data Alerts
with
validation
pre-analysis
oekom r e s e a r c h
ESG Ratings
A to G
1.2.1 Pre-Analysis Phase
The prerequisite of a security analysis:

t Identification of ESG issues by sector
t Creation of a set of criteria for the various sectors
t Choice of outside suppliers likely to meet those criteria. Using more than one
supplier has various advantages:
1. Obtaining several analysis points on the same business and the same
criterion: suppliers that may have a different approach to analyze the same
criterion, using several suppliers offers true complementarity and a 360
view of the issues and behavior of businesses in managing these issues.
2. Access to updated analysis with greater frequency than if using a single
supplier (since each supplier updates its analyses on a given sector but
according to a calendar that is unique to each).
t Weighting of criteria:
The ESG rating is a weighted average of E, S and G ratings. The weighting varies
according to the sector to which the security belongs. The choice of criteria and

their weightings made by the extra-financial analysis team and is the result of a
so-called performance vector method described hereinafter:
To model the influence of ESG criteria on the enterprise value, extra-financial
analysts consider that the impact of these criteria can be made via three
performance vectors:
1. Reputation: it is tied to the business image with consumers and investors.
2. Operational efficiency: this vector identifies the business capacity to improve
technologies, processes and behaviors that reduce the costs of production or
services, thanks to more efficient management of human and energy resources.
This vector is also tied to the business capacity to select, motivate and retain
qualified, competent staff, ensuring the long-term development of the business
and maintaining its know-how.
3. Regulation: this concerns activities that are subject or potentially subject to a
system of laws, regulatory obligations and fines.
Thus, for a given sector and criterion, the extra-financial analysis must answer the
following two questions:
1. How likely is it (on a scale of 0 to 5) that an event linked to a given criterion
will impact a performance vector?
2. What then will be its impact (on a scale of 0 to 5) on the business value?
The example below illustrates the case of the environmental criterion Water, in
the Utilities sector. The Water criterion may have an impact on the business
value via the three performance vectors:
1. The reputation vector: the issue of water leaks in routing circuits is
unknown by the populations, which exposes businesses to a slight risk to their
reputation (probability of 1 in 5). If there is a proven controversy, the impact on
the business value remains low (Impact of 2 in 5).
2. The Operational Efficiency vector: The issue of leakage rate is a
fundamental problem in the utilities sector. In cities like London, it can be
as high as 40%. Yet water lost in pipelines is water that has been previously
treated by businesses and has therefore incurred a cost. Therefore, the leakage
rate is also an environmental aberration, particularly in areas suffering from
water stress but also financial stress. We estimate more than $20 billion is lost
each year by utility companies. Thus the probability that businesses in this
sector are facing this phenomenon is maximum (5/5) and so is the impact on
the businesses value (5/5).
3. The regulation vector: European directives and other regulations around
the world impose strict standards on utility companies in terms of water quality.

After discussion with businesses in the sector and with the relevant financial
analyst, the extra-financial analyst may believe that new significant regulation is
likely to be introduced (probability of 3 in 5). If new regulation were put in place,
the cost generated for businesses would potentially be high (potential impact
on the business value of 3 in 5).
The table below shows the score for Water according to the three performance
vectors in the utilities sector:
Probability scale from

Criteria Value driver Impact Score
1 to 5
1 Reputation 2 2
Water 5 Operational efficiency 5 25
3 Regulation 3 9
36
By performing this exercise for all criteria, we get a score per criterion. For
instance, on the criteria of utilities E dimension, we get the following scores:
Criteria Score
Energy consumption & GHG emissions 18
Water 36
Biodiversity, pollution & waste management 10
These scores are used to define:

t The weight of the criteria: this is the ratio of the score of a criterion to the sum
of all criteria, E, S and G:
Criteria Weight of criteria

Energy consumption & GHG emissions 11%
Water 21%
Biodiversity, pollution & waste management 6
t The weight of the dimensions E, S and G: this is the ratio of the sum of the
scores of criteria for a given dimension to the sum of scores for all criteria. In
the utilities sector, we get:
E S G
Weights 38% 30% 32%

The same exercise performed in other sectors would give the following result:
Sector E S G
Automobile 37% 32% 31%
Bank 26% 33% 41%
Pharmacy 28% 42% 30%
Thus, the more potential impact a criterion has on the value of the business, the
more it will be weighted in the analysis model, in compliance with our desire to have
a pragmatic approach, i.e. based on the most tangible risks and opportunities.
1.2.2 Analysis Phase
A. A proprietary ESG data analysis and processing tool was developed with
the intent to:
t Collect and process data from extra-financial rating agencies so as to make
them comparable,
t Calculate the ratings,
t Generate alert signals for extra-financial analysts in case of insufficient,
obsolete or contradictory data,
t Include the ESG evaluation of businesses done by the extra-financial analysis
team and return it to the management teams,
t Spread the ESG rating of an issuer to all underlying issues.
B. A complementary qualitative approach
In addition to the automatic rating calculation, an active, in-depth analysis is done
on more than 250 securities. To enrich their analyses, extra-financial analysts rely
on several sources of extra-financial data:
t Meetings with businesses and their sustainable development ratio,
t Brokers, who are producing a greater quantity of increasingly refined studies
on the topic of SRI and sustainable development,
t NGOs,
t Analyses by Crdit Agricole Group,
t Media and public documents,
t Scientific reports...
To perform the analysis, each security is compared to securities belonging to a
sector whose ESG issues are homogeneous.
For example, the utilities sector is subdivided into three homogeneous sectors:
t Water utilities,
t Electric utilities,
t Networks.

Next, each criterion is broken down as a risk and risk management indicator:
For example, in the water utilities sector, the benchmarks water criterion is
analyzed as follows:
Example: the water issue in the Utilities sector

> The sector is divided into homogeneous subsectors
Water Utilities Electric Utilities Grid operators
> The water criteria is divided into risk KPIs and risk management KPIs
Water Utilities
Investments for infrastructure renovation
Score Score
Management % grid upgraded since 2008
0 ESG Average consumption 0
of ESG
L Risks L
Risks Objective for leakage rate Presence in water stress zone
Leakage rate
10 10
> Link between financial and extra-financial performance

' Reduction of operational losses: $14 bn per year
' Mitigation of environmental risks: Lake of water and pollution
Thus for each criteria in the benchmark, the businesses get a risk exposure/risk
management score of between 0 and 10.
On a given criterion, businesses that properly manage their extra-financial risks are
those who succeed in managing their financial risk as well. Indeed, in our example
of the water criterion, good risk management means lower operating losses due to
leaks (as the lost water has been previously treated by the business and therefore
incurred a cost) but also environmental risks (lack of water) in view of the rarity of
water, particularly in certain geographic areas.
Therefore there is a close link between financial and extra-financial risk.
Finally, each business is placed on a graph tracing its risk exposure to its risk
management (for compliance reasons, the businesses are not explicitly cited):
On the graph below, water utilities are shown in blue.
X-axis:
The business operating in the United Kingdom is not exposed to much of a water
risk - thus it is on the left in the graph (4/10 risk), while the American business
operates in water stress areas and is therefore given a risk score of 8.8/10.
Y-axis:
The business operating in the United Kingdom has the lowest leakage rate and
thus gets a risk management score of 9/10, while the other businesses, which are
down on this criterion, get a lower score.

Example: the water issue in the Utilities sector
Risk Management
10.00
English company
9.00
A
8.00
7.00
American
B
Company
6.00
Global Company 1
5.00 Global Company 2
4.00
C
3.00
2.00
1.00
D E F G
0.00
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00
Low exposure High exposure Risk exposure
Water Utilities Electric Utilities Grid operators
Ultimately, the businesses placed at the upper left of the graph will be given the
best ratings on the criteria analyzed (the English business rating on the water
criterion is B) while the businesses at the lower right of the graph will be given the
lower ratings (the US business rating on the water criterion is D).
Amundis extra-financial rating is based on a consensus of extra-financial

agencies, plus an internal analysis (weighting of criteria and specific ratings of
250 businesses).
1.2.3 Post-Analysis Phase
The post-analysis phase is based on the quantitative research teams expertise,

and includes:
t Refining the algorithm for calculating ratings and ensuring the calculations
traceability
t Identifying and understanding the causes of the most marked scoring dynamics
on securities
t Identifying, in tandem with the management teams, the securities that out- or
underperformed most, for any extra-financial reasons. This allows the analyst
to recognize ESG criteria that are performance vectors

t Analyzing biases (capitalization, country, etc.) and possibly correcting them.
The choice was made to partially correct the stock-market capitalization bias
inherent in the ESG rating, in order to be more demanding with biggest stock-
market capitalizations, as they have greater resources to set up a sustainable
development policy.
t Reconsidering the weightings of criteria by sector and therefore enriching the
pre-analysis phase.
This type of analysis is done by sector, criterion or geographic area.
Ultimately, ESG scores follow a normal distribution. In order to issue a signal that
can be used by management, ESG scores are successively transformed into
Z-scores, then ratings, as shown in the table below:
Min Z-score Max Z-score Rating

2.5 A
1.5 2.5 B
0.5 1.5 C
-0.5 0.5 D
-1.5 -0.5 E
-2.5 -1.5 F
-2.5 G
On the basis of these ratings, SRI managers must abide by these constraints:
t the overall portfolio ESG rating has to be over 0.5 (C rating)
t the overall portfolio ESG rating has to be over the benchmark rating
t the portfolio cannot be invested in stocks with an ESG rating below -0.5 (E, F
and G rated stocks)
II - SRI for investors

what added value?
2.1 Best-in-class and use of SRI ratings in stock picking
(from long shorts to model portfolios)
Because the choice of criteria included in ESG ratings is based on their possible
impact on the businesses economic performance, it is reasonable to expect
that portfolios that pick their stocks according to ESG criteria may outperform.
However, the demonstration of this potential outperformance comes up against
difficulties with the rating method.
One of the most traditional - and simple - ways to test the added value of a

signal is to prepare a long short portfolio in which you are buying the best-rated
stocks and selling the worst-rated. The ESG rating of companies is done with a
best-in-class approach within the sectors. As a result, you cannot establish a
hierarchy between two companies belonging to different sectors using an ESG
rating. Therefore, the sector biases resulting from the construction of a portfolio
and must be neutralized when the long short is created.
Amundi has chosen to keep geographic biases in the ESG ratings, because they
reflect differences identified by analysts for a philosophical reason: belonging
to a geographic area that is less successful on ESG criteria does not, within a
sector, excuse a company that is less attentive to those criteria. From a financial
performance standpoint, it is thus foreseeable that a long short portfolio
significantly reflects the geographic biases present in the ratings. Then there
is the issue of how these geographic biases are handled in the performance
analysis of the long short portfolio.
By way of illustration, the graph below shows the performance of an SRI
long short global equity portfolio, its systematic performance and its specific
performance.
LS portfolio - ESG performance & breakdown
140
120
100
80
60
01-2008 07-2008 01-2009 07-2009 01-2010 07-2010 01-2011 07-2011 01-2012 07-2012
ESG Systematic ESG Specific ESG

The two curves - total and systematic - are closely correlated, which alerts us
to the importance of these geographic biases in this long shorts performance.
It is hard to justify attributing the relative out- or underperformance of the
geographic areas to differences in ESG ratings. There are too many economic,
cyclical and political factors influencing such out- or underperformance.
Thus we have also carried out an ex ante risk analysis of the long short portfolio.

ESG - Ex-ante risk breakdown
100%
80%
60%
40%
20%
0%
01/08/2008 09/08/2008 05/08/2009 01/08/2010 09/08/2010 05/08/2011 01/08/2012 09/08/2012
Specific Country Other systematic factors

The graph above shows that the geographic allocation effect explains between
50% and 90% of the ex-ante risk. As such, this effect is quite dominant, and there-
fore totally masks the effect of the SRI in the long shorts gross performance.
For bond portfolios, additional challenges emerge in the construction of a long
short. Depending on the maturity of the bonds, sensitivity to rate fluctuations is
highly variable. A long short portfolio, created without precautions, is very likely to
be exposed to interest rate risk, moreover, with an unstable magnitude over time.
Furthermore, many issuers have several bonds with different characteristics; then
the question arises of choosing and weighting the bonds. An alternative is to use
fixed-maturity CDs instead of bonds, but the credit risk is very different depending
on the issuers rating, and varies in a much broader spectrum than on equities.
Thus a long short portfolio, even if built with CDs, has every chance of showing
strong systemic biases that are unstable over time.
We think the potential added value of ESG criteria is more likely to be detected in
the performance after stripping out the systematic effects, at the level of security
picking. Several methods can be used to measure this added value:
A. Create long short portfolios by geographic area, for equities

This is not a solution within Europe, because we see significant disparities in ESG
ratings between European countries (Northern Europe vs. Southern Europe).
Moreover, if we restrict to a geographic area, the ratings are no longer centered
within the sectors, which makes it more difficult to establish neutral portfolios on a
sector level. In any case, this solution cannot be transposed to systematic biases
other than geographic.
B. Remove the systematic biases of the long short by performance attribution
This method is preferable to a lack of treatment, but only removes the biases that
are specified straight away. It is suitable if the bulk of the systematic risk comes from

clearly identifiable geographic biases. Any systematic biases other than geographic
may persist, and it must at least be ensured that their contribution to long short risk
is minimal. The analysis done on the governance criterion in Section 2.3.2 shows
how the presence of a size bias - not initially detected - gives a misleading measure
of that signals added value. This method cannot be transposed in the case of bond
portfolios, where systematic biases are too numerous and unpredictable to be
managed in a performance attribution.
C. Control systematic biases in advance by building model portfolios
It is reasonable to assume that if the contribution of systematic biases to ex ante
risk is low, their potential impact on a portfolios outperformance will be low with
regard to pure stock picking. This can be achieved by building portfolios that
explicitly ensure systematic risk control via the objective function of optimization
or via the constraints. This method has its own drawbacks as the tested signal
may be affected by the constraints introduced during the optimization process,
as we will see in Section 2.4 on transfer coefficients. For instance, building a
benchmarked long-only portfolio while not allowing short positions means that
the signal cannot be exploited in case of poorly-rated small securities (because
they are virtually impossible to underweight). However, this method does have two
advantages: the first is to control systematic risk factors, whatever they are, since
we can specify that systematic risk must not exceed a certain % of the total risk.
The second is that the test recreates the conditions in which the signal is actually
applied in management. The test focuses on the portion of the signal that can
actually be used, which is more relevant for our clients. Moreover, it is possible,
to introduce turnover constraints or transaction costs into the simulation, in order
to see how much they affect results. This is the most general method and it can
be applied for both equities and bonds.
2.2 Does the constraint of being SRI penalize management?

2.2.1 Minimum TE: what is the threshold for different SRI indexes
(World, Europe, North America, Pacific)?
Measuring the minimum tracking error (TE) induced by an SRI process is critical
information regarding the development of SRI. Such information represents the
cost of SRI implementation from a risk perspective. If this minimum TE is low, there
is little risk turning a classic index into an SRI-compliant index while integrating
ESG convictions. On the contrary a high TE would be the sign of a greater risk
to undergo in order to be compliant and would then be less favorable for SRI
advocates. Additionally we have previously seen that geographic biases are of
paramount importance and as such further tests need to be carried over different
geographic zones.
To determine the minimum level of TE induced by transforming some well-known

indexes (MSCI World, MSCI Europe, MSCI North America and MSCI Pacific)
into portfolios respecting Amundis SRI rules, minimum TE optimizations are
performed through a risk-model, BarraOne, over the January 2005 June 2013
period.
The only constraints imposed to the optimizer are those to be respected to make
a portfolio compliant with Amundis SRI rules, meaning that at each and every
rebalancing date:
t the overall portfolio ESG rating has to be over 0.5
t the overall portfolio ESG rating has to be over the benchmark rating
t the portfolio cannot be invested in stocks with an ESG rating below -0.5
As the following chart indicates, the minimum ex-ante TE for an SRI compliant
process is low on average, but unsurprisingly it depends on the geographical
area:
ESG compliant portfolios ex-ante minimum TE (%)

2.50 2.50
2.00 2.00
1.50 1.50
1.00 1.00
0.50 0.50
0.00 0.00
World North America Pacific Europe
Median Source: Amundi Research
For Europe and World we can state that SRI compliance is consistent with low
tracking error processes, with respectively 0.30% and 0.55% on average.
However, this is no longer the case for the North American and Pacific zones where
minimum TE levels are higher, above 1%, and even reach 2% TE levels in some
specific conditions.
Ex-post levels of TE are globally in line with the ex-ante observations.
TE ex-post ESG
World 0.60%
North America 1.07%
Pacific 1.21%
Europe 0.35%
In general the implied TE of an SRI compliant portfolio is relatively low, in particular

for Europe and World. But more attention has to be paid to the American and

Pacific areas where risk implications cannot be ignored.
Even though the different TE levels are relatively low, they are not necessarily stable
over time. Specific market conditions or large capitalization exclusions may lead to
uncomfortable TE levels that we need to monitor.
ESG compliance portfolios ex-ante minimum TE (%)

2.5
2.0
1.5
1.0
0.5
0.0
01/01/05
01/04/05
01/07/05
01/10/05
01/01/06
01/04/06
01/07/06
01/10/06
01/01/07
01/04/07
01/07/07
01/10/07
01/01/08
01/04/08
01/07/08
01/10/08
01/01/09
01/04/09
01/07/09
01/10/09
01/01/10
01/04/10
01/07/10
01/10/10
01/01/11
01/04/11
01/07/11
01/10/11
01/01/12
01/04/12
01/07/12
01/10/12
01/01/13
01/04/13
-------- World North America Pacific Europe Source: Amundi Research
Looking at the previous chart, it is interesting to focus on some specific situations,

such as the MSCI North America optimization which suffers from the biggest hikes
and declines. Two periods are of particular interest:
t the strong and sudden increase in mid-2008 (1)
t the sharp decline in 2009 and the rebound that comes after in 2010 (2)
(1) Mid 2008, during the crisis, the large increase in ex-ante TE (+1.26%) was due to
the combination of different events:
Date TE
08/31/2008 1.12%
01/31/2009 2.38%
t new exclusions, most of which were large capitalizations making benchmark

replication more complicated (in January 2009, 39% of companies under
coverage were excluded versus 34% in August 2008). The exclusion of the
following companies contributed 0.54% to the TE increase over the period.
%CR %CR
Ptf weight Ptf weight ESG ESG Total Total
to Active to Active
Company as of as of score score risk risk
Total Risk Total Risk
08/2008 01/2009 08/2008 01/2009 08/2008 01/2009
08/2008 01/2009
Company 1 1.83% 0.00% -0.15 -0.60 23.1 37.1 0.09% 5.12%
Company 2 1.22% 0.00% 0.05 -0.57 37.1 65.4 0.06% 5.36%
Company 3 1.21% 0.00% 2.32 -0.70 43.6 101.8 0.05% 2.61%
Company 4 0.93% 0.00% 0.03 -1.18 43.4 59.1 0.02% 5.61%
Company 5 0.92% 0.00% 1.38 -0.53 36.9 67.2 0.06% 4.21%

t higher market volatility made this rise even bigger.
t some already excluded companies whose volatility sharply increased along
with their benchmark weights due to a strong relative performance compared
to the index. Such combination led to an additional TE increase of 0.55%.
Weight in Weight in %CR %CR
Total Total
the bench the bench to Active to Active
Company risk risk
as of as of Total Risk Total Risk
08/2008 01/2009
08/2008 01/2009 08/2008 01/2009
Company 6 1.37% 1.76% 26.17 43.2 7.11% 6.57%
Company 7 3.26% 4.81% 23.83 39.6 29.04% 33.51%
(2) What happened in 2010, with a sudden reduction of the TE level, is almost the
opposite of the previous situation:
Date TE
01/31/2009 2.38%
01/31/2010 0.57%
t the overall market volatility decreased leading to lower volatility of individual

stocks, in particular for large capitalizations.
t the exclusion rate was lower than it used to be (exclusion rate of 18% in the US in
January 2010 compared to 39% in January 2009) as many previously excluded
large companies were then rated above the exclusion threshold.
Company ESG score 08/2008 ESG score 01/2009 ESG score 01/2010
Company 1 -0.15 -0.60 -0.57
Company 2 0.05 -0.57 0.53
Company 3 1.38 -0.67 -0.04
Company 4 2.32 -1.18 -0.04
Company 5 0.03 -0.53 -0.85
Company 6 -0.97 -0.88 0.60
Company 7 -0.67 -0.96 0.53
t conversely, in late 2010 the minimum TE partially rebounded as some of the
previously mentioned companies were excluded again through qualitative
over-writing, bringing back the exclusion level to the 2009 situation.
To limit such TE fluctuations over time and avoid such ins and outs by large frontier
stocks, Amundi decided to further validate exclusions through a monthly ESG
rating committee with the following objectives:
t minimization of movements by large stocks companies that were alternatively
excluded and included due to small changes in their average ESG score;
t in-depth understanding of the ESG fundamentals for any of the excluded
stocks;
t specific follow-up for problematic issuers.

2.2.2 Minimum TE: what is the impact on performance?
We previously measured the impact of applying an SRI process from a risk

perspective through the minimum implied TE.
It is also worth measuring the impact from a performance point of view of creating
such a minimum TE portfolio using ESG scores.
For the same four geographic investment zones we calculated 1-year information
ratios (IR) and computed t-stats to assess the significance of these ratios. None
of them turned out to be significantly different from 0 at an acceptable level of
confidence.
IR 1 Y 2005 2006 2007 2008 2009 2010 2011 2012 t-stat

World 0.73 -0.55 0.51 -2.00 0.70 1.42 -0.04 -0.41 0.12
North America 0.38 -1.10 1.00 -2.07 0.54 0.86 -0.30 -0.75 -0.48
ESG
Pacific -0.19 1.47 -0.93 -1.31 1.23 0.94 0.47 0.56 0.79
Europe 0.24 -0.56 -0.36 -0.52 0.10 0.55 -1.50 -0.35 -1.36
Computing similar statistic tests on the full sample period IRs does not prove more
significant.
ESG IR 2005-2012 t-stat

World -0.14 -0.39
North America -0.26 -0.73
Pacific 0.32 0.91
Europe -0.38 -1.06
Therefore the relative performance of a minimum TE SRI portfolio cannot reasonably

be proven as different from 0. As a consequence the only cost of an SRI compliant
investment process with a minimized TE is the level of TE itself, as the added value
is neutral.
Minimum TE - ESG relative performance
120
115
110
105
100
95
90
85
80
01/01/05
01/05/05
01/09/05
01/01/06
01/05/06
01/09/06
01/01/07
01/05/07
01/09/07
01/01/08
01/05/08
01/09/08
01/01/09
01/05/09
01/09/09
01/01/10
01/05/10
01/09/10
01/01/11
01/05/11
01/09/11
01/01/12
01/05/12
01/09/12
01/01/13
01/05/13
-------- World North America Pacific Europe Source: Amundi Research

For a minimum TE SRI compliant portfolio the impact on performance is neutral,
whatever the geographic area.
2.3 The added value of SRI
We have seen so far that making an index SRI compliant proved to be neutral from
a performance perspective. We now investigate whether an investment process
aiming to maximize the ESG profile of a world developed markets equity portfolio
under some constraint of tracking error may improve performance relative to
a standard benchmark. While in the previous exercise we tried to quantify the
impact of ESG constraints on performance, we are now focusing on a more direct
exploitation of ESG rating.
From January 2005 to June 2013, with a monthly frequency, we maximize a
classical risk-return utility function as:
Utility=ESG - 2
Where ESG is the weighted average ESG score of the portfolio, is the risk aversion
coefficient, and is the tracking error of the portfolio relative to its benchmark.
The risk aversion parameter describes the tradeoff between the average ESG
rating of the portfolio and the relative risk of the portfolio: the lower the investor
tolerance to relative risk, the lower its ambition should be in terms of ESG rating;
the higher the portfolios required ESG rating, the higher the relative risk that the
investor should accept. In order to assess the impact on performance (as well as
on portfolio characteristics) of a different target tracking error level, we run several
simulations allowing the risk aversion parameter to vary.
Every month of our sample period from January 2005 to June 2013, we maximize
our utility function applying a risk aversion of 9 for the low tracking error portfolio,
and a risk aversion of 3 and 1 for the average and high tracking error portfolios,
respectively. The risk aversion parameters have been calibrated in order to obtain
average TE levels of roughly 2%, 3%, and 4.5% respectively.
Obviously, as in the previous simulation on tracking error minimization, we continue
to impose Amundis SRI rules:
t the overall portfolio ESG rating has to be higher than or equal to 0.5;
t the overall portfolio ESG rating has to be higher than benchmark rating;
t the portfolio cannot be invested in stocks with an ESG rating below -0.5.
We are aware that the ESG signal may be biased toward some styles and some
geographical areas. As we want to capture the specific component of the ESG
signal rather than some implicit (and involuntary) allocation effect, we use a risk
budgeting constraint: the common factors component of risk is limited to 10%,
leaving the remaining 90% of the risk budget available for specific risk.

We have computed relative returns, realized tracking error and information ratios,
for the full sample of our simulation and on an annual basis. As shown in the
table below, full sample information ratios are very close to zero and t-stats are
not statistically significant. Interestingly we notice that realized tracking error
are in line with target levels, calibrated through the choice of the risk aversion
parameter.
ESG IR 2005-2012 t-stat TE ex-post

min TE -0.14 -0.39 0.60%
AV = 9 -0.13 -0.36 1.99%
AV = 3 0.26 0.72 3.07%
AV = 1 0.26 0.72 4.32%
Although some observations may differ substantially from zero on an annual basis
(table below), none of the results turns out to be significant at an acceptable level
of confidence. The t-stats in the previous table have been computed over 100
monthly returns, while those in the table below have been computed over eight
annual information periods.
IR 1 Y 2005 2006 2007 2008 2009 2010 2011 2012 t-stat
min TE 0.73 -0.55 0.51 -2.00 0.70 1.42 -0.04 -0.41 0.12
AV = 9 0.60 0.33 -3.39 -0.37 1.42 -0.82 -1.06 -0.70 -0.98

ESG
AV = 3 0.45 1.08 -1.70 -0.54 2.35 -0.43 -0.74 0.03 0.14
AV = 1 -0.11 1.31 -2.40 -0.97 3.24 -0.29 -0.33 0.27 0.15
We can infer from this analysis that whatever risk aversion we use, the added value
of maximizing the ESG rating is neutral and does not depend on the level of tracking
error. Thus, we should expect an ESG portfolio to deliver performance in line with
that of a classic portfolio and its benchmark, at least with a tracking error ranging
from 2.0% - 4.5%.
2.3.1 Various SRI criteria (ESG / E / S / G - World)
However, stating that the added value of maximizing the ESG rating on a global equity
portfolio is neutral does not necessarily imply that none of the environmental /social/
governance (E/S/G) stand-alone ratings have some relationship with performance,
or that none of the stand-alone ratings or the aggregate ESG rating is neutral in any
specific geographical area.

For this reason, we first test the ESG signal on the North America, Europe, Euro
zone and Pacific regions separately, with a risk aversion of 3. We than repeat
the test with stand-alone components in geographical areas and in the World
developed markets as a whole.
Coherently with a back test on World developed markets, the cumulative value
added of the ESG score is very close to neutral, as well as on the main four
geographical areas.
AV3 - ESG relative performance

110
105
100
95
90
01/01/11
01/04/11
01/07/11
01/10/11
01/01/06
01/04/06
01/07/06
01/10/06
01/01/07
01/04/07
01/07/07
01/10/07
01/01/08
01/04/08
01/07/08
01/10/08
01/01/09
01/04/09
01/07/09
01/10/09
01/01/10
01/04/10
01/07/10
01/10/10
01/01/12
01/04/12
01/07/12
01/10/12
01/01/13
01/04/13
01/01/05
01/04/05
01/07/05
01/10/05
World North America Pacific Europe Euro Source: Amundi Research
In order to check the performance neutrality of each of the stand-alone E/S/G

components as well, we run 12 additional back tests (on specific geographical
areas and on the World developed markets separately), maximizing the same
utility function and applying the same constraints as in the previous case. This
time we maximize the E/S/G scores respectively instead of the overall ESG score;
similarly we impose exclusions based on any stand-alone component each time.
Again, we set the risk aversion parameter to 3, corresponding to an average target
tracking error of 3%.
Most of the information ratios are positive, even though none of them are statistically
significant (only the environmental criterion in Europe passed a t-test at a very weak
20% confidence level).

E IR 2005-2012 t-stat
World 0.09 0.25
Pacific 0.14 0.40
Europe 0.41 1.16
S IR 2005-2012 t-stat
World 0.29 0.83
North America 0.37 1.04
Pacific 0.10 0.27
Europe -0.24 -0.69
G IR 2005-2012 t-stat
World 0.34 0.95
Pacific 0.14 0.39
Europe 0.25 0.70
On an annual basis, some information ratios may be substantially different from

zero; however, none of them turns out to be statistically significant at an acceptable
level of confidence.
IR 1 Y 2005 2006 2007 2008 2009 2010 2011 2012 t-stat

World 0.40 2.32 -2.79 -0.43 1.47 0.24 -1.73 0.56 0.01
North America 1.06 1.62 -2.53 -1.33 1.10 -0.79 -1.15 -0.20 -0.54
E
Pacific -1.21 0.57 0.31 1.67 0.51 -0.81 -1.35 -0.22 -0.18
Europe 0.29 0.08 -0.95 0.88 1.02 1.26 -0.88 0.97 1.09
World 1.95 1.65 -1.34 -2.24 1.10 0.44 0.47 0.50 0.62
North America 3.28 0.35 -0.57 -0.56 1.02 -0.20 1.03 0.60 1.40
S
Pacific 0.25 -0.59 -1.28 0.51 0.59 1.33 -0.91 0.74 0.26
Europe 0.53 1.42 -1.00 -2.15 0.49 -0.19 -0.59 -0.25 -0.57
World 2.64 1.22 -0.69 -0.75 1.36 0.02 -0.04 -1.16 0.71
North America 1.52 0.73 -0.93 -0.76 0.45 -0.76 -0.48 -0.72 -0.37
G
Pacific -0.04 0.22 -0.18 -0.37 1.42 0.04 0.35 -0.66 0.44
Europe 0.88 1.28 0.66 -0.33 1.21 0.79 -2.08 -0.20 0.70
As a conclusion, we can state that whatever the extra-financial criteria and the
geographic areas we investigate, the added value of the SRI signal is neutral.
Combining our finding on ESG value added with those on typical tracking error
of an ESG equity portfolio construction, we can infer that investors may achieve
full compliance with SRI rules at the costs of reasonable tracking error (especially

in Europe and in the World developed markets), and may even tilt the portfolio
towards SRI criteria, while being confident that relative performance versus a
standard market benchmark will likely be neutral.
2.3.2 Focus on Governance (World developed markets)

While addressing the ESG investment process, many academic studies focused
on governance as it often appeared to be the ESG component delivering the best
returns. These findings have usually been considered intuitive at some instance,
since the governance rating is somehow addressed as the most economic
among the ESG components. The intuition being that good governance helps to
avoid operational and reputational risk, and that governance is linked to a general
effective functioning of the corporate structure.
In previous studies we have also documented a persistent and significantly positive
impact of the governance filter on performance.
On the other hand, our latest empirical evidence only partially confirms these
findings. As shown in some previous tables, in Europe, in the Pacific, as well as
in the global strategy, the cumulative active returns of a strategy based on the G
rating turn out to be positive (while in North America the cumulative active return is
slightly negative), even though it is not statistically significant in any of these areas.
What is apparently contradictory among these two different findings is explained by
the different utility function that we have recently employed in our back-tests, and
more precisely by the risk budget constraint that we have introduced: contribution
from regional, industry, and style allocation must not exceed 10% of total tracking
error.
In the chart below we present a back-test on the world developed markets, without
any risk budgeting constraint (in line with back tests presented in previous studies).
AV3 G relative performance, without risk budgeting

130
125
120
115
110
105
100
95
90
r-2 5
Ju 05
O 005
Ja 005
Ap 06
6
O 006
Ja 006
Ap 07
7
O 007
Ja 007
Ap 08
8
O 008
Ja 008
Ap 09
9
O 009
Ja 009
Ap 10
0
O 010
Ja 010
Ap 11
1
O 011
Ja 011
Ap 12
2
O 012
Ja 12
Ap 13
3
Ap v-0
00
00
00
00
01
01
01
01
0
20
20
20
20
20
20
20
0
20
l-2
-2
r-2
l-2
-2
r-2
l-2
-2
r-2
l-2
-2
r-2
l-2
-2
r-2
l-2
-2
r-2
l-2
-2
r-2
l-2
-2
r-2
n
n-
n-
n-
n-
n-
n-
n-
n-
ja
ct
ct
ct
ct
ct
ct
ct
ct
Ju
Ju
Ju
Ju
Ju
Ju
Ju
Total Active Currency Industry Style Specific Return Source: Amundi Research

In this case, up to 70%-80% of the active performance and its variability is explained
by common factors. Consequently, most of the cumulative performance is not
purely ESG specific but comes from an implicit (and most of the time involuntary),
regional, sectorial, or style allocation.
The table below reports that all the common factor groups have a positive
impact on performance but, among them, the only significant component is
style with an average annual active return of 0.96% (out of 2.80%), a full sample
IR of 0.86 and a t-stat of 2.50 (significant at a 1% confidence level).
Although not significant, country/industry and currency effects explain 0.90%
of annual outperformance while the pure specific Governance component
is positive, but rather weak (less than one fourth of the total active return is
specific) and quite volatile, thus not statistically significant.
Component Ann. AR IR t-stat

Total Active 2.80% 0.92 2.67**
Currency 0.52% 0.42 1.21
Industry 0.37% 0.24 0.71
Style 0.96% 0.86 2.50**
Specific 0.67% 0.27 0.79
** indicate return is significant at the 1% level
For more details, in the table and in the chart below we rank stand-alone styles
according to their contribution to performance.
Styles contributions to cumulative active performance over the 2005-2013 period,

without risk budgeting
12.0%
10.0%
8.0%
6.0%
4.0%
2.0%
0.0%
e
th
ue
ity
ge
ld
l
ta
ilt
er
rit
iz
rs
tio
ur
ie
w
tu
id
To
al
ra
at
S
ea
ve
th
os
ro
Y
ia
en
qu
V
ve
ol
O
ar
G
ni
in
xp
om
V
Li
Le
U
-L
V
on
M
n
gn
ng
io
N
at
ei
ni
e
im
r
ar
Fo
iz
st

-E
on
N

Common Factor Cumulative AR IR t-stat
Size 5.96% 0.80 2.33**
Volatilty 1.52% 0.42 1.21
Growth 1.51% 0.67 1.93*
Value 1.07% 0.78 2.28**
Liquidity 0.44% 0.33 0.95
Non-Estimation Universe 0.32% 0.64 1.86
Earnings Variation 0.23% 0.44 1.29
Others 0.06% 0.08 0.22
Foreign Exposure -0.01% -0.01 -0.02
Size Non-Linearity -0.45% -0.59 -1.72
Leverage -0.71% -0.38 -1.10
Momentum -0.96% -0.40 -1.15
Yield -1.27% -0.36 -1.04
* and ** indicate return is significant at the 5% and 1% levels respectively
By far, the best contributor to performance is the size factor (the portfolio is exposed
to small caps, with small cap risk factor performing extremely well over the last 8 years).
Some additional positive contribution comes from low risk stocks (negative
exposure to volatility), and growth (negative exposure, significant contribution).
Contribution from value is also significant (especially in the US and Australia).
Among the worst performers, yield and momentum have limited impact and are
not statistically significant.
If we impose a risk budget constraint on common factors, the picture changes
dramatically. Limiting the common factors to 10% of the total ex ante risk, the
cumulative active return of the Governance portfolio is much lower, and interestingly
it fits the unconstrained portfolios specific component very well.
G - Comparison of cumulative active returns

130
125
120
115
110
105
100
95
90
6 7 8 9 0 11 11 11 11 12 12 12 12 13 13
05 05 05 05 06 06 06 0 07 07 07 0 08 08 08 0 09 09 09 0 10 10 10 1
v- 20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 -20 -20 -20 -20 20 -20 20 -20 20 -20
jan pr- Jul- ct an- pr Jul- ct an- pr Jul- ct an- pr Jul- ct an- pr Jul- ct an- pr Jul- ct Jan Apr Jul Oct an- pr Jul- ct an- pr
A O J A O J A O J A O J A O J A O J A O J A
Specific no risk budgeting Total with risk budgeting Source: Amundi Research

Empirical evidence on governance is thus actually weaker than its consensus in
the literature would suggest.
In addition, empirical evidence may be time varying as well. Here we consider
the two sub-periods 2005-2009 and 2010-2012: the signal G is not statistically
significant, but it works well up to 2009 and quite poorly thereafter.
G IR 2005-2009 t-stat G IR 2010-2012 t-stat

World 0.44 0.99 World -0.47 -0.82
North America -0.06 -0.14 North America -0.67 -1.15
Pacific 0.20 0.45 Pacific -0.08 -0.14
Europe 0.53 1.19 Europe -0.49 -0.85
We believe that the governance filter has gained popularity in academic studies as
a well performing signal, thanks to both a favorable timeframe up to 2009-2010,
and thanks to some bias that should be taken into account properly. All in all, an
investment process that is compliant with governance criteria may be achieved
at a cost of low tracking error, and without negative impact on performance, but
we cannot state that the process is likely to benefit from superior performance.
2.4 Signal utilization within optimized portfolios

2.4.1 The average portfolio ESG score
We have seen that ESG investing is not expensive in terms of risk and is neutral as
regards to performance, especially in Europe and in the global developed markets.
Consequently there is no relationship between the added value of an ESG process
and the target tracking error level of the portfolio that maximizes the average ESG
score. However, as far as we allow higher tracking error, we can come out with a
higher average portfolio ESG score.
ESG score (max min and median) and risk aversion

2.50 2.50
2.00 2.00
1.50 1.50
1.00 1.00
0.50 0.50
0.00 0.00
minTE AV9 AV3 AV1

We might conclude that, as far as the impact on performance is null, and as far
as the relationship between TE and average portfolio ESG rating is positive, the
higher the TE the better the portfolios ESG profile.
We first have to keep in mind that higher tracking error is an explicit cost, as it
necessarily implies a higher variability of active returns themselves, thus a higher
probability of extremely negative (though unlikely) outcomes. Also, we may agree
that the average ESG score is a quick and transparent criterion to address the
ESG quality of a portfolio, yet we believe that some more clever indicators may be
used for this purpose.
In our view, a good ESG portfolio should respect some compliance criteria (such
as not being invested in very negatively ESG rated stocks), it should provide an
average ESG score that is higher than the standard market index, but should
also assure a generalized coherence between the ESG score and the portfolio
composition. In a well optimized ESG portfolio, the contribution to risk of any
constituent should be related and ideally proportional to its contribution to the
portfolios average score.
2.4.2 The Transfer Coefficient
The transfer coefficient is computed as the correlation between the alpha

employed in the optimization (ESG score) and the marginal contribution to risk of
any asset in the portfolio. Theoretically, in an unconstrained portfolio maximizing
a classical mean-variance utility function, the marginal benefit of any asset (its
alpha, its expected return, or its ESG score in our example) should be equal to its
marginal cost (the marginal contribution to risk). The correlation between marginal
cost and marginal benefit should thus be perfect, but in practice this is rarely the
case because mean-variance optimizations bear many constraints.
A general constraint applying to the vast majority of the investment process is
the long only constraint: optimized portfolios are generally underweighted on
negatively rated stocks (as this benefits the average ESG score). With no short
constraints, the optimal negative active weight on some of these stocks could
ideally be much higher (in absolute terms) than the weights of these stocks in
the benchmark, thus leading to some short positions in the portfolio. When short
positions are forbidden, under-weights are limited to the weight of the stocks in
the benchmark, the process is thus not exploiting the full potential of the signal,
and the transfer coefficient decreases.
The long-only constraint alone often prevents the transfer coefficient from
exceeding 75%.
In addition to that, all the other ESG constraints (exclusions, minimum average
ESG score, risk budget constraints, etc.) further reduce the transfer coefficient.

As we can see in the following chart, the impact of those constraints on portfolio
compositions is not the same for any utility function (minimizing tracking error or
maximizing average score), nor for any level of tracking error.
Transfer coefficient for different risk aversions/TE

1.00 1.00
0.90 0.90
0.80 0.80
0.70 0.70
0.60 0.60
0.50 0.50
0.40 0.40
0.30 0.30
0.20 0.20
0.10 0.10
0.00 0.00
minTE AV=9 AV=3 AV=1
While an investment process aiming to be ESG compliant while minimizing TE has

an average transfer coefficient of 0.45 (and ranging from a minimum of 0.37 to a
maximum of 0.52), portfolios optimized with an ESG maximization objective and
a risk aversion of 9, have an average transfer coefficient of 0.5 and more or less
the same variability as the minimum tracking error process: adding an ESG rating
maximization objective to the utility function (and accepting some higher level of
tracking error) allows the optimizer to better exploit the signal. But this is true only
up to a point: when risk aversion is equal to 3 and then to 1, the average transfer
coefficient is lower and its variability is higher.
In order to better understand how differently the three optimization methods
exploit the signal, we split the ESG signal into three categories: good (ESG notes
A, B & C), average (D) and bad (E, F & G), and we compute an average transfer
coefficient over the three groups separately. We then plot the minimum tracking
error portfolio first, and just one of the ESG maximizing portfolio thereafter, as we
will show that the three of them have similar profiles.

Minimum TE - Transfer coefficient breakdown into ESG score levels
10%
Marginal Contribution to Active Total Risk
5%
R = 0.00
Good
0%
R = 0.00 Average
R = 0.00
Bad
-5%
-10%
-4.00 -2.00 0.00 2.00 4.00
ESG score
As for the minimum tracking error portfolio, the average transfer coefficient is
very low among the three groups. Low transfer coefficients among any groups of
stocks, however, may result in a (artificially) positive transfer coefficient overall.
This depends on the highly asymmetrical distribution of the scatter plot of the
marginal contribution to risk and the ESG score. The negative contribution to risk
of stocks with an ESG rating lower than -0.5 (thus excluded, thus with negative
active weights, thus with negative contribution to risk), gives the charts an overall
positive slope (positive information coefficient).
The picture changes completely if we investigate one of the ESG maximizing
portfolios. Below we show the case for risk aversion equal to 3.

AV3 -Transfer coefficient breakdown into ESG score levels
10%
5%
R = 0.32
R = 0.00
R = 0.00 Good
0%
Average
Bad
-5%
-10%
-4 -2 0 2 4
ESG score Source: Amundi Research
Here the transfer coefficient is flat among bad and average stocks, while
it is significantly positive among good stocks. The positive overall transfer
coefficient is thus explained by an excellent transfer coefficient among good
stocks: the stocks that contribute the most to the portfolio tracking error are
those with the best ESG score (ideally those with the more pronounced positive
active weights).
Generally speaking, transfer coefficient is a correlation measure, and it is very
explicative when dealing with two normal distributions. When leading with
asymmetrical distribution (as the marginal contribution to risk of stocks in a long
only portfolio), we should better investigate the transfer coefficient among several
partitions of the scatter plot.
In order to allow for some generalization we compute transfer coefficients at any
date of our sample period, for any portfolio, and among any group of stocks. We
plot their median in the following chart.

Transfer coefficients for different risk aversions / TE
1.00
0.50
0.00
-0.50
Good Average Bad
minTE AV9 AV3 AV1 Source: Amundi Research
The ESG compliant portfolio with the lowest possible TE has a different profile with
respect to the three optimizations with an explicit ESG maximization objective.
These latter are quite similar to each other, since all of them better exploit the
signals relative to the minimum tracking error, especially among well-rated stocks.
Among them, the lower the tracking error, the better the transfer coefficient is.
In order to gain coherence between portfolio composition and stocks ESG rating,
an ESG process should not be limited to pure compliance, but should rather allow
for some further ESG rating improvement, even though it should ideally remain in
a low tracking error range.
However, the validity of this assumption may depend on how ambitious the ESG
compliance constraints are. If our investment universe has an average ESG rating
above our minimum threshold (0.5), and if in our investment universe there are only
a few small companies that must be excluded because of their negative rating,
ESG compliance would be respected by simply investing in the benchmark without
any further ESG improvement needed. On the opposite hand, if the average ESG
rating of our benchmark is much lower than our minimum threshold, and if many
stocks are excluded because of their negative rating, targeting ESG compliance
implies a significant ESG improvement, as it forces the optimizer to choose well
rated stocks to reach the minimum threshold.
In order to distinguish between periods where ESG compliance is straightforward
and periods where ESG compliance is harder to accomplish, we have monitored
the average ESG score of the minimum tracking error portfolio. We consider periods
where the average ESG rating is higher than our threshold, as easy periods, that
is periods where the average ESG score is naturally above the threshold. In the
same way, we consider periods where the average ESG rating is equal to our
thresholds as difficult periods as the optimizer matches the threshold constraint
while minimizing the tracking error.

We compute the transfer coefficient in difficult periods only and we compare the
different portfolios.
Transfer coefficient in "difficult" periods for different risk aversions / TE

1,00
0,80
0,60
0,40
0,20
0,00
Good Average Bad
minTE AV9 AV3 AV1 Source: Amundi Research
When the ESG profile of the investment universe is poor, and compliance is
an ambitious goal to accomplish, a minimum tracking error process forces the
optimizer to exploit the good and average stocks even better than in each of the
three processes with an explicit ESG maximization objective.
Minimum TE portfolio - Transfer coefficient breakdown in "difficult" periods
10%
5%
R = 0.27 R = 0.95
R = 0.00 Good
0%
Average
Bad
-5%
-10%
-6 -4 -2 0 2 4 6
ESG score

As a general rule, in order to exploit the ESG score of well rated stocks, we should design
a low tracking error process with some ESG maximization target, rather than a pure
compliance goal, coupled with a straight tracking error minimization. However, we should
be aware of how far from being compliant our investment universe is. If compliance is hard
to accomplish, we could favor minimum tracking portfolios, since in these circumstances,
the latter better exploit the ESG signals (at least from a transfer coefficient perspective).
2.5 Performance of SRI signals in corporate bond markets

2.5.1 Bond portfolios
At Amundi the usual framework of a best-in-class fund within the fixed income
universe is a bond picking process. A bond picking process consists of choosing
bonds for their intrinsic value and not for any bet on style, region or any other
systematic asset allocation. Thus such a process supposes a minimization of the
portfolios active systemic risk compared to its market proxied by a benchmark.
Put another way, the main active risk source is the bonds specific part of the risk.
By keeping the sector and regional weights neutral, such a process is in line with
the best-in-class idea of comparing each company to its direct competitors.
The asset selection process is a maximization of the forecast intrinsic value of each company.
In a socially responsible fund it comes down to integrating ESG ratings as expected
returns. This supposes that these criteria have an intrinsic value leading to a source of
alpha. Our bond picking process uses a maximization of the following utility function.

Where is a vector containing the weights of the bonds, expo is a vector containing
the exposures of all bonds to the risk factors, r is the vector of expected returns, VCV
is the covariance matrix of the systematic risk, is a matrix containing the specific
variances on the diagonal, is the risk aversion and the specific risk multiplier.
The specific risk multiplier is set to a very small value so that we have an almost
free floating active specific risk. Indeed, at any given point in time in our portfolio
simulations, more than 90% of the active risk is idiosyncratic.
We use a relatively high risk aversion that assures a very low active systematic risk
while keeping a high level of the portfolio ESG ratings. We optimize each portfolio
on a monthly basis between January 2010 and July 2013. Our benchmark is the
Merrill Lynch Large Cap corporate bond index. This index only contains investment
grade bonds. Furthermore, we use the Merrill Lynch Industry Level 3 that classifies
all issues according to 16 different sectors.
The modified duration and the option adjusted spread of each sector of the optimized
portfolio must be within +-5 basis points of their benchmark counterparts. We set
the same constraint on the entire portfolio. Furthermore, we set a constraint of the
weight of each currency of +-5 basis points. We track currency risk closely and

notice that it is negligible at all times (less than 5% of the very low active systematic
risk). No issue and no issuer may exceed one and two per cent in terms of weight
of the portfolio, respectively. Furthermore, we restrict the portfolio to 300 assets.
All optimizations are carried out with BarraOnes optimizer.
By looking at the graph we see that the portfolio containing European bonds
outperforms its counterparts by an annualized 0.4%, not a significant outperformance.
Indeed, the active performance of the North America and Asia/Pacific portfolios
is almost zero. On a global scale, the World portfolio exhibits a statistically non-
significant outperformance of 0.3% (annualized).
ESG Cumulative relative performance

102.5
102
101.5
101
100.5
100
99.5
99
98.5
01/01/10
01/03/10
01/05/10
01/07/10
01/09/10
01/11/10
01/01/11
01/03/11
01/05/11
01/07/11
01/09/11
01/11/11
01/01/12
01/03/12
01/05/12
01/07/12
01/09/12
01/11/12
01/01/13
01/03/13
01/05/13
01/07/13
------ ESG Europe ESG US ESG World ESG Asia/Pacific Source: Amundi Research
As one can see in the table below, the outperformance of the European portfolio is
mainly driven by the factors E and G. The portfolio optimized with E as the expected
return shows a non-significant outperformance of 0.4%. The portfolio optimized with G
as the expected return exhibits a significant outperformance at the 10% level of 0.6%.
Zone ESG E S G
Ann Act Ret 0.30% -0.20% -0.30% 0.20%
World IR 0.47 -0.31 -0.41 0.29
T-stat 0.9 (-0.58) (-0.77) 0.55
Ann Act Ret 0.40% 0.40% -0.20% 0.60%
Europe IR 0.6 0.65 -0.31 0.88
T-stat 1.13 1.23 (-0.59) 1.67
Ann Act Ret 0.00% -0.10% -0.30% -0.10%
US IR -0.05 -0.3 -0.36 -0.07
T-stat (-0.10) (-0.57) (-0.68) (-0.13)
Ann Act Ret 0.00% 0.10% 0.10% 0.00%
Asia/Pacific IR 0.13 0.23 0.3 0.06
T-stat 0.25 0.44 0.56 0.11
By looking at the graph, of all portfolios optimized with the factor G as the expected
return, we see that the governance factor has an impact on the performance only
in Europe.

G - Cumulative relative performance
104
103
102
101
100
99
98
97
01/01/10
01/03/10
01/05/10
01/07/10
01/09/10
01/11/10
01/01/11
01/03/11
01/05/11
01/07/11
01/09/11
01/11/11
01/01/12
01/03/12
01/05/12
01/07/12
01/09/12
01/11/12
01/01/13
01/03/13
01/05/13
01/07/13
-------- Europe US World Asia/Pacific Source: Amundi Research
The optimized portfolios incorporating the ESG rating do not exhibit any significant
outperformance. Only the portfolio containing European bonds incorporating the
governance factor has a significant positive outperformance.
2.5.2 Bond spreads

We run panel regressions to identify a potential link between our ratings and the bond
spread. We use the same period and the same data as for our back tests on bonds.
NB: The panel is unbalanced because of some bonds entering and exiting the
index. On the one hand, some unobserved variables such as management
capability are likely to affect the spread. On the other hand, the market as a whole
moves according to macroeconomic news. It is thus important to control both
dimensions. We opt for fixed effects on the industry level and time effects on the
monthly level. We calculate cluster robust standard errors on the industry level. The
panel is built according to the following function.

We use control variables specific to the issue and to the issuer. All issuer specific
data stems from FactSet. The control variables specific to the issuer are the
following: The Size is measured by the natural logarithm of the total assets of the
firm. Large firms are widely perceived as less risky and thus benefit from a lower
cost of debt. The Leverage is defined as total liabilities over total assets. It indicates
how much debt a firm has. The higher the leverage ratio, the riskier the firm. A
high leverage ratio should increase the cost of debt. The ROA is the accounting
return on assets. It represents the profitability of the firm and thus the ability to
pay back its debt. The Capital Intensity is the ratio of fixed assets to total assets.
Since the fixed assets could be claimed by a creditor in case of a default, a high
capital intensity should decrease the level of the spread. Loss is a dummy variable
that equals 1 if the firms net income before extraordinary items is negative in the
current and prior fiscal year. The control variables specific to the bond issue are

the Modified Duration and the Nominal. The Modified Duration is positively linked
to the spread since a bond with a higher maturity is perceived as riskier. We use
the Nominal as a proxy for liquidity. Large issues tend to be more liquid than small
ones. Larger issues should have a lower spread than small issues. Moreover, we
integrate dummy variables as a control for financial ratings.
The table below shows the results of a panel regression for all regions of the spreads
natural logarithm on the ESG rating and the control variables. Almost all control
variables are statistically significant and have the expected sign. Moreover, the adjusted
R squared is above 0.7, except for Europe where it is 0.57. Our control variables seem
to explain the biggest part of the spread and are in line with the academic literature.
The ESG rating has no significant impact in any region except for Europe where we
interestingly observe that a high level of ESG integration increases the cost of debt.
Observations included 16197 5361 9003 2255

Cross-sections included 660 227 344 67
Dependent Variable log(Spread) log(Spread) log(Spread) log(Spread)
Independent Variable World Europe North America Asia Pacific
ESG -0.012 0.061*** -0.006 -0.01

(0.014) (0.019) (0.019) (0.046)
ROA -0.004 0.005 -0.008** -0.01

(0.003) (0.003) (0.004) (0.017)
Loss Indicator 0.125*** 0.110*** 0.173*** 0.108
(0.042) (0.035) (0.064) (0.098)
log(Size) 0.01 -0.019 -0.034 -0.166**
(0.021) (0.026) (0.046) (0.072)
Capital Intensity -0.558 -0.542 0.294 0.377
(0.465) (1.174) (0.376) (1.345)
Leverage 0.006*** 0.005** 0.004 0.010***
(0.002) (0.002) (0.003) (0.004)
log(Nominal) -0.298*** 0.024 0.091** -0.397***
(0.026) (0.042) (0.038) (0.021)
Modified Duration 0.046*** 0.033*** 0.052*** 0.029**
(0.005) (0.005) (0.005) (0.015)
Industry Fixed Effects Yes Yes Yes Yes

Period Fixed Effects Yes Yes Yes Yes
Adjusted R-Squared 0.76 0.57 0.72 0.74

*, ** and *** indicate that the regression coefficient is significant at the 10%, 5% and
1% levels respectively. The standard deviation is in parentheses.
In the table below, we separate the environmental, social and governance
performance and we still do not observe any impact on the world level.
By looking only at Europe, we see a positive link between the spread and the social
rating as well as the governance rating. Both are significant at the one per cent
confidence level.
In North America no statistic is significant. We observe that the governmental
performance is negatively linked to the spread, but since we only observe a t-stat
of 1.59 and the panel is quite large, we cannot conclude that it is significant.
There is no evident link between the spread and the ESG ratings in Asia/Pacific.
Dependent Variable log(Spread)

Independent Variable Coefficient Standard Deviation
World
E 0.006 0.013
S 0 0.017
G -0.015 0.019
Europe
E 0.029 0.021
S 0.054*** 0.019
G 0.058*** 0.018
North America
E 0.018 0.015
S -0.009 0.022
G -0.039 0.026
Asia/Pacific
E -0.032 0.024
S -0.005 0.037
G 0.068 0.059
Industry Fixed Effects Yes

Period Fixed Effects Yes

*, ** and *** indicate that the regression coefficient is significant at the 10%, 5% and
1% levels respectively.
Only in Europe, we find a link between the cost of debt of a firm and the
environmental, social and governance ratings. Interestingly, the cost of debt
increases with a better ESG performance. It seems that the spreads of well rated
firms in Europe decrease during the 3 years of our backtests. This is mainly driven
by the factor governance. Portfolios using the factor governance as expected
return outperform their benchmark in Europe. Event though the overperformance
portfolios optimized using the ESG ratings is positive in Europe, it is not significant
according to our statistics.

Conclusion
This paper presents our viewpoint on the impact of SRI on management
performance. The measurements are taken using the SRI approach as seen by
Amundi. On the companys ESG ratings, the Amundi approach is a best-in-class
approach in the sectors, but not in countries or geographic areas. The ratings
depend on data by criteria coming from several outside providers, but Amundi
makes a choice about the weighting of the most appropriate ESG criteria for
each sector. Next, SRI portfolio management is based on exclusion criteria for
the worst-rated securities, as well as on the portfolios overall ESG rating, in
absolute and relative terms compared to its benchmark and its universe.
For large capitalizations, excluding the worst-rated securities may lead to
significant as well as unstable active risk, if these securities fall on either side
of the exclusion limit. Therefore, Amundi set up qualitative tracking of these
borderline securities at a Rating Committee meeting.
In terms of outperformance, in our historical backtests, we do not find any
significant added value, neither positive nor negative. Depending on the
periods, the geographic areas and the components of the signal, we can see
positive results, but they are not statistically significant. The good outcomes we
have seen in the past on governance are largely due to inadequate correction
of systematic biases (size effect). This means that while performing the tests,
the portfolios must be built limiting every component of the systematic risk
(geographic and size mainly).
Being SRI also has no significant cost in terms of risk, and for Europe or the
world, the tracking error that results from the constraints that are used to SRI-
ize a portfolio remains very limited compared to that of active management.
However, for the Pacific and, to a lesser extent, the US, the cost of SRI in terms
of incompressible tracking error is higher.
Finally, depending on the type of management being done - minimalist SRI
management with simply some constraints or more committed management
that endeavors to optimize the portfolios ESG rating - the ESG signal is not
used in the same way. Too many constraints in terms of average ESG rating
mean that the signal is only used on certain categories of ratings. To obtain the
greatest consistency between the securities ESG ratings and the portfolios
over- or underweights, a careful analysis of the optimized portfolios must be
done. From this viewpoint, the portfolio with the highest ESG rating is not
necessarily the most SRI.
SRI portfolios do not underperform or outperform significantly over the assessed
period. In addition, from an incompressible risk standpoint the inclusion of SRI

criteria, as practiced at Amundi, does not produce a significant cost on either
the European or global investment universes. Furthermore, some SRI factors
are likely to become more important in the future, with differences in corporate
practices having considerable impacts on profitability. SRI management can
therefore be a relatively cost-free way to benefit from this evolution. Choosing
to invest in SRI products also involves non-monetary considerations in terms of
reputation and investors responsibilities to future generations.

Bibliography
Amenc, N. et Le Sourd, V., 2008. Les performances de linvestissement socialement

responsable en France. Edhec Working Paper.
Bauer, R. et Hann, D., 2010. Corporate Environmental Management and Credit Risk.
Document de travail, European Centre for Corporate Management. Maastricht University.
Chen, H., Kacperczyk, M., Ortiz-Molina, H., 2007. Do non-financial stakeholders affect
agency costs of debt? Evidence from unionized workers. Working Paper, University of British
Columbia.
Clarke, R., Harindra de Silva, Steven Thorley, 2002. Portfolio Constraints and the
Fundamental Low of Active Management. Financial Analyst Journal, Vol. 58, n5 (September/
October), 48-66.
Fama, E., Jensen, M.C. 1983. Separation of ownership and control. Journal of Law and
Economics 26, 301-325.
Friedman, M., 1970. The Social Responsibility of Business is to Increase its Profits. The New
York Times Magazine, September 13.
Freeman, R.E., 1984. Strategic management : a stakeholder approach (Pitman Series in
Business and Public Policy). Pittman, Marshfield, MA.
Freeman, R.E., Harrison, J.S., Wicks, A.C., 2007. Managing for stakeholders : Survival,
Reputation, and Success. Yale University Press, New Haven CT.
Goss, A., Roberts, G.S., 2011. The impact of corporate social responsibility on the cost of
bank loans. Journal of banking & finance, 2011, vol 35, issue 7.
Jensen, M.C., 2002. Value maximization, stakeholder theory, and the corporate objective
function. Business Ethics Quarterly, 12, 235-256.
Jensen, M.C., 1976. Theory of the firm : managerial behavior, agency costs and ownership
structure. Journal of Financial Economics 3, 305-360.
Kacperczyk, A., 2009. With greater power comes greater responsibility? Takeover protection
and corporate attention to stakeholders, Strategic Management Journal, 30, 261-285.
Margolis, J., Elfenbein, H. et Walsh, J., 2007. Does it pay to be good? A meta-analysis
and redirection of research on the relationship between corporate social and financial
performance. Ross School of Business - University of Michigan.
Menz, K.M., 2010. Corporate social responsibility : Is it rewarded by the corporate bond
market? A critical note. Journal of Business Ethics 96, 117-134.
Orlitzky, M., Schmidt, F. et Rynes, S., 2003. Corporate social and financial performance :
A meta-Analysis. Organization Studies, 24, 403-441.
Renneboog, L., Ter Horst, J., Zhang, C., 2008. The price of ethics and stakeholder
governance : the performance of socially responsible mutual funds. Journal of Corporate
Finance, 14 : 302-322.
Sharfman, M., Fernando, C., 2008. Environmental risk management and the cost of capital.
Strategic Management Journal, 29, 569-592

DP-02
Risk-Free Assets:
What Long-Term Normalized
Return?
Sylvie de LAGUICHE,
Head of Quantitative Research
March 2014
This article offers some insight into determining long-term expected

returns on risk-free assets which are then used as a basis for risky
asset forecasts and strategic allocations. It first examines the
notion of risk-free assets in themselves in order to demonstrate
that, even if regularly renewed short-term investment does present
some relevant characteristics, there is in fact no such thing over the
long term. It then analyses the impact of different economic and
financial theories on short-rate forecasts and includes a series of
tests over a long-term period from 1930 to 2013 in the United States.
The simple operational solution derived from these findings consists
in a compromise between:
- the current short rate given the strong autocorrelation between
short rates (25%),
- inflation, which avoids endless extrapolation from either very
restrictive or extremely accommodating monetary policies linked
to a specific backdrop (25%),
- long rates over the full horizon as the yield curve provides some
information on future short rates (50%).

RISK-FREE ASSETS:
WHAT LONG-TERM
NORMALIZED RETURN?
I - Why this question?

When building a strategic allocation over the long term, investors assess an assets
appeal according to its risk-return trade-off. To do so, they often reason in terms
of its expected excess return in excess of a risk-free investment. One could,
therefore, be forgiven for thinking that risk premiums on assets are the only thing
that matter and that the risk-free rate is of little importance.
This is not, however, always the case since long-term solvency forecasts or
accurate simulations for a product or portfolio very often require a hypothesis in
terms of cash returns which is then used as a basis in determining the profitability
of assets by adding a risk premium. This report focuses on long-term horizons of
at least 5 to 7 years.
Normalized returns are taken to mean returns that are coherent but that are not
based on any macroeconomic forecasts. The aim is to establish a set of rules
that use market information that is readily available when determining normalized
returns. First we analyse the notions of cash and risk-free rates, then we review
the most commonly-used approaches to establish the normalized returns on
cash. We use historical data from the US to ascertain how accurate these different
approaches are, and offer a compromise that is easy to implement.
II - Cash, a risk-free rate that really

does mean no risk?
Determining whether or not cash is entirely risk-free depends on the investment
horizon and instruments in play, and should prompt us to question the notion of
risk itself which can take a number of forms over a long investment period. For
investors who choose not to spend in the immediate term but to invest over the
long term, the risk is perceived as the uncertainty linked to what the money they

have invested will be worth in the future. This uncertainty can have several causes.
The first thing we naturally think of is the volatility of the markets. However, over the
long term, volatility is not the only source of risk. There is also the risk of erosion
via inflation, the risk of default, systematic risk and, even in the absence of these
elements, the uncertainty linked to the future level of rates means there is a risk of
opportunity at the time of the reinvestment of coupons, dividends, or repayments.
Over longer horizons, the hierarchy of risks can also differ from over the short term.
Over the long term, the risk of reinvestment is far from negligible when it
comes to cash. The graph below shows the standard deviation in the annualised
return on cash over rolling periods of 1, 3, 7, 10 and 30 years. The calculations
apply to the US between 1929-2013.
Dispersion of annualised cash return according to horizon: 1929-2013

15%
10%
5%
0%
1 YR 3 YR 7 YRS 10 YRS 30 YRS
Dispersion of cash return Volatility of a classical asset class yielding the same dispersion
It is clear that, over long and forward-looking timeframes, cash is not without its
risks. If, for each horizon, we determine risk in the form of an equivalent normal
volatility which offers the same dispersion (i.e. by multiplying it by the root of the
horizon), we see that equivalent volatility increases.
This means that, over a 30-year horizon, the risk of reinvestment linked to cash
is high. In fact, dispersion is similar to that of an asset with a volatility of 12%
(namely around 60-80% of equities). While the natural response, to avoid this risk
of reinvestment, is therefore to define zero-coupon investments with a nominal rate
over the horizon in question as risk-free, this definition poses several problems.
First, although it is true that nominal returns are known, this is no longer the case
if we reason in real terms. A zero-coupon investment over the long term is
seriously exposed to the risk of erosion by inflation. As such, a regularly
renewed short-term investment (i.e. for which there is a certain adjustment in short
rates in relation to inflation) is therefore less risky in real terms.

Dispersion of annualised real return - US
3.5%
3%
2.5%
2%
1.5%
1%
0.5%
0
1929-2013 1959-2013 1989-2013
Cash ZC buy and hold Source: Amundi Research
Second, depending on the issuer, such an investment may carry a credit risk.
Before the crisis, and at least for strong currencies, the tendency was to consider
government bond rates as risk-free. Within the eurozone, the sovereign debt crisis led
to a serious discrepancy in government borrowing rates depending on the country,
much in line with the drop in ratings for peripheral countries. Unless we accept the
highly debatable view that returns on assets in euros would be different according
to country, a common reference curve is a fundamental requirement. Swap rates are
of course an option as their collateralisation means they incur very little credit risk.
Having said that, the crisis in confidence in the banking system and the increase in
Libor rates led to a distortion in swap rates, linked not to the risk of these instruments,
but rather to the very strong distortion of the Libor 3-month rates used as a reference
for variable rate swaps. One remedy was to replace the Libor 3-month rate by a daily
rate that was therefore less distorted. OIS rates (overnight interest swaps) do indeed
appear to bear less credit risk even if there are no long-term time series on OIS rates.
Libor and OIS 10y swap rate: Euro

6 0.6
5 0.5
4 0.4
3 0.3
2 0.2
1 0.1
0 0
01-07 01-08 01-09 01-10 01-11 01-12 01-13
OIS 10yrs euro Libor 10yrs swap Difference (right hand scale)
Lastly, if we compare the risky nature of a cash investment in relation to a ze-

ro-coupon investment, the latter is exposed to a capital risk in the event of early
redemption.

These findings suggest that there are in fact no universally risk-free assets
over the long term. An investment decision must factor in the known or unknown
horizon and the outlook in nominal or real terms. Insofar as our goal is to use risk-free
investment returns as a reference to which we can then add a risk premium on assets,
we think it preferable to define this risk-free reference as what the markets appear
to consider as less risky as they are less demanding in terms of returns:
cash. In practise, proxies can be used as long as they do not present any substantial
risk in relation to a day-to-day investment, and they correspond more to what would
be achieved via a regularly renewed short-term investment. It is nonetheless prudent
to remember that, while this asset may bear the least risk, it is not entirely risk-free,
except over short-term horizons. Wherever possible, the rate for cash investments
must be close to an OIS rate over a short horizon or to a government rate as long as
the country has an AAA rating or genuine control over its currency.
III - Approaches
3.1 Macroeconomic equilibrium model
In a balanced economy, the interest rate at which the economy finances itself is
equal to the sum of the equilibrium values for (real) GDP growth and inflation. This
approach is particularly useful in establishing a reasonable target value for a short
rate associated with a short-term investment in the far distant future. This perfect
equilibrium is obviously not achieved today which is why this approach cannot be
used to directly forecast future cumulative returns for cash starting immediately.
Furthermore, the rate at which the economy finances itself is a mix of short and long
rates. Ascertaining the cumulative return on cash over a long period starting today
using this method means establishing a path between the current and target rates.
Finally, this method also requires an estimation for growth and inflation at equilibrium.
3.2 - Yield curve
The great merit of methods that use the yield curve is that they are based on observable
data and not on forecasts. A first solution is a pure and simple extrapolation of the
current short rate. The virtue of this solution is its simplicity, but it is only justified if
investors anticipate that rates will remain stable which will mean a flat yield curve. The
slope between long and short rates is, however, much too variable over time to make
it an acceptable hypothesis. Another approach is to use the zero-coupon rate over
the investment horizon as a normative forecast for the return on cash. This approach
is based on two arguments. The first is to say that, in nominal terms at least, the
return on a buy and hold bond investment is known in advance and can therefore
be considered as risk-free. We have seen that this point of view is debatable and, in
any case, in no way guarantees the link between the initial long rate and the return

on cash as the investment is not made on the same asset. The second argument is
to tie the return on cash to long rates on the basis that long rates are an indication of
the future change in short rates (Lutz [1940]), i.e. to say that the best forecast for the
behaviour of short rates is that they will evolve towards forwards. This also enables
us to define a coherent trajectory for short rates. Despite its simplicity, this model is
criticised for various reasons. The first is that it implicitly supposes that short rates
evolve in a deterministic manner and fails to factor in their random nature. In actual fact,
there is a risk premium between long and short rates which explains why the slope
of the yield curve is most often positive. The second is that short rates are essentially
administrated and do not factor in agents future forecasts. Long rates, on the other
hand, are market data which do factor in short rate forecasts but which are also
affected at certain times by other criteria (fly to quality on the German curve, regulatory
changes which affect the appetite of buyer pension funds for very long bonds).
3.3 Arbitrage pricing theory

This theory was presented for the first time by Black and Scholes to value options on
stocks. Over the last 30 years, countless variations have been applied to different
asset classes. The first to apply the theory to interest rates was Vasicek [1977] for
short rates, after which adjustments over the entire yield curve were introduced by
Heath, Jarrow, Morton [1989] and Hull and White [1997]. This theory is used as the
basis for the valuation of option derivatives and explicitly takes into account the
uncertainty linked to future rates. It also provides a framework which allows for
the clear separation, for long or forward rates, between what is linked to rates
forecasts and what constitutes a risk premium. The risk premium depends on
the implicit volatility of rates and a coefficient in terms of risk aversion which can be
mapped suing the Sharpe ratio for the asset class. For our purposes, this theory
is used to establish that future short rates are lower than forward rates. Correction
in absolute terms is the more significant so as volatility is high and risk aversion is
strong. As illustrated in the graph below, when it comes to current standard levels,
the size of the correction is significant, particularly for long horizons.
Expected annualised cash return - correction to be applied to spot ZC rate

1
11
16
21
26
31
36
41
46
0.00%
DIFFERENCE MODEL - SPOT
-0.10%
-0.20%
-0.30%
-0.40%
-0.50%
-0.60%
-0.70%
-0.80%
-0.90%
-1.00%
Horizon (yrs)
HJM Source: Amundi Research

Over a given horizon, the level of volatility affects the risk premium. The graph be-
low shows how the level of volatility on long rates affects the correction. All other
parameters were kept constant for these calculations. Over a 7-year horizon, the
effect becomes really significant when volatility of long term rates is high. Today,
for a volatility of around 0.5%, the correction is around 35 basis points.
Correction to be applied to initial spot ZC rate in order

to forecast annualised cash return as a function of volatility
0.00%
Difference model - spot ZC
-0.20%
-0.40%
-0.60%
-0.80%
-1.00%
-1.20%
%
1%
2%
25
50
50
0.
0.
1.
Volatility
While these approaches are interesting, they require complex calculations and the
calibration of a substantial number of parameters. It is nonetheless useful to keep
in mind the implications of this theory.
IV - Historic analysis of long series in the US

We have examined, over a long period of time, the relationship between long and
short rates in the US which is the country for which we have the longest historical
series, and whose long and short rates are not too distorted by credit risk and
the government administration of foreign exchange rates. This is more delicate
when it comes to the eurozone where, in the 1980s and 1990s, short rates
were affected by German reunification and the currency crises in the European
monetary system.
Forecasts based on the initial short rate alone are not very credible, except
over short horizons. We looked at several horizons ranging from 3 months
to 30 years. For the 7-year horizon, for example, we calculated the cumulative
performance over 7 years of a short-rate investment and compared it with the short
rate at the start of the period - the difference constituting the forecast error. To
ascertain whether the use of the initial short rate is actually relevant, we compared
the dispersion of the forecast errors with the actual unconditional dispersion of the
annualised returns on cash over the same horizon.
The following graph shows the difference between dispersion figures.

Dispersion of forecasting error using initial short term rate: 1929-2013
4%
3%
2%
1%
0
3 months 1 YR 3 YRS 7 YRS 10 YRS 30 YRS
Observed dispersion Dispersion of forecasting error
Source : Amundi Research
Unsurprisingly, over short horizons, the dispersion on forecast errors is much

lower than the dispersion on the actual unconditional returns for cash. In fact, it
is virtually nil which is how cash earned its reputation as a risk-free asset. This
is not verified over long horizons for which it is therefore much less important to
know the initial short rate.
We examined whether it is more useful to know the initial long rate, and the
long rate for which we have very long series is the 10-year bond yield. However,
because a coupon bond with a 10-year maturity has a shorter duration (around
7 years), we examined the cumulative returns on a short-rate investment over
7-year periods, and analysed the forecast errors made using the initial long rate
for each sub-period.
The dispersion is a little lower than with the initial short rate but is still strong.
Dispersion of cash performance

3,5 %
Annualised dispersion
3%
2,5 %
2%
1,5 %
1%
0,5 %
0
Observed Knowing initial Knowing initial
short terme rate long term rate
Horizon 7 yrs Source : Amundi Research
Here, we employ a 7-year horizon as an approximate duration for the long rate
used. Our aim is to identify the relationships between the forecast errors using
the short rate or the initial long rate and other variables.

In the first instance, the graph below shows that when we use the initial short
rate alone, this forecast error is linked to the slope.
Forecasting error on cash return and initial yield curve slope: 1930-2013
7.00%
FORECASTING ERROR
5.25%
3.50%
1.75%
0%
-1.75%
-3.50%
-5.25%
SLOPE
-7.00%
-2.00% -1.00% 0% 1.00% 2.00% 3.00% 4.00%
VForecasting error using initial short term rate Linaire (forecasting error using initial short term rate)
Source : Amund Research
While this relationship is significant and robust, it is also variable over time with a
sensitivity of around 50% on average.
More specifically, we limited our analysis to periods where initial conditions are
not fundamentally different (initial inflation between 0% and 5%). The graph below
shows the relationship between the forecast errors and the slope depending on
whether we use the long rate or the initial short rate.
Forecasting error on cash return and initial slope: 1929-2013 with filter
8,00
FORECASTING ERROR
6,00
4,00
2,00
0
-8,00 -6,00 -4,00 -2,00 0 2,00 4,00 6,00
-2,00
-4,00
-6,00
-8,00
-10,00
SLOPE -12,00
Foracasting error using initial long term rate Forecasting error using initial short term rate
Linaire Linaire
(forecasting error using initial short term rate) (foracasting error using initial long term rate)
What we find is that there is a positive relationship between the forecast error
using the initial short rate and the slope. Using the initial long rate on its own
creates an error with a negative and often downward bias in relation to the slope.
This means that the initial long rate globally overestimates the future return on
cash, all the more so when the slope is steep.
The graph below shows that there is also a link with the volatility of rates.

ForFcBsting error on cash return and initial volatility: 1929-2013 with filter
6.00%
FORCASTING ERROR
4.50%
3.00%
1.50%
0%
-1.50%
-3.00%
-4.50%
INITIAL VOLATILITY OF LONG TERM RATE
-6.00%
0 0.75 1.50 2.25 3.00
VForacasting error using initial long term rate Linaire (foracasting error using initial long term rate)
Using the initial long rate means we overestimate future returns, all the more so
when volatility is high. The scope of the phenomenon is also significant since,
for 1% volatility, using the initial long rate overestimates the return on cash by an
average 0.8%. While this is consistent with the arbitrage pricing theory, it is even
higher than what is predicted by the theory with a Sharpe ratio of 0.3. There are
two possible explanations here. The first is that the decline in rates during the
period in question meant that the Sharpe ratio was higher. The second is that the
increase in risk aversion during periods of strong volatility appears to accentuate
the phenomenon.
Given todays extremely positive curve and the low volatility in rates, using short
and long rates together therefore seems more effective than using either one or
the other on their own. The correction linked to volatility, which involves complex
calibration, is less useful.
We can also see that the forecast error is linked to the real short rate.
Forecasting error on cash return and initial real rate: 1929-2013 with filter
7.00%
FORECASTING ERROR
5.25%
3.50%
1.75%
0%
-1.75%
-3.50%
INITIAL REAL RATE
-5.25%
-6.0% -4.0% -2.0% 0% 2.0% 4.0% 6.0% 8.0%
VForecasting error using initial short term rate Linear (forecasting error using initial short term rate)
Source: Amundi Reseach
Mitigating the short rate with inflation means we can improve the forecast as we
avoid having to extrapolate a short rate level, distorted due to a very restrictive or

accommodating monetary policy, which, in the long term, is not consistent with the
fundamentals.
Our forecast model mixes the short and long rates and initial inflation. This enables
us to eliminate a substantial part of the dependency between the forecast error
and the slope and real initial rate, even if it does remain highly sensitive to a change
in inflation regime. The graph shows that there is a strong relationship with the
variation between initial and cumulative inflation.
Forecasting error on cash return

and change in inflation: 1929-2013 with filter
6.00%
FORECASTING ERROR
4.50%
3.00%
1.50%
0%
-1.50%
-3.00%
OBSERVED - INITIAL INFLATION
-4.50%
-5.3% -3.5% -1.8% 0% 1.8% 3.5% 5.3% 7.0%
VForecasting error using mixture Linaire (forecasting error using mixture)

The preferred alternative, where available, is forward inflation.

We have found no link between the forecast error and growth.
V - Operational implications
In order to marry operational simplicity and coherence with theory and observation,
we recommend using a combination of short rates (25%), inflation (25%) and
zero-coupon long rates (50%) over the horizon as a normative forecast for
returns on short-term investment.
The initial long rate segment (50%) factors in the data on the future trend in rates
extracted from the yield curve.
The initial short rate segment (25%) reflects the fact that, on average, short
rates are lower than their forecast using long rates and present a strong level of
autocorrelation.
The inflation segment (25%) is used to correct the distortion between short rates
and the macroeconomic fundamentals linked to a temporary episode of very
accommodating or very lax monetary policy.
Today, we recommend using the OIS curve rates or, failing that, government
bond rates where governments have an acceptable rating or are in control of their
currency.

7 year annualised rolling cash return and forFcBst at begining of period (with filter)
12%
ANNUALISED RETURN
10%
8%
6%
4%
2%
0%
1930
1940
1949
1956
1960
1964
1968
1983
1987
1992
1996
2000
2004
2008
2012
Annualised cash return over 7 yrs Forecast
We then suggest extracting the forward short rates from the curve obtained for the
future performance of cash using the same mechanism as that used to calculate
forward rates.
This rule should however be reviewed in the event of:
t a notable change in inflation forecasts,
t a significant increase in the volatility of long rates,
t an inversion in the yield curve.

Acknowledgements
I would like to thank my colleagues Jean Gabriel Morineau, Gianni Pola and Eric
Taz-Bernard at Amundi for their valuable input in improving this paper.
REFERENCES
Black F, Scholes M [1972] The Valuation of Option Contracts and a Test of Market Efficiency,
Journal of Finance, Vol. 27.
Heath DC, Jarrow BA, Morton A, [1992] Bond Pricing and the Term Structure of Interest
Rates: A New Methodology for Contingent Claims Valuation, Econometrica, Vol. 60.
Hull JC, White A [1997] Options, Futures and Other Derivatives, 3rd Edition, Prentice-Hall.
Musiela M, Ruthowski M [2004] Martingale Methods in Financial Modelling, 2nd Edition,
Springer.
Ltz FA [1940] The Structure of Interest Rates, Journal of Economics, Vol 55.
Modigliani F, Shiller R [1973] Inflation, Rational Expectations and the Term Structure of
Interest Rates, Economica, Vol 40.
Vasicek O [1977] An Equilibrium Characterization of the Term Structure, Journal of Financial
Economics, Vol 5.


WP-034
Managing Uncertainty with DAMS.

Asset Segmentation in Response
to Macroeconomic Changes
Gianni Pola,
Balanced Quantitative Research, Amundi
May 2013
A novel challenge in strategic asset allocation looking at assets

as vehicles of more fundamental factors offers a new language
to decipher financial markets. This new way forces us to rethink
asset segmentation in terms of macroeconomic and market
stress scenarios. Traditional approaches look at nominal bonds,
commodities and equities as representative of (respectively)
deflation, inflation and growth. We show that this interpretation
is not adequate: asset classes do not constitute good axes and
a rotation in an abstract three-dimensional space is needed in
order to get reliable proxies. We firstly analysed a large investment
universe in the US (139 assets including traditional and alternative
investments) and then we segmented them in order to get the best
hedge for each macroeconomic and market stress environment.

0. Introduction
A new approach in asset allocation consists in looking at asset classes as vehicles of more
fundamental factors. According to this method, fundamental factors govern the majority of
asset class dynamics, and hence asset allocation should be rephrased in terms of risk
allocation of fundamental factors. This approach allows portfolio managers to relate their
portfolio to factors risk premia. Whether the factors approach is superior to the traditional
asset class method is still controversial.
Indeed defining factors in the market is not obvious: common practice consists, for example,
in identifying nominal bonds, commodities, and equities as proxies of (respectively) deflation,
inflation and growth. Two complementary approaches have been developed in order to detect
fundamental factors: statistical approaches and economic scenario methods. The main virtue
of the former, mainly based on principal component analysis, is to provide (by-construction)
orthogonal axes, the main benefit of the latter consists in identifying meaningful and stable
factors that are easily related to macroeconomic dynamics and relevant financial indicators. In
this work we pursue the second approach.
The factors approach forces us to rethink asset segmentation. Taxonomy of assets is

traditionally related to the asset type (e.g. bonds, stocks, commodities), or mainly to risk
arguments (e.g. assets in traditional personal-financial-planning programs are clustered
together according to their CVaR) and correlation measures (correlated assets are considered
similar). The recent financial crisis clearly showed the limitations of these approaches. What
is relevant for us in asset segmentation is to bring together assets that exhibit similar
behaviour in response to factors dynamics.
The key words in the title of the manuscript are dams and macroeconomic changes. We
briefly clarify what we mean by them below.
Dams is an acronym and stands for Diversification Across Macroeconomic Scenarios, and it
corresponds to an investment process in place at Amundi Italy since December 2011 to design
strategic asset allocations. In the last decade the overconcentration of portfolio risk on equity
markets in traditional pension fund allocations 1 led to poor performance, large draw-downs,
and slow recovery; the traditional bond-equity portfolio is implicitly designed for disinflation
and rising growth scenarios. The aim of DAMS is to provide a more balanced allocation,
1
60% equity 40% bond portfolio allocation implies more than 90% risk concentration on equities.

limiting risk in a wrong assessment of the future macroeconomic environment. This document
constitutes the foundation of the DAMS investment process 2.
The second key words are macroeconomic changes. Standard approaches investigate
relationships between asset returns and levels of macroeconomic variables (e.g. high or low
inflation). In this document we follow a different approach. We strongly believe that asset-
return dynamics can be mostly explained by variations of expectations, rather than the levels
themselves of the macroeconomic variables 3: markets move based on shifts in conditions
relative to the conditions that are priced in (Bridgewater). We individuated three factors that
are particularly relevant in determining asset prices: inflation, growth and market stress.
Growth and inflation are crucial because the value of an investment is mainly affected by the
volume of economic activity (growth) and its pricing (inflation). Indeed we added the market
stress because it often plays a major role in asset dynamics, being more relevant than purely
macroeconomic variables like inflation and growth, as e.g. in 2008 when financial stress was
mainly due to the liquidity problem. We identified six macroeconomic scenarios: rising
growth, falling growth, rising inflation, falling inflation, rising stress and falling stress, rising
and falling referring to changes in expectations of the fundamental variables.
Asset behaviour is represented through an abstract three-dimensional cube representing

polarization of assets to factors variations (see Chart 1, left panel). In order to simplify the
representation let us consider the two-dimensional projection along the market stress axis:
Chart 1 (right panel) sketches the polarization of traditional asset classes (nominal bonds,
inflation-linked bonds, equities, and commodities) to changes in expectations of inflation and
growth 4. Asset classes are placed about in the corners of the chart 5, meaning that they present
mixed behaviour. They do not constitute good axes, and a rotation in the plane is needed in
order to find proper proxies for fundamental factors. This heuristic approach is supported by
rigorous statistical tools (the polarization coefficient P).
2
In Pola and Facchinato (2013) we will discuss the portfolio construction and DAMS investment process.
3
We are aware that this is an approximation. In addition, as we recalled in Pola and de Laguiche 2012, asset
returns are certainly impacted by supply-demand effects, markets liquidity condition, market inefficiencies due
to investor irrationalities, and market frictions. However we prefer firstly to design strategic asset allocation
according to asset polarization to macroeconomic changes, and then to further refine allocations taking into
account the above mentioned effects.
4
The closer the assets are to the borders of the box, the more significant their response to variations of
macroeconomic variables; conversely the closer to the centre the more uncertain their response.
5
The chart indicates that: nominal bonds polarized to falling inflation and/or falling growth scenarios; equity to
falling inflation and/or rising growth scenarios; commodity to rising inflation and/or rising growth scenarios;
inflation-linked bonds to rising inflation and/or falling growth scenarios.
6

Chart 1a. Polarization cube Chart 1b. Two-dimensional projection
We investigated asset segmentation through empirical analyses. The point of view is that one
of a US investor who is managing a portfolio of domestic and international investments
including traditional and alternative asset classes; in total we investigated 139
assets/strategies. The aim of the paper is to segment them in order to identify the best hedge
for each macroeconomic scenario.
We conclude the section with a remark. In this work we investigate the relationship between
long positions of assets and fundamental factors. Short positions on assets are certainly to
reverse the relationship with factors, hence in principle providing interesting opportunity to
diversify. We work only with long positions because we want to be long on asset risk
premia 6. Indeed our horizon is long term (we deal with strategic asset allocation), and net
short positions may lead to structural negative risk premia over time.
Investigating the behaviour of asset classes to factors dynamics is particularly crucial today as
e.g. we expect inflation will become extremely relevant in the near future. The optimal
portfolio to hedge inflation risk has been studied by Atti and Roache (2008), Amenc et al.
(2009) in developed countries and Brire and Signori (2011) in Brazil.
The paper is organized as follows. In section 1 we briefly review traditional approaches for
asset segmentations. In section 2 we introduce the fundamental factors, and investigate firstly
6
The investment universe is made up mainly by long positions on assets/strategies that deliver a risk premium in
the long term. However we included also some assets/strategies that are not supposed to bear any long-term
return as fx-rates, equity spread indices (long equity sectors and styles and short the S&P500).

whether they are orthogonal, then if the relationship is stable over time. Section 3 illustrates
briefly the methodology. Section 4 shows the main empirical findings. Section 5 illustrates
some implications of the present study. A conclusive section delineates the take-home
messages. Annexes include details on database, technical notes on the Kullback-Leibler
distance, the definition of rising and falling scenarios, the polarization coefficient P
(parametric and non-parametric), tables illustrating all the empirical findings, and finally an
additional analysis section.
1. Traditional schemes for asset segmentation
Traditional approaches for asset segmentations consist in clustering together assets according
to:
x similarity of asset type (bonds, equities, commodities, etc.);
x risk argument: assets are mainly segmented according to their volatility, VaR, or CVaR;
x correlation measure: assets are similar if they are correlated (metric mainly based on
standard Pearsons coefficient).
The recent crisis highlighted some limitations of the above mentioned methods. Let us
consider for example the Italian BTP:
x The debt crisis in the Eurozone was characterized by the decoupling of the Euro core
bonds from the peripherals: despite the similarity of asset type between the Italian BTP
and Euro core bonds (e.g. German Bund), it is evident that their dynamics during the
debt crisis have been profoundly different.
x As shown in Pola and de Laguiche 2012, the volatility of BTP increased dramatically
during the debt crisis. Any classification of asset classes based on risk argument suffers
from the instability of asset volatility.
x Another effect of the decoupling of the Eurozone is the change in the correlation
between bond and equity markets: in Pola and de Laguiche 2012, we showed clearly
that the effect of the debt crisis on BTP was to make it positively correlated to the
equity market.
We worked in a different direction: our goal is to cluster together assets that exhibit similarity
in their response to changes in macroeconomic variables and stress indicator. Assets
8

behaviour will be represented according to their position in an abstract three-dimensional
cube, geometric proximity between assets indicating similarity in their response to factors
changes.
We do not claim here the superiority of the present approach on the above-mentioned
methods: our goal here is to investigate asset polarization, thereby posing the basis to build
more robust asset allocation (see Pola and Facchinato (2013)).
2. Fundamental factors
In this section we introduce the fundamental factors, and investigate their orthogonality over
time with empirical analysis in US.
2.1 Defining rising and falling scenarios
Whichever methodology is used to describe asset behaviour in terms of factors, it requires:
x firstly to individuate a set of factors (from statistics or from a priori decisions),

x secondly to verify if they are orthogonal 7, and
x finally to identify the relationship between factors and asset returns.
Two main approaches have been developed to address this issue: statistical methods and
economic scenario approaches.
Statistical methods are mainly based on principal component analysis (PCA). PCA allows one
to compute orthogonal directions (eigenvectors) under the covariance matrix metric. Principal
components are expressed in terms of long-short combinations of asset returns 8. Pros are that
factors are by-construction orthogonal, and that the relationship between factors and assets is
explicitly expressed by a transformation matrix (eigenvectors matrix). Cons are that
eigenvectors are sometimes counterintuitive, or at least difficult to relate to fundamental
macroeconomic variables. Moreover, updates of the covariance matrix make eigenvectors no
more orthogonal, and hence it requires finding new eigenvectors which diagonalize the
7
Independence among factors guarantees a more clear representation of asset price dynamics. However, as it
will be clarified in the following, a compromise is needed between orthogonality and the choice of meaningful
and stable directions. We prefer quasi-orthogonal stable directions which are simple to interpret, and that are
directly related to macroeconomic variables.
8
In general within the standard approach, there is no guarantee that principal components sum up to one (or
zero) and that the weights are positive (long-only constraints). Meucci (2009) performed PCA with constraints
on eigenvectors specified by linear-matrix inequalities; moreover he showed how the linear-matrix inequality
approach can locally describe more complex non-linear inequalities.

updated covariance matrix. In general, factors estimated with PCA are not stable over time
especially in case of strong changing regimes.
In the economic scenario methods, directions are ex-ante identified and are usually related to
macroeconomic variables, factors affecting asset prices (e.g. equity size, style, and liquidity),
and market stress indicators. The virtue of this approach is that directions are ex-ante
meaningful and that factors are by-construction stable over time; the main cons are:
x the choice of factors is arbitrary,

x independence should be verified (it is not a free-lunch as in the statistical approach), and
x the relationship between factors and assets is not obvious 9; it should be fixed either by
statistical inference, or by strong ex-ante assumptions.
Our research moves within the economic scenario framework. Fundamental factors are ex-
ante identified according to what, we believe, are mostly relevant in determining asset prices
(see Chart 2 and Annex A for more details on time-series):
x US inflation: US Consumer Price Index (seasonally adjusted),
x US growth: US real GDP and PMI (seasonally adjusted) 10,
x US market stress indicator 11.
Chart 2. Macroeconomic and stress indicators in US
9
Moreover we remark that, in general, the relationship between factors and assets is not even invertible. Usually
the number of assets is greater than the number of factors. If the linear relationship holds, transformation matrix
is not a square matrix, and thus not invertible, at least, in the canonical way. A generalization might be found
within the Moore-Penrose pseudo-inverse.
10
We included PMI in order to have a monthly evaluation of growth.
11
Growth and inflation themselves can represent stress in financial markets (e.g. unexpected inflationary
shocks). Market stress indicators signal financial stress coming from a broader set of drivers, including liquidity
issues.
10

Traditionally, dependency between fundamental factors and asset returns is assessed
comparing asset returns with fundamental factors levels (e.g. in Pola and de Laguiche 2012,
we question whether high inflation or low/negative inflation hurt asset returns). In this
document we follow a different approach. We believe that asset price movements can be
mostly explained in terms of variations of expectations of macroeconomic variables and stress
indicators: all asset classes have environmental bias (Bridgewater). Stocks move not
because of low or high growth but mainly because growth is above or below expectations.
Measuring expectations in macroeconomic variables like inflation and growth 12 is tricky,
moreover it is difficult to find a very deep historical database; hence we preferred to study the
polarization of asset returns towards historical variations of macroeconomic variables. Each
asset is evaluated since inception: our aim here is to individuate a long term relationship.
We performed the analysis with quarterly and monthly observations: this allows one to
investigate the relationship in different time-frames, and to increase the data sample. The
comparison between assets and factors is synchronous: we compared them in the same time
period (month or quarter); in Annex F we addressed the issue of introducing a temporal-lag
between the factor measurement and asset performance.
In order to represent factors changes, we preferred to follow a parsimonious approach: for
each fundamental factor we define rising and falling scenarios. More explicitly we related to
each factors time-series a Boolean time sequence (+1, -1), indicating in each time period a
rising or falling scenario 13 (see Annex C for more details). According to this procedure we
identify six scenarios: rising inflation, falling inflation, rising growth, falling growth, rising
stress, and falling stress. The aim of the present work is to segment asset classes with respect
to the six scenarios.
Chart 3 plots the alternating regimes between rising and falling scenarios for US inflation
since 1950. Despite the CPI index rising sharply from 23.61 in December 1949 to 231.75 in
October 2012, rising and falling scenarios have been about equally likely: rising and falling
scenarios express (respectively) acceleration and deceleration of fundamental factors; indeed
they deal with the second derivatives of macroeconomic variables.
12
Inflation swaps and ZEW indicators permit one to evaluate expectations on inflation and growth.
Unfortunately time-series are very short, and do not allow one to perform a robust statistical analysis.
13
It is worth mentioning that our point here is simply to name a specific time interval as rising or falling; the
evaluation is ex-post and we do not need to forecast market scenarios.

Chart 3. Rising and falling inflation scenarios in the US
2.2 Are fundamental factors orthogonal? Is the relationship stable over time?
The first issue to address is whether the identified fundamental factors are independent or
redundant. For the sake of simplicity let us name the four factors (CPI, real GDP, PMI, and
market stress) as f1, f2, f3, f4. We remind the reader that fundamental factors are expressed
by Boolean time-series of +1 and -1, hence the standard Pearsons correlation coefficient is
not applicable 14.
We quantify the dependency between fundamental factors as the distance between the joint
probability P(f1, f2, f3, f4) and the product of marginal probabilities P(f1)P(f2)P(f3)P(f4):
x if the distance is zero, factors are independent of each other;
x a non-zero distance signals a statistical relationship among some factors. The Kullback-
Leibler distance (KL distance in the following) permits one to quantify the distance
between two probability measures (Cover and Thomas 2006; Annex B for a brief
review), and hence the dependency among factors.
This approach does require large amount of data samples (see Annex B for a discussion on
data sampling); moreover it is not very appropriate for our sample because time-series
inceptions are not homogeneous across factors. For this reason we preferred to investigate
pair-wise orthogonality 15.
14
The standard Pearsons correlation coefficient measures the relationship between two continuous variables; it
cannot be applied to discrete random variables.
15
Indeed pair-wise orthogonality does not guarantee full independence among factors. But usually, pair-wise
relations are more relevant than higher order ones.
12

First of all we computed the probability of each scenario. Chart 4 summarizes the result:
The scenarios are well distributed; this means that rising and falling scenarios are about
equally likely (see Annex A for series inceptions).
In Chart 5 we compute the six pair-wise joint probabilities. Probabilities are estimated from
the frequency table. In order to have perfect orthogonality between factors, we expect to find
a joint probability close to 25%, this corresponding to rising and falling scenarios of the two
factors being completely unrelated each other. Among the computations, those involving the
stress scenario are less significant because of the short data sample.
The main findings are:
x CPI vs. real GDP. The equilibrium is slightly moved towards off-diagonal elements:
this means that rising (resp. falling) inflation scenarios are more likely linked to falling
(resp. rising) growth ones;
x CPI vs. STRESS. The off-diagonal figures are more likely than the diagonal ones;
x CPI vs. PMI. Almost uniformly distributed;
x Real GDP vs. STRESS. The most likely joint scenario is falling real GDP and falling
stress;
x Real GDP vs. PMI. The equilibrium is clearly moved towards the diagonal elements,
indicating a strong (positive) relationship between them;

x STRESS vs. PMI. The equilibrium is clearly moved in favour of the off-diagonal
elements.
In the second last row in Chart 5 we reported the KL distance: we remind the reader that the
KL distance is zero when the variables are independent (i.e. the joint probability is identical to
the product of marginal probabilities), and greater than zero otherwise (see Annex B for more
details). Test of significance has been achieved through a non-parametric approach 16.
The analysis shows that:
x there is clear redundancy between real GDP and PMI, and
x orthogonality among real GDP, CPI and market stress holds only as a first
approximation.
In order to further investigate the non-negligible dependency observed among factors, we

computed the historical joint probability for CPI and real GDP scenarios in rolling windows
(10-year trailing windows; 40 quarterly observations). Chart 6 reports the main findings. The
upper panel plots the joint probability of CPI and real GDP: the 70s exhibited a marked
increase of scenarios characterized by rising inflation. This outcome is confirmed by the
lower panels which report the marginal probabilities for CPI and real GDP. Chart 6 suggests
that part of the deviations from independence comes from the 70s inflationary period, and
hence dependency among CPI and real GDP might be due to sample specificity. As we
showed in Pola and de Laguiche 2012, the last century has been characterized by inflationary
scenarios; however, investigating the likelihood of inflationary and deflationary scenarios in a
deeper historical analysis, it emerges that the last century itself represented an anomaly.
The main take-home message of this section is that fundamental factors are not strictly
orthogonal: the analysis shows a deviation from independence in some pairs taken from CPI,
real GDP, PMI, and market stress. We believe that, at least, part of the anomaly can be
explained by the sample specificity of the analysed period: in order to assess the significance
of the result, a deeper historical analysis is advisable.
16
In order to clarify the approach, let us consider the first pair: CPI vs. real GDP. In order to compute the KL
distance we only need the two Boolean sequences indicating rising or falling scenarios. In order to test
significance, we firstly randomly reshuffled the Boolean sequence of GDP, and computed the KL distance. We
performed this procedure 10,000 times, in order to get 10,000 computations of the KL distance. It is worth
stressing that computations from reshuffled pairs should not signal any relationship between CPI and real GDP;
any possible relationships are certainly spurious and originate from the data sample. This way allows one to
evaluate if the KL distance is significant or spurious.
14

Indeed it is worth stressing again that we prefer to work with quasi-orthogonal axes that are
easily related to macroeconomic dynamics, than perfectly orthogonal factors (from statistical
tools) which are difficult to interpret and unstable over time. On the other way round we
should point out that independence might hold only as a first approximation: possible
deviations from orthogonality should be taken into account in the portfolio construction, as
second-order correction to the main frame.
Chart 6. Historical joint and marginal probabilities (focus on CPI and GDP)
Joint Probability of rising-falling scenarios on CPI and real GDP - 10 yr rolling analysis in US
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1963
1965
1966
1968
1969
1971
1972
1974
1975
1977
1978
1980
1981
1983
1984
1986
1987
1989
1990
1992
1993
1995
1996
1998
1999
2001
2002
2004
2005
2007
2008
2010
2011
(- ; -) (- ; +) (+ ; -) (+ ; +)
Marginal Probabilities of rising-falling scenarios on CPI - 10 yr rolling analysis in US
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1963
1965
1966
1968
1969
1971
1972
1974
1975
1977
1978
1980
1981
1983
1984
1986
1987
1989
1990
1992
1993
1995
1996
1998
1999
2001
2002
2004
2005
2007
2008
2010
2011
CPI - CPI +
Marginal Probabilities of rising-falling scenarios on real GDP - 10 yr rolling analysis in US
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1963
1965
1966
1968
1969
1971
1972
1974
1975
1977
1978
1980
1981
1983
1984
1986
1987
1989
1990
1992
1993
1995
1996
1998
1999
2001
2002
2004
2005
2007
2008
2010
2011
real GDP - real GDP +

3. Methodology to assess polarization of asset returns towards fundamental factors
In this section we illustrate the methodology to assess the relationship between asset returns
and factors scenarios. Ideally it is desirable to investigate the relationship between assets and
each factor separately (e.g. studying the dependency of nominal bonds versus movements of
inflation, keeping the dependency on growth and market stress fixed). This is not possible
because in financial markets all factors act synchronously to determine asset prices; moreover
factors are orthogonal only as a first approximation. In small data sample, it might be the case
that some factors are more important than other (e.g. inflation in 70s, growth in the Great
Depression, market stress in 2008), leading potentially to misleading results. A longer data
sample enables this problem to be alleviated.
We start with a nave approach: for each asset and factor we divide asset return in
correspondence of rising and falling scenarios. In Chart 7 we perform this exercise for the US
treasury. Chart 7 (left panel) reports the (conditional) average of excess-returns over 3-month
T-bills in different scenarios 17; the chart includes also the unconditional average 18. Two main
findings emerge:
x nominal bonds present positive risk premia;
x nominal bonds exhibit clear characteristics in terms of factors scenarios: they are good
hedge against falling inflation, falling growth and rising stress scenarios.
Chart 7. Conditional analysis of US Treasury across macroeconomic scenarios

US Treasury. Average Excess Return across factors' scenarios US Treasury. Dispersion of Excess Return across factors'
scenario
1.80% 15%
1.60%
1.40% 10%
1.20%
1.00% 5%
0.80%
0.60% 0%
0.40%
0.20% -5%
0.00%
-0.20% -10%
CPI GROWTH STRESS CPI rising CPI falling GROWTH GROWTH STRESS STRESS
rising falling rising falling
rising avg falling
17
Cash level is mostly determined by central bank, whereas asset excess return over cash responds mainly to
surprises in expectations of macroeconomic variables and market stress.
18
The average figure for the stress indicator is significantly different from those of CPI and growth because of a
shorter data sample. Growth in this chart refers to real GDP. Observations are quarterly.
16

However, a more detailed analysis shows that the picture is not very neat, and some further
investigation is needed. Chart 7 (right panel) reports the dispersion 19: results are noisy, hence
we need to quantify how robust the polarization is towards a specific factor scenario.
First of all we consider two statistical tests: a t-test on the difference of means between rising
and falling samples (parametric test), and a non-parametric test based on the historical
probability that the rising sample outperform the falling sample (see Annex D for more
details). While the parametric test is the standard in the industry, it presents two limitations:
x hypothesis on gaussianity of returns is certainty violated in many asset classes, and
x it does infer about the mean of the two samples, whereas we are much more interested
in the entire distributions.
We believe that the coupled application of the two tests may help us to conduct a more robust
analysis.
The parametric and non-parametric tests allow one to define the polarization coefficient P 20.
(We name with P param and P non-param respectively the parametric and the non-parametric
indicators; see Annex D for all the technical details regarding the derivation of the indicators.)
The main properties of P are:
x sign of P indicates a preference of the asset class for scenario +1 (rising) or -1 (falling);
x abs(P) corresponds to the probability to polarize to specific rising/falling scenarios

(obtained by a t-test for the parametric computation, and a bootstrap technique for the
non-parametric one).
We define significance bands, as follows:
x 0.99P<1 preference for rising scenarios with a high confidence level;
x 0.90P<0.99 preference for rising scenarios with a good confidence level;
x 0.70P<0.90 preference for rising scenarios with a moderate confidence level;
x 0.60P<0.70 weak tendency to prefer rising scenarios;
19
We report the maximum and minimum observations, as well as figures within +1 standard deviation and -1
standard deviation.
20
We label the polarization measure with capital rho P: this is to indicate the analogy with the standard
Pearsons correlation coefficient . P shares with to be in between -1 and 1, main difference is that P measures
the relationship between a continuous variable (asset returns) and a Boolean variable (+1 and -1 standing
respectively for rising and falling scenarios), whether compares two continuous variables.

x -0.60P<0.60 blend response (uncertainty);
x -0.70P<-0.60 weak tendency to prefer falling scenarios;
x -0.90P<-0.70 preference for falling scenarios with a moderate confidence level;
x -0.99P<-0.90 preference for falling scenarios with a good confidence level;
x -1P<-0.99 preference for falling scenarios with a high confidence level.
In the following we report P param and P non-param for the US Treasury in Chart 7:
x Falling Inflation: P param = -99.88% ; P non-param = -99.85%;
x Falling Growth: P param = -99.65% ; P non-param = -99.85%;
x Rising Stress: P param = 99.67%; P non-param = 99.53%.
The figures above demonstrate that nominal bonds are sensitive to falling inflation, falling
growth and rising stress with a high confidence level. Indeed the parametric and non-
parametric measures are coherent.
4. Asset segmentation: empirical results
In this section we compute the polarization coefficient P for a large investment universe from
a US investors perspective. We firstly introduce a two-dimensional representation (the
polarization plan), then we move towards a more comprehensive three-dimensional chart that
expresses polarization of assets to the three fundamental factors (the polarization cube).
4.1 Asset class universe
We consider the point of view of a US investor investing in domestic bonds (nominal bonds,
inflation-linked bonds, corporate bonds), international emerging market bonds (local currency
and hard currency), domestic equity, domestic equity styles (value, growth, large, small) and
sectors, commodities, volatility, hedge fund strategies (including CTAs, long-short trend
follower strategies on long term nominal bonds and fx-rates, long-only trend follower
strategies on a diversified investment universe), and a global risk parity balanced strategy.
Moreover we considered zero-duration credit indices (in order to disentangle credit dynamics
from interest rate ones), alpha equity indices (long equity sectors and styles, short S&P 500),
and beta-neutral equity indices (long equity sectors and styles, short S&P 500 to neutralize
beta to S&P500). The point of view is that one of a US investor who wants to understand the
relationship of asset returns (domestic and international) versus his own countrys
18

macroeconomic dynamics and market stress. Going into more details, we analyzed 139
assets/strategies:
x Nominal bond: four indices with different maturities;
x Inflation-linked bond: three indices (two US, one global emerging);
x Credit markets: eight indices (including Investment Grade, High Yield, Emerging
Market Debt in Hard currency and Local currency) and three zero-duration indices 21
(Investment Grade, High Yield, Emerging Market Debt in Hard currency);
x Equity market: 25 (including sectors and styles), 24 alpha indices 22 (including sectors
and styles), and 25 beta-neutral indices 23 (including sectors and styles);
x Commodity markets: six indices;
x Volatility: three indices;
x FX-rates: 19 indices;
x Hedge Funds: 11 indices;
x Trend-follower and CTAs: six indices/strategies (including long-short trend follower on
nominal bonds 24, long-short trend follower on fx-rates24, two long-only trend follower
strategies on a diversified investment universe 25);
x Dynamic strategy: two strategies (including a Risk Parity strategy26, and a dynamic
strategy27 selecting securities according to their historical sharpe ratio).
21
Indices obtained taking the excess returns over government bond of same maturity (Merrill Lynch
computation).
22
Synthetic indices obtained going long on the sector or style and going short 100% on the S&P 500; calculation
on a monthly basis.
23
Synthetic indices obtained going long on the sector or style and going short on the S&P 500 for a quantity
equal to the index beta to the S&P 500; calculation on a monthly basis. Beta estimation is on the whole data
sample.
24
Long-short strategy is designed on a monthly basis (the last Thursday of the month), and invests in 7-10-yr
bonds from US, Germany, UK, Japan, Australia, Canada, Italy. Portfolio construction is given by asset historical
volatility for the previous year (risk budget is the same for all the markets). Long-short signals are given
comparing the index with the moving average for the last 100 days: if the price is above (resp. below) the signal
is long (resp. short). Portfolio is not leveraged. Long-short trend follower on fx follows the same investment
scheme. Investment universe is EUR vs. USD, CAD, AUD, GBP, JPY, and USD vs. CAD, AUD, JPY, NZD,
NOK, BRL, MXN, KRW, TWD, ZAR (carry returns indices). Our hypothesis is that strategies are implemented
through derivatives instruments. We considered transaction costs equal to 15 bps for 100% portfolio turn-over.
25
They are labelled in the following as riskparity8 trend and cstra95.
26
Risk-parity allocation is derived within a monthly rebalancing. Risk budget is equally divided between
nominal 7-10-yr bonds (US, Germany, UK, Japan, Australia, Canada), Equity (Euro, US, UK, China, Japan),
Commodity (Gold, Copper, Agriculture, Oil WTI). Each month the portfolio is determined according to asset
historical volatility for the previous year; the portfolio then is leveraged in order to match a target volatility of
8% (estimated on the previous year historical observations). Risk allocation is rebalanced once a month (the last
Thursday of the month), volatility targeting is on a daily basis.

Annex A reports the details (Bloomberg code, inception, description) of the indices.
4.2 Getting started: the polarization plan.
An effective way to represent polarization of asset classes to two factors is to scatter plot their
polarization coefficients P. Chart 8 shows the behaviour of a selection of asset classes versus
CPI and real GDP as a proxy of growth (here, we report P param for quarterly observations).
This section is devoted to presenting the main ideas; in the next section we present a more
comprehensive analysis. We remind the reader that, according to the definition of P, the
closer the spots are to the border, the more significant the asset response to macroeconomic
changes.
Let us summarize the main findings:
x Nominal bonds are excellent hedges in case of falling growth and falling inflation.
x Equity is an excellent hedge in case of rising growth, and to a lesser extent to falling
inflation scenarios. The equity energy sector tracks commodity investments, performing
better in rising growth and rising inflation, whereas real estate stocks are very sensitive
to growth, but weakly related to inflation changes.
x Corporate bonds do not present a homogeneous response. Investment grade bonds are
an excellent hedge in case of falling inflation, to a lesser extent in case of falling
growth. High-yield bonds offer a good preference for rising growth scenario; they are
rather insensitive to inflation changes. Emerging Market Debt in Local currency
(EMDL) behaves like commodity CRB, preferring rising growth and rising inflation.
On the other hand, Emerging Market Debt in Hard currency (EMDH) results almost
blend versus changes in inflation and growth.
x The 1-5 segment of Inflation-linked bonds is an excellent hedge for rising inflation
scenarios (and rather insensitive to growth). The Inflation-linked all maturity index
responds well to falling growth scenarios, and, surprisingly, do not offer a very
significant hedge against rising inflation (in this case the short data sample may have
had a strong impact in the estimation).
27
The strategy invests in a large, diversified, and global investment universe including more than 150 indices
(fx-rates, nominal bonds, equities, commodities). The strategy is designed to invest on the best ten indices in
terms of realized historical Sharpe ratio on a given time interval. Explicit constraints allow us to control portfolio
turnover. The strategy is long-only on all assets, with the exception of fx-rates where we take long-short
positions.
20

x Commodities CRB, Industrial Metals, and Crude Oil WTI are excellent hedges for
rising inflation and growth. Gold presents a tendency to rising growth and rising
inflation, even if the result is less significant. Agriculture is very sensitive to inflation,
but rather independent from growth.
x The VIX index is similar to nominal bonds in terms of growth, to a less extent with
respect to inflation.
Chart 8. Polarization chart to CPI and Growth in US
EMDL
S&P 500 ENERGY OIL
100%
SPX 500 DJ US REAL ESTATE CRB
INDUST
CORPORATE HY GOLD
80% METALS
AGRICULTURE
60%
US INFLATION 1-5
EMDH
Polarization to Growth ( P )
40%
20%
0%
-20%
-40%
-60%
CORPORATE IG
-80%
US INFLATION ALL MATURITY
GOVY VIX
-100%
-100% -50% 0% 50% 100%
Polarization to Inflation (P )
The analysis above shows clearly that traditional asset classes present mixed characteristics
(e.g. commodities are commonly reported as proxy of inflation, indeed they are sensitive to
growth as well). In order to individuate diagonal proxies 28 for the rising/falling scenarios, it
might be wiser to consider combinations of assets, as follows:
x proxy rising growth: portfolio of equities and commodities;
x proxy falling growth: portfolio of nominal bonds and inflation-linked bonds;
x proxy rising inflation: portfolio of commodities and inflation-linked bonds;
x proxy falling inflation: portfolio of nominal bonds and equities.
28
With this term we mean proxies that exhibit a clear relationship versus a factor, but are rather insensitive to
other factors: e.g. commodity is not a proper proxy of inflation, because it is strongly polarized to growth as
well.

These combinations allow one to get well polarized proxies versus growth and inflation
without any relevant biases towards inflation and growth, respectively. The above scheme for
proxies can be interpreted, from a geometrical point of view, as a rotation in the polarization
chart of about half of a right angle.
We conclude this section with three remarks:
x Comparison between assets and factors is synchronous. In Annex F1 we investigated
whether a temporal lag may lead to more robust indications or not. We computed the
polarization coefficient P for the main asset classes and considered three months lag for
quarterly series and one month lag for monthly series. The effect of lagging is to reduce
the significance, and in some cases in reverting the relationship in a counterintuitive
way.
x In Annex F2 we verified the robustness of the analysis, slightly changing the definition
of rising and falling scenarios. We introduced the blend scenario in case of weak
signals. Results are robust and coherent.
x The mathematics behind the polarization coefficient P is not linear: this means that if
we take two assets in the polarization plan, ask what is the position of a portfolio of
those assets, the result will generally not lie within the linear interpolation of the two
points.
In the next section we present the results in a more exhaustive way.
4.3 The polarization cube.
In order to express the polarization of asset returns to the three factors, we scatter plot the
polarization coefficients P in a three-dimensional space. By construction all assets will be
represented by spots in the interior of a cube (see Chart 9; top panel). Different colours refer
to different asset classes, as follows: nominal bond (blue), inflation-linked bonds (magenta),
credit markets (light blue), zero-duration credit markets (cyan), volatility (red), equity
(orange), equity alpha indices (light orange), equity beta-neutral indices (yellow),
commodity (green), fx-rates (brown), hedge funds, trend-follower, CTAs and dynamic
strategies (black). Chart 9 lower panel reports the two-dimensional projection along the stress
axis. Coherently with Chart 8, the plot refers to the parametric evaluation of the polarization
coefficient P for quarterly data; moreover we consider the real GDP as proxy of growth.
22

Chart 9
The main findings emerging from Chart 9 are:
x asset classes are placed about in the corners of the inflation-growth two-dimensional
projection;
x the rising-inflation falling-growth corner is scarcely populated;

x equity sectors can be approximately divided between equity-like sectors (rising-growth
falling-inflation corner) and commodity-like sectors (rising-growth rising-inflation
corner); alpha and beta-neutral indices permits us to populate the falling-growth falling-
inflation corner;
x despite the behaviour of credit markets is not homogeneous, the zero-duration indices
are definitely placed in the rising-growth rising-inflation corner.
In order to clarify the picture we considered three slices of the cube:
x the rising stress slice is defined clustering together assets with P stress between 0.70 and
1;
x the blend stress slice includes assets with P stress between -0.70 and 0.70;
x the falling stress slice is defined for assets exhibiting P stress between -1 and -0.70.
For the sake of simplicity we plot the three slices integrating away the dependency on the
market stress factor (see Charts 10, 11, and 12). At a first glance it is evident that the rising
stress scenario is less populated than the falling stress one: most of the assets rarely hedge
against market stress. In the following we describe each slice in detail, sorting asset classes
according to their joint polarization to inflation and growth.
Chart 10 indicates assets that perform well in case of rising-stress:
x Rising-inflation AND rising-growth corner. The best hedge is given by the Information
Technology equity-sector alpha index (the beta-neutral index is placed in the same
corner, but it is less significant in terms of polarization to growth) and, to a lesser
extent, Russell 1000 Growth and S&P 500 Growth (alpha and beta-neutral indices), and
the equity spread Growth vs Value.
x Rising-inflation AND falling-growth corner: this class is empty.
x Falling-inflation AND rising-growth corner: this class is empty.
x Falling-inflation AND falling-growth corner. The best hedge is given by Nominal

Bonds (intermediate, all and long maturity), Trend-follower strategy on Nominal Bonds
(dynGOVY), VIX Futures Enhanced Roll index, Telecommunication Services equity-
sector (alpha and beta-neutral indices), and USDCAD fx-rate. Less significant assets in
this scenario are USDNOK fx-rate, USDSEK fx-rate, the Health-Care and Consumer-
24

Staples equity-sectors (alpha and beta-neutral indices), Long maturity Corporate
Investment Grade, Long-only Trend follower on a diversified basket (riskparity8 trend),
VIX, EURUSD fx-rate, HFRI Short bias, CTA long term.
Chart 10
Chart 11 includes assets that are rather independent from market stress factor:
x Rising-inflation AND rising-growth corner. The best hedge is given by the Information
Technology equity-sector. Less significant are Inflation-linked 1-5 years (high
confidence level on inflation axis, but poor polarization to growth direction), and
Agriculture.
x Rising-inflation AND falling-growth corner: Inflation-linked all maturity and USDJPY

fx-rate; polarizations are weak in this case.
x Falling-inflation AND rising-growth corner. The best hedge is given by Consumer

Staples, and to a lesser extent Health Care equity-sector and Emerging Market Inflation-
linked bond.
x Falling-inflation AND falling-growth corner. The best hedge is given by the S&P 500
High Dividend alpha index. Less significant are the Corporate Investment Grade (all

maturity and intermediate maturity), USDDKK fx-rates, and Utility equity-sector alpha
index.
Chart 11
Chart 12 reports assets that polarize to falling-stress scenarios:
x Rising-inflation AND rising-growth corner. The best hedge is given by the TWDUSD
fx-rate, Industrial Metals, Commodity CRB, Emerging Market Debt in Local currency,
Corporate Investment grade zero-duration index, KRWUSD fx-rate, ZARUSD fx-rate,
Gold Miners stocks, High Yield zero-duration index, Asian dollar index, Crude Oil
WTI, HFRI Equity Hedge, S&P500 Energy equity-sector (pure and beta-neutral
indices), EURUSD fx-rate, HFRI Fixed Income Convertible arbitrage, Gold miners
stocks beta-neutral index, and Small Cap 600 growth (alpha and beta-neutral index). See
Annex E for more details.
x Rising-inflation AND falling-growth corner: RUBUSD fx-rate; the polarization is very

weak in this case.
x Falling-inflation AND rising-growth corner. The best hedge is given by the S&P 500,
and to a lesser extent, Industrials equity-sector (alpha and beta-neutral indices), S&P
5000 High Dividend, Russell 2000 Value (alpha and beta-neutral indices), Consumer
Staples and Financials equity-sectors. See Annex E for more details.
26

x Falling-inflation AND falling-growth corner. The best hedges are Financial equity-
sector beta-neutral index and Russell 1000 Value alpha index, and, to a lesser extent,
S&P 500 High Dividend and Russell 1000 Value (beta-neutral indices).
Chart 12
We remind the reader to refer to Annex E for more details.
5. Implications of DAMS segmentation: duality and synthetic assets
From the previous section, it emerges that nominal bond, inflation-linked bond, equity, and
commodity CRB are placed about in the corners of the polarization cube. This plot suggests
us two relationships among them:
Nominal bond + Commodity CRB 0 (1)
Inflation-OLQNHGERQG(TXLW\ (2)
The equations above indicate that an ad-hoc combination of nominal bonds and commodities
might lead to a blend polarization versus inflation, growth and market stress changes;
similarly inflation-linked bonds and equity might be combined together to neutralize their
environmental biases. On this basis, we will say that nominal bonds and commodity are dual
with respect to inflation, growth and market stress; the same for equity and inflation-linked

bonds, even if in this case the relationship is less robust. These relationships can be useful, at
least, for two purposes.
Firstly, in designing balanced funds which are less sensitive to specific macroeconomic
scenarios. Macroeconomic trends may last for many decades; hence it might be wiser to
design allocations that are more balanced across macroeconomic scenarios. The previous
section shows that even if the traditional bond-equity allocation might be neutral versus
growth (and if and only if they are well balanced), they are not equipped to manage rising
inflation scenarios: the first practical implication of the study would suggest replacing
equities with commodities or nominal bonds with inflation-linked bonds in traditional bond-
equity portfolios.
Secondly, relationships in Eqs (1) and (2) may help us to design synthetic asset classes. Let us
consider for example the case of Commodities. Many funds (e.g. pension funds in some
countries) are not allowed to invest in commodities. The relationships above offer a possible
solution. According to Eq. (1), the macroeconomic behaviour of commodities might be
represented by a structural short position in nominal bonds; however this position would be
also structural short of the risk premia of nominal bonds. Indeed it might be wiser to combine
Eqs (1) and (2):
Commodity CRB Inflation-linked bonds + Equity Nominal bonds. (3)
Portfolio weights in Eq. (3) have been determined according to the following receipt:
x portfolio weights are inverse proportional to the long term volatility of assets,
x and the portfolio leverage has been set in order to target the long term volatility of CRB.
We then computed P param for the synthetic series (quarterly data) computed according to Eq.
(3):
P param (inflation)=94.21%,
P param (growth)=89.80%,
P param (stress)=-99.41%.
The same calculations on monthly time-series read:
P param (inflation)=99.61%,
28

P param (growth)=97.35%,
P param (stress)=-99.99%.
We conclude that the basket accurately represents the response to macroeconomic changes of
CRB, with the differences coming from market specificities that are not captured by the
factors dynamic.
6. Conclusion
The recent crisis poses serious doubts on the effectiveness of diversification to reduce draw-
downs in balanced portfolios: diversification failed when it was mostly needed was the
leitmotif of many institutional investors after 2008. Nevertheless recent portfolio construction
schemes, like risk parity and maximum diversification, make of diversification the kernel of
asset allocation. While most approaches diversify on asset class level, the new challenge in
asset allocation suggests diversifying on fundamental factors that are believed to be the main
drivers of asset price dynamics.
The factors approach provides a new, challenging, and powerful way to interpret financial
markets. Even non-stationary patterns of correlation between asset classes might be brought
back to the dominance of one factor over the others (e.g. inflation in the seventies, growth in
the Great Depression, and stress in 2008).
This new way forces us to rethink asset segmentation. The main assumption of our approach
is:
asset price dynamics can be largely explained in terms of changes
in expectations of macroeconomic variables and market stress.
This statement suggests a new way to segment asset classes according to their similarity in
responding to changes in expectations. In this work, we found that each asset class presents
mixed relationships with fundamental factors. Asset polarization to factors is represented
through their position in a three-dimensional cube:
(i) assets placed in the corners present strong polarization towards the three factors
(e.g. nominal bond, equity, commodity);
(ii) assets at the edges present high polarization to two factors, and rather blend the
response to the third factor (e.g. high yield and real estate stocks);

(iii) assets in the centre of the cubes faces respond strongly to one factor only (e.g.
inflation-linked 1-5 yr bond);
(iv) assets in the centre of the cube are rather insensitive to factors dynamics (e.g.
dynamic strategies like trend follower and risk parity).
Whether assets with specific polarization to factors are particularly appealing for asset
allocation, it is desirable for portfolios to be placed in the centre of the cube especially in
uncertain macroeconomic and financial market conditions.
Practical implications of the present study can be found within Quantitative Finance and
Portfolio Management:
(i) building asset allocations that are better equipped to navigate different
macroeconomic environments (in Pola & Facchinato 2013 we will explore the
implications in terms of portfolio construction, and we will illustrate the DAMS
investment process);
(ii) providing new insights in crisis management to protect portfolios against

extreme financial events and stress;
(iii) constructing asset allocations better adapted to face perspective macroeconomic

scenarios like inflationary ones 29;
(iv) investigating the underlying macroeconomic bias in already managed balanced
portfolios;
(v) deepening our understanding of asset prices dynamics, correlation among asset
classes, and the relation of equity sectors and styles to macroeconomic
dynamics29;
(vi) building a macro equivalent synthetic asset allocation which tracks precise
macro behaviour;
(vii) determining expected returns better adapted to the current uncertainty in
financial markets;
(viii) generating Monte Carlo synthetic assets that are sensitive-or-blend to specific
macroeconomic environments.
We believe that rigorous portfolio construction which is explicitly related to the macro view
can help us to manage portfolios in uncertain financial markets: strategic decisions should be
29
We briefly addressed these issues in Rethinking strategic asset allocation in terms of diversification across
macroeconomic , G Pola, Amundi Special Focus.
30

mostly rephrased in terms of asset environmental biases towards macroeconomic and stress
factors, rather than on standard mean-variance optimization packages that need forecasting
returns.

Acknowledgement
I would like to thank S. Facchinato and C. Casadei for very illuminating discussions and
suggestions regarding this research topic. The ideas investigated in this paper constitute the
foundation of the DAMS investment process, which has been developed by Amundi Milan
Investment Management and G. Pola (Amundi Paris Quant Research); DAMS is in place at
Amundi Milan within a range of flexible funds (diversified DAMS), absolute return (income
DAMS) and equity funds (equity DAMS) since December 2011. Moreover I would like to
thank Ph. Ithurbide and S. de Laguiche for their contribution which improved the research
study and quality of the manuscript.
32

Annex A Database description
We considered four fundamental factors:
x Inflation. US CPI Seasonally Adjusted (Bloomberg code CPI INDX Index). Time-series
from December 1953;
x Real Growth. US Real GDP Seasonally Adjusted (Bloomberg code GDP CHWG
Index). Time-series from September 1948;
x PMI. ISM Manufacturing PMI Seasonally Adjusted (Bloomberg code is NAPMPMI

Index). Time-series from December 1948;
x Stress Index. Bloomberg US Financial Conditions Index 30 (Bloomberg code is BFCIUS

Index). Time-series from September 1996.
We investigated 139 assets/strategies as detailed in tables A1, A2 and A3. Time-series are up
to December 2012.
30
The Bloomberg U.S. Financial Conditions Index combines yield spreads and indices from U.S. Money
Markets, Equity Markets, and Bond Markets into a normalized index. The values of this index are Z-scores,
which represent the number of standard deviations that current financial conditions lie above or below the
average of the January 1994-June 2008 period.

34

36

Annex B Technical note: the Kullback-Leibler distance
The Kullback-Leibler distance (KL distance in the following) measures the distance between
two probability distributions (Cover and Thomas, 2006). Let us consider two random
variables x and y taking values from a discrete set 31; we do not need the set to be of finite
dimension. Let P(x) and P(y) be (respectively) their probability distributions. Let P(x,y) be the
joint probability of outcomes (x,y). x and y will be independent if and only if
P(x,y)=P(x)P(y). This notion of independence is stronger than that one given by the standard
Pearsons correlation coefficient 32. The KL distance permits one to measure the distance
between P(x,y) and P(x)P(y), hence to quantify the degree of dependence. The KL distance is
defined as follows:
P ( x, y )
DP ( x, y ) || P( x) P( y ) P( x, y) log P( x) P( y) , (B1)
x, y
ZKHUHWKHVXPLVH[WHQGHGWRDOO[DQG\VXFKWKDW3[DQG3\,WFDQEHSURYHQWKDW

the KL distance is always greater than zero, and equal to zero if and only if P(x,y)=P(x)P(y);
hence any dependence between x and y is certainly to increase the KL distance. This property
justifies why this measure is called distance 33. Equation (B1) can be rearranged as
DP ( x, y ) || P ( x) P ( y ) H ( x) H ( y ) H ( x, y ), (B2)
where H(x), H(y) and H(x,y) stand respectively for the Shannon Entropy 34 of x, y, and the
joint variable (x,y). This formulation expresses the KL distance as the difference between the
sum of the variability of variables x and y minus the variability of (x,y). This representation
allows one to write down the following inequality:
DP ( x, y ) || P( x) P( y ) d max( H ( x), H ( y )),
31
We present here the KL distance for discrete random variables, but the formalism is more general and can be
applied to continuous random variables (Cover and Thomas, 2006).
32
In fact the standard Pearsons correlation coefficient can even wrongly interpret two deterministic related
random variables: let x be uniformly drawn in the interval [-1, 1], and let y=x2. It can be proven easily that the
Pearsons correlation coefficient is zero in this case. The other important aspect is that the statement according to
which positive correlated assets imply that once one asset increases (decreases) the other increases (resp.
decreases) as well is wrong, as shown clearly in Lhabitant 2011. Finally the Pearsons correlation coefficient is
a linear measure; it completely misses non-linear terms like x2y, xy2, x2y2, etc. The KL distance enables us to
alleviate many of these drawbacks. The only con, as we will see in this section, is that computing a reliable
estimate of KL distance requires large data samples.
33
However it should be stressed that this measure is not symmetric. There are several ways to get a symmetric
KL measure. We prefer here to work with the simplest version of the KL distance.
34
The Shannon entropy for random variable x is defined as H(x)=-x P(x) log P(x).

the equality holding if and only if x and y are deterministically related. This upper bound is
useful to understand the range of variability of the KL distance. The second consequence of
equation (B2) is that it permits one to re-write the KL distance as:
N N
DP ( x, y ) || P( x) P( y ) log x y ,
N
xy
where N x , N y and N xy are respectively the average number of relevant states 35 of P(x), P(y),
and P(x,y).
The generalization of the KL distance to more random variables is straightforward:
N
P ( x1 ,..., x N )
D P ( x1 ,..., x N ) || P ( xi ) P( x ,..., x
1 N ) log N
. (B3)
i 1 x, y
P( x )
i 1
i
Analogously it is easy to derive the upper bound and the formulation of the KL distance in
terms of relevant states.
We conclude this section briefly discussing the estimation problem in Eq. (B3). For the sake
of simplicity let us assume that random variables x i can take only two values (+1 and -1). In
this case the dimension of the response space is 2N. Let us come back to the practical example
with four factors (CPI, real GDP, PMI, market stress). If we require at least 30 observations
(on average) for each state we come out with 3024=480, which is equivalent to 120 years of
quarterly observations. This simple evaluation clarifies why measuring the KL distance on all
the factors might be difficult in practice, and why we prefer to evaluate pair-wise dependency
(in this case we can lower 120 years of quarterly observations to 30 years; i.e. 3022=120).
Nevertheless it is worth pointing out that there is a vast literature which faced the sampling
problem of the KL distance: many contributions might be found within the field of
Computational Neuroscience, where the KL distance was extensively used to study neural
coding of sensory information. A brief review can be found in Pola et al 2002. In Pola et al
2005 we proposed a tight lower bound of the mutual information which can be useful to
delineate robust bounds for the KL distance.
35
Given an entropy measure H=-i Pi log Pi , index i running from 1 to N, the average number of relevant states
DUH GHILQHG DV H[S+ ,W FDQ EH SURYHQ HDVLO\ WKDW LI 3i= 1/M for each i= 1, , M1 0 7KLV DQDO\WLFDO
result justifies the definition. Meucci (2009) introduced an analogous definition.
38

Annex C Technical note: defining rising and falling scenarios
The aim of this section is to provide the methodology according to which we define rising and
falling scenarios starting from the time-series of the fundamental factors (CPI, real GDP,
PMI, market stress). We will follow a heuristic approach 36. In case of quarter data, the
approach consists in:
x CPI. Take the quarterly variations. At each point compare the current quarter variation
to the average of the last four variations (including the current figure). If the last
observation is above the average, we will name the quarter as a rising-inflation quarter
(symbol +1), otherwise as a falling-inflation (symbol -1) quarter.
x GDP. Take the quarterly variations. At each point compare the current quarter variation
to the average of the last four variations (including the current figure). If the last
observation is above the average, we will name the quarter as a rising-growth quarter
(symbol +1), otherwise as a falling-growth (symbol -1) quarter.
x PMI. At each point compare the current quarters figure (average of the three monthly
observations) to the average of the last twelve observations (including the current
quarters figure). If the last observation is above the average, we will name the quarter
as a rising-PMI quarter (symbol +1), otherwise as a falling-PMI (symbol -1) quarter.
x STRESS 37. At each point compare the current quarters figure (average of the three
monthly observations) to the average of the last twelve observations (including the
current quarters figure). If the last observation is above the average, we will name the
quarter as a rising-stress quarter (symbol +1), otherwise as a falling-stress (symbol -1)
quarter.
In case of monthly data, the approach consists in:
x CPI. Take the monthly variations. At each point compare the current months variation
to the average of the last twelve variations (including the current figure). If the last
observation is above the average, we will name the month as a rising-inflation month
(symbol +1), otherwise as a falling-inflation (symbol -1) month.
36
Recently Kritzman et al 2012 investigated this issue in the context of Markov switching models.
37
In order to simplify notation, we preferred to compute rising/falling scenarios to the negative of the Bloomberg
Stress indicator: in this way rising stress would be linked to symbol +1 and falling stress to symbol -1.

x GDP. Not available with monthly frequency.
x PMI. At each point compare the current months figure to the average of the last twelve
observations (including the current figure). If the last observation is above the average,
we will name the month as a rising-PMI month (symbol +1), otherwise as a falling-PMI
(symbol -1) month.
x STRESS35. At each point compare the current months figure to the average of the last
twelve observations (including the current figure). If the last observation is above the
average, we will name the month as a rising-stress month (symbol +1), otherwise as a
falling-stress (symbol -1) month.
40

Annex D Technical note: the polarization coefficient P
The aim here is to define a polarization indicator, quantifying how much an asset class
responds to rising or falling scenarios. In order to answer to this question we perform a
conditional historical analysis.
Let us consider that we want to verify if there is a relationship between asset A and
fundamental variable V. Firstly we define symbol +1 and -1 in correspondence of a rising or
falling scenario for indicator V. Secondly we divided the performance sample of A in
correspondence of symbol +1 and -1, we will name them respectively as A + and A - . We
question whether there is an ordering between sets A + and A - . If we prove that A + is on
average significantly greater (resp. lower) than A - , then we conclude that asset A is a good
candidate to hedge against rising (resp. falling) scenario of fundamental factor V.
We considered two statistical tests:
x Parametric. t-test of the null hypothesis that data in the samples A + and A - are
independent random samples from normal distributions with equal means and different
and unknown variances (Behrens-Fisher problem), against the alternative that the means
are not equal. The t-statistics, under the null hypothesis, is distributed according to an
approximate Students t distribution with a number of degrees of freedom given by
Satterthwaite's approximation (see Timm, 2002). The p-value 38 is reported as p param.
x Non-Parametric. Let a+ and a- be respectively two elements of A + and A - . We estimated

from the frequency tables the probability that a+>a-; we will name this probability
P(a+>a-). In order to quantify how reliable the estimate is, we randomly bootstrapped
the order of the symbol +1 and -1 to get A + (i) and A - (i) for each reshuffling (i). We
indicate by A + (0) and A - (0) the true sets, and by P(0)(a+>a-) the true probability. We
performed 10000 reshufflings. For each reshuffling we estimated P(i) (a+>a-). Looking
at the entire estimates of P(i) (a+>a-) we will be able to extract the distribution of
P(a+>a-). At this point we can assess if P(0)(a+>a-) is significant or not. We will name
the p-value estimated from bootstrapping technique as p non-param .
It is worth stressing that p param >0.50 (resp. p param <0.50) indicates a preference of asset A for
scenario +1 (resp. -1); the same holds for p non-param .
38
We considered the general case of different variances for sample A+ and A-, and one-tailed distribution.

Finally we define the polarization coefficients as:
P param = p param +0.50 sign(p param -0.50)-0.50,
P non-param = p non-param +0.50 sign(p non-param -0.50)-0.50.
By-construction P is bound between -1 and +1: the closer is to the borders (+1 or -1) the more
sensitive is the asset class to changes of that fundamental variable.

Annex E Empirical Results
This section includes eight tables corresponding to the following scenarios: rising inflation,
rising growth (real GDP), rising growth (PMI), rising market stress, falling inflation, falling
growth (real GDP), falling growth (PMI), and falling market stress. For each scenario we
report only assets which present a moderate polarization coefficient. In each table we report
the parametric and non-parametric estimations of P for quarterly and monthly data. Asset
classes are sorted according to the average P. In general we found coherence among the four
metrics (parametric and quarterly data, parametric and monthly data, non-parametric and
quarterly data, non-parametric and monthly data). Differences between the parametric and
non-parametric evaluations are small, and may come from small data-sample effects, and
relevant non-gaussian features in the return distributions. Differences between the quarterly
and monthly data estimations are more relevant: this evidence may suggest that asset classes
respond to factors dynamics according to different time frames. For lack of space we did not
investigate this issue further in this work.

Annex F Additional Analysis
In this section we investigate some side issues related to the polarization analysis. Data
analysis has been developed on the main asset classes (US Treasury, US Inflation-linked
bond, S&P500, Commodity CRB).
F1. Effect of temporal lag on polarization analysis.

We performed the polarization analysis introducing a temporal lag of one month for monthly
data, and three months for quarterly data 39. Table F1 permits one to compare the synchronous
measurements to the temporal lagged ones: lag 1 indicates three months lag for quarterly
series, and one month lag for monthly series. The effect of temporal lag is in weakening some
relationships, and in some cases reversing the relationship in a counterintuitive way.
F2. Polarization analysis with rising, falling and blend scenarios.

In order to study robustness of the polarization measure between assets and factors we
introduce a new category: the blend scenario brings together weak rising and falling
scenarios. As described in Annex C, rising and falling scenarios are defined comparing the
value of an index to a given moving average, if the ratio is greater than 1 we name the
scenario as rising, falling otherwise. We slightly changed this rule introducing a band around
1. So the new rule reads as:
x If the ratio is greater than 1+ then rising scenario;
x If the ratio is less than or equal to 1- then falling scenario;
x If the ratio is in (-,+] then blend scenario.
We investigated the polarization indicator P for the main asset classes as a function of
different threshold . Results are reported in Chart 11. It is evident that P is rather stable for
different .
39
Asset performance is measured at time k+1, whereas macroeconomic scenarios are evaluated at time k.

References
Amenc, N., Martellini, L., Ziemann, V., 2009. Alternative Investments for Institutional
Investors, Risk Budgeting Techniques in Asset Management and Asset-Liability
Management. The Journal of Portfolio Management, 35(4), p. 94-110.
Atti, A.P., Roache, S.K., 2009. Inflation Hedging for Long-Term Investors. IMF Working
Paper 09-90, April.
Brire, M, Signori, O., 2011. Hedging Inflation Risk in a Developing Economy: the case of
Brazil, Amundi Working Paper WP-012-2011.
Cover, T. M., Thomas, J. A., 2006. Elements of Information Theory, Wiley series in
Telecommunications and Signal Processing.
Lhabitant F. S., 2011. Correlation vs. Trends: A Common Misinterpretation, EDHEC-Risk

Institute Working Paper.
Meucci, A, 2009. Managing Diversification, Working paper, symmys.com.
Kritzman, M., Page, S., and Turkington, D., 2012. Regime Shifts: Implications for Dynamic
Strategies, Financial Analysts Journal, 68 (3), 22-39.
Pola, G., and de Laguiche, S., 2012. Unexpected Returns. Methodological considerations on
Expected Returns in Uncertainty, Amundi Working Paper WP-032-2012.
Pola, G., and Facchinato, S., 2013. Managing uncertainty with dams. Balancing
macroeconomic scenarios, in preparation 2013.
Pola, G., 2013. Rethinking strategic asset allocation in terms of diversification across
macroeconomic scenarios, Amundi Special Focus.
Pola, G., Petersen, R. S., Thiele, A., Young, M. P., Panzeri S., 2005. Data-robust tight lower
bounds to the information carried by spikes times of a neuronal population, Neural
Computation, 17, 1-44.
Pola, G., Schultz, S. R., Petersen, R. S., Panzeri, S., 2002. A practical guide to information
analysis of spike-trains. Book chapter in A practical guide to neuroscience databases and
associated tools, Kluwer, London.
Timm, N. H., 2002. Applied Multivariate Analysis, Springer.

WP-039
Optimal Asset Allocation

for Sovereign Wealth Funds:
Theory and Practice
Zvi Bodie,
Boston University
Marie Brire,
Amundi, Paris Dauphine University, Universit Libre de Bruxelles
October 2013
This paper addresses management of sovereign wealth from the

perspective of the theory of contingent claims. Starting with the
sovereigns balance sheet, we frame sovereign fund management
as an asset-liability management (ALM) problem, covering all
public entities and taking explicit account of all sources of risks
affecting government resources and expenditures. Real-life SWFs
asset allocations differ strongly from theoretical ones. Financial
management of the sovereign balance sheet is hampered by a lack of
aggregate data, which compromises the coordination of sovereign
wealth management with fiscal policy, monetary policy and public
debt management. In this framework, we suggest institutional
arrangements that could overcome this obstacle and enable efficient
coordination.

1. Introduction
Interest in sovereign wealth funds (SWFs) as key players in financial markets has grown
rapidly over the last years. A large number of sovereign wealth funds (SWFs) have been set
up to collect and manage the tax revenues that states receive from natural resources or
exports. SWFs serve various economic objectives, such as budget stabilization, diversification
from commodities, saving for future generations. They may also pursue political strategies,
such as controlling politically sensitive industries, or supporting the domestic economy
(Avendano and Santiso, 2009; Ang, 2012). SWFs can be managed by different institutional
structures, from central banks to independent financial corporations.
A large body of empirical research has analysed the public investment strategies of sovereign
wealth funds and their performance. Although this takes into account only a fraction of SWF
investments, mainly equity stakes in listed firms, it shows that SWFs tend to invest in large
foreign firms, often in the finance and energy sectors, with low diversification and poor
medium-term performance (Bernstein et al., 2013; Chhaochharia and Laeven, 2009; Dyck and
Morse, 2011; Bortolotti et al., 2013). SWFs also served as investors of last resort during the
last crises, intervening to support their domestic financial markets (Clark and Monk, 2010;
Raymond, 2010). Research on optimal sovereign wealth management is scarcer. Scherer
(2009a and b), Brown et al. (2010), Martellini and Milhau (2010) have addressed the optimal
allocation for an SWF by examining non-tradable commodity wealth in the SWF or
exogenous liabilities set by the government and proxied by an inflation-linked investment
benchmark. But the example of the recent crisis clearly shows that other sovereign liabilities
have to be taken into account: debt, contingent liabilities, etc. Moreover, when a government
is short of liquidity to meet its debt payments, the SWFs assets are often available to
substitute for the funds initially earmarked for this purpose. In 2010, for example, in the wake
of the subprime crisis, Russia, Ireland, Kazakhstan and Qatar used SWFs or public pension

fund assets to invest in banks or shore up equity markets. In a recent paper (Bodie and Brire,
2014), we proposed estimating the whole sovereign economic balance sheet using the theory
of contingent claims and considering the joint management of all sovereign assets and
liabilities in an ALM framework. The sovereign is considered in the broad sense, including
all the related institutions (budgetary government, central bank, SWFs, pension funds and
public entities placed under the sovereign's authority).
Managing the wealth of a sovereign is not very different from managing the wealth of an
individual (Merton, 1969; Bodie et al., 1992; Bodie et al., 2008), a pension fund (Bodie et al.,
2009) or a foundation (Merton, 1993). The central government receives tax revenues each
year. Part of this income can be spent, and the residual saved in the SWF, central bank
reserves, or the public pension fund. How much should be saved and how it should be
invested is a classic ALM problem. The optimal allocation and expenditures of the sovereign
will crucially depend on the nature and size of its assets and liabilities, and the sources of their
uncertainty. Merton (1993) solved a similar problem for a university endowment fund. In our
sovereign case, the optimal sovereign allocation differs slightly. It can be broken down into a
performance-seeking portfolio and three additional portfolios hedging for the variability of
the fiscal surplus and external and domestic debt. Financial management of government
resources and expenditures raises difficult issues in practice. Standard macroeconomic tools
are ill-suited to estimating sovereign economic balance sheets. Most of the macroeconomic
variables monitored at present describe flows, not stocks, and are unsuitable for valuing
intangible assets such as human and natural capital (Aglietta, 2010). Moreover, traditional
macroeconomic data lack a significant dimension, namely risk (Gray et al., 2007). This lack
of aggregate data makes it difficult to coordinate sovereign wealth management with fiscal
policy, monetary policy and public debt management.

In this paper, we review the literature on SWF investment, both from a theoretical and an
empirical point of view, and we show how real-life SWFs asset allocations differ from
theoretical ones (Section 2). We present our conceptual framework for optimal sovereign
wealth management (Section 3). We then discuss its practical implementation, giving country
examples and suggesting possible institutional arrangements that would enable efficient
coordination (Section 4). We finally conclude (Section 5).
2. Sovereign Wealth Fund Investment
While there is an abundant literature on the allocation of foreign-exchange reserves, there are
only a few papers devoted to SWF optimal asset allocation. The two topics are nevertheless
interlinked, since the funds invested in SWFs often come from foreign exchange reserves. We
start this section with a state of the art review for these two topics.
Caballero and Panageas (2005a and b), Beck and Rhababi (2008), Beck and Weber (2011)
examine the optimal allocation of foreign exchange reserves in the event of a sudden
slowdown in private capital inflows (sudden stop). The central bank uses its reserves to
repay the short-term foreign debt and minimize the variance of its portfolio in real terms. In
this framework, optimal portfolio weights depend, in addition to the standard minimum
variance demand term, on the extent to which the assets can be used to hedge against sudden
stops. In their empirical investigation, Caballero and Panageas (2005b) suggest the use of
assets based on the S&P 500 implied volatility index, providing efficient protection against
sudden stops in emerging markets, often linked to global liquidity crises. Beck and Rhababi
(2008) show that dollar-denominated assets are a better hedge for global stops and for
regional stops in Asia and Latin America, whereas the euro is a better hedge in Emerging
Europe.

The authors do not have a uniform view of SWFs objectives. This reflects the different roles
that governments assign to SWFs in practice. Aizenman and Glick (2010) compare the
optimal allocations of foreign-exchange reserves by the central bank and by an SWF, which
have different objectives: (1) reducing the probability of sudden stops for the central bank,
and (2) maximizing the expected utility of a domestic representative agent for the SWF. In
this framework, the authors show that the SWF must hold a riskier foreign-asset allocation
than the central bank. Brown et al. (2010) propose an allocation model for different types of
SWFs, with either a pure return objective or a fiscal smoothing objective. Scherer (2009a and
b) considers that SWFs of commodity-producing countries implicitly possess a stock of non-
tradable wealth, and shows that in this case the optimal asset allocation of the SWF should
include a hedging demand against commodity price variations. Martellini and Milhau (2010)
propose a dynamic asset allocation framework for SWFs having liabilities exhibiting inflation
indexation. In a recent study (Bodie and Briere, 2013), we proposed a framework for optimal
asset allocation of sovereign wealth, taking explicit account of all sources of risk affecting the
sovereigns balance sheet. We used Mertons approach (1974) to estimate the process of the
country's assets, and then we optimized the balance sheet using the ALM approach.1 This
framework expanded previous results on SWFs optimal asset allocations by introducing three
additional sources of risk affecting the sovereign balance sheet. We showed that the optimal
composition of sovereign wealth should involve a performance-seeking portfolio and three
hedging demand terms for the variability of the fiscal surplus and external and domestic debt.
Comparing theory on optimal SWF asset management with real-life data could provide
interesting insights. Unfortunately, a large portion of SWF investments remains private, and
most authors concentrate on SWFs equity interests in listed companies. Dyck and Morse
(2011) and Bernstein et al. (2013) show that SWF portfolios tend to be insufficiently
1
Das et al. (2012) offer a literature review on the use of ALM techniques applied to sovereign fund
management.

geographically diversified, with a strong home bias. SWFs tend to have significant holdings
in large companies in politically sensitive industries, like energy,2 finance and
telecommunications (Bertoni and Lugo, 2012, Bortolotti et al, 2013; Chhaochharia and
Laeven, 2009), contradicting the principles of sound diversification. They also tend to take
large stakes in companies facing financial difficulties, both abroad and domestically
(Raymond, 2010). During the subprime crisis, some SWFs3 played the role of investor of last
resort, rescuing major Western banks or recapitalizing their home equity markets. The
performance of those investments is generally poor in the long run, even if the announcement
of SWF investments yields positive abnormal stock-price returns in the very short run
(Bortolotti et al, 2013).
3. Conceptual Framework
We consider the concept of sovereign in the broad sense, including not just the states
budgetary institutions and monetary authorities (central bank), but also the other institutions
related to it, such as pension funds, SWFs and state-owned enterprises.4 The sovereign has a
multitude of objectives. Some are purely financial, such as debt repayment and setting aside
foreign exchange reserves to cope with liquidity crises. Others are social, including pensions
and financing of social services (infrastructure such as hospitals, roads, education, defence,
etc.). Still others are economic, such as investment in key sectors or industries for future
growth. To achieve its objectives, the sovereign has a variety of resources, particularly future
2
Even when the country is producing commodities
3
For example in China, Hong Kong, Kuwait, Qatar, Russia, Saudi Arabia and Singapore.
4
Distinctions among various state entities are less and less meaningful, as recent crises have shown. In 2010
several countries turned to public institutions for assistance in coping with the crisis-related credit crunch. Some
countries used the assets of SWFs or national pension funds to invest in bank deposits (Russia and Kazakhstan)
or to support equity-market liquidity (Kuwait). Others used the resources to directly recapitalise ailing banks
(Ireland, Kazakhstan and Qatar). For this purpose, states modified their funds investment rules on a
discretionary basis, exposing them to new risks. Finally, in some countries with greater borrowing capacities, the
state tweaked the funds regulations to allow them to buy a larger share of the sovereign debt. These recent
examples clearly show that a state facing a crisis can elicit contributions from the off-budget entities that it
owns or controls in order to meet its short-term obligations without unduly worsening the fiscal deficit.

tax revenues, as well as income from other sources such as state-owned enterprises, fees,
seigniorage, and possibly a stock of financial assets (foreign exchange reserves, SWF assets,
public pension funds, etc.).
Defining the Sovereign Economic Balance Sheet
The sovereigns global economic balance sheet is key to a full understanding of its situation
and risks (Gray et al., 2007). The idea is to estimate all the states assets and liabilities at
market price, and to measure the risks (volatility and sensitivity to economic shocks)
associated with each balance sheet item. Just as a companys balance sheet is regularly used
to assess the risk of bankruptcy (Merton, 1974 and 1977; KMV, 2002), the same analytical
framework may be applied to a state. This is useful not only with regard to the states debt
repayment capacity (Gray et al., 2007; Gray and Malone, 2008), which is obviously a minimal
objective, but more generally, as we shall see, with regard to its ability to meet its long-term
social and economic objectives. Table 1 gives a simplified example of a sovereign balance
sheet.
Table 1: Simplified Presentation of a Sovereign Balance Sheet
ASSETS LIABILITIES
Foreign reserves, gold, Special Drawing Base money
Rights
Local currency debt
Pension fund assets
Foreign currency debt
SWF
Pension fund liabilities
Other public-sector assets (state-owned
enterprises, real estate) Contingent claims: implicit guarantees (to
banks, etc.)
Present value of future taxes, fees,
seigniorage Present value of expenditures on economic
and social development, security,
government administration, benefits to other
sectors
Present value of target wealth to be left to

future generations

An initial approach to measuring a sovereigns economic balance sheet is to estimate the
market price and volatility of all its component assets and liabilities separately. However, to
do this, the present value of future income and expense flows has to be estimated. An
alternative method is to estimate the markets valuation of the balance sheet, as described by
Merton (1974, 1977) and Gray et al. (2007). An implied value for the sovereign's assets can
be estimated from the observed prices of liabilities. To do this, it is necessary to rearrange the
balance sheet entries and adopt an integrated presentation, subtracting the present value of
expenses from the present value of income, and subtracting the value of contingent liabilities
from assets. The two liabilities can then be valued as contingent claims on sovereign assets.
The foreign currency debt is considered as a senior claim, and the local currency debt plus
base money as a junior claim, which can be modelled as a call option on the total value of
the sovereign's assets. The value of the sovereigns assets and their volatility can then be
estimated as a function of the default barrier (promised payments in foreign currencies), (Gray
et al, 2007; Bodie and Brire, 2014).
Optimal Sovereign Wealth Management
From a theoretical standpoint, managing the wealth of a sovereign is similar to managing the
wealth of an individual (Merton, 1969; Bodie et al., 1992; Bodie et al., 2008), a pension fund
(Bodie et al., 2009) or a foundation (Merton, 1993). The sovereign receives tax revenues each
year. Part of these revenues are spent, and the residual is saved in SWFs, central bank
reserves, or public pension funds. Determining how much should be saved and how it should
be invested is a standard ALM problem.
We assume that the sovereigns objective is to maximise its expected utility, which is a
function of its Global Sovereign Surplus (GSS),5 depending on the allocation of the
5
Measured as sovereign assets minus sovereign liabilities.

sovereigns assets.6 The optimal allocation and the optimal expenditures of the sovereign
crucially depend on the nature and size of the fiscal asset and unconditional liabilities, and the
sources of their uncertainty. Bodie and Brire (2014) solve this problem analytically and show
that the optimal portfolio w* can be broken down into a performance-seeking portfolio and
three hedging demand terms for the variability of the fiscal surplus and external and domestic
debt:
1 (1 ) 1 (1 ) 1
w* = FA1 FA ,t FA FA, FS + FA1 FA, FL + FA FA, DL (5)
( 1)
with FA the vector of annualized expected returns of the n financial assets in the portfolio
over the investment horizon, FA their covariance matrix, the fraction of total sovereign
assets dedicated to financial wealth (the remainder is the fiscal surplus), the fraction of total
sovereign liabilities dedicated to foreign debt (the remainder is domestic debt), FA, FS FA, FL ,
FA, DL the covariance of the financial asset returns with the fiscal surplus, foreign liabilities
and domestic liabilities respectively.
These results shed new light on the optimal allocation of the sovereigns wealth. We
generalize previous results on SWFs asset allocations by introducing three additional sources
of risk affecting the sovereign balance sheet. Martellini and Milhau (2010) express the SWFs
preference in real terms and observe a hedging demand against realized inflation. Scherer
(2009a and b) identifies the optimal asset allocation of an SWF with non-tradable wealth and
observe a hedging demand against oil price variations. In a more general framework, taking
explicit account of all sources of risk affecting the sovereign balance sheet, three hedging
demand terms are added to the speculative portfolio. We recommend taking into account not
only the risks from inflation and fluctuations in natural resource prices, which both influence
6
We disregard other potential macroeconomic decision variables (tax rate, etc.), considered as constant, in order
to concentrate on the asset allocation choice.

the variability of the fiscal surplus, but all the risks stemming from the fiscal surplus, and
from foreign and domestic liabilities. Moreover, the fiscal surplus variability is influenced not
only by commodity prices and inflation volatility, but also by the sovereigns policies on
natural resource extraction, taxation, and so on.7
4. Practical Implementation
The practical implementation of sovereign ALM raises several difficulties. Traditional public
finance data are often incomplete and ill-suited to accurately estimation of the sovereign
economic balance sheet. This lack of data compromises the coordination of sovereign wealth
management with fiscal policy, monetary policy and public debt management. We discuss
institutional arrangements that could enable efficient coordination.
Traditional Public Finance Data and their limitations
To implement sovereign ALM, what really needs to be measured is the actual nature of
macroeconomic and financial risks, with their non-linear features (contingent liabilities
modelled as options, etc.), and the accumulation phenomena that lead to systemic risks. Flow
of funds statistics available in many countries provide balance sheet estimates of the
government sector but do not fully correspond to what is actually needed. The definition of
the government entity differs between countries8 and may not correspond exactly to our
broad definition of the sovereign. The IMF's GFS database, created in 2001, remedies these
differences with a unified base of 153 countries data on government balance sheets, with a
7
This leads to another important difference from the previous literature. In our framework, the variability of the
flow of revenues from the sale of natural resources needs to be hedged, not the fluctuations in commodity prices
themselves (Scherer (2009a, 2009b)). This has important implications, as the fiscal surplus may not have a
sensitivity of one to natural resource prices, as we will see in our estimation for Chile in Section 3.
8
In the US, the Flows of Funds statistics consider state and local governments (excluding employee retirement
funds), the federal government (including government-owned corporations and agencies that issue securities
individually) and the monetary authority. In Europe, the European Central Bank and Eurostat Euro Area
Accounts have a more restrictive definition. The general government sector comprises only central, state
(regional) and local government and the social security or pension funds belonging to it. It does not include
public enterprises, which are included in the corporate or financial sector and cannot be disentangled from it.

particularly broad scope for the sovereign.9 The IMFs GFS data nonetheless have significant
limitations. There is no evaluation of the present value of future tax revenues, or expenditures.
Moreover, there are no estimates of contingent liabilities, such as too-big-to-fail guarantees to
the financial sector and implicit guarantees to provide social benefits when various needs
arise. Finally, these data, which are purely accounting-based and generally available on an
annual basis, are not sufficient to measure the risks associated with each item. In the case of
sovereign balance sheets, risks are related on the one hand to market price fluctuations (for
commodities, exports, wage costs, etc.) that cause the governments income and expenditures
to fluctuate, and on the other hand to inventory changes (natural resource depletion,
population growth, etc.).
In 2000 the World Bank took the unprecedented step of measuring the wealth of nations
(World Bank, 2006 and 2011). The total wealth of each nation is estimated as the present
value of future flows of consumption. Consumption levels are based on past historical data
but are adjusted to be sustainable.10 Total wealth is broken down into: (1) produced capital
(machinery, structures and urban land), (2) natural capital (energy resources, mineral
resources, timber resources, non-timber forest resources, cropland, pastureland and protected
areas) and (3) intangible capital (human, etc.), calculated as a residual, the difference between
total wealth and the sum of produced and natural capital. These data are a very useful
supplement to the existing figures because they provide an estimate of stocks11 of natural
resources and intangible assets. World Bank estimates of natural and human capital can be
used to estimate the present value of the fiscal surplus, given a certain level of desired
9
It comprises not just the central government budgetary authority but also the central bank, SWFs, pension
funds, deposit insurance funds, state-owned enterprises, subnational governments and other government
agencies.
10
For years when adjusted net savings are negative, the actual consumption rate is added to adjusted net savings.
11
Flow variables are also available: depletion of natural resources, investment in education, domestic net
investment.

taxation. Unfortunately, these data were estimated in 2000 and 2006 for the World Banks
2006 and 2011 reports and are not available as a historical series.
The need for central coordination
To implement sovereign ALM in practice, there needs to be a high level of coordination
between institutions that control sovereign assets and sovereign liabilities (at least the central
bank, the debt management office, the treasury and the ministry of finance). What the most
efficient institutional arrangement would be is still an open question, and the few country
examples show that very different organizations are possible. New Zealand, Canada,
Denmark, Britain, South Africa and Turkey are the handful of countries that have made
significant steps in the direction of developing an ALM framework. In New Zealand and
South Africa, there is a specialized asset-liability management unit that analyses the
sovereigns balance sheet. In New Zealand, the mandate of the debt management office is to
keep the net foreign currency position close to zero, explicitly matching foreign currency
assets and liabilities and hedging exchange-rate movements. In Canada, ALM was introduced
for the tactical management of foreign reserves in 1997, with the goal of minimizing currency
and interest-rate risks by matching the assets to the liabilities funding them. In Turkey, debt
management is also defined in an ALM framework, in close cooperation with the reserve
management office.
In most of the example countries cited (Canada being an exception), the ALM exercise has
been performed by the debt management office, already responsible for cash management and
treasury services. This is not without drawbacks since the issuance of government debt might
also respond to other, possibly conflicting, objectives. Government debt has public good
characteristics, including setting the risk-free yield curve and providing highly liquid
securities. In Australia and Norway, for example, the government decided to continue debt
issuance even though there was no need for government borrowing, because of the

importance of sustaining functioning capital markets. On the other hand, assigning ALM to
the asset management offices would also make sense. The example of Canada, which gave the
central bank tactical reserves management office an ALM mandate, is a good example of this.
But responsibility for the wider government balance sheet would sit uneasily with central
bank independence, and there could be potential conflicts of interest with monetary policy.
The sovereign wealth fund would actually be an excellent candidate for the job of
implementing the sovereign ALM. In many countries, this may be facilitated by the fact that
the finance ministry is responsible both for debt issuance and fiscal policy and for
determining the SWFs strategic asset allocation.
In any case, a coordinated approach to the management of the national balance sheet would
necessitate central responsibility. Probably the most realistic scenario would be to encourage
more links and consultation between the different agencies, with detailed instructions from
the ministry of finance. South Africa has organized such a framework with a common
committee bringing together the South African reserve bank and the treasury. When South
Africa had a net negative forward currency position in the late 1990s, a strategy was
developed jointly by the reserve bank and the treasury to bring down this exposure. The
finance ministry might be the best candidate to lead this coordination, but the optimal
institutional arrangement may in the end depend on the political organization of each country.
5. Conclusion
This paper presents an analytical framework for sovereign wealth and risk management,
extending the theory of contingent claims analysis, and discusses its practical implementation.
A complete approach to the sovereign balance sheet is necessary to fully understand the
country's risks and determine how it can best manage its wealth. This supposes the broadest
possible definition of the sovereign, including, in particular, entities subordinated to the state,

such as the central bank, SWFs, pension funds, government agencies and state-owned
enterprises. The reason is that the funds, even if located in different entities, become fungible
if a crisis arises. This approach also requires all balance sheet items, both assets and liabilities,
as well as their risks, to be measured precisely. To do this, it is necessary to measure not only
the sovereigns financial wealth, but also its human and natural capital. Similarly, a relatively
precise understanding of the governments economic objectives and an accurate estimate of
contingent liabilities are also needed. A sovereign ALM strategy can thus be developed for
managing asset risks in a way that is consistent with the sovereign entitys liabilities. One
significant application of this analytical framework is the management of financial wealth
under direct state control. The optimal allocation of sovereign wealth should involve a
performance-seeking portfolio and three hedging portfolios for the variability of the fiscal
surplus and external and domestic debt.
Our ambitious approach has limitations. First, to concentrate on asset allocation, we consider
macroeconomic variables as exogenous. In practice however, the sovereign benefits from
many more policy instruments, including taxation level. It can also inflate or repudiate its debt
(Landon-Lane and Oosterlinck, 2006). A general equilibrium model endogenizing all of the
states decision variables would be more realistic, but also much more complex. Second, the
practical implementation of our ALM framework requires reliable macroeconomic data on a
regular basis. Moreover, strong coordination is needed between the sovereign entities. This
coordination involves the institutions that manage both sides of the balance sheet: the central
bank and sovereign wealth fund on the asset side, and the debt management office on the
liability side. The ministry of finance is particularly well positioned as a central institution to
facilitate this coordination. However, even if the implementation of the ALM framework for
SWF asset allocation is an unfeasible first-best solution for many countries, far removed from
current practice, it can nevertheless be thought of as providing useful guidelines for efficient

management of sovereign wealth. In particular, it should help to improve the diversification
of sovereign assets and the hedging of important risk factors affecting the sovereign balance
sheet.

Acknowledgment
The authors are grateful to Ariane Szafarz for her comments on a previous version of this
paper.
References
Alfaro, L., Kanczuk, F., 2009. Optimal reserve management and sovereign debt, Journal of
International Economics, 77(1), 23-36.
Aglietta M. (2010), Sustainable Growth: Do We Really Measure the Challenge?, mimeo.
Aizenman J. and Glick R., Asset Class Diversification and Delegation of Responsibilities
between Central Banks and Sovereign Wealth Funds, NBER Working Paper, N16392.
Ang A. (2012), The Four Benchmarks of Sovereign Wealth Funds, in Bolton P., Samama
F., and Stiglitz J., eds., Sovereign Wealth Funds and Long-Term Investing, p 94-105,
Columbia University Press.
Avendano R. and Santiso J., 2009, Are Sovereign Wealth Funds Investments Politicall
Biased? A Comparison with Mutual Funds, OECD Development Center Working Paper, N
283.
Bagliano F. C., Fugazza C. and Nicodano G. (2009), Pension Funds, Life-Cycle Asset
Allocation and Performance Evaluation, Univesita di Torino Working Paper, March.
Beck R. and Rahbari E. (2008), Optimal Reserve Composition in the Presence of Sudden
Stops, the Euro and the Dollar as Safe Haven Currencies, European Central Bank Working
Paper, No 916, July.
Beck R. and Weber S. (2011), Should Larger Reserve Holdings Be More Diversified?,
International Finance, 14(3), p. 415-444.
Bernardell C., Cardon P., Coche J., Diebold F.X. and Manganelli S. (2004), Risk
Management for Central Bank Foreign Reserves, European Central Bank Publication, May.
Bernstein, S., Lerner J. and Schoar A. (2013), The Investment Strategies of Sovereign
Wealth Funds, Journal of Economic Perspectives, 27(2), April, p. 219-238.
Bodie Z. and Brire M. (2014), Sovereign Wealth and Risk Management: A Framework for
Optimal Asset Allocation of Sovereign Wealth, Journal of Investment Management,
Forthcoming.
Bodie Z., Merton R. and Samuelson W.F. (1992), Labor Supply Flexibility and Portfolio
Choice in a Life Cycle Model, Journal of Economic Dynamics and Control, 16, p. 427-449.
Bodie Z., Ruffino D., and Treussard J. (2008), "Contingent Claims Analysis and Life-Cycle
Finance." American Economic Review, 98(2), p. 29196.

Bodie Z., Detemple J. and Rindisbacher M. (2009), Life-Cycle Finance and the Design of
Pension Plans, Annual Review of Financial Economics, 1, p. 249-286.
Bortolotti B., Fotak V., Megginson W.L. (2013), The Sovereign Discount in Sovereign
Wealth Fund Investments, University of Oklahoma Working Paper.
Brown A., Papaioannou M. and Petrova I. (2010), Macrofinancial Linkages of the Strategic
Asset Allocation of Commodity-Based Sovereign Wealth Funds, IMF Working Paper, No
10/9, January.
Caballero R.J. and Panageas S, (2005 a), Contingent Reserves Management: an

Applied Framework Journal Economa Chilena (The Chilean Economy), Central Bank of
Chile, 8(2), p. 45-56.
Caballero R.J. and Panageas S, (2005b), A Quantitative Model of Sudden Stops and External
Liquidity Management, NBER Working Paper, N 11293.
Chhaochharia V. and Laeven L., (2009), The Investment Allocation of Sovereign Wealth
Funds, SSRN Working Paper, N1262383, available at:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1262383
Clark G.L. and Monk A.H.B. (2010), Government of Singapore Investment Corporation:
Insurer of Last Resort and Bulwark of Nation State Legitimacy, Pacific Review, 23(4), p.
429-451.
Das U.S., Lu Y., Papaioannou M. G., and Petrova I. (2012), Sovereign Risk and Asset and
Liability ManagementConceptual Issues, IMF Working Paper, No 12/241, October.
Dyck I.J. and Morse (2011), Sovereign Wealth Fund Portfolios, Chicago Booth Research
Paper, N11-15.
Gray D.F., Merton R.C. and Bodie Z. (2007), Contingent Claim Approach to Measuring and
Managing Sovereign Credit Risk, Journal of Investment Management, 5(4), p. 1-24.
Gray D.F. and Malone S. (2008), Macrofinancial Risk Analysis, Wiley Finance, UK.
KMV (2002), "Modelling Default Risk", Moodys KMV.
Landon-Lane J., K. Oosterlinck (2006). Hope Springs Eternal: French Bondholders and the
Soviet Repudiation (1915-1919), Review of Finance, 10, p. 507-535.
Martellini L. and Milhau V. (2010), Asset-Liability Management Decisions for Sovereign

Wealth Funds, Edhec-Risk Institute Publication, October.
Merton, R.C. (1969), Lifetime portfolio selection under uncertainty: The continuous-time
case, Review of Economics and Statistics, 51, p. 247-257.
Merton, R.C. (1974), On the Pricing of Corporate Debt: The Risk Structure of Interest
Rates, Journal of Finance, 29, p. 449-470.

Merton, R.C. (1977), An Analytic Derivation of the Cost of Loan Guarantees and Deposit
Insurance: an Application of Modern Option Pricing Theory, Journal of Banking and
Finance, 1, p. 3-11.
Merton, R.C. (1993), Optimal Investment Strategies for University Endowment Funds,
Studies of Supply and Demand in Higher Education, Clotfelter C.T. and Rothschild M. ed.,
University of Chicago Press, p. 211-242.
Raymond H. (2010), Sovereign Wealth Funds as domestic investors of last resort during
crises, International Economics, 123, p. 121-159.
Scherer B. (2009a). A Note on Portfolio Choice For Sovereign Wealth Funds, Financial
Markets and Portfolio Management, 23(3), 315-327.
Scherer B. (2009b). Portfolio Choice for Oil Based Sovereign Wealth Funds, Journal of
Alternative Investments, 13(3), 24-34.
World Bank (2006). Where is the Wealth of Nations? Measuring Capital for the 21s Century,
The International Bank for Reconstruction and Development/The World Bank.
World Bank (2011). The Changing Wealth of Nations, The International Bank for
Reconstruction and Development/The World Bank.

WP-032
Unexpected Returns.
Methodological Considerations on
Expected Returns in Uncertainty
Sylvie de Laguiche,
Head of Quantitative Research, Amundi
Gianni Pola,
Balanced Quantitative Research, Amundi
November 2012
The recent crisis has brought increasing uncertainty in the exercise

of forecasting long-term returns due to:
- significant changes in risk levels and observed risk premia which
make calibration on recent history more difficult;
- unpredictable effects of non-conventional monetary policies on
macro-economic variables, especially inflation;
- the fact that some market variables (interest rates) are in uncharted
territory.
For strategic allocation of diversified portfolios, despite these
major changes, the traditional Sharpe ratio approach, equal for all
asset classes, may still be of interest because it provides good risk
diversification.
As for establishing a forward looking expected return which
can serve as a reference for discounting liabilities, a more careful
approach would be required. It would have to take into account
the observed dependency on macro-economic scenarios and the
distortion in favour of less risky or less liquid assets which has been
observed and documented in literature.
Complementary approaches, mixing skills in macroeconomics,
econometrics and quantitative finance may be required now in
order to overcome the challenge.

0. Introduction
Expected returns are closely related to the portfolio allocation: according to the Markowitz
portfolio selection (1952), once agreed on the market risk model and the investors risk
aversion, expected returns unambiguously determine the portfolio, and, conversely, reverse
optimisation techniques (Cantaluppi; 1999) allow us to relate a given (optimal) portfolio to a
set of expected returns. While the finance industry and academia are aware of the inadequacy
of the Markowitz model, it is clear that estimation of expected returns is a key issue for
strategic asset allocation. A pension fund, for example, needs expected returns on asset
classes for:
x building a strategic allocation; and
x assessing returns on assets and setting an appropriate discount rate for liabilities.
While in the former, risk-adjusted hierarchy between assets is more relevant, in the latter the
level of returns itself is crucial to determine discount rates. In this document, we refer to long
horizons - at least ten years - somewhat in line with the average liabilities in pension funds.
Unfortunately, while volatility is somewhat predictable as "large changes tend to be followed

by large changes, of either sign, and small changes tend to be followed by small changes"
(Mandelbrot, 1963), returns are un-correlated over time and probably unobservable 1. This is
particularly true today: the recent crisis in the Eurozone and the previous sub-prime crisis in
2008, remind us that returns dynamics are not stationary in time, and that prevailing market
conditions can strongly affect returns and cause them to deviate significantly from the long-
term average.
Many financial variables are in a peculiar territory, never reached going back to many
decades, and more than a century in some cases:
x the 10-year US Treasury yield is at its lowest level since 1871 (see chart 1);
x the two-year yield synchronously negative in many countries in 2012 (Germany,

Netherlands, Switzerland, Denmark; see chart 2, left panel); and
x the decoupling of the Eurozone: the two-year yield divergence (see chart 2, right
panel).
1 Historical averages are good estimators if and only if the (underlying) stochastic process is stationary. Rapidly changing regimes and non-stationary dynamics prevent us
from estimating returns from historical averages, making them effectively unobservable.

Chart 1
10-year US Treasury yield
16
12
0
1871
1874
1878
1882
1886
1890
1894
1898
1902
1906
1910
1914
1918
1921
1925
1929
1933
1937
1941
1945
1949
1953
1957
1961
1965
1968
1972
1976
1980
1984
1988
1992
1996
2000
2004
2008
2012
Chart 2
Low Yield (two-year bond yield) Eurozone Divergency (two-year bond yield)
10 10
8 8
6
6 EURO
4 PERIPHERALS
4
2
2
0
EURO
Sep-90
Sep-92
Sep-94
Sep-96
Sep-98
Sep-00
Sep-02
Sep-04
Sep-06
Sep-08
Sep-10
Sep-12
0
-2 CORE
Jun-10
Jun-11
Jun-12
Dec-09
Feb-10
Apr-10
Aug-10
Oct-10
Dec-10
Feb-11
Apr-11
Aug-11
Oct-11
1
ec-11
Dec-11
eb-12
Feb-12
Apr-12
Aug-12
ug-12
-2
u
GERMANY FRANCE NETHERLANDS
BELGIUM AUSTRIA JAPAN GERMANY FRANCE ITALY
UK CANADA SWITZERLAND NETHERLANDS BELGIUM AUSTRIA
DENMARK SWEDEN US SPAIN PORTUGAL GREECE
During the fifties and sixties, expected returns were considered to be time-dependent. They
were estimated from asset fundamentals (e.g. the dividend discount model for stocks, the
yield for a bond, etc.). In the next two decades, thinking on the subject changed according to
the work of Ibbotson and Sinquefield (1976a and 1976b). They modelled equity expected
returns as a time-varying baseline given by cash or bonds, plus a constant term, the long-term
equity risk premium. In the eighties, the financial community came back to the origin: risk
premia were believed to be time-varying quantities themselves. This counterrevolution started
with Campbell and Shiller (1988a, 1988b) and continued with the works of Asness (2000),
Arnott (2002), and Fama & French (1989). Recently, Ilmanen (2011) faced the problem of
estimating expected returns on major asset-classes broadening the investments to non-
traditional assets (commodities, real estates), investment styles (value, trend, carry,
volatilities), and underlying factors (growth, inflation, illiquidity, and tail risks).

Current thinking today is more complex and closer to experimental evidences. Modern
approaches:
x derive asset-prices from asset returns co-variation with bad times;
x are based on multiple risk-factor models;
x incorporate time-varying risk premia, skewness and liquidity preferences;
x take into account supply-demand effects on asset prices; and
x model market inefficiencies due to investor irrationalities and market frictions.
We refer to Ilmanen (2011) as a comprehensive review on the subject.
Time-varying risk premia can be profitable if and only if investors are able to predict them.
Nevertheless the recent crises again call into question the predictability of asset-class risk
premia: neither a very simple normative approach like the Sharpe ratio can be applied easily
given the uncertainty on risk-free and high volatility of risk premia, nor can equilibrium-based
models be easily estimated given the rise in macroeconomic volatility. Today, researchers and
practitioners prefer to incorporate uncertainty and estimation errors in expected returns,
leading to Bayesian approaches (Black & Litterman model, 1990) and robust asset allocation
models (Meucci, 2011). Recently more extreme approaches emerged in the financial arena,
aiming to construct portfolios without any specific views on expected returns (e.g. minimum
variance, maximum diversification, risk parity approaches). Indeed, while the needs for
expected returns might be questionable in an optimisation process, it is evident that the long-
term estimates are absolutely crucial for pension funds and insurance companies to design
investment strategies to match their liabilities.
Rather than providing a specific recipe for estimating expected returns and computing return
figures, our aim here is to investigate this issue, stimulating the reader with relevant questions,
considerations, and few side empirical analyses to sustain our ideas. This manuscript is about
expected returns. Nevertheless, we will complement our considerations with historical
performance figures: even if the recent history is probably too peculiar to make extrapolations
from long-term time-series, there are always lessons that can be learned from history.
The paper is organised as follows. In section 1 we will start investigating how the crisis
changed the relationship between risk and return, and then questioning the possibility today of

calibrating a model on past history. Hence in section 2 we will introduce the two main classes
of estimation methods (Statistical and Equilibrium-based approaches) highlighting the pros
and cons, and stressing the peculiarity of the estimation problem in this specific era for
financial markets: in sections 3, 4, 5 and 6 we respectively critically evaluate the Sharpe ratio
approach, factors-model focusing mainly on the consequence of deflationary or inflationary
scenarios on asset-returns, reverse optimisation techniques to get implied returns from
equilibria allocation, and finally the equilibrium-based models. In section 7 we will then
illustrate how portfolio construction can help to handle uncertainty in expected returns to
deliver robust strategic asset allocation. A conclusive section delineates the main take-home
messages that we think are relevant to address expected return estimates in uncertainty. The
annex includes all the details about the charts and data-analysis presented in sections 1 and 3.
1. The financial crisis questions:
1.1. The relationship between risk and return
Past average returns are rarely related to future performance. Nevertheless past bad
performance can be somehow indicative of future returns: risk premia correspond to the
premium that an investor requires in order to be compensated for poor performance in bad
times. Chart 3a (redrawn from Ilmanen 2011) plots the compound average real returns from
1960 to 2009 in some US markets as a function of the average returns in bad times (1974,
1981, 2008) for financial markets and global economy. Even if the relationship is not clearly
linear, it is evident that assets that performed poorly in crisis periods delivered higher returns
in the long-run. This intuition is the key idea of modern theories on asset-pricing which, rather
than deriving prices from future discounted cash flows, build expected returns from asset
returns co-variation with bad times.
Given this argument, we measured the volatilities, maximum draw downs (MDD) 2, and
performance of some asset classes in two historical periods: pre-crisis sample (1990-2007)
and the full sample (1990-2012); see chart 3b in annex A1 for more details. The aim is to
show how the crisis period (2008-2012) brought new stress that was not priced in the pre-
2
While the MDD is related to the tail-risk and is sometimes related to specific historical events (local measure),
the volatility is a more general measure of returns dispersion (global measure).

crisis sample: higher volatility and worst MDD should suggest a higher premium for future
returns due to uncertainty.
The Eurozone debt crisis determined a marked increase in MDDs and volatilities in Euro
peripherals nominal bonds (the MDD of Italian bond doubled) and Euro inflation-linked
indices. The 2008 crisis was dramatic for spread markets: all volatilities increased markedly,
MDD for US investment grade more than doubled, World high-yield doubled its MDD, Euro
credit markets suffered less with respect to the pre-crisis period. The worst equity markets in
the crisis period were US, Euro peripherals, Australia and Emerging markets; the volatilities
of equity markets increased but only marginally with respect to the pre-crisis period.
Commodity indices registered new MDDs in the crisis sample, gold performed better during
the crisis (the average compound return increased); volatilities increased.
The crisis period increased long-term volatility in most of the asset classes. Most volatility
estimation models for strategic asset allocation are mainly based on Exponential Smoothing
techniques which overweight the recent history with respect to more distant history.
According to these approaches, the ex-ante volatility of many asset allocations is likely to
increase especially in the low-medium risk investor profiles where bond markets are mostly
allocated.
Chart 3 a
Redrawn from Ilmanen (2011)
8%
Compound Average Real Return, 1960-2009
7%
Small-Cap Stocks
6%
Stock Markets 5%
Real Estate
Commodity Futures 4%
Treasury Bonds High Yield Corporate Bonds

3%
Investment Grade Corporate Bond
2%
Treasury Bills
1%
0%
5% 0% -5% -10% -15% -20% -25% -30%
Average Real Return in Bad Times (1974, 1981, 2008)

1.2. The possibility of calibrating a model on past history
While recent history (few months) can moderately reflect near future performance (as trend-
follower strategies and CTAs demonstrated in the long run), it is rather dangerous to
extrapolate over multi-year time windows when reversals often take place.
In historical estimates, the sample period is crucial: long time windows reduce sample
specificity and allow for more robust statistical inference, but may miss structural changes
happening in the market and may not be able to catch up non-stationary risk premia dynamics.
On the other hand, small samples may lead to nonsignificant results, and give an incomplete
picture of the financial dynamics. Given the level of many financial indicators today, and the
emergence of structural changes in the world economy, it is rather difficult to select an
appropriate sample to estimate econometric models: recent history may or may not be
relevant. Chart 4 reports the rolling historical volatilities of German, Italian, and Spanish 7-10
year bonds (upper panel), and their rolling historical correlation to the EMU equity markets
(lower panel): the charts clearly show that the Eurozone moved from divergence to
convergence in the new millennium, and recently from convergence back to divergence again
with the debt crisis. The most relevant sample to estimates markets returns in this case is not
obvious. Financial markets are moving from pre-crisis equilibrium to a new one: at the
moment we are in between, and uncertainty prevents us from tracing a clear picture.
We calculated the new millennium performance (2000-2012) of US 10-year bonds and US

equity, and compared them with the past performance since 1871. In chart 5a we computed
the annualised compound returns for each decade. The average nominal figures for bonds and
equity are respectively 4.45% and 8.19% (real figures are respectively +2.48% and 6.14%).
Equity performance in the new millennium was very poor compared to the post-war period
1950-2000; however we should admit that the post-war period was misleadingly favourable to
equity, if compared with the full sample starting from 1871. In chart 5b we rearranged the
data in a scatter plot of the real performance of bonds and equity.

Chart 4
BOND 7-10 (historical volatility 1 yr rolling)
20%
18%
16%
14%
12%
10%
8%
6%
4%
2%
0%
Jan-93
Jun-97
Jan-98
Jun-02
Aug-93
Apr-94
Nov-94
Oct-96
Oct-03
Jan-05
Jun-09
Jan-10
Sep-98
Apr-99
Dec-99
Aug-00
Nov-01
Aug-05
Apr-06
Dec-06
Jul-95
Oct-08
Mar-96
Mar-01
Feb-03
Sep-10
Dec-11
Aug-12
May-04
Jul-07
Mar-08
May-11
germany 7-10 italy 7-10 spain 7-10
BOND 7-10 (historical correlation to Equity EMU; 1 yr rolling)
1
0,8
0,6
0,4
0,2
0
-0,2
-0,4
-0,6
-0,8
-1
Oct-94
May-95
Mar-94
Jul-96
Oct-97
May-98
Jul-99
Feb-97
Feb-00
Apr-01
Jul-02
Feb-03
Apr-04
Apr-07
Mar-10
Oct-10
Dec-95
Dec-98
Nov-07
Jan-93
Aug-93
Sep-00
Nov-01
Sep-03
Nov-04
Jun-05
Jan-06
Sep-06
Jun-08
Jan-09
Aug-09
Jun-11
Jan-12
Aug-12
germany 7-10 italy 7-10 spain 7-10
Chart 5a
Nominal returns US 10-year bond & equity (1871-2012)

Annualized performance
20%
15%
10%
5%
0%
-5%
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2012 avg
bond equity cpi

Chart 5b
Real performance per decade (1871-2012)
20%
1930
15%
1960 2000
1870
10%
Equity perf (ann)
1950 1900 1990

AVG
5% 1910 1890
1970
1880
1940
1980
0%
2012
-5% 1920
-10%
-10% -5% 0% 5% 10% 15% 20%
Bond perf (ann)
Equity in the new millennium did not offer any spread over inflation: this pattern on equities
was similar to the seventies when high inflation caused a bear equity market 3. On the bond
side, performance in the new millennium was higher than the long-term average: bonds
benefited from the marked fall in interest rates (see chart 1). A final remark on correlation
between performance of bond and equity: the scatter plot shows that, on a ten-year horizon,
they were on average positively correlated. This paradigm was not verified only from the
beginning of 1941 to the end of 1980, when bonds suffered from the increase in 10-year
interest rates from 1.95% to 12.84%, delivering a negative real performance.
This analysis highlights the importance of extending the historical sample. In addition, in
order to get more robust estimates it might be wiser to complement model-free historical
figures with models from financial theories, models from behavioural theories, forward-
3
In the 1970-1980 decade the annual inflation rate was +8.05%.

looking market indicators (e.g. valuation ratios), discretionary views may help as well in case
of strong structural changes.
2. Should expected returns reflect an unbiased view about the market, or should
they incorporate economic scenario expectations and/or recent history
information such as valuation?
The approaches to estimate expected returns can be divided in two main categories: statistical
methods and equilibrium-based models. The separation between them is not very neat. Both
of them are certainly statistical as most of the models are derived according to econometric
relations, and both of them can be built on macroeconomic relationships among variables. The
main differences between them are that the former do not assume any specific
macroeconomic views, and are not equipped with valuation tools. We would say that while
statistical models estimate returns mainly looking at the past history, equilibrium-based
models forecast future returns incorporating forward looking macro variables.
2.1 Statistical methods.
We intend with these approaches:
x a normative method based on Sharpe ratio hypothesis;
x factors-model approaches based either on statistical analysis tools (e.g. principal

component analysis, independent component analysis), or on more fundamental
methods looking at asset-classes as vehicles of more systematic factors (e.g. CPI,
GDP, Illiquidity, Risk Aversion); and
x reverse optimisation techniques (Cantaluppi; 1999) to get expected returns from an

equilibrium portfolio.
The above approaches:
x lead to unbiased and, hence, more robust expected returns,

x are good candidates for portfolio optimisation leading to well-posed solutions and
diversified portfolios 4;
x but are not very appropriate for computing discount rate for pension funds and
insurance companies.
2.2 Equilibrium-based methods.
We mean models based on macroeconomic assumptions (e.g. long-term expectations on

inflation and GDP) and valuation arguments. They are:
x good candidates for computing discount rate;
x superior to forecast future returns;
x sometimes they lead to corner solutions in an optimiser.
In this document we do not face the class of Behavioural models. These approaches enable us
to explain many observed anomalies in risk premia 5, and might be useful to diversify
statistical and equilibrium models. This way may lead to more robust estimations.
3. Sharpe ratio, is a simple constant figure still relevant?
Sharpe ratio is a risk-adjusted performance measure, and it is usually used to assess the
efficiency of an asset-class, investment strategy or fund. Sharpe ratio is usually computed as
the average excess return of a risky asset over the risk-free rate and normalised for historical
volatility. We prefer instead to compute the Sharpe ratio as the difference between the
annualised compound returns of the risky asset and the risk-free rate divided by the
annualised volatility. According to this formulation the expected return can be expressed as
the risk-free plus a term given by the ex-ante volatility times the Sharpe ratio.
Sharpe ratio presents the following difficulties:
x What is the risk-free rate?
4
Maximum-diversification portfolios are obtained maximising the (ex-ante) Sharpe ratio under the hypothesis of
constant Sharpe ratio across all asset-classes (Choueifaty and Coignard, 2008).
5
The equity premium puzzle indicates the difficulty of explaining the observed equity risk premium within
standard macroeconomic models. Behavioural models provide an explanation (see Bernatzi and Thaler 1995, and
Barberis and Huang 2001).

Usually it corresponds to a short-term or long-term government yield, but which one now?
Government debt may no longer be risk-free. Chart 6 reports the 10-year CDS for Italy and
Germany from 2005.
x Instability over time
Chart 7 plots the Sharpe ratio for the US, Japanese, UK, German and French equity markets.
Computations have been performed on rolling windows of 10- and three-year horizons. The
longer the time window, the more stable the Sharpe ratio. Table 1 in annex A2 reports some
descriptive statistics: the median and the average figures are reported for each country and
three-, five- and 10-year time horizons. For the equity markets, 0.25 is a reasonable estimate
for the 10-year horizon.
Chart 6
CDS 10 yrs (USD)
600
500
400
300
200
100
0
04/01/2005
04/04/2005
04/07/2005
04/10/2005
04/01/2006
04/04/2006
04/07/2006
04/10/2006
04/01/2007
04/04/2007
04/07/2007
04/10/2007
04/01/2008
04/04/2008
04/07/2008
04/10/2008
04/01/2009
04/04/2009
04/07/2009
04/10/2009
04/01/2010
04/04/2010
04/07/2010
04/10/2010
04/01/2011
04/04/2011
04/07/2011
04/10/2011
04/01/2012
04/04/2012
04/07/2012
04/10/2012
italy germany

Chart 7
Historical Sharpe ratio (10yrs rolling)
2,5
1,5
0,5
-0,5
-1
-1,5
-2
juil.-65
juil.-67
juil.-69
juil.-71
juil.-73
juil.-75
juil.-77
juil.-79
juil.-81
juil.-83
juil.-85
juil.-87
juil.-89
juil.-91
juil.-93
juil.-95
juil.-97
juil.-99
juil.-01
juil.-03
juil.-05
juil.-07
juil.-09
juil.-11
USA JAPAN GERMANY FRANCE UK
Historical Sharpe ratio (3yrs rolling)
2,5
1,5
0,5
-0,5
-1
-1,5
-2
juil.-65
juil.-67
juil.-69
juil.-71
juil.-73
juil.-75
juil.-77
juil.-79
juil.-81
juil.-83
juil.-85
juil.-87
juil.-89
juil.-91
juil.-93
juil.-95
juil.-97
juil.-99
juil.-01
juil.-03
juil.-05
juil.-07
juil.-09
juil.-11
USA JAPAN GERMANY FRANCE UK
x Differences among asset classes. Does it make sense to make a normative assumption
of equal Sharpe ratios for all asset classes?
Recently, practitioners and researchers (see Frazzini & Petersen 2010) documented some
anomalies regarding the Sharpe ratio. We report in the following the most relevant:
x On historical basis low volatility asset-classes delivered higher Sharpe ratio6. This
evidence inspired minimum-variance investments in the equity markets. We tested this
hypothesis in different baskets, as follows:
6
As reported by Asness et al. (2011), if investors are leverage averse, low-beta assets will offer higher risk-
adjusted returns, and high-beta assets lower risk-adjusted returns. Leverage aversion breaks the standard CAPM.

Diversified basket. We considered a diversified investment universe in the US market
since 1970 (see annex for details). Chart 8 (top left panel) reports the Sharpe ratio as a
function of the volatility. Each spot stands for a specific asset-class. The plot shows
clearly a reverse relation of the Sharpe ratio to the volatility (linear regression; R2 is
0.9401). When dealing with diversified investment universe, this relation is
particularly sample specific: restricting the analysis within each macro asset-class can
alleviate this problem.
Equity markets. In chart 8 (top right panel) we investigated the bias of Sharpe ratio
towards low volatility segments measuring the Sharpe ratios for each sector of US
equity market, and plotting them against the volatilities. The plot confirms the
anomaly, even if the relation is noisy (linear regression; R2 is 0.2210). In chart 9
(lower panels) we report two pictures taken from Baker et al. 2011. They studied the
US equity market from January 1968 to December 2008.
Chart 8
US Diversified (1971-2012) US MSCI Sectors (1995-2012)
0,60 0,50
0,50 0,40
0,30
0,40
sharpe (ann)
sharpe (ann)
0,20
0,30
0,10
0,20
0,00
0,10 -0,10
0,00 -0,20
0% 5% 10% 15% 20% 25% 0% 10% 20% 30% 40%
vol (ann) vol (ann)

Each month they sorted stocks into five groups according to their historical volatilities
calculated on five-year trailing windows. They did the analysis twice: left panel
reports the analysis performed on all the stocks, right panel on the 1,000 largest stocks
according to their market capitalisation. The plots show clearly the outperformance of
low volatility stocks over high volatility ones. They complement the analysis replacing
volatility as a risk measure with beta, and they found similar results.
Bond markets and Credit markets. As reported in Ilmanen (2011) a similar rule holds
for bond and credit markets: lower maturity bonds reported a higher Sharpe ratio than
longer maturity ones, and similarly higher quality credit buckets historically delivered
higher Sharpe ratios than lower rated credit ones.
Diversified indices delivered higher Sharpe ratios. We tested this argument in chart 9.
We measured the Sharpe ratio for the EMU equity markets from 2002 to 2007. The
Sharpe ratio is 1.30 which is greater than most of each countrys ones. The arithmetic
average of the Sharpe ratio cross-countries is 1.09 (the weighted average according to
the market cap is 1.19; we made an approximation of keeping the market cap constant
over time). The result confirms the benefit of diversification.
Illiquid asset classes historically delivered higher Sharpe ratios with respect to liquid
assets (see Ilmanen, 2011). In the long run, less liquid assets offer a compensation for
the higher trading costs and lower flexibility to rebalance portfolio positions. From an
estimation point of view, illiquid assets tend to exhibit smoothed prices thus leading to
underestimation of the true risk; this estimation error is certain to artificially increase
the Sharpe ratio.
Chart 9
Sharpe ratio EMU area 2002-2007
2,00 1,73
1,60 1,32 1,30
1,09 1,06 1,09 1,19
1,20 0,85 0,94
0,80 0,62
0,40
0,00
EMU avg
SPAIN
NETHERTLAND
BELGIUM
FINLAND
EMU
GERMANY
ITALY
EMU avg2
FRANCE

The observation that Sharpe ratio of asset-classes is higher for low volatility and more
diversified assets seems to be sufficiently documented in order to be taken into account.
However, the level of this anomaly may be sample specific; as an example, the exceptionally
favourable Sharpe ratio of bonds has been driven by a significant decline in interest rates.
A more detailed approach than the constant Sharpe ratio method may be required to set up the
expected returns to calculate the discount rate. However when it comes to portfolio
construction we should note that taking advantage of high Sharpe ratio for low volatility asset
classes is only possible when leveraging portfolios or for very low risk profiles. In addition,
the constant Sharpe ratio approach leads to a better diversified portfolio in terms of risk;
therefore except in very particular case this approximation may be kept to building strategic
allocation in diversified portfolios.
x Non-normality of asset classes
The Sharpe ratio implicitly assumes volatility as the risk measure. In case of strong
asymmetric and fat-tailed asset classes, the approximation is too rough, and the Sharpe ratio
becomes unreliable. In this case many practitioners and researchers prefer to correct the
formula by replacing the volatility with the downside volatility (Sortino ratio), or maximum
draw down. Nevertheless we should remind the reader that volatility misses some important
aspects of risk: liquidity risk, default risk, higher order statistics and tail risk, model risk,
timing risk, valuation risk, and fundamental risk, and that is going to have implications on the
estimation of the Sharpe ratio.
4. New trends in asset allocation look at assets as vehicles of more fundamental

macroeconomic variables, stress and liquidity indicators. It is evident that
potential inflation caused by unconventional monetary policy of central banks on
one hand, and possible extended recession and hence deflation on the other hand,
make inflation a key variable in order to decipher the future market scenario.
Hence, what are the implications of inflationary or deflationary scenarios on
asset returns?
Inflation refers to a rise in the consumer price level and consequently to a reduction in the real
value of money. While rising inflation is usually related to poor conditions in the real
economy, disinflation (i.e. a slowdown of inflation to lower levels) is commonly associated

with better economic conditions. Deflation consists in a fall in consumer prices: it hurts the
real economy because of a potential spiral in consumer prices, and the consequent recession
and depression. Deflation affected many countries in the past: the USs Great Depression in
the 30s, Japan from 90s to present times, and Hong Kong from 1997 to 2004.
The first issue we want to address is the likelihood of deflationary and inflationary scenarios
on a historical basis. Chart 10a reports the annual inflation rates for US (from 1871) and
Sweden (from 1900).
It is evident that the last millennium was mainly characterised by positive inflation rates;
hence it is questionable whether deflationary periods were less likely than inflationary ones.
As Reinhart and Rogoff (2011) demonstrated, on longer historical perspective, inflation and
deflation were both well represented: according to their study the 20th century itself represents
an anomaly. Secondly we analysed the historical performance of asset classes conditional to
contemporaneous inflation rate variation (on an annual basis). We divided the inflation rates
into buckets to separate different scenarios:
x deflation: negative inflation rates;
x low inflation: inflation rates between 0% and 2%;
x moderate inflation: inflation rates between 2% and 5%;
Chart 10a
Inflation rates US (1871-2012) and Sweden (1900-2012)
50%
40%
30%
20%
10%
0%
-10%
-20%
-30%
1871
1876
1881
1886
1891
1896
1901
1906
1911
1916
1921
1926
1931
1936
1941
1946
1951
1956
1961
1966
1971
1976
1981
1986
1991
1996
2001
2006
2011
us sweden

x high inflation: inflation rates between 5% and 10%;
x very high inflation: inflation rates higher than 10%.
Then we computed the annualised real performance of major asset classes conditional to the
above inflation scenarios. Chart 10b reports the outcome of the analysis for US 10-year bonds
and equity from 1871 to September 2012, and for Swedish bonds and equity from 1900 to
September 2012. For each bucket we plot the average, the minimum and maximum
performance as a proxy of the dispersion. Three main messages emerge:
x nominal bonds are a good hedge for deflationary scenarios;
x high inflation hurts nominal bonds: inflation rates greater than 5% lead to negative
average performance in US and close to zero in Sweden;
x the high dispersion of returns for equity markets highlights that inflation is not the main
driver in this case.
In chart 10c we performed the same exercise for US corporate investment grade all maturities,
US corporate investment grade intermediate maturity, commodity GSCI and gold. We
investigated the dependency with respect to the US inflation rates. Time series are shorter in
this case: our sample is from 1970 for investment grade corporate and commodity GSCI, and
from 1927 for gold 7. The main take-home messages are:
x corporate investment grade mimics the behaviour of bonds, hence inflation hurts corporate
investment grade;
x commodity GSCI is a good hedge against inflationary scenarios;
x gold does not reward significantly over inflation on high inflation scenarios (as noted in
footnote 7, the very high inflation scenario is not significant in this case).
In Pola (2013) we will more carefully illustrate the dependency of asset returns on
macroeconomic variables, showing how to segment asset classes according to their attitude to
polarise with respect to variations of macroeconomic variables (rising or falling scenarios).
7
In this case results on extreme scenarios (deflation and very high inflation) should be taken with care. The
sample from 1970 only has three entries for very high inflation scenarios and zero for deflationary scenarios. The
sample from 1927 only presents four entries for very high inflation scenarios and seven for deflationary
scenarios.

Chart 10b
US Bond 1871-2012 conditional to CPI rate buckets US Equity 1871-2012 conditional to CPI rate buckets
60% 60%
50%
40%
40%
real performance (annualized)

30% 20%
20%
0%
10%
0% -20%
-10%
-40%
-20%
-30% -60%
<=0% (0%; 2%] (2%; 5%] (5%; 10%] >10% <=0% (0%; 2%] (2%; 5%] (5%; 10%] >10%
min avg max min avg max
Sweden Bond 1900-2012 conditional to CPI rate buckets Sweden Equity 1900-2012 conditional to CPI rate buckets
80% 100%
80%
60%
60%
40%
40%
20%
20%
0%
0%
-20%
-20%
-40% -40%
-60% -60%
<=0% (0%; 2%] (2%; 5%] (5%; 10%] >10% <=0% (0%; 2%] (2%; 5%] (5%; 10%] >10%
Chart 10c
US Corporate IG 1970-2012 conditional to US CPI rate buckets US Interm Corporate IG 1970-2012 conditional to US CPI rate
buckets
40% 40%
30%
30%
20%
20%
10%
10%
0%
0%
-10%
-10%
-20%
-20% -30%
<=0% (0%; 2%] (2%; 5%] (5%; 10%] >10% <=0% (0%; 2%] (2%; 5%] (5%; 10%] >10%
Commodity GSCI 1970-2012 conditional to US CPI rate Gold 1927-2012 conditional to US CPI rate buckets
buckets
80% 60%
60%
40%
40%
20%
20%
0%
0%
-20%
-20%
-40%
-40%
-60% -60%
<=0% (0%; 2%] (2%; 5%] (5%; 10%] >10% <=0% (0%; 2%] (2%; 5%] >5% >10%

5. Reverse optimisation, what can it be used for?
The Markowitz model strongly depends on expected returns assumptions, and sometimes
produces results that are extreme and not particularly intuitive. The remedy of Black &
Litterman (1990) was to firstly identify a reference point for expected return assumptions
(equilibrium expected return), and then to elaborate a consistent Bayesian framework to
integrate qualitative views.
The main assumption beyond the equilibrium expected return is that, according to the Capital
Asset Pricing Model (CAPM), prices will adjust until the expected returns of all assets in
equilibrium are such that if all investors hold the same belief, the demand for these assets will
exactly equal the outstanding supply (He & Litterman 1999). The procedure to determine the
implicit returns in the market portfolio consists in reverse engineering the mean-variance
method: the aim is to determine the expected returns according to which the market portfolio
is optimal; this procedure is known in literature as reverse optimisation (Cantaluppi 1999).
This approach is more general than its application in the Black & Litterman (1990) model:
according to the reverse optimisation technique, implicit returns can be extracted from any
strategic asset allocation.
Reverse optimisation is a valuable tool for achieving consistency between model portfolios
and the fund managers portfolio. The iterative process that enables the allocation to be
transformed in implied views (reverse optimisation), and then the expected views in a
portfolio (optimisation), allows fund managers to build their portfolios more coherently.
Reverse optimisation implied returns suffer the following difficulties:
x requires calibration of a risk aversion parameter
The risk aversion parameter corresponds to the expected risk-return trade-off. It is the rate at
which more return is required for more risk, and it can be expressed as the ratio of risk
premium and variance of the market portfolio. This approach becomes decreasingly intuitive
when dealing with a diversified investment universe including bonds, equities, commodities
and alternative assets. In this case the most popular approach consists in determining the risk
aversion parameter according to a strong prior information or conviction (e.g. the return of a
short-term bond). Given the debt crisis in the Eurozone, this issue remains a critical one.

x favours overshot assets (no mean reversion mechanism) when applied to the market
cap
According to the reverse optimisation technique, returns are proportional to the market cap,
hence the approach suffers from increasing the returns of assets which are in a bubble. No
mean reversion mechanisms are present to alleviate the problem.
x assumes no segmentation in markets and no constraints
The model does assume that investors can buy any markets without any regulatory
constraints. Indeed it does not include some bias, such as the home bias, for example,
according to which investors tend to overweight assets of their own country.
6. Equilibrium-based model
The financial crisis questions the possibility of setting up reasonable forward looking
equilibrium figures due to uncertainty on long-term macroeconomic prospect trends.
What is the impact of this uncertainty on asset class return forecasts? And what is the
confidence on expected returns?
One of the most common approaches for forecasting returns in an equilibrium-based

framework consists in the building block approach. According to this model, we need to
forecast: the long-term inflation rate, the real risk-free rate (which corresponds to the
premium of the risk-free rate over inflation) and the risk premium (which corresponds to the
risk premium of risky assets over risk-free assets). Equilibrium-based models usually assume
that the valuation of markets against fundamentals converge to some historical average or
fundamental level.
In the past this approach was able to produce coherent figures. The developed world economy
had been characterised by decreasing inflation and decreasing volatility in growth and
inflation in the '90s (see chart 11). This allowed monetary policy rates to be generally lower
and more stable than in the '70s and the '80s. In particular, inflation expectations stabilised at
fairly low levels over the '90s. Therefore, in this more stable environment, rates tended to also
be more stable and more in line with nominal growth: this represented the first assumption.

Chart 11
US GDP & CPI rolling volatility (5 yrs rolling)
3,50%
3,00%
2,50%
2,00%
1,50%
1,00%
0,50%
0,00%
dc.-57
dc.-59
dc.-61
dc.-63
dc.-65
dc.-67
dc.-69
dc.-71
dc.-73
dc.-75
dc.-77
dc.-79
dc.-81
dc.-83
dc.-85
dc.-87
dc.-89
dc.-91
dc.-93
dc.-95
dc.-97
dc.-99
dc.-01
dc.-03
dc.-05
dc.-07
dc.-09
dc.-11
GDP CPI
The second assumption was represented by the empirical evidence of a positive relationship
between performances and risks measured by volatility over a long period. This environment
favoured equilibrium-based model approaches. Current market conditions are now far more
uncertain, and they make it difficult to apply this approach because econometric relations to
estimate risk premia might not be robust in non-stationary markets.
Furthermore, the monetary policy regime changed as well: extraordinary measures like
quantitative easing were introduced over recent years by many central banks as a new policy
response to the crisis. At the same time the assumption on the link between return and risk
over the long term has also come under scrutiny.
Even if forward-looking indicators, such as valuation ratios, have a better track record in
forecasting asset class returns than rearview-mirror measures (Ilmanen 2011), we should
remind the reader that returns sometimes may never revert over a practical time frame. Chart
12 reports the time-series of US bonds, US equity and Japanese equity in specific time
periods. The figures are contrasted to a hypothetical long-term average and dispersion (see
annex for details). The plots show clearly that the return-to-the-mean sometimes is very slow,
and that risk premia can exhibit robust trends which can last for many decades.
The best way to overcome difficulties in equilibrium-based models and valuation approaches
rely on diversifying the estimation process with complementary models.

7.
Cumulative Performance (logarithm) Cumulative Performance (logarithm) Cumulative Performance (logarithm)
-1,5
-1
-0,5
1,5
-2
-1,5
-1
-0,5
-0,2
0
0,5
1
2
2,5
3
3,5
0
0,5
1
1,5
2
0
0,2
0,4
0,6
0,8
1
1,2
1989,12 1929,12 1964,12
Chart 12
1990,07 1930,05 1965,06

1991,02 1965,12
1930,1
1991,09 1966,06
1931,03
1992,04 1966,12
1992,11 1931,08
1967,06
1993,06 1932,01
1967,12
1994,01 1932,06
1968,06
1994,08 1932,11 1968,12
1995,03
1933,04 1969,06
1995,10
rid of expected returns?

1933,09 1969,12
1996,05
1934,02 1970,06
1996,12
1934,07 1970,12
1997,07
1971,06
mkt
mkt
mkt
1998,02 1934,12
1998,09 1971,12
1935,05
1999,04 1972,06
1935,10
1972,12
median
1999,11
median
median
1936,03
2000,06 1973,06
2001,01 1936,08 1973,12
2001,08 1937,01 1974,06
+2sigma
+2sigma
+2sigma
2002,03 1937,06 1974,12
2002,10 1975,06
1937,11
2003,05 1975,12
1938,04
-2sigma
-2sigma
-2sigma
US Bond 1965-1981 (Cumulative Return)
2003,12
US Equity 1929-1942 (Cumulative Return)
1976,06
1938,09
Japan Equity 1990-2012 (Cumulative Return)

2004,07
1976,12
Amundi Investment Strategy Collected Research Papers

2005,02 1939,02
1977,06
2005,09 1939,07
1977,12
2006,04 1939,12
1978,06
2006,11
1940,05 1978,12
2007,06
1940,10 1979,06
2008,01
2008,08 1941,03 1979,12
2009,03 1941,08 1980,06
2009,10 1942,01 1980,12
2010,05 1981,06
1942,06
2010,12 1981,12
The last decade in the equity market has been particularly dramatic for diversified asset
allocation such as pension funds and balanced portfolios: poor performances, large draw-
How to handle uncertainty in expected returns? Can strategic asset allocation get
199
downs and very slow recovery again call into question the predictability of risk premia, and
the effect of incorrect assumptions in the portfolio optimisation process.
In the eighties and nineties, Jobson and Korkie, (1980), Best and Grauer (1991), Chopra and
Ziemba (1993) demonstrated that the sub-optimality due to estimation risk can be dramatic.
This evidence led researchers and practitioners to investigate more robust portfolio
construction schemes which can alleviate the estimation risk in the portfolio construction
process. The main approaches are the Bayesian methods and the robust models. The most
famous Bayesian method is the Black-Litterman (1990) model. Robust allocations deal with
uncertainty of input parameters by choosing the best allocation in the worst market condition
within a given uncertainty range. Indeed, the choice of this range is quite arbitrary. Robust
Bayesian allocations enable the uncertainty region for the input parameters to be defined in a
more coherent way. Moreover, the approach allows investors to modify the region according
to their specific views. Meucci (2011) consider robust Bayesian allocations that also account
for the estimation errors in co-variances.
More recently, the financial industry moved even further, elaborating investment processes
that can completely neglect assumptions on expected returns. The most popular were the
minimum variance, the maximum diversification portfolios, and the risk parity approach (see
Clarke et al. 2012). While the former has been used for equity portfolios, the seconds are
good candidates for diversified allocation. These portfolio construction schemes allow the
investor to make allocation according (only) to a risk model for the asset-classes. The
portfolio construction implication of these approaches is the overweight of low risk assets,
and thus the need for leveraging allocations in order to match medium-high risk profile
without losing equilibrium among portfolios bets.
Risk-parity approach is proven to be superior to the traditional mean-variance solution in case

of high uncertainty in the input parameters. Meucci (2009) proposed a diversification measure
which allows investors to diversify portfolios on independent factors (principal components).
More recently the need for diversification led researchers and practitioners to explore new
directions, thus complementing traditional asset classes (bonds and equities) with
commodities, real estate, investment styles (value, trend, carry, volatilities), and factors
(growth, inflation, illiquidity, and tail risks). The virtue of the risk-parity approach is
robustness in portfolio construction, the limit is that it is difficult to match it with a specific

macroeconomic view. In some cases, risk-parity models end up being pure statistical tools to
diversify unobservable information.
In Pola and Facchinato (2013), we illustrate a new approach for strategic asset allocation
(DAMS) which can be helpful to build more robust estimates for expected returns in
uncertainty. The key assumption of the Black-Litterman approach is to identify a reference
point for expected returns from reverse engineering the market portfolio (according to the
CAPM hypothesis). Given the limits of the CAPM hypothesis, it might be wiser to define the
reference expected returns from a different perspective. In Pola and Facchinato (2013) we
introduce a new reference portfolio where different macroeconomic scenarios are in
equilibrium, thus expressing the current uncertainty in financial markets.
8. Conclusion
The recent crisis exhibited major changes in the risk level, in the observed risk premia and the
risk-return relationship among asset classes. Many financial variables are in an uncharted
region, never reached going back many decades, and more than a century in some cases:
whether they would revert to the mean or they would stabilise to new levels is not clear (e.g.
Euro peripherals bond market). The uniqueness of the level reached by many financial
variables should at least convince us to lower our confidence on predictability of asset
classes risk premia.
We investigated the effectiveness of statistical methods, and equilibrium-based approaches:

high non-stationary patterns for the Sharpe ratio on one hand, the increase in macroeconomic
volatility on the other hand, make it difficult to use recent historical time-series to calibrate
econometric models. Increasing the length of history and identifying the observed risk-return
under different market regimes may help to clarify the picture. We briefly investigated the
dependency of main asset classes on inflation. Main results are:
x in inflationary scenarios nominal bonds suffer, gold does not provide excess return over
inflation, whereas more diversified commodity indices offer better real performances;
x in deflationary scenarios, nominal bonds are the best hedge.
Even if inflation hurt US stocks in the seventies and eighties, the dependency of equity on
inflation is not clear: going back to 1871 in the US and 1900 in Sweden, high dispersion of

conditional equity returns on inflation buckets, suggests to us that inflation is not the primary
risk driver.
Investigating the Sharpe ratio as a function of volatility, we show a rich phenomenology

according to which low volatility investments, higher quality credit bonds, less liquid assets,
and more diversified indices delivered, on historical basis, higher Sharpe ratios. Despite these
empirical evidences, we should point out that a normative hypothesis of equal Sharpe ratio for
each asset class allows us to build more diversified portfolios. Indeed whenever the portfolio
expected return is relevant per se (e.g. determining the discount rate for pension funds and
insurance companies), we prefer a more detailed description of asset classes Sharpe ratio. On
a more practical basis, 0.25 is a reasonable estimate for the Sharpe ratio of equity markets for
single countries and a ten-year horizon; global equity indices are likely to deliver higher
Sharpe ratios due to index diversification.
The directions which we would like to advice to tackle the expected return issue in this new
environment are:
x mix several complementary approaches to determine long-term expected returns, relying

on pluri-disciplinary skills (macroeconomics, econometrics and quantitative finance,
behavioural models are important as well to explain many documented anomalies in the
markets);
x relying more on risk to construct portfolios (e.g. minimum variance, maximum

diversification and risk parity portfolio construction).
This pluri-disciplinary approach may overcome some of the difficulties experienced by both
statistical and equilibrium-based approaches, leading to better adapted figures both to build
robust allocations and set up a realistic future level of expected returns and discount rate of
liabilities.

Acknowledgements
We would like to thank Eric Taz-Bernard, Sergio Bertoncini, Jean-Renaud Viala, and Marc-
Ali Ben Abdallah for very stimulating discussions and very constructive suggestions.

ANNEX A1
In chart 3b we measured the volatilities, maximum draw downs (MDD), and performance of
some asset classes in two historical periods: pre-crisis sample (1990-2007) and the full sample
(1990-2012). The grey shaded areas in chart 3b signal markets that exhibit a worse MDD than
before, bold indicates an increase in volatility.
Bond markets. The issue is the decoupling of the Eurozone: Italys MDD doubled in the
crisis period. Nevertheless Germany suffered between the end of 2011 and first quarter of
2012. In the Eurozone, French bonds did not exhibit any draw-downs not priced before. With
the exception of Australian, Canadian, and Japanese bonds, all volatilities increased.
Inflation linked markets. Eurozone debt crisis drove the main changes: the euro index
presents the higher increase in volatility. The emerging inflation-linked market is not very
significant due to small data sample: anyway data indicate that they deliver the best
performance per unit of volatility across all asset classes.
Credit markets. The 2008 crisis was a dramatic year for most of the spread markets. MDD
for US investment grade more than doubled, euro investment grade suffered less. World high
yield doubled its MDD, even though the euro high yield did not suffer its worst draw down in
the crisis period (the MDD in the full sample was from February 2000 to September 2002).
All volatilities increased except for the emerging market debt in hard currencies.
Equity markets. The crisis increased the MDDs in US, Euro peripherals, Australia and
Emerging markets. Germany, France, UK, Japan, and Canada did not register worst draw
downs. The volatilities increased but only marginally with respect to the pre-crisis period.
Commodity markets. CRB and GSCI indices registered new MDDs in the crisis sample,
gold performed better in the crisis (the average compound return increased). The increase in
volatility of CRB was more marked with respect to the GSCI and gold; the difference between
CRB and GSCI indices is mainly due to the larger exposure of CRB to energy commodities.

Chart 3b

ANNEX A2
In the following we report the median and average Sharpe ratio computed in different equity
markets and according to different-sized rolling windows (10 years, 5 years, 3 years)
Table1
MEDIAN 10 yrs 5 yrs 3 yrs AVERAGE 10 yrs 5 yrs 3 yrs

USA 0,29 0,27 0,35 USA 0,27 0,33 0,40
Japan 0,19 0,21 0,19 Japan 0,20 0,24 0,29
Germ any 0,21 0,09 0,10 Germ any 0,18 0,19 0,24
France 0,30 0,14 0,10 France 0,23 0,21 0,28
UK 0,26 0,28 0,32 UK 0,24 0,29 0,38
ANNEX A3
Chart 1
The plot reports the US Treasury yield 10-year. Data provider is Schiller database.
Chart 2
The plot reports the two-year yields for international countries: Germany, France,
Netherlands, Belgium, Austria, Italy, Spain, Portugal, Greece, Japan, UK, Canada,
Switzerland, Denmark, Sweden, US. Data provider is Bloomberg.
Chart 3a
Chart 3a is redrawn from Ilmanen (2011). It reports the performance of various US markets
from 1960 to 2009. The scatter plot relates the compound average real return (Y-axis) to the
average real losses in the three worst years (1974, 1981, 2008) for financial markets and the
global economy.
Chart 3b
Chart 3b contrasts the performance, volatility, and Maximum Draw Down of various indices
in two samples: pre-crisis sample (1990-2007) and the full-sample (1990-September 2012).
All time-series are monthly and in local currency. A few indices are shorter due to their

inception: World Inflation Linked since Dec. 1996, Euro Inflation Linked since Dec. 1999,
Emerging Inflation Linked since Dec. 2003, Credit Euro IG since Dec. 1995, Credit Euro HY
since Dec. 1997, Emerging Debt in Hard currency since Dec. 1993, Emerging Debt in Local
currency since Dec. 1993, Spanish equity since Dec. 1998, Australian equity since May 1992,
commodity CRB since January 1994. Data provider is Bloomberg.
Chart 4
The plot reports the historical 1-year rolling volatility (top panel) for German, Spanish and
Italian 7-10 year bonds, and their historical 1-year correlation to the EMU equity market.
Time-series are daily, data provider is Bloomberg.
Chart 5a & 5b
The left chart reports the performance of US bonds, US equity, and compares them to the
inflation rates in the same period. Performances are presented in decades (except for the last
decade which includes 12 years). The vertical bar divides the sample in three periods: pre-
war, post-war and new millennium. The right panel reports a scatter plot between real
performance for US bonds and US equity. Each spot corresponds to the annualised
performance of bonds and equity for a specific decade. Data provider is Schiller database.
Chart 6
The plot reports the 10-year CDS for Italy and Germany since 2005. Data provider is
Bloomberg.
Chart 7
The plot reports computations of Sharpe ratio over rolling windows (10 and 3 years) of
various equity markets (US, Japan, UK, Germany, France). Time-series are monthly. Table 1
in annex A2 reports the summary of the results: median and average for countries and
horizons (10, 5, 3 years). Data providers are Datastream and Bloomberg.
Chart 8
In chart 8 (top left panel) we computed the historical Sharpe ratio and plot against historical
volatility in a diversified investment universe in the US. The risky assets are: short-term
treasury, all-maturity treasury, long-term treasury, short-term corporate investment grade, all-
maturity corporate investment grade, long-term corporate investment grade, all-maturity

x constant drift. We assumed a constant Sharpe ratio (0.20) and a risk-free rate given by
the 10-year bond yield at the beginning of the sample;
x constant volatility. Historical volatilities in the previous decade (monthly

observation).
Data providers are Bloomberg and Schiller database.

References
Arnott, R. D., and Bernstein, P. L., 2002. What Risk Premium Is Normal?, Financial
Analysts Journal, vol. 58, no. 2 (March/April): 6485.
Asness, C. S., 2000. Stocks vs. Bonds: Explaining the Equity Risk Premium., Financial
Analysts Journal, vol. 56, no. 2 (March/April): 96113.
Baker, M., Bradley, B., and Wurgler, J., 2011. Benchmarks as Limits to Arbitrage:
Understanding the Low-Volatile Anomaly., Financial Analysts Journal, vol. 67, no. 1.
Barberis, N., and Ming H., 2001. Mental Accounting, Loss Aversion, and Individual Stock
Returns., Journal of Finance, vol. 56, no. 4 (August): 12471292.
Benartzi, S., and Thaler, R., 1995. Myopic Loss Aversion and the Equity Premium Puzzle.,
Quarterly Journal of Economics, vol. 110, no. 1 (February): 7392.
Best, M. J., and Grauer, R. R., 1991. On the sensitivity of mean-variance-efficient portfolios
to changes in asset means: Some analytical and computational results.. Review of Financial
Studies 4, 315-342.
Black, F. and Litterman, R.. 1990. Asset Allocation: Combining Investors Views with
Market Equilibrium., Fixed Income Research, Goldman Sachs.
Cantaluppi, L., 1999. The Reverse Optimization., Financial Markets and Portfolio
Management.
Campbell, J. Y., and Shiller, R., 1988a. Stock Prices, Earnings and Expected Dividends.,
Journal of Finance, vol. 43, no. 3 (July): 661676.
Chopra, V., and Ziemba, W. T., 1993. The effects of errors in means, variances, and
covariances on optimal portfolio choice., Journal of Portfolio Management pp. 6-11.
Choueifaty, Y., and Coignard, Y., 2008. Towards Maximum Diversification., Journal of
Portfolio Management, Vol. 35, No. 1 (Fall).
Clarke, R., De Silva, H., and Thorley, S., 2012. Risk Parity, Maximum Diversification, and
Minimum Variance: An Analytic Perspective., working paper.
Fama, E. F., and Kenneth, R. F., 1989. Business Conditions and Expected Returns on Stocks
and Bonds., Journal of Financial Economics, vol. 25, no. 1 (November): 2349.
Frazzini, A., and Lasse, H. P., 2010. Betting Against Beta., Working Paper, AQR Capital
Management, New York University and NBER Working Paper 16601.
He, G., and Litterman, R., 1999. The Intuition Behind Black-Litterman Model Portfolios.,
Investment Management Research, Goldman, Sachs & Company.

Ibbotson, R. G., and Sinquefield, R. A., 1976a. Stocks, Bonds, Bills and Inflation: Year-by-
Year Historical Returns (19261974)., Journal of Business, vol. 49, no. 1 (January): 1147.
Ibbotson, R. G., and Sinquefield, R. A., 1976b. Stocks, Bonds, Bills, and Inflation:
Simulations of the Future (19762000)., Journal of Business, vol. 49, no. 3 (July): 313338.
Ilmanen, A., 2011. Expected returns. An Investors Guide to Harvesting Market Rewards,
Wiley Finance
Jobson, J. D., and Korkie, B., 1980. Estimation for Markowitz efficient portfolios, Journal
of the American Statistical Association 75, 544-554.
Markowitz, H. M., 1952. Portfolio Selection. The Journal of Finance.
Mandelbrot, B., 1963. The variation of certain speculative prices. The Journal of Business.
Meucci, A., 2007. Risk and Asset Allocation, Springer Finance.
Meucci, A., 2009. Managing Diversification, working paper, symmys.com
Pola, G., 2013. Managing uncertainty with dams. Assets segmentation in response to
macroeconomic changes, in preparation.
Pola, G., and Facchinato, S., 2013. Managing uncertainty with dams. Balancing
macroeconomic scenarios, in preparation.
Rebonato, R, 2008. Plight of the Fortune Tellers: Why We Need to Manage Financial Risk
Differently. Princeton, NJ: Princeton University Press.
Reinhart, C. M., Rogoff, K. S., 2011. From Financial Crash to Debt Crisis, American
Economic Review, 101: 1676-1706.

WP-033
Low Risk Equity Investments:

Empirical Evidence, Theories,
and the Amundi Experience
Alessandro Russo,
Head of Equity Quantitative Research
March 2013
The outperformance of low volatility stocks over the last 50 years

has been among the equity markets most puzzling anomalies. At
the same time, low risk investing has recently gained a remarkable
interest, due to its documented performance coupled with the
unprecedented volatility experienced during the last global
financial crisis.
Researchers have been documenting such anomalies since the early
nineties. We find that most of the relevant empirical studies focus
on systematic risk; while some of them state that the anomaly holds
regardless of whether total or systematic risk is considered, only a
few exceptions refer to idiosyncratic volatility.
According to theories referring to leverage constraints, investors
willing to leverage but facing leverage constraints buy high beta
stocks, inflating their prices and thus lowering subsequent returns.
Some other explanations rather refer to behavioral bias, delegated
portfolio management, benchmarking, and the utility function of
fund managers. Some theories focus on the distribution of equity
returns: skewness and convexity.
In the last section of our research, we provide a formal description
of two investment processes that successfully exploit this low risk
anomaly: the minimum variance and the risk parity portfolio.

1. Low Risk Equity Investing, empirical evidence of success
The first axiom of investing is that higher risk is compensated -on average- by higher returns.
In his Portfolio Selection (1952), Markowitz defines portfolio risk as the volatility (or
variance) of its returns and states that, while taking investment decisions, investors maximize
expected returns for a given amount of portfolio risk: this leads to a positively sloped and
concave efficient frontier.
In Sharpes CAPM (1964) risk is defined as portfolio beta relative to the Market Portfolio.
The idea behind the CAPM is that any assets average return may be expressed as the sum of
the risk-free rate and a risk premium, the latter being proportional to the assets covariance
with the Market Portfolio (its beta).
According to the theory, both the Minimum Variance portfolio and the Market Portfolio
belong to the efficient frontier; the latter exhibiting a higher expected return than the former.
As mentioned, our goal is not to take a position in the everlasting dispute in favor or against
the CAPM. However it is worth mentioning some preliminary tests that provided some
comforting results during the early seventies, as well as many subsequent studies proving the
outperformance of low volatility stocks versus high volatility stocks, and its persistence
during recent decades.
Black, Jensen and Scholes (1972), and Fama and MacBeth (1973) form portfolios on the basis
of in sample estimations of beta, and then look at realized Beta and stock returns, finding that
higher beta is positively correlated to higher returns. Yet, the slope of the linear relationship is
flatter than anticipated by CAPM. For instance, in Fama and MacBeth, the hypothesis that
beta is uncorrelated with returns is significantly rejected only in the full sample period (1935-
1968) while it is not rejected in any of the 10-year sub-periods.
Despite some statistical weakness, these contributions held for a long time as a general
confirmation of the traditional positive relationship between risk and return.
In fact, more recent evidence shows that well-constructed low risk portfolios contradict this
axiom, as they deliver higher returns than riskier portfolios.
In 1992, Fama and French showed that value and size factors generate large returns in the US
markets and, quite surprisingly, that the relationship between beta and returns is rather
negative, when corrected for size effect.

Ang et al (2006) find strong evidence of negative relationship between stocks returns and
their sensitivity to changes in aggregate volatility (daily and monthly changes in VIX index
and daily and monthly returns of an equity portfolio, replicating the VIX itself). They explain
such evidence by the premium that investors pay (thus reducing subsequent returns) for stocks
that are positively correlated with volatility jumps, thus providing a hedge in case of market
drops (as typically market drops take place together with volatility jumps). They also find a
strong negative relationship between idiosyncratic risk (residual component in a Fama French
environment) and future returns.
Intuitively, as the aggregate volatility (VIX factor) is significant indeed, but missing in the
Fama French model, companies with greater sensitivity to the VIX factor should exhibit
higher idiosyncratic risk in a Fama French framework.
Surprisingly, after controlling for aggregate volatility, the negative relationship between
returns and idiosyncratic risks is little changed, and completely unexplained.
As a consequence, the authors associate the low-risk high-return anomaly to idiosyncratic

risk, rather than systematic and total risk.
Blitz and Vliet (2007) focus on systematic and total volatility. They state that the low-risk
anomaly holds regardless of which of the two measures is used for stock selection: low
variance stocks exhibit low beta, while high variance stocks exhibit high beta.
The authors controls for factors such as value, momentum, and size, are both via linear
regressions with those factors returns as explanatory variables (with little if any variance
explained), and via cross basket analysis: returns of the volatility factors are investigated
among stocks with similar momentum, size or value characteristics. The significance of the
low risk effect is little reduced, and only within baskets of stocks with similar market
capitalization.
They also argue (probably underestimating the cost in terms of loss of flexibility and risk of
incurring an undesired exposure) that there is no need to rely on sophisticated risk models and
that an equal weighted portfolio of stocks with low historical volatility is all investors need to
achieve superior returns.

Baker, Bradley and Wurgler (2011), agree that there is no major change in addressing the
superior risk adjusted performance of low risk stocks whether we rank our universe (CRSP)
by total volatility or by beta.
We have replicated these transparent tests on the MSCI World constituents from January 2003
to June 2012 (excluding stocks with less than two years of presence in the index), and we
confirm that grouping stocks for (ex-post) beta rather than (ex-post) volatility does not have a
big impact: in both cases average risk-adjusted returns decrease with risk measure, with
baskets built according to beta showing somehow better regularity and monotony (Annex 1,
tables 1 and 2, chart 1).
Baker and Haugen (2012) implement deciles analysis for each of the 21 developed and 12
emerging countries of their sample: they group stocks from 1990 to 2011 according to their
historical volatility and they find that lowest risk stocks exhibit higher returns and better
Sharpe ratio, compared to high risk stocks, in any observed country. They also point out that
only rarely do the highest volatility deciles outperform lowest deciles, in a rolling window of
three years.
In their intentionally simple approach, they use an equally weighted basket and 24 months
historical volatility as a measure of risk, apparently supporting argument by Blitz and Vliet
(2007) for no need for a sophisticated risk model. Actually no control is done for valuation,
size, or any other factor that could help to explain the anomaly.
In 1991, Baker and Haugen first investigate Minimum Variance portfolios in the US equity
market, pointing out a 30% reduction in portfolio volatility, compared to both a common US
index and randomly selected portfolios, with no reduction in average returns.
Clarke, de Silva and Thorley (2006) build Minimum Variance portfolios on the largest 1000
US stocks over the period 1968 2005. They estimate a covariance matrix with Bayesian
methods for shrinkage -in order to avoid error maximization problems- as well as principal
component analysis.
They first detail some well-known portfolio characteristics like typical concentration (75 to
250 stocks with 3% cap on a single company), turnover (143% with monthly rebalancing),
positive exposure to size and value factors, and zero mean but rather volatile exposure to
momentum factor. They confirm Baker and Haugens evidence (1991) of 30% reduction in

volatility with substantially no reduction in average returns. Results are little changed after
constraining fundamental factors exposures and reducing turnover to 56% through quarterly
rebalancing.
In their more recent (2011) Minimum Variance Portfolio Composition, Clarke, de Silva and
Thorley update previous (2006) statistics with basically no change. However their work is
worth mentioning mostly because, while proposing an interesting analytical solution of the
Minimum Variance portfolio composition, they point out that systematic risk dominates in the
construction of such a portfolio: in their simplified single factor model, stocks with beta
higher than a threshold are strictly excluded from the portfolio, while high idiosyncratic risk
only contributes to lowering the stocks weight in the portfolio.
Carvalho, Lu, and Moulin (2011) discuss typical low beta and small cap exposure of
Minimum Variance portfolios. Furthermore in their view, changes in the correlation matrix
generate higher than justified turnover, thus they suggest constructing portfolios by applying
stable weighting schemes based on stocks beta, rather than optimizing.
In our opinion, a weighting scheme may be a reasonable solution, even though optimization
remains our optimum for Minimum Variance portfolio construction, as an optimization
package enables turnover to be reduced, while controlling many other constraints.
Thomas and Shapiro (State Street Global Advisors, 2007) show their encouraging results on a
Minimum Variance portfolio built on the Russell 3000, using a standard optimization package
(BARRA). Their contribution is relevant to us because -as we will discuss deeper in section 3-
they recognize the advantage of using such an optimization package, in order to avoid the
typical drawbacks of a Minimum Variance investment (excessive concentration in a few low
risk sectors and stocks, lack of control for involuntary factor exposure), while enhancing their
performance by tilting the portfolio toward some long term successful alpha strategy, like the
historical dividend yield.
2. Why low risk stocks outperform
Most of the recent literature addressing the low risk anomaly has also offered some theoretical
framework.

Explanations account for borrowing constraints and other market frictions, behavioral
hypothesis, equilibrium models addressing the utility function of fund managers (rather than
of investors) in delegated portfolio management, or taking into account higher moments (and
their relative premium) than the mean and the variance of returns.
Among the others, theories referring to leverage constraints are probably the most
represented.
2.1. Leverage Constraints
Blake identifies borrowing restriction as one of the possible sources of low risk stocks
outperformance in his Beta and Returns (1995). Investors are supposed to pass through such
a limitation, investing massively in high beta stocks thus lowering subsequent returns.
Frazzini and Pedersen (2011) develop a model of asset equilibrium populated by two
categories of investors: investors with no leverage constraints but with margin requirements,
and investors for which leverage is forbidden. They argue that while the latter overweight
high beta assets, causing those assets to offer lower subsequent returns, the former buy low
beta stocks and leverage their holdings by short selling the high beta stocks. Intuitively both
the low beta and the high beta stocks have their demand in the market, thus it is not
immediately clear why only high beta stocks should be overpriced, thus delivering poorer
returns than the CAPM would predict. However the analytical solution of the required rate of
return in equilibrium shows that the alpha of each security monotonically decreases with its
beta, thus reducing the slope of asset returns relative to beta itself. This slope is flatter as the
tightness of the funding constraint increases.
The authors build a betting against beta factor by going long on low beta stocks and
shorting a smaller amount of high beta stocks, thus obtaining a beta neutral factor. They
provide empirical evidence for alpha decreasing with security beta, and for the betting
against beta factor exhibiting positive returns in equity, treasury, credit, and currency
markets.
Carvalho, Lu, and Moulin (2011) agree that leverage constraint has a major impact on the
preference towards high beta stocks. However, they add that, as they simply seek high returns,
investors prefer riskier stocks, and create a demand imbalance.

Blitz and Vliet (2007) state that as long as leverage is limited and leveraging a low beta
portfolio is needed to match the volatility of the market (in order to obtain a volatility neutral
strategy), the anomaly is not fully arbitraged away.
However, among the possible explanations of the anomaly, they take into account a
behavioral theory by Shefrin and Statman (2000), where investors allocate their wealth
according to two layers: a low risk layer designed to avoid poverty, and a high aspiration
layer, for which they are much less risk averse. In this case, investors will overpay for (often
few and badly diversified) risky stocks, which are perceived to be similar to lottery tickets.
The authors finally mention another well represented family of theories that will be further
discussed in the next section, which relates the low risk anomaly to the utility function of the
fund manager: as the largest inflows go toward outperforming fund managers and to well
performing asset classes, these portfolio managers may seek to maximize outperformance in
the upward markets, thus systematically preferring high beta stocks.
2.2. Delegated portfolio management, benchmarking, and fund managers utility

function
Cornell and Roll (2005) recognize that the considerable market share of investment being
nowadays delegated to professional fund managers, impose asset pricing models to
incorporate the objective function of the agents, together with the traditional utility function
of the final investors. The objective function of the fund manager is not the investors wealth
maximization, but the maximization of active return versus the benchmark of the delegated
mandate.
Without a direct implication on performance of low risk versus high risk stocks, they show
how a pricing model based on delegated investments (where fund managers objective
function dominates the investors utility function), in equilibrium implies some cross sectional
relationship between stocks alpha and their beta relative to the benchmark. These
relationships violate CAPM.
Baker, Bradley and Wurgler (2009 and 2011), make a step forward focusing explicitly on the
low risk anomaly.
The anomaly itself is indeed a consequence of irrational behavior of non-professional and

professional investors, who (1) favor lottery-type investments, (2) consider few cases of

success of high volatile stocks as representative of all the other high volatility stocks, (3) are
overconfident, trade more actively when optimistic than when pessimistic, and are likely to
disagree on future stocks returns, especially on high volatility stocks. However, like Cornell
and Roll (2005), the authors recognize the prominent role of delegated and benchmarked
investment, as a limit to arbitrage: behavioral bias result in high absolute risk stocks
delivering similar returns than low absolute risk stocks, but the latter still generate a higher
tracking error relative to the mandate benchmark, and consequently fund managers are
prevented from arbitraging the anomaly away (as they have low incentive to seek similar
returns for a higher tracking error).
Backer and Haugen (2012), state that, as the risk return relationship is rather inverted, fund
managers would have a concrete incentive in buying low risk stocks.
Their explanation for fund managers not actually exploiting the anomaly is that they seek to
maximize the probability of receiving a bonus, that is usually paid when performances are
positive in absolute terms, and higher than a threshold (benchmark return plus the
management fees, or an absolute discretionary threshold).
Portfolio managers exchange higher expected returns of low beta stocks, for a higher
probability of beating their benchmarks (or a given target), provided by high volatility stocks.
In other words they exchange a higher mean of returns distribution (low volatility stocks) for
a higher expected value in the right-end tail of the distribution (high volatility stocks).
The authors also mention some additional incentives to hold volatile stocks, this time related
to the delegated portfolio construction process: in order to impress colleagues, analysts are
often willing to recommend stocks in the news, or stocks whose news flow is quite intense
and that tend to exhibit higher than average volatility. Finally fund managers may find it
easier to justify holdings or turnover of newsworthy stocks.
Supporting their intuition, the authors show that for 1000 US stocks grouped in 10
homogeneous classes of market capitalization, from 2000 to 2011 companies with higher
institutional ownership exhibit higher volatility than stocks with low institutional ownership.
Finally, they find that analysts coverage (number of recommendations) is positively
correlated with volatility.

2.3. Return distribution of stocks return: may convexity and skewness have an impact?
Chunhachinda et al (1997) find that the returns of the world's 14 major stock markets are not
normally distributed. Optimal portfolio compositions are computed (as allocations of 14
international stock indexes) incorporating investors preferences for skewness. The empirical
findings suggest that the incorporation of skewness into the investor's investment process
causes a major change in the construction of the optimal portfolio. The evidence also suggests
that investors exchange expected return for positive skewness.
Similar findings are provided by Prakasha et al. (2003).
Kraus and Litzenberger (1976) find that, while having aversion to variance, investors exhibit a
preference for positive skewness. As a consequence, if the capital asset pricing model is
extended to include systematic skewness, this latter is associated with a positive price (instead
of a discount, as it is the case for variance), and the zero intercept for the security market line
is not rejected.
The intuition behind these three contributions is that, as skewness has a positive price, it
should be higher among high beta or high risk stocks, so that these latter are priced at a
premium compared to a CAPM equilibrium, and finally deliver lower average returns.
We tested this hypothesis in the last decade on the constituents of the MSCI World Index in
the period from January 2003 to June 2012. We regressed stocks weekly returns (excluding
only those stocks with less than two years of available data) on the index returns; we then
formed equally populated baskets according their betas. We then computed differences and T-
statistics of the average skewness, for any pair of baskets. The signs and their significance
prove some interesting monotony: higher skewness for higher beta stocks.
Results are summarized in tables 3 and 4 in Annex 2: skewness generally increases with beta,
sometimes significantly, especially for baskets whose differences in beta are high enough.
We repeated the exercise using total ex-post volatility as a measure for sorting and grouping
stocks (tables 5 and 6). Results are even more significant, as skewness increases almost
monotonically with volatility.
We recognize our analysis is completely in sample and misses some predictive power;
however, it is transparent and easily replicable. The hypothesis of skewness increasing with
beta and volatility is confirmed, and this suggests that the premium that investors pay for a

positive skewness may explain lower subsequent risk-adjusted returns for high beta and high
volatility stocks.
Cowan and Wilderman (2011) find that high beta stocks exhibit a positive convexity relative
to broad market index returns, while low beta stocks have negative convexity.
They explain that high beta stocks provide a call option payoff. In the case of positive market
returns, high beta stocks deliver market returns, multiplied by a factor roughly equal to their
beta (whatever the market return is, exactly like a leveraged position). Conversely, in case of
negative market returns, losses are limited to 100% of invested capital, at worst.
The difference with leveraged investments is straightforward as the latter generate payoffs
exactly equal to the returns of the unleveraged positions times the leverage, with theoretically
no limit to downside.
The authors argue that investors exchange future returns for having this call-type convex
payoff: they pay an additional premium for high beta stocks just like they paid a premium to
buy a call option.
In our view this explanation may be convincing only for very extreme market returns, that is
quite rarely (the convex profile takes place in the form of a stop loss, that is activated in case of
market returns of, lets say, -33% in the case of a stock beta higher than 3, or -50% in case of
beta higher than 2).
However, we apply the methodology described for skew in the previous paragraph, to test if
we find increasing convexity, for increasing beta. For each MSCI World constituent (from
January 2003 to June 2012, excluding companies with less than 100 weekly returns) we run
two OLS regressions: the first with the series of MSCI World returns (in USD) as the only
explanatory variable; and a second one with the series of MSCI World squared returns as an
additional explanatory variable:
Ri = i RMsci + ui (1)
and
2
Ri = i R Msci + i R Msci + vi (2)

where a positive estimate of implies positive convexity of stocks returns, relative to market
returns.
We group stocks according to their estimate of from OLS regression 1, and we compute the
average and the standard deviation of the estimate for any basket. We then test the
significance of differences in average basket convexities.
Results are summarized in Annex 3 and seem to support Cowan and Wildermans intuition.
Table 7 reports average convexity and standard deviation across any basket; while table 8
reports differences in average convexity for any pair of baskets.
The higher the distance of betas, the larger and more significant the difference in convexity:
high beta stocks exhibit a higher and thus more profitable convexity than low beta stocks.
Differences are often significant and this may explain some premium paid by investors for
high beta stocks.
Table 9 and 10 report results for the same experiment, when we rank and group stocks
according to total ex-post volatility. Results are even more significant as convexity increases
almost monotonically with volatility.
3. Amundi NextGen Equities
We have lived for several years in very challenging markets: international equity indexes have
exhibited high realized volatility and quite disappointing returns, during the last decade.
Equity investors have been faced with a major and unfavorable change in traditional risk
return payoffs.
In the last few years, Amundi has strongly invested in order to meet investors needs in such a
challenging market context, developing a range of innovative solutions aiming at Sharpe ratio
improvement. Their risk-return profile differs as well as their behavior in up and down
markets. Over the last decade, these strategies have all succeeded in enhancing risk return
trade-off (Sharpe ratio has been systematically superior to that of relevant equity index as the
MSCI World). They all belong to the absolute risk category: away from the notion of tracking
error or information ratio, they focus on Sharpe ratio or risk-adjusted return, and volatility
metrics. They are based either on the use of instruments providing favorable asymmetry
(options and other derivatives), or on portfolio construction techniques as maximum

diversification, minimum variance, and risk parity (Smart Beta -or simply Smart- from
now on).
The latter two are particularly relevant to this document as both exploit the low risk anomaly
discussed previously.
On an ex-post basis, the volatility of the Global Minimum Variance and Global Smart
portfolios is significantly lower than the market index, with a 10 to 20% reduction for Smart,
and up to a 35% reduction for Minimum Variance (Annex 5.1 and 5.2). However, most of the
relevant literature strictly identifies low risk strategies with the selection of low risk stocks and
-among several measures of risk- systematic risk and beta are by far the most significant. For
this reason it is worth investigating whether Minimum Variance and Smart processes limit
portfolio volatility mainly by selecting low risk stocks, or rather by enhanced diversifications.
In Annex 4, we provide some empirical evidence on two back-tested Minimum Variance and
Smart portfolios, during the period December 2003 December 2011.
Table 11 shows that the ex-post beta of the Minimum Variance portfolio on full data sample is
only 0.55 (significantly lower than 1 at a 1% confidence level). On an ex-ante basis with
quarterly observations, portfolio beta ranges from 0.48 to 0.65. As for the Global Smart
portfolio, table 11 shows that the ex-post beta relative to MSCI World is 0.85 (again
significantly lower than 1 at 1% confidence level) and the ex-ante beta ranges from 0.7 to
0.87.
In table 12 we further investigate the systematic risk characteristics of the two portfolios and
the benchmark in a multifactor framework. For each of them at every quarter of our back-test,
we compute the weighted average common factor risks extracted from the BIM model by
BARRA1. Weighted average common factor risk of the Minimum Variance portfolio is lower
than that of Global Smart; both of them are lower than the MSCI World.
1
Weighted average common factor risk, at any time t of our sample, is computed as follows:
CF(t) = i wi(t) CFi(t)
where CFi(t) is the common factor risk of the ith stock at time t, and wi(t) is its weight at time t.

Charts 2 and 3 show average portfolio weight over stocks grouped by beta and common factor
risk2. Minimum Variance and Smart portfolios overweight stocks with lower than average
beta, compared to the MSCI World. The same result holds for average common factor risk.
Aggregate weights decrease with increasing risk for Minimum Variance and Smart, while
weights are distributed much more uniformly for the MSCI World, with slightly higher
frequency on stocks with average risk.
This evidence supports the intuition by Carvalho, Lu, and Moulin (2011), who infer that the
exposure to low beta and low systematic risk plays a major role in explaining the superior
performance of Minimum Variance portfolios; we further extend these findings to Smart Beta
as well.
In fact, such exposure of Smart portfolios to low risk stocks is somehow intuitive if we
consider the sector allocation process (sector weights are inversely proportional to their
marginal contribution to risk), and the stock weighting scheme (inversely proportional to their
total volatility). The two of them are described in detail in section 3.2.
3.1. Amundi Global Minimum Variance: an enhanced process
As shown, the most straightforward way to exploit the low risk anomaly is certainly building a
portfolio of stocks with the lowest possible risk.
In this section we describe our Minimum Variance approach on a Global Developed Equity
universe. We claim several years of experience in Minimum Variance management, with two
Europe portfolios (since 2007 and 2009 respectively), a very recent Global portfolio, and our
paper portfolios on developed World, Japan, Emerging markets, Pacific ex Japan, and other
customized universes. However in this document, we focus on the Global Minimum Variance
as it is the most complete case for descriptive purposes.
Efficient frontier and the Minimum Variance portfolio
The efficient frontier represents the set of portfolios that earn the maximum rate of return for
every given level of risk. We use an optimization process to build a portfolio sitting on the
very edge of the efficient frontier. In building such a portfolio, expected returns are not needed
as the only requirement is to minimize volatility, while being fully invested.
2
Every quarter we build three equally populated baskets of stocks, according to their beta and common factor
risk (high, median, and low risk). We then compute the aggregate weight of stocks in each group, for any
portfolio. Finally we compute the historical average of aggregate weights.

Our simple objective function is thus:
Min (wTVw)
Such that eTw = 1
where w is the vector of the optimal portfolio weights, V is the variance-covariance matrix,
and eT is a vector of ones.
Although we recognize the advantage of such a process being transparent and intuitive, we are
conscious of some typical drawbacks that may arise from Minimum Variance portfolios: as
shown in Clarke, de Silva and Thorley (2006), Minimum Variance portfolios may be quite
concentrated on a few low volatility stocks, may exhibit rather high turnover, positive
exposure to value and small capitalization stocks (with some relevant implications on
liquidity), and some volatile exposure to momentum factor.
Similarly, Thomas and Shapiro (2007) highlight the risk of the Minimum Variance portfolio
being excessively concentrated on few low risk sectors, and the lack of control for involuntary
factor exposure. They also express their preference for tilting portfolios toward some
successful stock ranking criteria.
We agree that most of those are relevant issues in portfolio construction and we do believe that
handling them through an optimization package is strictly needed in order to come out with
reasonable and investable portfolios, as only few of them might be addressed correctly with
some clever weighting schemes (like in the case of limiting turnover through an equally
weighted basket of low beta stocks, as suggested by Carvalho, Lu, and Moulin).
Thats why we implement our enhanced portfolio construction process in Barra One, as
described below.
Quality Stocks
We believe that fundamental equity selection can provide some valuable enhancement in the
risk return profile of equity portfolios, at least in the long run. At the same time we dont want
to renounce an optimization process which is completely independent from expected returns.
Expected returns are very noisy in forecast and thus responsible for well-known error
maximization problems.

For this reason, we apply a qualitative filter to our investment universe, excluding the lowest
quality stocks from the optimization. Basically, each quarter we rank the constituents of the
MSCI World Developed Markets according to a Piotroski (2000) score and we exclude the
two bottom quintiles.
Keeping 60% of constituents available for investments, the optimizer is left with a high degree
of freedom and it tilts the optimal portfolio toward good quality stocks, without using explicit
expected returns.
Table 11 in Annex 4 reports the balance sheet, income statement and corporate actions
employed in the Piotroski score, and table 4 shows that -in the last decade- the top quality 33%
of MSCI constituents (equally weighted) have outperformed the market index with lower
volatility. At the same time, the median basket has performed in line with the market, and the
bottom basket has underperformed with even higher volatility.
Turnover and liquidity
High turnover is a critical issue in many systematic investment strategies like Minimum
Variance.
In our case, turnover in the investment universe is limited as the Piotroski score is based on
balance sheet data that varies very little during one quarter. Furthermore we also rebalance our
portfolio quarterly, as suggested by Baker and Haugen (1991).
Nevertheless, more than turnover itself, our concern is liquidity indeed: we aim to avoid small
illiquid companies as we want to be able to liquidate our portfolio in a reasonable time lag,
without incurring significant market impact costs.
To address this requirement, we limited the amount held in any stock to the following
percentage:
ADVi
UBi = 25% D
NOT
where UBi is the upper bound on the ith stock, D is the number of days that we accept to
liquidate the fund, ADVi is the average daily volume over the last quarter, and NOT is a
notional amount of assets under management of USD 1 billion: quite conservative as it is still
far above the current size of our fund.

Sector, country, and stock concentration
As mentioned above, Minimum Variance portfolios may tend to be poorly diversified across
sectors, countries or single stocks. We have thus applied some constraints at these levels,
without preventing the optimizer from choosing solutions that are far enough from a market
index.
On countries and sectors we accept deviations from the market index of 500 to 1000 bp, while
for single stocks we apply a general upper bound (GUB), thus modifying the actual upper
bound as follows:
ADVi
UB i = min GUB;25% D
NOT
Management of asymmetries in factor returns
Furthermore, we are conscious that Minimum Variance portfolios may be exposed to

fundamental factors as size, value or momentum.
We observe that much of our size exposure is corrected away by the liquidity constraints. As
for other factor exposures, we have decided not to manage them systematically as again we
dont want to excessively constrain the optimization process.
On the other hand, we regularly monitor the behavior of all the risk factors of the BARRA
model (size, value, growth, momentum, leverage). The goal of this monitoring is to detect
bubbles or suspicious asymmetries like excessive positive skewness in recent performance
(Sornette, 2003; Morel, Malongo, and Lambinet, 2013): in the case of significant alerts, we
punctually hedge the risk of an exploding bubble, imposing a neutral exposure to the
suspected factor.
As for performance (Annex 5.1, table 15), results are very satisfactory indeed: from the
beginning of 2003 to the third quarter of 2012, Minimum Variance portfolios outperformed
the standard index by a minimum of 3.6% in Pacific (All Countries) ex Japan, to a maximum
of 5% in the World developed Markets, while volatilities and draw-downs were reduced by

one fourth to one third. As a consequence, risk adjusted returns are twice as high as for the
standard index3.
Minimum Variance portfolios belong to the absolute return investment category, as it is

confirmed by ex-post tracking errors ranging from 9 to 10% roughly.
Real money performance of our two Europe portfolios (from end 2007 and mid 2009
respectively) confirms the results of our back test with both of them outperforming the
standard index in absolute and risk-adjusted terms4.
3.2. Amundi Global Smart Beta: A systematic risk parity approach
In this section we discuss our risk parity approach on global developed markets. We recall that
we currently manage several risk parity portfolios (Euro Area, Europe, World), and we are
investigating the behavior of such strategies in many others areas (World ex Japan, Pacific ex
Japan, Japan, Emerging Markets, Emerging Markets with Sharia filter). Our approach is
consistent across the regions with the Developed World being probably the best example for
descriptive purposes.
Risk parity means that each asset (asset class, equity sector, single stock) has an equal
contribution to the total risk of the portfolio.
As Maillard, Roncalli, and Teiletche (2009) have pointed out, full risk parity cannot be
obtained in a closed formula unless some unrealistic hypotheses (such as equal correlation
among all the assets in the investment universe) are made, and may not be achieved either
through optimization, if the number of assets involved is somehow relevant, and correlations
are very heterogeneous.
As for the number of assets involved, we typically deal with 1,500 to 2,000 constituents of the
MSCI World. For this reason we have decided to split our portfolio construction process into
two steps: the region-sector allocation, and the stock weighting in each regional sector basket.
Optimization (for instance, the minimization of the cross-section standard deviation of assets
contribution to risk) does not guarantee a full risk parity solution. Furthermore, it can be
3
As for the Emerging Markets portfolio (from December 2005), in order to be eligible to minimum variance optimization,
stocks must be held on the Amundi emerging market flagship fund, or must be top-ranked (33%) according to a Piotroski
score.
4
The former of the two European portfolios has been designed in order not to exceed the ex-ante volatility of 10%. Thats the
main source of performance difference between them.

time-consuming, and should rely on a robust numerical algorithm. For these reasons we have
investigated alternative solutions among some reasonable systematic weighting procedures.
While defining the region and sector allocation of our risk parity portfolio, we distinguish
three regions: North America, Europe, and Pacific. Then, within each geographical region, we
operate on the 10 regional sectors according to GICS definition (Level 1) as homogeneous
groups of stocks.
In the allocation of each economic sector within one region, as well as of each region within
the global portfolio, we reject the very popular method of weighting baskets by the inverse of
their volatility: this procedure ignores correlations that should be taken into account explicitly
instead, as they may be highly heterogeneous across sectors and regions.
One way to account for them is to use the measure of marginal contribution to total risk.
If W is the vector of weights of portfolio P, and is the volatility of portfolio P, MCi is the
marginal contribution to risk of each asset, and is equal to:

MCi =
Wi
In order to come out with full risk parity, the following relation must hold:
MCiWi = MC jW j for any i j
In other words the risk contribution should be the same for any basket:
RC = MC W = Ke
where RC is the vector of risk contributions, k is a constant, MC is the vector of marginal

contributions, e is a vector of ones, and is the element by element product operator.
Marginal contributions are function of volatilities and correlations of any basket with the rest
of the portfolio, with correlations depending on portfolio composition itself.
In our equation, weights are the unknowns and should lead to a constant vector when
multiplied by MC, which depends on weights themselves: the problem is clearly recursive,
and the solution is endogenous.

Intuitively, we should set the target weight of each basket as proportional to the inverse of its
marginal contribution. Starting from a discretionary non-optimal initial portfolio composition
WINIT:
1
W TGT ~
MC INIT
Unfortunately, moving from WINT to WTGT:
MC TGT MC INIT
Marginal contributions change as a consequence of weights change, and as the change in

marginal contributions is different for any asset, the risk budget is no longer equal across all
the assets, and the weights calculated accordingly no longer guarantee risk parity. As a
consequence, WTGT may be somewhat far from optimal.
It is clear that the choice of the starting point where we compute marginal contributions is
crucial. For instance, a standard market index, where sectors and regions are weighted by
market cap, is not a good configuration for estimating marginal contributions, as they may be
exacerbated by index characteristics (very low weight in some extremely volatile sector like
IT in the Euro zone, may lead to an artificially low contribution).
In order to come out with a satisfactory solution in a reasonable time, we have chosen to
observe marginal contribution in the most neutral portfolio composition: the equally weighted
composition.
Equal weights as starting point have the advantage of not being far from the (still unknown)
optimal solution: in this way the marginal contributions that we use for target weight
calculation are a very good proxy for the marginal contribution that we will observe after
weight calculation, thus ensuring a truly well balanced risk contribution.
In order to check for the accuracy of our solution, we have computed percentage contribution
(PCi), for any basket, at any date of our back test.
Wi MCi
PCi =


As for the second step, that is the methodology at a stock level, we weight stocks inversely
proportionally to their total volatility. We have already evoked that, with equal correlations,
inverse of total volatility is a weighting scheme that guarantees risk parity across stocks, while
we observe that for similar correlation this solution is a very efficient proxy. Within the same
region and sector, correlations among stocks are very similar and most of the dispersion in
stocks returns is explained by their difference in beta and by idiosyncratic risk.
Also, we prefer using total risk as the relevant metric because it is intuitive and easy to
estimate, even without resorting to a complex risk model: while it is easy to challenge or
validate an existing risk model in re-estimating marginal contributions on 10 sectors and 3
regions (or even on 30 region-sector baskets together), it is not such an easy task to do the
same with the roughly 2,000 constituents of an index.
In a focus on the Euro zone, Chart 4 of Annex 5.2 reports the highest and lowest percentage
contribution for any sectors, during the 10 years of our sample, and Chart 5 reports full sample
means.
If we implement sector allocation after re-building the sector with stocks weights inversely
proportional to their volatility (we refer to this procedure as bottom up as step 2 is
performed before step 1), percentage risk contribution ranges from 9.9% to 10.1%, leading to
almost perfect risk parity. Performing step 1 before step 2 (top down), risk parity is slightly
less accurate, as marginal contributions are estimated on sector baskets where stocks are
weighted for free float adjusted market cap, thus diverging from the actual baskets (where
stocks are finally weighted for the inverse of volatility). Both solutions are definitely
satisfactory, if we consider that, within the MSCI Index, risk contributions range from 3% to
25%.
As for performance (Annex 5.2, table 16), data are again extremely good: from the beginning
of 2003 to third quarter of 2012, Smart portfolios outperformed MSCI indexes by roughly
5% annually in World Developed markets, World Emerging markets, EMU area, and Pacific
All Countries ex Japan. As for Japan, outperformance is 3.5% when the Smart portfolio is
built on MSCI index constituents, and it rises to 5% if it is built on TOPIX index constituents5.
Volatilities are reduced by 10 to 25% and draw-downs by 20 to 25%.
5
With MSCI Japan constituents, we perform risk parity on the 10 GICS level 1 sectors. With TOPIX constituents, we involve
17 sectors, according to TSE classification.

Real time data in Europe and in the EMU area (from mid-2010) are really encouraging indeed,
as both portfolios outperform the standard index in absolute and risk-adjusted terms, with
volatilities and draw-downs reduced similarly to the back-tests.
The global portfolio (available since January 2012 only) so far exhibits the same risk-adjusted
return as the MSCI index.

Acknowledgments
We are grateful to Corentin Bouzac for his help in literature review and statistical tests, and to
Amundi Equity Quant Research for back-tests.
References
Ang A., Hodrick R.J., Xing, Y., Zhang, X. 2006. The Cross Section of Volatility and Expected
Returns. The Journal of Finance. Vol. LXI, No. 1 (February)
Baker M., Bradley B., Wurgler J. 2009. A Behavioral Finance Explanation for the Success of
Low Volatility Portfolio. New York University, Working Paper
Baker M., Bradley B., Wurgler J. 2011. Benchmarks as Limits to Arbitrage: Understanding
the Low Volatility Anomaly. Financial Analysts Journal, Vol. 67, No. 1, CFA Institute
Baker N. L., Haugen R. A. 2012. Low Risk Stocks Outperform within All Observable Markets
of the World. Social Science Research Network, Working Papers Series (April)
Baker N. L., Haugen R. A. 1991. The Efficient Market Inefficiency of Capitalization-

Weighted Stock Portfolios. The Journal of Portfolio Management, Vol. 17, No. 3 (March)
Black F., Jensen M.C., Scholes M. 1972. The Capital Asset Pricing Model: Some Empirical
Tests. Studies in the Theory of Capital Markets, No 81
Blitz D., Vliet P. 2007. The Volatility Effect: Lower Risk without Lower Return. The Journal
of Portfolio Management, Vol. 17, No. 1 (Fall)
Clarke R., De Silva H., CFA, Thorely S., CFA. 2011. Minimum Variance Portfolio
Composition. The Journal of Portfolio Management, Vol. 37, No. 2 (Winter)
Clarke R., De Silva H., CFA, Thorely S., CFA. 1991. Minimum Variance Portfolios in the US
Equity Market. The Journal of Portfolio Management, Vol. 33, no. 1 (Fall)
Clarke, R., De Silva, H., and Thorley, S. 2012. Risk Parity, Maximum Diversification, and
Minimum Variance: An Analytic Perspective. Working Paper
Carvalho R.L., Lu X., Moulin P. 2011. Demystifying Equity Risk-Based Strategies: A Simple
Alpha Plus Beta Description. The Journal of Portfolio Management, Vol. 38, no. 3 (Spring)
Chunhachinda P., Dandapani K., Hamid S., Prakash AJ. 1997. Portfolio Selection And
Skewness: Evidence From International Stock Markets. The Journal of Banking and Finance,
No. 21 (February)

Choueifaty, Y., Coignard, Y. 2008. Towards Maximum Diversification. The Journal of
Portfolio Management, Vol.35, No. 1 (Fall)
Cornell, B., Roll, R. 2005. A Delegated-Agent Asset-Pricing Model. Financial Analyst

Journal - CFA Institute, Vol. 61, No. 1
Fama E., French K. 1992. The Cross-Section of Expected Stock Returns. The Journal of
Finance. Vol. XLVII, No. 2 (June)
Fama E., MacBeth J.D. 1973. Risk, Return, and Equilibrium: Empirical Tests. The Journal of
Political Economy, Vol. 81, No. 3 (May)
Falkenstein E. 2009. Risk and Return in General: Theory and Evidence. Social Science
Research Network, Working Papers Series (June)
Frazzini, A., Pedersen L. H. P. 2010. Betting Against Beta. Working Paper, AQR Capital
Management, New York University and NBER Working Paper 16601
Kraus A., Litzenberger R.H. (1976). Skewness Preference and the Valuation of Risk Assets.
The Journal of Finance Vol XXXI, No. 4 (September)
Markowitz H. 1952. Portfolio Selection. The Journal of Finance, Vol. VII, No.1 (March)
Maillard S., Roncalli T., Teiletche J. 2009. On the Properties of Equally-Weighted Risk
Contributions Portfolios. Social Science Research Network, Working Papers Series
(September)
Morel T., Malongo, H., Lambinet, R. 2013. Bubbles and Regimes: Two Complementary
Approaches. Amundi, Cross Asset Investment Strategy Special Focus (Forthcoming)
Prakash A. J., Chang C. H., Pactwa T. E. 2003. Selecting a Portfolio with Skewness: Recent
Evidence from US, European,
and Latin American Equity Markets. Journal of Banking and Finance No. 27
Piotroski J. D. 2000. Value Investing: The Use of Historical Financial Statement Information
to Separate Winners from Losers. Journal of Accounting Research, Vol. 38, Supplement 2000
Sharpe W.F. 1964. Capital Asset Prices: A Theory of Market Equilibrium under Conditions of
Risk. Journal of Finance, Vol. XIX, No. 3 (September)
Shefrin H., Statman M. 2000. Behavioral Portfolio Theory. Journal of Financial and
Quantitative Finance, Vol. 35, No. 2 (June)
Sornette, D. 2003. Why Stock Markets Crash (Critical Events in Complex Financial Systems).
Princeton University Press
Thomas R., CFA, Shapiro R., CFA. 2007. Managed Volatility: A New Approach to Equity
Investing. State Street Global Advisors

ANNEX 1
(Ex-post risk and returns in World Developed markets)6
6
Source for all figures: Amundi, Factset

ANNEX 2
(Skewness in World Developed markets) 7
7

ANNEX 3
(Convexity in World Developed markets)8
8

ANNEX 4
(Amundi NextGen Equities and low risk exposure)9
9

ANNEX 5
(Amundi NextGen Equities)10
Annex 5.1 Minimum Variance
10

Annex 5.2 Smart Beta (risk parity)11
11

WP-035
Determining the Maximum

Number of Uncorrelated
Strategies in a Global Portfolio
Ling-Ni Boon,
Doctoral Student, Research Analyst Investor Research Center, Amundi
Florian Ielpo,
Portfolio Manager, University of Paris 1, Panthon-Sorbonne
April 2013
The maximum number of uncorrelated strategies to be included in

a portfolio is a pertinent issue when building cross-asset strategies.
This question can be formulated as the search for the optimal
number of factors in a factor model, a task for which various criteria
have been proposed, many of them in the framework of Principal
Component Analysis. Using a refined information criterion, we
estimate the number of factors to be associated to five datasets. In
the case of US Treasury Bond rates, we find the usual three factors,
while credit spreads display one factor. Commodity prices are
affected by two common factors, whereas currencies demonstrate
only one factor. When bringing all these assets together, we find that
a total of five factors are at work. Factors are identified, interpreted
and their stability over time is investigated via testing the
significance of correlation between factors over a rolling window.
During recessions, the number of uncorrelated strategies drops for
US Treasury Bond rates. Yet for commodities, the concentration
of correlation is weaker, providing evidence to support their
diversification potential, even during economic downturns.

1. Introduction
In a portfolio comprised of different assets from the same class, or numerous asset
classes, the drivers of return variation may appear elusive. While there are idiosyncratic
factors influencing return variation, there a r e a ls o common factors that account for
the portfolios collective variation. Uncovering and decomposing the importance of these
common drivers help in cross-asset strategy building in portfolio management - a topic
of interest to investment managers. One method o f formalizing the task of determining
the maximum number of uncorrelated strategies to include in a global portfolio is
selection of the number of factors in a large-dimensional factor model.
Factor models have been widely studied, mainly in macroeconomics and asset pricing.
In macroeconomics, they are used to determine the factors that influence measures of
the economy, or in policy analyses. For example, Bernanke et al. (2005) introduced the
FAVAR model to analyse monetary policy, Forni et al. (2003) study the structure of the
macroeconomy, while Favero et al. (2005) compare static and dynamic principal
components in estimating macroeconomic variables. In consumer demand theory,
Lewbel (1991) applied factor models to budget share data to reveal information about
the demand system. In finance, Chamberlain and Rothschild (1983) extend Arbitrage
Pricing Theory using the factor model, which has since been used not only to
decompose risk and return into explicable and inexplicable components, but also to
describe the returns covariance structure, prediction, and to construct portfolios with
desired characteristics, among others. More recently, Merville et al. (2001) analysed
the factor structure of equity returns.
Factor models are categorized by type of factors. There are (i) macroeconomic factors
(observable economic or financial time series), (ii) fundamental factors (observable
asset characteristics), or (iii) statistical factors (unobservable asset characteristics). An
example of a well-known single factor model is the Capital Asset Pricing Model

(CAPM), which describes the relationship between risk and return. The observation
that stocks with a small capitalization and a high book-to-market ratio tend to perform
better led Fama and French to refine CAPM as a three-factor model. In this paper, the
focus is on statistical factor models, and the factors are computed using Principal
Component Analysis.
The operational value of exposing common factors that drive variances in these markets
is immense. In addition to theoretical interest regarding the suitability of methodologies
on empirical data, investors would be better able to evaluate t h e risk-adjusted
performance of portfolios allocated to various kinds of assets, and to build cross-asset
strategies. For instance, we find that the risk appetite factor in the Global Macro Hedge
Fund accounts for up to 45% of the datas variance, hence i t deserves heightened
attention. Conversely, Asian market movements explain only about 4% of variances, so
tracking them yields only a tenth of the insights o b t a i n e d b y tracking investor
sentiment.
Numerous methods to determine the number of factors in the case of unobserved
factors have been proposed. Arguably the most popular is via information criteria
(IC). IC is based on the idea that an (r + 1)-factor model has to fit at least as well as
an r-factor model, but is less efficient. The well- known Akaike Information Criterion
(AIC) and the Bayesian Information Criterion (BIC) cannot be directly adopted as they
are functions of N or T alone, hence they fail to consistently estimate the number of
factors when the factors are unobserved. Bai and Ng (2002) (BN) propose a set of
six penalty functions to replace the ones in AIC and BIC, and established conditions
ensuring the consistency of their methods. Despite its wide empirical adoption, BNs
criterion often does not converge, as demonstrated in Forni et al. (2007). Alessi et al.
(2009) (ABC) refine BNs criterion and demonstrate that their criterion has superior
performance. Alternate approaches include analysing the factor loadings (Connor and

Korajczyk, 1993), tests on the eigenvalues of the covariance matrix of returns (Onatski
(2009), Onatski (2009b), Kapetanios (2005)), numerous tests on the rank of the
covariance matrix (Lewbel (1991), Forni and Reichlin (1998)), and a graphical method
that is rarely used due to its lack of theoretical basis (Donald, 1997).
By comparison of the criteria in a Monte Carlo study, ABCs is deemed to be the
overall best in terms of accuracy and precision. T h e a pplication of ABCs criterion
to five datasets yields the following number of factors: five for Global Macro Hedge
Fund (GMHF), three for US Treasury Bond Rates (USTB), two for Commodity Prices,
and one each for US Credit Spreads (USCS) and currencies. T h e t otal variation
explained by the factors varies, as it is 74% for GMHF, 94% for USTB, 49% for USCS,
27% for commodity prices, and 59% for currencies. Economic interpretation is attached
to the factors according to correlation between the factors and the r e t u r n s o f assets
whose variances they describe. The five factors for GMHF are associated with risk
appetite, commodities, t he US dollar, the Japanese market and Asian stock markets.
Those for USTB fit the description of factors found in previous research and are
labelled as level, slope and curvature. USCSs sole factor corresponds to mid-range risky
assets while that for currencies is labelled as the carry factor. The pair of factors for
commodity prices is linked to energy and metal.
The stability of the number of factors over time is investigated by testing the
significance of correlation between factors. Even though by definition of principal
components, the factors are, in theory, orthogonal to each other, the estimated factors
may not be so. Instantaneous correlation between these estimated factors can be
uncovered by considering a rolling window over the time dimension. Correlation
between the factors over rolling windows does not yield an overarching conclusion for
all datasets regarding the hypothesis that during periods of economic downturn, fewer
factors are required to explain variances in the data due to increased cross-market

correlation. For traditional asset classes, such as USTB, it is evident that correlation
between factors increased. Yet for commodity prices, it is less clear. When all asset
classes are considered together in the GMHF dataset, the different factor interactions
that are observed within each asset class manifest in a more complex manner, and
suggest that cross-asset correlation may be a leading indicator of economic cycles, as a
spike in the correlation is observed prior to the 2007 to 2010 financial crisis.
This paper is organized as follows. We first present the factor model and briefly explain
the numerous criteria proposed to determine the number of factors. After a Monte Carlo
Study for selected methods, we apply the methods to datasets relevant to the investment
management industry. The results are interpreted a n d t h e i r stability over time is
analysed.
2. Methodology
2.1. Presentation of the Factor Model
An r-factor approximate factor model is specified as:
Xit = t Fit + eit
i = 1 . . . N and t = 1...T (1)
whereby Xit is the observed data for the ith cross section at time t, F t is the r 1 vector
of common factors, i is the r 1 vector of factor loadings, eit is the idiosyncratic
component, t Ft is the common component of Xit , and t denotes the complex
conjugate transpose of the matrix. Ft and et are assumed to be uncorrelated and the
matrix comprised of cov(ei, ej) is not necessarily diagonal (i.e. allows non-
pervasiveness in asset-class-uncertainty), but the largest eigenvalue of the idiosyncratic

components covariance matrix is bounded to limit the degree of correlation. Hence, Xt
is explained by both common components and specific factors.
The tests considered in this paper are:
1. Bai and Ng (2002) (BN)
The estimated number of factors by BN is the integer corresponding to the lowest
value of loss function V (r, F r )+rg(N, T ), or log(V (r, F r ))+r 2 g(N, T ), with
whereby F is the matrix of r factors, =
(1 . . . N), g(N, T ) is the penalty for over-fitting, r is a constant, and is a
consistent estimate of . In practice, can be replaced
by , with as the maximum number of factors considered.
Two examples of g(N, T ) are ) which is
frequently used in empirical works, and which has
been shown to possess good properties when errors are cross-correlated (Bai and
Ng, 2008).
2. Alessi et al. (2009) (ABC)
Alessi et al. (2009) propose a refinement of BN that multiplies a constant, c, to the
penalty function as follows, V (r, F r ) + rcg(N, T ), or log(V (r, F r )) + rc 2 g(N, T ).
The number of estimated factors remains as the one yielding the lowest value for
these modified loss functions. Furthermore, the authors suggest evaluating the
loss functions over random subsamples of the data to find an estimate that is
insensitive to the sample size, and neighboring values of c. Detailed explanations
of the role of c are provided in Hallin and Liska (2007), while the generation of the
random subsamples is described in Alessi et al. (2009). This criterion has been
shown to provide a solution when BNs criterion fails, and it is not any more

complex in implementation because it requires, in essence, multiple repetitions of
BN.
3. Connor and Korajczyk (1993) (CK)
An alternate approach developed by Connor and Korajczyk (1993) (CK) is based
on the idea that an r factor models (r + 1)st factor can have nontrivial factor
loadings for some assets, but only a small proportion of them. A statistical test
for this is developed to test whether the (r + 1)st factor is pervasive. It proceeds by
running two regressions by Ordinary Least Squares (OLS), one with factors
while the other with r + 1 factors. The adjusted squared residuals,
with as the OLS estimated residuals, are computed. A cross-
sectional mean for both it s, defined as , is calculated next, for both regression
models. Then the even months for the regression with r + 1 factors is
subtracted from the odd months for the r factor model, giving a value N .
Under the null hypothesis that the model has r factors, N , with as the
1
covariance matrix of is asymptotically standard normal as n , hence in
practice, a t-test is carried out on the estimates . In order to establish the
distribution of the idiosyncratic components, the authors made the assumption
of homoscedasticity across time periods - pointed out in Bai and Ng (2002) to be
undesirable - and that ei = (ei1 , ei2 , . . . , eiT ), i = 1, . . . is a mixing process.
4. Onatski (2009b)
Drawing upon the property that an r factor panel of data has unbounded first r
largest eigenvalues of the covariance matrix of Xt, and bounded (r + 1)st
eigenvalue, Onatski (2009b) developed a statistical test for Ho : r = r0 versus

H1 : r0 < r r1 , with r as the number of factors, r1 and r0 are the upper and
lower bounds for the number of factors, which are determined by prior
knowledge. Beginning from r0 , for each successive r, the Discrete Fourier
Transforms (DFTs), are computed at pre-specified frequencies j .
I n El Karoui (2006), Xt is shown to be asymptotically distributed as Tracy-
Widom. The test statistic is whereby i is the largest
eigenvalue of the covariance matrix of Xt. R is essentially a measure of the
curvature at the would-be breakpoint of the frequency-domain Scree plot
postulated by the alternative hypothesis that the model has more r factors, but
fewer r1 factors. Critical values for the test are provided in Onatski (2009b) for
up to r = 18 factors. Similar to the case of CK, imposing a valid distribution
requires fairly strong assumptions such as having idiosyncratic components that
follow a Gaussian distribution.
2.2. Monte Carlo Study
Before choosing one method over another, we perform a Monte Carlo test to evaluate their
relative performance on simulated data with various qualities. The experimental design
employs seven data-generating processes (DGPs) that differ in their relationship
between elements of the idiosyncratic components of the following model:
Ftj and ij are normally distributed with zero mean and unit variance. This is similar to
the DGPs used in Alessi et al. (2009).

1. Homoscedastic idiosyncratic component, same variance for the common an
diosyncratic component: eit N (0, 1) and r =
2. Heteroskedastic idiosyncratic component, same variance for the common and
idiosyncratic component:
3. Homoskedastic idiosyncratic component, the common component has a larger
variance than the idiosyncratic component: eit N (0, 1) and r = 2
4. Homoskedastic idiosyncratic component, the common component has a smaller
variance than the idiosyncratic component: eit N (0, 1) and .
5. Small cross-section correlation across idiosyncratic parts, same variance for the
common and idiosyn cratic compone nt:
6. Seri al corr elation across idiosy ncrat ic parts, t h e com mon component has a
smaller variance than the idiosyncratic component:

7. Serial and small cross-section correlation across idiosyncratic parts, the common
component has a larger variance than the idiosyncratic component.
For all seven DGPs, we test for the pairs of time and cross-section dimension (N, T ) =
(70, 70), (100, 120), (150, 500). The true number of factors, r is chosen to be 1, 3, 5,
8, 10, and 15, all of which are consistent with the requirement r < minN, T . The
corresponding rmax for BN and ABC, and the upper bound on CK and Onatski test is
rmax = 8 when r = 1, 3, 5; rmax = 15 when r = 8, 10; rmax = 20 when r = 15. CK and
Onastkis test always began with the lower bound of 1, to suggest that in many financial
datasets, it is unlikely to have prior knowledge beyond the belief that there should be at
least one factor, given that the data does indeed have a factor structure. The
correlation between the common and idiosyncratic components is = 0.5, = 0.2,
while H = max . 500 Monte Carlo replications are performed for each instance.
Additionally, to test ABCs criterion, the parameters to determine the random
subsamples a r e :
nJ = (see Alessi et al. (2009)), and cmax = 13 with step size 0.01.
BNs criterion has perfect performance for DGPs 1-4, correctly identifying the number
of factors. However, when the DGP demonstrates cross-section or serial correlation
across the idiosyncratic components, such as in DGP 5 to 7, the criterion slightly
overestimates the number of factors. There is no obvious effect of dimensions (i.e. N and
T ) on the results, which is consistent with BNs claim that their criterion yields precise

estimates for minN, T > 40. While ABC does not display perfect performance, like
BN d o e s for DGP 1-4, b e c a u s e it is generally plagued by mild overestimation, its
performance is more accurate for DGP 5. This result is similar to ABCs own Monte
Carlo study. Even though BN has stellar performance in most cases, the adoption of
ABC is justified since most financial portfolio time series demonstrate cross-section and
serial correlation.
Onatskis criterion performance pales in comparison with the other criteria in almost all
cases, as it estimates that the true number of factors is 1 close to 60% of the time, 2
about 30%, and 3 about 10% of the time, being insensitive to the true number of factors.
This could be due to having the lower bound of the test always set at one to reflect the
case that when the test is implemented on actual data, no prior knowledge is available
to determine the lower bound. An upper bound, however, can be set since it must be
fewer than minN, T , and methods such as ABC are developed to be less sensitive to
the upper bound, hence a larger upper bound can always be selected.
CKs test has a similar tendency of underestimating the true number of factors as 1
close to or exceeding 50% of the time when r 5, when the time dimension is small.
However, for the sample with N = 150, t = 500, i.e. large cross-section and time
dimensions, the test performs reasonably well for all DGPs except for DGP 2 and 7,
correctly identifying the number of factors at least 50% of the time for DGP 1-4 and 6,
and overestimating the factor by 1 for DGP 5. In the case of DGP 2, the number of
estimated factors is 1 more than 80% of the time, regardless of the actual number of
factors, time and cross-section dimensions.
In general, the Monte Carlo study substantiates BNs criterion as superior not only in
terms of accurate estimates, but also in terms of its ease of implementation2 . In cases

where BNs criterion does not perform well, either due to cross-sectional or serial
dependence, or difficulty in estimating rmax , then ABCs criterion should be executed.
CKs test may perform well when the time and cross-section dimensions are large.
The three best criteria: BN, ABC and CK are implemented on five sets of data comprising
equities, commodities, credit spreads, interest rates, and currencies between 1997 and
2012.
3. Empirical Results
In this section, ABCs criterion is applied to five datasets covering major asset classes
such as equities, US Treasury Bonds, credit spreads, currencies and commodities. The
estimated number of factors are first identified and labelled. Then, the stability of
factors over time is analysed by testing the significance of correlation between them,
and re-estimating the number of factors after splitting each dataset along the time
dimension according to the economic cycle.
3.1. Global Results
The Global Macro Hedge Fund (GMHF) dataset is comprised of major indices,
government bonds, currency exchange rates, currency exchange rate and oil futures,
dated between January 1999 and March 2012. The data is of weekly frequency due to
its cross-boundary nature, so the difficulty i n r e c o n c i l i n g market closing times
worldwide is mitigated. Next is a dataset of daily closing rates of US Treasury Bonds
(USTB) w i t h a maturity o f 3 months to 30 years, from January 1997 to March 2012.
The US Credit Spreads (USCS) data is comprised of daily closing rates categorized by
industry (e.g. financial corporation, insurance, and energy), financial rating (e.g. AAA,
AA, BBB), and duration (e.g. 1-3 years, 3-5 years), from January 1997 to March 2012.
T h e d aily closing prices of commodities such as gold, aluminium, natural gas, corn,

wheat, the S&P GSCI (Goldman Sachs Commodity Index) etc. from October 1998 to
March 2012 are compiled in the dataset of Commodities. The fifth dataset is called
Currencies, and includes the daily prices of the Euro, Great Britain pound, Swiss Franc,
Japanese Yen, Canadian Dollars, Australian Dollars, New Zealand Dollars, Norwegian
Krone, and the Swedish Krona in terms of US Dollars from January 1999 to December
2012. BN, ABC and CK criteria are applied to log prices or log spread variations for
GMHF, USCS, Commodity Prices and Currencies datasets, whereas for USTB, rate
variation is used since with two-year rates close to zero or negative, their percentage
variation are extreme. The composition of each dataset is presented in Exhibit 1.
Aside from applying BN, ABC and CK criteria to the datasets, since the Monte Carlo
study indicates that the selected criteria generally have poorer performance when cross-
section correlation exists, the influence of such dependencies on the accuracy of results
obtained are evaluated by fitting a Vector AutoRegressive (VAR) model to remove
dynamic linear dependence, then applying the same criteria on the residuals. Using the
AIC criterion to determine the number of lags to include in the VAR model, USTB
and USCS datasets are fitted with VAR(3), commodity prices with VAR(2), and GMHF
and currencies with VAR(1). BN and ABC criteria provide the same outcome when
the analysis is done on the residuals as on the returns data. CKs criterion yields
slightly different estimates. Despite its commendable performance in the Monte Carlo
study, BNs criterion fails to converge on all datasets - rmax is always estimated. CKs
estimates do not always coincide with ABCs estimates, but the latter is taken to be
more accurate due to better performance in the Monte Carlo study, and it is invariant
to whether linear dependencies exist in the data.
In Exhibit 6, the number of factors estimated and the proportion of variance explained by
each factor,

i.e. ratio of the ith eigenvalue and the sum of all eigenvalues of the covariance matrix
of the returns, are presented. The proportion of variances explained by the estimated
number of factors suggests the concentration of correlation. An asset class or portfolio
for which correlations across asset variations are high should have a high percentage
explained. Using ABCs criterion, the GMHF is estimated to have five factors, which
collectively explain about 74% of variances in the dataset. USTB has three factors
that explain 94% of the variances. The commodity prices dataset is estimated to have
two factors. Together they explain 27% of the variances - the lowest among the
datasets considered. Commodities low concentration of correlation is consistent with
the findings of Gorton and Rouwenhorst (2004) on the asset classs diversification
potential, and becomes the rationale for investors to increase portfolio allocation to
commodity assets (Daskalaki and Skiadopoulos, 2011). Next, o n e factor each is
estimated for US credit spreads and currencies, accounting for around 49% and 59% of
the variances respectively.
After obtaining the number of factors, the sensitivity of each assets return to the factor,
i.e. factor loadings, is investigated and presented in bar charts. This notion of sensitivity
can be extended to that of a portfolio, as it is merely the sum of the corresponding asset
in the portfolio. Moreover, portfolios that are insensitive to a particular factor can be
constructed by selecting assets such that their weighted sum (i.e. loadings treated as
weights) is zero. Factor loadings also help in identifying the factor, that is, by
labelling the latent, hypothetical factors according to their relationship with the assets.
Absolute correlation of the factors with the assets returns is first placed in
descending order, and progressively added to the selection, i.e. adding those with the
largest absolute correlation first, until the selection achieves at least 95% R2 when used
as explanatory variables in a simple regression model for the asset returns. These

correlations are presented as bar plots in Exhibits 2 to 5. A s ummary of the results is
provided in the final table in Exhibit 1.
Global Macro Hedge Fund
Exhibit 2 shows that factor 1 of GMHF is highly correlated to major equity indices
worldwide. This suggests that factor 1 corresponds to a risk appetite factor. During
bullish periods, investors are more willing to take risks, hence the y prefer to invest in
equities. Conversely, during bearish periods, investments are diverted into bonds, which
have lower risk. Factor 2 relates to oil and GSCI, hence it is labelled a commodities
factor. Factor 3 is a dollar factor due to its association with the US dollar. Factor 4
represents the Japanese market, whereas factor 5 is linked to Asian markets.
US Treasury Bonds
Prior research by Dai and Singleton (2000) and Litterman and Sheinkman (1991) has
shown that observed variation in bond prices can be explained by three factors: level,
slope, and curvature. These names describe the shift of the yield curve in response to a
shock. A level shock shifts the curve in a parallel manner, resulting in an almost equal
effect on bonds of all maturity. The slope factor implies larger shocks for bonds with
s h o r t maturities, as compared to bonds with longer maturities. Its name is derived
from the effect of the yield curve becoming less steep as a result of a slope shock.
Curvature affects medium-term interest rate, hence i t s h o w s as a hump on the
yield curve. Indeed, Exhibit 3 shows that factor 1 has relatively uniform correlation
across bonds of all maturity, just as a level factor would. T h e c orrelation for factor 2
changes sign once, and has a larger correlation, a n d hence impact, on bonds with
shorter maturities, as should a slope factor. Factor 3 has a hump in its correlation
figure, fitting the description of a curvature factor. Thus, the findings are consistent
with existing results.

US Credit Spreads
The sole factor for USCS is correlated to A-rated investments, financial corporates and
industrial credit spreads that dominate those with utilities and consumer cyclicals, as
shown in Exhibit 3, implying that it is associated with assets with mid to low credit
risk. Investments with low credit risks and conventionally small credit spreads, such as
AAA-rated assets and treasury bonds, are not among those that possess the highest
correlation with the factor, suggesting that they play a relatively smaller role in the
movement of returns. Similarly, industries responsible for providing goods with low
substitutability, such as utilities, healthcare and energy, along with those that are highly
dependent on the state of the economy, such as consumer cyclicals (e.g. entertainment,
automotive etc.), possess lower correlation with the factor. In comparison with other
datasets, assets in USCS have a relatively more uniform correlation (i.e. all above 50%)
with the factor.
Commodity Prices
Interpretation for the two factors of commodity prices is straightforward: factor 1 is
the energy factor. Factor 2, being correlated with numerous metals, is the metals
factor, as is evident in Exhibit 4. Similar to the case of USTB, the number of factors
in commodity datasets is commonly studied. Daskalaki et al. (2012) investigate the
common components of a cross-section of commodity futures data using numerous asset
pricing models intended for equities, macro and equity-motivated factor models, and
principal component factor models to find that none of them satisfactorily prices
commodity assets. The authors attribute this poor pricing model performance to
heterogeneity in commodity markets, and segmentation of equity and commodity
markets. Our results are consistent with those of Daskalaki et al. since among all
datasets, commodity price factors collectively explain the least amount of variance in
the data, an observation that supports the commodity diversification effect, made

popular by Gorton and Rouwenhorst (2004), and which motivated investors to increase
their portfolio allocation in this asset class (Daskalaki and Skiadopoulos, 2011).
Furthermore, the identification of factors by commodity group and the low correlation
across these groups suggest a sectorial framework when studying the commodity
market.
Currencies
For currency datasets, ABC estimates a single factor. It is labelled as the carry factor
as it is correlated to currencies with high average rates, such as the Australian Dollar,
Norwegian Krone and Swedish Krona, as shown in Exhibit 5. It could also be
understood as the dollar factor as all currency prices are positively correlated with the
US dollar. To the best of our knowledge, there is no empirical evidence in the
literature regarding the number of factors in the currency datasets. Thus, by applying
ABCs criterion to the datasets, the number of salient factors is determined. These
factors are estimated and identified by analysing their correlation with returns.
3.2. Stability Analysis
Since all datasets span ten years, including the most recent financial crisis in 2008, it is
of interest to know whether the number of estimated factors is stable over time.
Literature on t h e stability of factor models is sparse. Most authors have either
assumed stability or relied on some graphical method. Bliss (1997) divided his sample
into three sub-periods and investigated the factor loadings on each. His hypothesis is
that if the factor loadings appear to be similar a c r o s s all sub periods, then the
factor structure is stable. Prignon and Villa (2002) found that factor loadings are
stable, but factor volatility varies over time. Chantziara and Skiadopoulos (2008)
analysed the term structure of petroleum futures, also by splitting the sample into two.
Since the PCA results are similar in both samples, the authors conclude that the factor
structure is stable. Attempts to devise formal tests include those by Audrino et al.

(2005), and Philip et al. (2007), who focused on inte re st rate term structure. The
former relies on testing t h e equivalence of the factor loadings on sub-periods, whereas
the latter involves constructing a bootstrap distribution for the test statistic. An
evaluation of the aforementioned research is a free-standing topic worthy of a full-length
research paper, so we have not done it here.
To investigate the stability of the factors on our dataset, two approaches are attempted.
The first assumes that the factor structure is stable, but the correlation between the factors
may not be. Correlation between factors when the factor structure is stable is closely
related to the concentration of eigenvalues. To reveal the evolution of the relationship
between factors, the t-test for significance of correlation is used. Next, discarding the
assumption of a stable factor structure, the dataset is split into expansion and
contraction economic periods, and the corresponding number of estimated factors in each
sub-period is obtained. To take into consideration t h e f a c t that financial market
performance is a leading indicator of macroeconomic situation, the periods of contraction
and expansion are lagged by a negative number of months, from -1 down to -12. In other
words, the number of factors prior to the start of a contraction or expansion period is
computed to determine if the change in number of factors occurs before the economy
takes a turn in its cycle. The key difference between the two approaches is that by
assuming the factor structure is stable in the former, the estimate of the number of
factors is done only once, and the focus is on their dynamics, a n d specifically t h e i r
correlation, over time. In the latter, the number of factors is estimated for each variation
in the economic cycle using ABCs criterion.
3.2.1. Significance of Correlation Between Factors
In this section, the factor structure estimated in the previous section is assumed to be
stable but the dynamics between factors evolve over time. Correlation between the factors
is tested over six-month rolling windows for GMHF, and one-year rolling windows for

US Treasury Bond rates as well as commodity prices, using the t-test for significance of
correlation. Exhibits 7 and 8 plot the number of uncorrelated factors as a result of the
t-test using the test statistic:
Under the null hypothesis that r = 0, i.e. the correlation between the factors is
insignificant, t Student with degree of freedom equalling N 2.
Plots of the number of uncorrelated factors suggest that the factors tend to be
correlated during recession periods. Increased correlation between factors is
particularly obvious for USTB. Exhibit 7 shows a marked drop in the number of
uncorrelated factors to only one factor during the most recent financial crisis. This is
attributable to historically low interest rates as investors sought safe investments. Since
the factors were identified by their impact on bonds of different maturity, i.e. factor 1
affects bonds of all maturity evenly, factor 2 has a strong impact on short maturity
bonds while factor 3 has the highest influence on bonds of mid-term maturity, having all
factors collapse onto a single factor suggests that the prolonged near-zero short-term
interest rates and low long maturity rates have removed the disproportionate impact
that shocks have on the yield curve of bonds of varying maturity. Hence, the level,
slope and curvature factors merged into a single factor.
Factors for commodity prices appear to be insensitive to the economic situation, as
the number of uncorrelated factors remains stable at two throughout the period
studied. This is in line with the result in Kat and Oomen (2006) that there is weak
correlation across commodity groups. Indeed, energy and metals, the two factors
identified, are distinct commodity groups. Moreover, energy and metals are essential to

many industries and are not substitutable; they represent immutable drivers of return
among commodity assets. Thus, the sectorial view of the commodity market remains
valid throughout ups and downs in the economy.
Since USCS and currencies have only one factor, stability analysis is not applicable.
Out of curiosity, we over-estimate the number of factors at five each, and apply the
same t-tests to investigate the evolution of the number of uncorrelated factors over time.
Exhibit 8 for USCS shows that the number of uncorrelated factors fluctuates between
one and two thus does not provide conclusive evidence that the number of factors is
lower during economic crisis. As for currencies, Exhibit 8 demonstrates that the number
of factors r e m a ins constant at one for most of the time period, except in early 1999,
when occasionally two factors are estimated. This period coincides with the introduction
of the Euro, which could have given rise to an interim factor influencing the returns.
For the 5% test on GMHF, the number of uncorrelated factors is never five - the
number estimated using ABCs criterion over the entire horizon - during recession. Since
it is also fewer than five in many other instances, it is less clear whether higher
correlation between factors is a unique feature of recessions. To further investigate this,
the mean absolute correlation between factors is plotted in Exhibit 9, on which there is
an indisputable spike in correlation in 2007. Hence, prior to the most recent financial
crisis, the correlation between factors rises substantially, followed by fluctuations in
correlation. It could also be argued that the number of factors reduced before recession,
an observation that motivated the analysis in the next sub-section. With as many as five
factors, the dynamics between factors do evolve over time, as shown by the mean
absolute correlation plot, but t he y do not necessarily materialize c l e a r l y as a lower
number of factors during recession.
The stability of the factors is dependent on the asset class. For USTB, which is
considered to be the lowest risk among those considered, the number of factors that

affects its return is lower during recessions, especially for that which occurred
between late 2007 and 2010. Commodity price factors seem invariant to the economic
climate. When combined in the GMHF portfolio, these interactions become more
complex and are realized as fluctuating correlations between factors.
3.2.2. Estimated Factors by Economic Cycles
As financial markets are often thought of as leading indicators of economic cycles, it
is possible that the change in the number of factors occurs prior to contractions in
the economy. To investigate this, the contraction periods, as determined by NBER, are
lagged by a negative number of months. For example, Lag = -1 month of the most
recent financial crisis between December 2007 and June 2009 refers to the interval
November 2007 to May 2009. The results are presented in Exhibit 10. For USTB,
the number of factors fluctuates between one and five during expansion periods,
supporting the view that the correlation between factors changes prior to recessions.
On the contrary, GMHF and US Credit Spreads have a constant number of factors
throughout expansion periods. As for commodities, the number of factors is stable,
coherent with the t-test for significance of correlation results. The number of factors in
the currencies dataset falls to zero prior to contraction periods. Therefore, the view of
financial market performance as a leading indicator of economic cycles is supported by
USTB rates only, and to a lesser extent by currencies.
4. Conclusion
This paper investigates the number of cross-asset uncorrelated strategies that are
available to portfolio managers. After reviewing the literature on existing approaches to
determine the number of factors, four methods are selected and tested. The Monte
Carlo analysis suggests that the criterion by Alessi et al. (2009) is most reliable. T h e

a p p l i c a t i o n of the criteria t o five datasets yields the corresponding estimated
number of factors by ABC in parenthesis: Global Macro Hedge Fund (5), US Treasury
Bond Rates (3), US Credit Spreads (1), Commodity Prices (2), and Currencies (1).
Plots of the number of uncorrelated factors do not all support the hypothesis of
increased cross-market correlation during economic recession. The evidence is strongest
for US Treasury Bond rates, which is most likely due to US post-financial crisis
macroeconomic policies, but i s w e a k e s t for commodity prices, aligned with the
observation of low correlation between commodity groups claimed in previous studies.
GMHF, comprised of a mix of these assets, ba r ring credit spreads, demonstrate a
combination of the observed outcomes on the rest of the datasets, yielding fluctuating
correlation between the factors during economic downturns. However, the results
regarding stability have to be considered with caution, as not every financial crisis is
followed by a recession period. Thus, there could be a change in correlation between
assets occurring outside the NBER-determined recession periods. With more information
on the factors t h a t drive the returns o f different asset classes, investors would have a
better understanding of the common sources of risk. Depending on the asset class in
mind, a strategy that is built upon exploiting these common sources of risk may have to
take the economic climate into account.

Exhibits
Exhibit 1: Composition of All Datasets and Results Summary

Range: January 1999 to March 2012
List of Assets in t h e Global Macro Hedge Fund Dataset
tock indices (13 assets) Dow Jones, Nasdaq 100, Eurostoxx, FTSE, CAC 40, DAX, IBEX, AEX, SMI, Nikkei, TOPIX,
Hang Seng, MSCI Singapore Free Index (Singapore).
Bonds (12 assets) US 30Y, US 10Y, US 5Y, US 2Y, CAN 10Y, GE 10Y,
GE 5Y, GE 2Y, UK 10Y, JP 10Y, OZ 10Y, ED 4 (Euro-Dollar bond).
EUR-USD, USD-JPY, GBP-JPY, USD-CHF, USD-CAD, AUD-USD, EUR-GBP, EUR-CHF, EUR-
Currencies (29 assets) SEK, EUR-NOK, EUR-PLN, EUR-AUD, EUR-CAD, CHF-JPY, GBP-JPY, AUD-JPY, AUD-
NZD, AUD-CAD, USD-BRL, USD-SGD, USD-KRW, USD-TWD, USD-CNY, CAD-JPY, EUR-
JPY, NOK-SEK, NZD-USD, USD-ZAR, USD.
Futures (5 assets) CHF futures, JPY futures, AUD futures, CAD futures, GBP futures.
Commodities (2 assets) Oil, S&P GSCI.
Others (1 asset) Mini SP 500.

List of Assets in t h e US Treasury Bond Rates Dataset
(Maturity) 3M, 6M, 1Y, 2Y, 3Y, 4Y, 5Y, 7Y, 8Y, 9Y, 10Y, 15Y, 20Y, 25Y, 30Y.

List of Assets in the US Credit Spreads Dataset
Master index, financial corporates, banks, insurance, industrials, capital goods, energy, utilities, consumer cyclicals,
consumer non-cyclicals, healthcare, AAA, AA, A, BBB, 1-3 years, 3-5 years, 5-7 years, 7-10 years.
Range: October 1998 to March 2012

List of Assets in the Commodity Prices Dataset
Gold, silver, platinum, aluminum, copper, nickel,
zinc, lead, WTI, Brent, gas-oil, natural gas, heating oil, corn, wheat, coffee, sugar, cocoa, cotton, soybean, rice,
S&P GSCIs: agriculture, energy, industrial metals, precious metals.
Range: January 1999 to December 2012

List of Assets in t h e Currencies Dataset
Daily price in USD of: EUR, GBP, NOK SEK, CHF, JPY, AUD, NZD, CAD.
Summary Table of Results

Number of Factors Estimated by Method
Criterion
Connor and
Bai and Ng (BN) Korajczyk (CK)* essi et al. (ABC)
Global Macro 12 (5) 5

US Treasury 4 (5) 3
US Credit Spreads rmax is always estimated. 1 (2) 1
Commodity Prices 3 (6) 2
Currencies 12 (9) 12
*Values in parentheses are the estimated number of factors when analysis is performed on residuals of a VAR model.

Exhibit 2: Factor Identification - Global Macro Hedge Fund Data
Factor 1 (Global Macro) Factor 2 (Global Macro)
Eurostoxx
SMI
EURJPY
Nikkei CHFJPY
SMI USDCAD
Hang.Seng
CAD.fut
Dow.Jones
AUD.fut
IBEX NZDUSD
USDSGD AUDUSD
AEX
GBPJPY
Mini.SP500
GBP.fut USDCHF
FTSE
CHF.fut EURUSD
DAX
CAC.40 USD
GSCI
Eurostoxx
Oil
0.0 0.2 0.4 0.6 0.8 0.8 0.0 0.2 0.4
Correlation Correlation
Factor 3 (Global Macro) Factor 4 (Global Macro)
CAC.40 Nasdaq.100 GE.5Y EURAUD GE.10Y USDKRW
USDBRL Hang.Seng OZ.10Y US.5Y US.30Y USDTWD US.10Y EURGBP

Singapore ED.4
CADJPY GSCI
CAD.fut
USDCAD
USDTWD
USDKRW
Oil
USDSGD
EURCAD
AUDCAD CHFJPY EURCAD EURUSD USDSGD USDCHF
AUDJPY CHF.fut EURJPY AUDJPY USD GBPJPY.1 CADJPY USDJPY

NZDUSD JPY.fut
GBPJPY.1
AUDUSD
AUD.fut
GBPJPY
GBP.fut USD
CHF.fut
EURJPY
USDCHF
EURUSD
CHFJPY
0.4 0.0 0.2 0.4 0.6 0.5 0.0 0.5
Factor 5 (Global Macro)
Eurostoxx
CAD.fut
Mini.SP500
USDCAD
IBEX
EURGBP
CHFJPY
EURJPY
CHF.fut
USDCHF
USD
EURUSD
Hang.Seng
Singapore
Nikkei
Topix 16
0.4 0.0 0.2
Correlation

Exhibit 3: Factor Identification for US Treasury Bond Rates and US Credit
Spreads
Factor 1 (Treasury Bond) Factor 2 (Treasury Bond)
30Y 30Y
25Y 25Y
20Y 20Y
15Y 15Y
10Y 10Y
9Y 9Y
8Y 8Y
7Y 7Y
5Y 5Y
4Y 4Y
3Y 3Y
2Y 2Y
1Y 1Y
6m 6m
3m 3m
0.2
0.0
0.2
0.4
0.6
0.8
0.0
0.2
0.4
0.6
Factor 1 (Credit Spreads)

Factor 3 (Treasury Bond)
30Y Master Index
25
20Y A
15Y
10Y Financial Corporates
9Y
8Y Industrials
7Y
5Y Years
4Y
3Y BBB
2Y
1Y Years
6m
3m AA
0.4
0.3
0.2
0.1
0.0
0.1
0.2
0.0 0.2 0.4 0.6 0.8

Correlation
Correlation
A partial characterization of USTB factors is the number of sign changes; zero, one and two, respectively. This is
satisfied by the above figures.

Exhibit 4: Factor Identification - Commodity Prices
Factor 1 (Commodities) Factor 2 (Commodities)
GSCI.Prec..Metals GSCI.Ind..Metals GSCI.Energy GSCI.Agri.

Rice Soybean Cotton Cocoa Sugar Coffee Wheat Corn Heating.Oil
GSCI.Agri.
Natural.Gas
Gasoil Brent WTI
Lead Zinc Nickel Copper Aluminium Platinium Silver Gold
WTI
GSCI.Energy
Heating.Oil
Natural.Gas
0.7 0.6 0.5 0.4 0.1 0.0 0.00.2 0.4 0.6 0.8
Exhibit 5: Factor Identification - Currencies
Factor 1 (Currencies)
SEK
NOK
NZD
AUD
CAD
JPY
CHF
GBP
EUR
0.0 0.2 0.4 0.6 0.8
Correlation

Exhibit 6: Results Summary All datasets
Global Macro hedge Fund
Factor 1 2 3 4 5
Label Global Equities Commodities US dollar Japanese Market Asian Stock Markets
Eigenvalue 0.135 0.041 0.022 0.015 0.014
Proportion (%) 45 13 7 4.8 4.2
Cumulative (%) 45 58 65 69.8 74
US Treasury Bond Rates

Factor 1 2 3
Label Level Slope Curvature
Eigenvalue 12.305 1.469 0.676
Proportion (%) 80 10 4
Cumulative (%) 80 90 94
US Credit Spreads
Factor 1
Label Mid-range risky assets
Eigenvalue 0.895
Proportion (%) 49
Commodity Prices
Factor 1 2
Label Energy Metals
Eigenvalue 0.206 0.181
Proportion (%) 15 13
Cumulative (%) 15 27
Currencies
Factor 1
Label Carry factor
Eigenvalue 0.11
Proportion (%) 59

Exhibit 7: Number of Uncorrelated Factors - GMHF, USTB and Commodity
Prices
Top to Bottom: 1% level t-test for significance of correlations for GMHF, USTB, and Commodity
5 Prices
No. of uncorr. fact.
4
3
2
1
2000 2002 2004 2006 2008 2010 2012
Time
1.0 1.5 2.0 2.5 3.0
2000 2005 2010
Time
5
4
3
2
1
1998 2000 2002 2004 2006 2008 2010 2012
Time
The shaded areas mark the intervals of contraction (peak to trough) of business cycles, i.e. March
to November 2001, and December 2007 to June 2009 (National Bureau of Economic Research
(2012)).

Exhibit 8: Number of Uncorrelated Factors - US Credit Spreads (10% level t-test,
top) and Currencies
2.0
1.5
1.0
2000 2005 2010
Time
2.0
1.8
1.4
1.2
1.0
1998 2000 2002 2004 2006 2008 2010 2012
Time
(2012)).

Exhibit 9: Mean Absolute Correlation of Factors - Global Macro Hedge Fund
0.35
0.30
Mean Absolute Correlation of Factors
0.25
0.20
0.15
0.10
2000 2002 2004 2006 2008 2010 2012
Time
(2012)). The horizontal dotted line indicates the median.
Exhibit 10: Number of Estimated Factors by Dataset, by Business Cycle

Expansions and Contractions, lagged by negative number of months
Dataset Global Macro Hedge Fund US Treasury Bonds Rate US Credit Spreads
Lag (No. of months) Expansion Contraction Expansion Contraction Expansion Contraction
0 5 6 3 4 1 1
-3 6 6 5 3 1 1
-6 5 5 2 3 1 1
-12 5 4 2 3 1 2
Dataset Commodities Currencies

Lag (No. of months) Expansion Contraction Expansion Contraction
0 2 2 1 1
-3 2 2 1 0
-6 2 2 1 1
-12 2 2 1 0

References
Alessi, L., Barigozzi, M., and Capasso, M. (2009). A robust criterion for determining
the number of factors in approximate factor models. Technical report.
Audrino, F., Barone-Adesi, G., and Mira, A. (2005). The stability of factor models of
interest rates.
Journal of Financial Econometrics, 3(3):422441.
Bai, J. and Ng, S. (2002). Determining the number of factors in approximate factor
models. Econometrica, 70(1):191221.
Bai, J. and Ng, S. (2008). Large dimensional factor analysis. Foundations and Trends in Econometrics,
3(2):89163.
Bernanke, B., Boivin, J., and Eliasz, P. S. (2005). Measuring the effects of monetary
policy: A factor- augmented vector autoregressive (favar) approach. The Quarterly
Journal of Economics, 120(1):387 422.
Bliss, R. R. (1997). Movements in the term structure of interest rates. Economic Review,
(Q 4):1633.
Chamberlain, G. and Rothschild, M. (1983). Arbitrage, factor structure, and mean-

variance analysis on large asset markets. Econometrica, 51(5):1281304.
Chantziara, T. and Skiadopoulos, G. S. (2008). Can the dynamics of the term
structure of petroleum futures be forecasted? E vidence from major markets. Energy
Economics, 30(3):962985.
Connor, G. and Korajczyk, R. A. (1993). A test for the number of factors in an
approximate factor model. Journal of Finance, 48(4):126391.
Dai, Q. and Singleton, K. J. (2000). Specification analysis of affine term structure
models. The Journal of Finance, LV(5):19431979.
Daskalaki, C., Kostakis, A., and Skiadopoulos, G. (2012). Are there common factors
in commodity futures returns.
Daskalaki, C. and Skiadopoulos, G. (2011). Should investors include commodities in
their portfolios after all? N ew evidence. Journal of Banking and Finance, 35(10):2606
2626.
Donald, S. G. (1997). Inference concerning the number of factors in a multivariate
nonparametric relationship. Econometrica, 65(1):103132.
El Karoui, N. (2006). Tracy-widom limit for the largest eigenvalue of a large class tracy-
widom limit for the largest eigenvalue of a large class of complex wishart matrices.
Annals of Probability, 35(2).
Favero, C. A., Marcellino, M., and Neglia, F. (2005). Principal components at work:
the empirical analysis of monetary policy with large data sets. Journal of Applied
Econometrics, 20(5):603620.

Forni, M., Giannone, D., Lippi, M., and Reichlin, L. (2007). Opening the black box -
structural factor models with large gross-sections. Working Paper Series 712,
European Central Bank.
Forni, M., Lippi, M., and Reichlin, L. (2003). Opening the black box: Structural
factor models versus structural vars. CEPR Discussion Papers 4133, C.E.P.R.
Discussion Papers.
Forni, M. and Reichlin, L. (1998). Lets get real: a factor analytical approach to
disaggregated business cycle dynamics. ULB Institutional Repository 2013/10147,
ULB Universite Libre de Bruxelles.
Goff, J., editor (2003). Economic Letter - What Makes the Yield Curve Move?, number 2003-15.
Federal Reserve Bank of San Francisco,
Gorton, G. and Rouwenhorst, K. G. (2004). Facts and fantasies about commodity

futures. NBER Working Papers 10595, National Bureau of Economic Research, Inc.
Hallin, M. and Liska, R. (2007). Determining the number of factors in the general
dynamic factor model.
Journal of the American Statistical Association, 102:603617.
Kapetanios, G. (2005). A testing procedure for determining the number of factors in

approximate factor models with large datasets.
Kat, H. M. and Oomen, R. C. A. (2006). What every investor should know about
commodities, part ii: Multivariate return analysis. Technical Report 33, Cass Business
School.
Lewbel, A. (1991). The rank of demand systems: Theory and nonparametric
estimation. Econometrica, 59(3):71130.
Litterman, R. and Sheinkman, J. (1991). Common factors affecting bond returns. The Journal of
Fixed Income, pages 5462.
Merville, L. J., Hayes-Yelken, S., and Xu, Y. (2001). Identifying the factor structure
of equity returns.
Journal of Portfolio Management, 27(4):5161.
National Bureau of Economic Research (2012). US business cycle expansions and
contractions. Onatski, A. (2009a). Determining the number of factors from empirical
distribution of eigenvalues. To
appear in the Review of Economics and Statistics.
Onatski, A. (2009b). Testing hypotheses about the number of factors in large factor
models. Economet- rica, 77(5):14471479.
Prignon, C. and Villa, C. (2002). Permanent and transitory factors affecting the
dynamics of the term structure of interest rates. Technical report, International Center
for Financial Asset Management and Engineering.
Philip, D., Kao, C., and Urga, G. (2007). Testing for instability in factor structure
of yield curves. Technical report, Cass Business School.

WP-009
Social Responsibility
and Mean-Variance
Portfolio Selection
Bastien Drut, PhD, Fixed Income
and Forex Strategist Amundi
April 2010, Revised: January 2012
In theory, investors choosing to invest only in socially responsible

entities restrict their investment universe and should thus be
penalized in a mean-variance framework. When computed, this
penalty is usually viewed as valid for all socially responsible investors.
This paper shows however that the additional cost for responsible
investing depends essentially on the investors risk aversion. Social
ratings are introduced in mean-variance optimization through
linear constraints to explore the implications of considering a
social responsibility (SR) threshold in the traditional Markowitz
(1952) portfolio selection setting. We consider optimal portfolios
both with and without a risk-free asset. The SR-efficient frontier
may take four different forms depending on the level of the SR
threshold: a) identical to the non-SR frontier (i.e. no cost), b) only the
left portion is penalized (i.e. a cost for high-risk-aversion investors
only), c) only the right portion is penalized (i.e. a cost for low-risk
aversion investors only) and d) the whole frontier is penalized (i.e.
a positive cost for all the investors). By precisely delineating under
which circumstances SRI is costly, those results help elucidate the
apparent contradiction found in the literature about whether or not
SRI harms diversification.

1. Introduction
In Markowitzs (1952) setting, portfolio selection is driven solely by financial parameters and
the investors risk aversion. This framework may however be viewed as too restrictive since,
in the scope of Socially Responsible Investment (SRI)1, investors also consider non-financial
criteria. This paper explores the impact of such SRI concerns on mean-variance portfolio
selection.
SRI has recently gained momentum. In 2007, its market share reached 11% of assets under
management in the United States and 17.6% in Europe.2 Moreover, by May 2009, 538 asset
owners and investment managers, representing $18 trillion of assets under management, had
signed the Principles for Responsible Investment (PRI)3. Within the SRI industry, initiatives
are burgeoning and patterns are evolving rapidly.
In practice, SRI takes various forms. Negative screening consists in excluding assets on
ethical grounds (often related to religious beliefs), while positive screening selects the best-
SR rated assets (typically, by combining environmental, social, and governance ratings).
Renneboog et al. (2008) describe negative screening as the first generation of SRI, and
positive screening as the second generation. The third generation combines both screenings,
while the fourth adds shareholder activism.
SRI financial performances are a fundamental issue. Does SRI perform as well as
conventional investments? In other words, is doing good also doing well? A large body of
empirical literature is devoted to the comparison between SR and non-SR funds. According to
Renneboog et al. (2008), there is little evidence that the performances of SR funds differ
1
SRI is defined by the European Sustainable Investment Forum (2008) as a generic term covering ethical
investments, responsible investments, sustainable investments, and any other investment process that combines
investors financial objectives with their concerns about environmental, social and governance (ESG) issues.
2
More precisely, the Social Investment Forum (2007) assessed that 11% of the assets under management in the
United States, that is $2.71 trillion, were invested in SRI, and according to the European Sustainable Investment
Forum (2008), this share was 17.6% in Europe.
3
The PRI is an investor initiative in partnership with the UNEP Finance Initiative and the UN Global Compact.
The six principles for responsible investment advocate deep consideration for ESG criteria in the investment
process (see PRI, 2009)

significantly from their non-SR counterparts. Conversely, Geczy et al. (2006) find that
restricting the investment universe to SRI funds can seriously harm diversification. Taken at
face value, those statements seem hard to reconcile.
Within Markowitzs (1952) mean-variance theoretical framework, negative screening implies
that the SR efficient frontier and the capital market line will be dominated by their non-SR
counterparts because asset exclusion restricts the investment universe. Farmen and Van Der
Wijst (2005) notice that, in this case, risk aversion matters in the cost of investing responsibly.
Positive screening corresponds to preferential investment in well-rated SRI assets without
prior exclusion, with each investor being allowed to choose her own SR commitment (Landier
and Nair, 2009). This translates into a trade-off between financial efficiency and portfolio
ethicalness (Beal et al., 2005). Likewise, Dorfleitner et al. (2009) propose a theory of mean-
variance optimization including stochastic social returns within the investors utility function.
However, to our knowledge, easily implementable mean-variance portfolio selection for
second-generation SRI is still missing from the literature. Moreover, the impact of risk
aversion on the cost of SRI has not been investigated so far. Our paper aims at filling those
two gaps. By delineating the conditions under which SRI is costly, it will furthermore help
elucidate the apparent contradiction found in the literature regarding SRIs influence on
diversification.
This paper measures the trade-off between financial efficiency and SRI in the traditional
mean-variance optimization. We compare the optimal portfolios of an SR-insensitive investor
and her SR-sensitive counterpart in order to assess the cost associated with SRI. Our
contribution is twofold. First, we extend the Markowitz (1952) model4 by imposing an SR
threshold. This leads to four possible SR-efficient frontiers: a) the SR-frontier is the same as
the non-SR frontier (i.e. no cost), b) only the left portion is penalized (i.e. a cost for high-risk-
4
See Steinbach (2001) for a literature review on the extensions of the Markowitz (1952) model.

aversion investors only), c) only the right portion is penalized (i.e. a cost for low-risk aversion
investors only), and d) the full frontier is penalized (i.e. a cost for all investors). Despite its
crucial importance, practitioners tend to leave the investors risk aversion out of the SRI story.
Our paper on the other hand offers a fully operational mean-variance framework for SR
portfolio management, a framework that can be used for all asset classes (stocks, bonds,
commodities, mutual funds, etc.). It makes explicit the consequences of any given SR
threshold on the determination of the optimal portfolio. To illustrate this, we complement our
theoretical approach by an empirical application to emerging bond portfolios.
The rest of the paper is organized as follows. Section 2 proposes the theoretical framework for
the SR mean-variance optimization in the presence of risky assets only. Section 3 adds a risk-
free asset. Section 4 applies the SRI methodology to emerging sovereign bond portfolios.
Section 5 concludes.
2. SRI portfolio selection (risky assets)
In this section, we explore the impact of considering responsible ratings in the mean-variance
portfolio selection. To do so, we first assess the social responsibility of the optimal portfolios
resulting from the traditional optimization of Markowitz (1952). Then we consider the case of
an SR-sensitive investor who wants her portfolio to respect high SR standards, and we
explore the consequences of such a constraint for optimal portfolios.
Consider a financial market composed of n risky securities5 (i 1, ... , n) . Let us denote by
P >P1 , ..., P n @' the vector of expected returns and by V

ij the n u n positive-definite
covariance matrix of the returns. A portfolio p is characterized by its composition, that is its
associated vector Z p >Z p1 Z p 2 ... Z pn @' , where Z pi is the weight of the i th asset in
portfolio p , L >1 ... 1@' and Z p 'L 1.
5
Notations of Lo (2008) are used here.

In the traditional mean-variance portfolio selection (Markowitz, 1952), the investor
maximizes her portfolios expected return P p Zp 'P for a given volatility or
variance V p2 Z p ' 6Z p , Let O ! 0 be the parameter accounting for the investors level of risk
aversion. The problem of the SR-insensitive investor is then written6:
Problem 1
O
max Z ' P Z' Z
^Z ` 2 (1)
subject to Z 'L 1
The solutions to Problem 1 form a hyperbola in the mean-standard deviation plane ( P p , V p )
and will be referred to here as the SR-insensitive efficient frontier.
Let us now add an SR rating independent from expected returns and volatilities. Typically,
this is an extra-financial rating relating to environmental, social, or governance issues. It can
also combine several ratings (Landier and Nair, 2009). Let Ii be the SR rating associated with
the i-th security and I >I1 I 2 ... I n @' . We assume that the rating is additive.
Consequently, the rating I p of portfolio p is given by:
n
Ip Z p 'I Z I
pi i (2)
i 1
This linearity hypothesis (see Barracchini, 2007; Drut, 2009; Scholtens, 2009) is often used
by practitioners to SR-rate financial indices7. The representation in eq. (2) holds for positive
as well as negative screening8.
6
For sake of simplicity, short sales are allowed here.
7
See for instance the Carbon Efficient Index of Standard & Poors with the carbon footprint data from Trucost
PLC.
8
For negative screening, Ii is binary and I p denotes the proportion of portfolio p invested in the admissible
assets.

Even when investors are SR-insensitive (thus facing problem 1), their optimal portfolios can
be SR-rated. Proposition 1 expresses those ratings I p associated with SR-insensitive efficient
portfolios.
Proposition 1
(i) Along the SR-insensitive frontier, the SR rating I p is a linear function of the
expected return P p :
Ip G 0 G1 P p (3)
( P ' 1 P )(L ' 1 I ) (L ' 1 P )( P ' 1 I )

with G0
(L ' 1 L )( P ' 1 P ) (L ' 1 P ) 2
( P ' 1 I )(L ' 1 L ) (L ' 1 P )(L ' 1 I )

and G1 .
(L ' 1 L )( P ' 1 P ) (L ' 1 P ) 2
L ' 1 I
(ii) If G 1 ! 0 , I p ranges from (for the minimum-variance portfolio) to f
L ' 1 L
(when the expected return tends to the infinite).
L ' 1 I
(iii) If G 1 0 , I p ranges from (for the minimum-variance portfolio) to f
L ' 1 L
(when the expected return tends to the infinite).
Proof: see Appendix 1
Thus, Proposition 1 gives the SR rating I p of any portfolio lying on the SR-insensitive
frontier. From the optimality conditions comes the fact that, along the efficient frontier, both
1
the SR rating I p and the expected return P p are linear functions of the quantity , so it is
O
straightforward that the SR rating I p can be written as a linear function of the expected
return P p , as in eq. (1). The direction of this link is determined by the sign of the
parameter G 1 . The parameter G 1 can take both signs because, for instance, the assets with the
highest returns can be the best or the worst SR-rated. Furthermore, the sign of the parameter
G 1 is crucial because it represents where the trade-off appears between risk aversion and SR

rating. If G 1 ! 0 , resp. G 1 0 , the riskier the optimal portfolio, the better, resp. the worse, its
SR rating. In other words, if G 1 ! 0 , resp. G 1 0 , the best SR-rated portfolios are at the top,
resp. at the bottom, of the SR-insensitive frontier.
Consider now the case of an SR-sensitive investor. For instance, she requires a portfolio that
is well-rated for environment. Henceforth, the SR rating I p is introduced in the mean-
variance optimization by means of an additional linear constraint imposing a given
threshold I 0 on I p . For positive screening, the threshold value is left to the investors
discretion (Beal et al., 2005, Landier and Nair, 2009). The SR-sensitive optimization is
summarized by Problem 2.
Problem 2
O
max Z ' P Z' Z
^Z ` 2
subject to Z 'L 1 (4)
Ip Z 'I t I 0
We derive the analytical solutions to this problem by following Best and Grauers (1990)
methodology. Proposition 2 summarizes the results.

Proposition 2
The shape of the SR-sensitive efficient frontier depends on the sign of G 1 and on the threshold
value I 0 in the following way:
G1 0 G1 ! 0
For O O0 , the SR-sensitive

frontier is another hyperbola lying
L ' 1 I below the SR-insensitive frontier. The SR-sensitive frontier is
! I0
L ' 1 L identical to the SR-insensitive
For O ! O0 , the SR-sensitive frontier
frontier is identical to the SR-
insensitive frontier.
For O O0 , the SR-sensitive
frontier is identical to the SR-
insensitive frontier.
L ' 1 I The SR frontier differs totally
I0 from the SR-insensitive frontier. For O ! O0 , the SR-sensitive
L ' 1 L
frontier is another hyperbola
lying below the SR-insensitive
frontier
( P ' 6 1L )(I ' 6 1L ) (L ' 6 1L )(I ' 6 1 P )

With O0 .
(I 'I 0L ' )6 1L
The associated expected return E 0 and the expected variance V0 are:
1 1
E0 ( P ' 6 1L (( P ' 6 1 P )(L ' 6 1L ) ( P ' 6 1L ) 2 )
L' 6 L
1
O0
1 1
V0 V 02 (1 2 (( P ' 6 1 P )(L ' 6 1L ) ( P ' 6 1L ) 2 )
L' 6 L
1
O0
Proof: see Appendix 2.
Proposition 2 makes explicit the situations in which there is an SRI cost. The impact of the
constraint on the SR ratings depends on the parameter G 1 and on the strength of the constraint.
As showed in Proposition 1, if G 1 ! 0 , resp. G 1 0 , the best SR-rated portfolios are at the top,
resp. at the bottom, of the SR-insensitive frontier: by consequence, the SR constraint impacts

first the efficient frontier at the bottom, resp. at the top. In addition, the more the investor
wants a well-rated portfolio, that is to say the higher the threshold I 0 , the bigger the portion of
the efficient frontier being displaced. In the case where the threshold I 0 is below the minimum
rating of the SR-insensitive frontier, the efficient frontier is even not modified at all. We
illustrate the four possible cases through Figures 1 to 4.
L ' 1 I
In the case where G 1 ! 0 and ! I 0 (see Figure 1), the SR-sensitive and the SR-
L ' 1 L
insensitive frontiers are the same and there is no SRI cost at all. This is the most favourable
case.
L ' 1 I
Figure 1 SR-sensitive frontier versus SR-insensitive frontier with G 1 ! 0 and ! I0
L ' 1 L
Mean
Standard Deviation
SR-insensitive frontier SR-sensitive frontier
L ' 1 I
In the case where G 1 ! 0 and I 0 (see Figure 2), the SR-sensitive and the SR-
L ' 1 L
insensitive frontiers are the same above the corner portfolio defined by its expected return
E 0 and its expected variance V0 V 02 . For portfolios with lower expected returns and
variances, the SRI constraint induces less efficient portfolios. There is only an SRI cost for
investors whose risk aversion parameter is above the threshold O0 .

L ' 1 I
Figure 2 SR-sensitive frontier versus SR-insensitive frontier with G 1 ! 0 and I0
L ' 1 L
E0
Mean
0
Standard Deviation
L ' 1 I
In the case where G 1 0 and ! I 0 (see Figure 3), the SR-sensitive and the SR-
L ' 1 L
insensitive frontiers are the same below the corner portfolio ( E 0 , V0 ) . For portfolios with
higher expected returns and variances, the SRI constraint induces less efficient portfolios.
There is only an SRI cost for investors whose risk aversion parameter is below the
threshold O0 .
L ' 1 I
Figure 3 SR-sensitive frontier versus SR-insensitive frontier with G 1 0 and ! I0
L ' 1 L
E0
Mean
0
Standard Deviation

L ' 1 I
In the case where G 1 0 and I 0 (see Figure 4), the SR-sensitive and the SR-
L ' 1 L
insensitive frontiers are totally different. The SRI constraint induces less efficient portfolios
for every investor. There is an SRI cost for everybody. This is the worst case.
L ' 1 I
Figure 4 SR-sensitive frontier versus SR-insensitive frontier with G 1 0 and I0
L ' 1 L
Mean
Standard Deviation
To sum up, while the investors risk aversion is generally left out of the story in the SRI
practice, we show in this Section that this parameter matters in the cost of responsible
investing9. Indeed, we show that this SR cost depends on the link between SR ratings and
financial returns and on the investors risk aversion; four cases being possible: a) the SR-
sensitive frontier is the same as the SR-insensitive frontier (i.e. no cost), b) only the left
portion of the efficient frontier is penalized (i.e. a cost for high-risk-aversion investors only),
c) only the right portion of the efficient frontier is penalized (i.e. a cost for low-risk aversion
investors only), and d) the full frontier is penalized (i.e. a cost for all the investors).
9
This section highlights the impact of a constraint on the portfolio rating in the mean-variance optimization.
However, the cost of investing responsibly, if non-zero, may be non-significant. The significance of the mean-
variance efficiency loss may be assessed using the test of Basak et al. (2002) or any spanning test (see de Roon
and Nijman (2001) for a literature review).

3. Portfolio selection with a risk-free asset
In this section, we assume the existence of a risk-free asset and we explore, in this case, the
impact of considering responsible ratings in the mean-variance portfolio selection. Indeed, the
social responsibility of this risk-free asset should also be taken into account. So, we assess
first the social responsibility of the optimal portfolios obtained by an SR-insensitive investor.
And then we study whether an SR-sensitive investor is penalized by requiring portfolios with
high SR standards.
Denote by r the return of the risk-free asset and by Zr the fraction of wealth invested in this
risk-free asset. The standard mean-variance portfolio selection in the presence of a risk-free
asset has been extensively studied by Lintner (1965) and Sharpe (1965). It corresponds to
Problem 3.
Problem 3
The investor wants to solve the following program:
O
max Z ' P Z r r Z ' Z
^Z ` 2 (5)
subject to Z 'L Z r 1
In the mean-standard deviation plan, the set of optimal portfolios is referred as the well-
known Capital Market Line (CML). As the investor does not consider responsible ratings in
her optimization, we refer it here as SR-insensitive capital market line.
As in Section 2, we add responsible ratings to the story. Henceforth, we denote I * as the
responsible rating of the risk-free asset and the portfolio rating is defined as I p Z 'I Z r I * .
In the following, we seek to determine the portfolio ratings I p of the optimal portfolios on the
SR-insensitive capital market line.

Proposition 3
(i) Along the SR-insensitive capital market line, the responsible rating I p is a linear
function of the expected return P p :
Ip G 0* G 1* P p (6)
(I I *L )' 1 ( P rL )
with G 0* I * r
( P rL )' 1 ( P rL )
(I I *L )' 1 ( P rL )
and G 1*
( P rL )' 1 ( P rL )
(ii) If G 1* ! 0 , I p ranges from I * (for the minimum-variance portfolio) to f (when
the expected return tends to the infinite.)
(iii) If G 1* 0 , I p ranges from I * (for the minimum-variance portfolio) to f (when
the expected return tends to the infinite).
Proposition 3 attributes an SR rating of any portfolio of the SR-insensitive capital market line.
From the optimality conditions comes the fact that, along the capital market line, both the SR
1
rating I p and the expected return P p are linear functions of the quantity It is therefore
O
straightforward that the SR rating I p can be written as a linear function of the expected
return P p as in eq. (6). It is striking that this relationship expressed by eq. (6) has the same
form as eq. (3) in the case without a risk-free asset. Note that the portfolio of an infinitely risk
averse investor would be fully invested in the risk-free asset and would have its SR rating I * .
Here, the direction of this link is determined by the sign of the parameter G 1* . In the same way
as in Section 2, the sign of the parameter G 1* is crucial because it represents where the trade-
off appears between risk aversion and SR rating. If G 1* ! 0 , resp. G 1* 0 , the riskier the optimal

portfolio, the better, resp. the worse, its SR rating. In other words, if G 1* ! 0 , resp. G 1* 0 , the
best SR-rated portfolios are at the top, resp. at the bottom, of the SR-insensitive capital market
line.
Similarly to Section 2, we now consider the case of SR investors wishing high SR standards
and so, requiring the portfolio rating I p Z 'I Z r I * to be above a threshold I0 . This
corresponds to Problem 4.
Problem 4
The investor wants to solve the following program:
O
max Z ' P Z r r Z ' Z
^Z ` 2
subject to Z 'L Z r 1 (7)
Ip Z 'I Z rI * t I0
In Problem 4, the constraints in the mean-variance optimization are also linear. Thus, we
employ Best and Grauers (1990) methodology, as we did for Problem 2. Proposition 4
summarizes the results.

Proposition 4
The shape of the SR-sensitive capital market line depends on the sign of G 1* and on the
threshold value I 0 in the following way:
G 1* 0 G 1* ! 0
For O O*0 , the SR-sensitive

capital market line is a hyperbola
I* ! I0 capital market line. The SR-sensitive capital market
line is the same as the SR-
For O ! O*0 , the SR-sensitive insensitive capital market line
capital market line is identical to
the SR-insensitive capital market
line.
For O O*0 , the SR-sensitive
capital market line is identical to
the SR-insensitive capital market
The SR-sensitive capital market
I* I0 line.
line differs totally from the SR-
insensitive capital market line and
becomes a hyperbola. For O ! O*0 , the SR-sensitive
capital market line is a hyperbola
capital market line.
( P rL )' 6 1 (I I *L )
With O*0
(I0 I *)
The associated expected return E0* and the expected variance V0* are:
1
E 0* r ( P rL )' 1 ( P rL )
O*0
1
V0* V 02* 2
( P rL )' 1 ( P rL )
O*0
Proposition 4 makes explicit the situations in which there is an SRI cost in the presence of a
risk-free asset. As in Proposition 2, the impact of the constraint on the SR ratings depends on

the parameter G 1* and on the strength of the constraint. As showed in Proposition 3, if G 1* ! 0 ,
resp. G 1* 0 , the best SR-rated portfolios are at the top, resp. at the bottom, of the SR-
insensitive capital market line: by consequence, the SR constraint impacts first the capital
market line at the bottom, resp. at the top. However, contrary to the case without a risk-free
asset, the modified part of the capital market line has a different mathematical form: for this
segment, the capital market line becomes a hyperbola in the mean-standard deviation plan.
Figures 5 to 8 illustrate the four cases.
In the case where G 1* ! 0 and I * ! I0 (see Figure 5), the SR-sensitive and the SR-insensitive
capital market lines are the same and there is no SRI cost at all. This is the best case.
Figure 5 SR-sensitive capital market line versus SR-insensitive capital market line with
G 1* ! 0 and I * ! I0
Mean
Standard Deviation
SR-insensitive Frontier SR-insensitive Capital M arket Line

SR-sensitive Capital M arket Line
In the case where G 1* ! 0 and I * I0 (see Figure 6), the SR-sensitive and the SR-insensitive
capital market lines are the same for portfolios below the corner portfolio defined by its
expected return E0* and the expected variance V0* V 02* . Below this portfolio, the SR-
sensitive capital market line becomes a hyperbola. There is an SRI cost only for investors
with cold feet, that is to say with a risk aversion parameter above the threshold O*0 .

G 1* ! 0 and I * I0
Mean
E0*
0*
Standard Deviation

In the case where G 1* 0 and I * ! I 0 (see Figure 7), the SR-sensitive and the SR-insensitive
capital market lines are the same for portfolios above the corner portfolio ( E 0* , V0* ) . Above
this portfolio, the SR-sensitive capital market line becomes a hyperbola. There is an SRI cost
only for investors with a risk aversion parameter below the threshold O*0 .
G 1* 0 and I * ! I 0
Mean
E0*
0*
Standard Deviation

In the case where G 1* 0 and I * I0 (see Figure 8), the SR-sensitive and the SR-insensitive
capital market lines differ entirely. The SR-sensitive capital market line is no longer a line but
a hyperbola. This is the most disadvantageous case: there is an SRI cost for all the investors.

G 1* 0 and I * I0
Mean
Standard Deviation

In the mean-variance portfolio selection in the presence of a risk-free asset also, the investors
risk aversion matters in the SRI cost. In order to be ready-to-use for practitioners, we make
explicit the four cases are possible: a) the SR-capital market line is the same as the SR-
insensitive capital market line (i.e. no cost), b) only the left portion is penalized (i.e. a cost for
high-risk-aversion investors only), c) only the right portion is penalized (i.e. a cost for low-
risk aversion investors only), and d) the full capital market line is penalized (i.e. a cost for all
the investors).
4. Application to an emerging bond portfolio
This section illustrates the results of Sections 2 and 3 by considering the case of a responsible
US investor on the emerging bond market.
We consider the EMBI+ indices from JP Morgan as proxy for emerging bond returns. These
indices track total returns for actively traded external debt instruments in emerging markets.10
The indices are expressed in US dollars and taken at a monthly frequency from January 1994
10
Argentina, Brazil, Bulgaria, Ecuador, Mexico, Panama, Peru, Philippines, Russia, Venezuela.

to October 2009. They are extracted from Datastream. Descriptive statistics are available in
Appendix 5.
In the same way as Scholtens (2009), we use the Environmental Performance Index (EPI) as
responsible ratings. The EPI is provided jointly by the universities of Yale and Columbia in
collaboration with the World Economic Forum and the Joint Research Centre of the European
Commission. EPI focuses on two overarching environmental objectives: reducing
environmental stress to human health and promoting ecosystem vitality and sound
management of natural resources. These objectives are gauged using 25 performance
indicators tracked in six well-established policy categories, which are then combined to create
a final score. EPI scores attributed in 2008 are reported in Table 3.
Table 3 Environmental Performance Index 2008
ARGENTINA 81.78
BRAZIL 82.65
BULGARIA 78.47
ECUADOR 84.36
MEXICO 79.80
PANAMA 83.06
PERU 78.08
PHILIPPINES 77.94
RUSSIA 83.85
VENEZUELA 80.05
Mean 81.00
Standard Deviation 2.44
UNITED STATES 81.03
Sources: Universities of Yale and Columbia.
Here, the portfolio EPI is defined in the same way as in eq. (2). We start by estimating the
portfolio EPI along the SR-insensitive frontier, which corresponds to estimating the
relationship (3) of Proposition 1. We obtain the following estimates for the parameters G 0
and G1 :
G0 76.00 G1 0.30

As G1 ! 0 , the portfolio EPI increases with the expected return on the SR-insensitive efficient
frontier: a 1%/year increase in expected returns corresponds to an increase of 0.30 in the EPI
portfolio. The minimal EPI portfolio on the SR-insensitive frontier is obtained for the
L ' 1 I
minimum-variance portfolio and is equal to 78.26 .
L ' 1 L
As an illustration of Problem 2, we seek to determine the impact of SR attempts on the
efficient frontier and we impose a set of constraints I p ! I 0 on the portfolio EPI. Figure 9
exhibits the SR-sensitive frontiers for several thresholds I 0 . The corner portfolios for which
there is a disconnect between the SR-sensitive and SR-insensitive frontiers are in Table 4.
Figure 9 SR-sensitive frontiers versus SR-insensitive frontier for the EMBI+ indices,
January 1994 to October 2009
30.00%
25.00%
Ann. Mean (%/year)
20.00%
15.00%
10.00%
5.00%
0.00%
0.00% 10.00% 20.00% 30.00% 40.00% 50.00%
Ann. Standard Deviation (%/year)
SR-insensitive Frontier SR 80
SR 82 SR 84
SR 86 SR 88

Table 4 Corner portfolios for which the SR constraint is binding
I0 O0 Expected return E 0 Expected volatility V 0
(%/year) (%/year)
82 3.14 20.13 22.22
84 2.05 26.84 32.17
86 1.52 33.55 42.48
88 1.21 40.26 52.93
L ' 1 I
As expected from Proposition 2, for I 0 78.26 , the SR-sensitive frontier is the same
L ' 1 L
L ' 1 I
as the SR-insensitive frontier. For I0 ! 78.26 , the SR-sensitive frontier differs from
L ' 1 L
the SR-insensitive frontier at the bottom and is the same at the top. For instance, with a
threshold equal to 84 on the portfolio EPI, the SR-sensitive and the SR-insensitive frontiers
are the same for expected returns above 26.84%/year and differ for expected returns below
26.84%/year. In the case of emerging bonds, improving the portfolio EPI costs more for
investors with high risk aversion.
In order to illustrate Problems 3 and 4, we rely on the US 1-month interbank rate as a risk-free
asset.11 Its responsible rating corresponds to the EPI of the United States I * 81.03 . Then, we
estimate the parameters G 0* and G 1* :
G0* 80.95 G1* 0.02
As G1* ! 0 , the portfolio EPI increases with the expected return of the SR-insensitive capital
market line. According to the estimations, for a 1%/year increase in expected returns, the EPI
portfolio is 0.02 higher. The minimal EPI portfolio on the SR-insensitive frontier is obtained
11
This variable is extracted from Datastream.

for the minimum-variance portfolio and is equal to I * 81.03 . The SR-insensitive frontier is
shown in Figure 10.
Figure10 SR-insensitive capital market line for the EMBI+ indices,
January 1994 to October 2009
30%
25%
Ann. Mean (%/year)
20%
15%
10%
5%
0%
0% 10% 20% 30% 40% 50%
Non-SR Frontier Non-SR Capital Market Line
From now on, we consider SR investors wishing to adopt high environmental standards: thus,
we impose a set of constraints I p ! I 0 in the same way as in Problem 4. Figure 11 exhibits the
SR-sensitive capital market lines for several thresholds I 0 , and Table 5 displays the corner
portfolios.

Figure 11 SR-sensitive capital market lines versus SR-insensitive capital market lines for the
EMBI+ indices, January 1994 to October 2009
30%
25%
Ann. Mean (%/year)

20%
15%
10%
5%
0%
0% 10% 20% 30% 40% 50%
SR-insensitive CML SR 82 SR 84 SR 86 SR 88
Table 5 Corner portfolios for which the SR constraint is binding
I0 O0 Expected return E 0 Expected volatility V 0*
(%/year) (%/year)
82 1.16 50.01 63.06
84 0.38 144.97 193.08
86 0.23 239.93 323.11
88 0.16 334.89 453.13
As expected, for I 0 I * 81.03 , the SR-sensitive capital market line is the same as the SR-
insensitive capital market line. For I0 ! I * 81.03 , the SR-sensitive capital market line
differs from the SR-insensitive one at the bottom and is the same at the top. Here also, the SRI
cost appears for investors with high risk aversion. We notice that the corner portfolios have
particularly high expected returns and volatilities (see Table 5): this can be explained by the

particularly low sensitivity G1* . For example, if we consider a threshold I 0 82 , the corner
portfolio has an expected return of 50.01%/year and an expected volatility of 63.06%/year,
meaning that for expected returns below 50.01%/year, the SR-sensitive and SR-insensitive
capital market lines are disconnected. However, we observe in Figure 11 that the SR-
insensitive and SR-sensitive capital market lines are very close for expected returns slightly
below 50.01%/year.
This numerical application highlights that the cost implied by high environmental
requirements in an emerging bond portfolio differs according to the investors risk aversion.
In this particular case, it costs more to be green for investors with cold feet. Let us now focus
on a typical investor. Sharpe (2007) suggests that the representative investor has a risk
2
aversion parameter O in the traditional mean-variance optimization of eq. (1). We seek
0.7
to determine the consequences of SR thresholds for this representative investor by
computing the optimal portfolios for different thresholds on the portfolio EPI. Figure 12
displays these portfolios (the means and variances of the optimal portfolios are given in
Appendix 6).
Figure 12 Displaced optimal portfolios for the representative investor
30.00%
25.00%
Ann. Mean (%/year)
20.00%
15.00%
10.00%
5.00%
0.00%
0.00% 10.00% 20.00% 30.00% 40.00% 50.00%
SR-insensitive Frontier SR 82
SR 84 SR 86
SR 88 SR 90

or the representative investor, the constraint on the portfolio EPI has no cost while the
threshold is below 82.37, which is slightly above the average EPI rating of the samples
countries. When the threshold is above 82.37, an SRI cost appears and the optimal portfolio is
no longer on the SR-insensitive frontier. This SRI cost rises with the strength of the
constraint. In this case, the representative investor is directly concerned by the disconnect
between SR-sensitive and SR-insensitive frontiers for reasonable SR thresholds.
5. Conclusion
The rapid growth of the SRI fund market has given birth to a burgeoning academic literature.
Most academic studies show that there is little difference between financial returns of SRI
funds and conventional funds. However, Geczy et al. (2006) highlight that limiting the
investment universe to SRI funds can seriously harm diversification. To shed light on this
debate, our paper aimed at modelling SRI in the traditional mean-variance portfolio selection
framework (Markowitz, 1952).
In our study, SRI is introduced in the mean-variance optimization as a constraint on the
average responsible rating of the underlying entities. Our results are detailed so that they are
ready to use by practitioners. Indeed, we show that a threshold on the responsible rating may
impact the efficient frontier in four different ways, depending on the link between the returns
and the responsible ratings and on the strength of the constraint. The SR-sensitive efficient
frontier can be: a) identical to the SR-insensitive efficient frontier (i.e. no cost at all), b)
penalized at the bottom only (i.e. a cost for high risk-aversion investors only), c) penalized at
the top only (i.e. a cost for low risk-aversion investors only), d) totally different from the SR-
insensitive efficient frontier (i.e. a cost for every investor). In other words, if portfolio ratings
increase (resp. decreases) with the expected return along the traditional efficient frontier, the
SRI cost arises first at the bottom (resp. at the top) of the frontier. The results are the same in

the presence of a risk-free asset. Our work highlights the fact that the investors risk aversion
clearly matters in the potential cost of investing responsibly, this cost being zero in some
cases. We strongly believe that this finding could help portfolio managers of SRI funds.
As the calculations in our paper are very general, we believe it could find other applications in
the asset management industry, notably for portfolio selection with asset liquidity constraints.
However, one limitation of our study is that it assumes expected returns to be independent
from responsible ratings: further research could focus on modelling the impact of an SRI
constraint in the mean-variance optimization when expected returns and volatilities depend on
responsible ratings.

Acknowledgements
The author thanks Marie Brire, Benjamin Lorent, Valrie Mignon, Kim Oosterlinck, Hugues
Pirotte, Ombretta Signori, Ariane Szafarz and Kokou Topeglo for their helpful comments.
References
Barracchini, C., 2007. An ethical investments evaluation for portfolio selection., Journal of
Business Ethics and Organization Studies 9, 1239-2685.
Basak, G., Jagannathan, R., Sun, G., 2002. A direct test of mean-variance efficiency of a
portfolio. Journal of Economic Dynamics and Control 26, 1195-1215.
Beal, D., Goyen, M., Philips, P., 2005. Why do we invest ethically? Journal of Investing 14,
66-77.
Best, M., Grauer, R., 1990. The efficient set mathematics when mean-variance problems are
subject to general linear constraints. Journal of Economics and Business 42, 105-120.
de Roon, F., Nijman T., 2001. Testing for mean-variance spanning: a survey. Journal of
Empirical Finance 8, 111-156.
Dorfleitner, G., Leidl, M., Reeder J., 2009. Theory of social returns in portfolio choice with
application to microfinance. Working Paper, University of Regensburg.
Drut, B., 2009. Nice guys with cold feet: the cost of responsible investing in the bond
markets, Working Paper No 09-34, Centre Emile Bernheim, Universit Libre de Bruxelles.
European Sustainable Investment Forum, 2008. European SRI Study. European Sustainable
Investment Forum, Paris. Available at: http://www.eurosif.org/publications/sri_studies .
Farmen, T., Van Der Wijst, N., 2005. A cautionary note on the pricing of ethics. Journal of
Investing 14, 53-56.
Geczy, C., Stambaugh, R., Levin, D., 2006. Investing in socially responsible mutual funds.
Working Paper, Wharton School of the University of Pennsylvania.
Landier, A., Nair V., 2009. Investing For Change. Oxford University Press, Oxford.
Lintner, J., 1965. The valuation of risky assets and the selection of risky investment in stock
portfolios and capital budgets. Review of Economics and Statistics 47, 13-37.
Lo, A., 2008. Hedge Funds: An Analytic Perspective. Princeton University Press, Princeton.
Markowitz, H., 1952. Portfolio selection., Journal of Finance 7, 77-91.

Principles for Responsible Investment, 2009. PRI Annual Report. Principles for Responsible
Investment, New York. Available at:
http://www.unpri.org/files/PRI%20Annual%20Report%2009.pdf .
Renneboog, L., Ter Horst, J., Zhang, C., 2008. Socially responsible investments: institutional
aspects, performance, and investor behaviour. Journal of Banking and Finance 32, 1723-1742.
Scholtens, B., 2009. Measuring sustainability performance of investments; the case of Dutch
bond funds. Paper prepared for the PRI Academic Conference 2009. Available at
http://www.unpri.org/files/Scholtens_PRI2009.pdf .
Social Investment Forum, 2007. Report on socially responsible investing trends in the United
States. Available at:
http://www.socialinvest.org/resources/pubs/documents/FINALExecSummary_2007_SIF_Tre
nds_wlinks.pdf .
Sharpe, W., 1965. Capital asset prices: a theory of market equilibrium under conditions of
risk. Journal of Finance 19, 425-442.
Sharpe, W., 2007. Expected utility asset allocation, Financial Analysts Journal 63, 18-30.
Steinbach, M., 2001. Markowitz revisited: mean-variance models in financial portfolio

analysis. Society for Industrial and Applied Mathematics Review 43, 31-85.

Appendix 1
Following Best and Grauer (1990), in the standard mean-variance case, the weights vector
that solves the problem is:
1 L 1 1 L ' 1 P
Z ( P L )
L' L O
1
L ' 1 L
The corresponding expected return P P stands as:
1 1
PP P 'Z L ' 1 P (( P ' 1 P )(L ' 1 L ) (L ' 1 P ) 2 )
L ' 1 L O
And the expected variance V P2 stands as:
1 1
V P2 Z ' 6Z 1 2 (( P ' 1 P )(L ' 1 L ) (L ' 1 P ) 2 )
L ' 1 L O
1
The corresponding responsible rating I P is also a linear function of :
O
1 1
Ip I 'Z L ' 1 I (( P ' 1 I )(L ' 1 L ) (L ' 1 P )(L ' 1 I )
L ' 1 L O
1
As the expected return is also a linear function of , it is possible to express the responsible
O
rating as a linear function of the expected return:
( P ' 1 P )(L ' 1 I ) (L ' 1 P )( P ' 1 I ) ( P ' 1 I )(L ' 1 L ) (L ' 1 P )(L ' 1 I )
Ip P
(L ' 1 L )( P ' 1 P ) (L ' 1 P ) 2 (L ' 1 L )( P ' 1 P ) (L ' 1 P ) 2
p
As a consequence, it possible to write I p G 0 G1 P p with

( P ' 1 P )(L ' 1 I ) (L ' 1 P )( P ' 1 I )
G0
(L ' 1 L )( P ' 1 P ) (L ' 1 P ) 2
( P ' 1 I )(L ' 1 L ) (L ' 1 P )(L ' 1 I )
G1 .
(L ' 1 L )( P ' 1 P ) (L ' 1 P ) 2

Appendix 2
The mean-variance optimization with the linear constraint on the portfolio rating has the same
portfolio solutions as the standard mean-variance optimization, while the constraint is not
binding. This is verified if the portfolio rating of the efficient portfolio from the traditional
mean-variance optimization is above the threshold I 0 . Two cases have to be distinguished:
where G 1 ! 0 and where G 1 0 .

Case G 1 ! 0
In this case, the portfolio rating increases linearly with the expected return in the traditional
L ' 1 I L ' 1 I
mean-variance optimization. The minimal portfolio rating is . If ! I 0 , the
L' L1
L ' 1 L
constraint on the portfolio rating is inactive and the solutions are the same as in the traditional
L ' 1 I
mean-variance optimization. If I 0 , the constraint on the portfolio rating is binding
L ' 1 L
at the bottom of the efficient frontier and inactive at the top. Indeed, the constraint is binding
for O ! O0 with:
( P ' 6 1L )(I ' 6 1L ) (L ' 6 1L )(I ' 6 1 P )

O0
(I 'I 0L ' )6 1L
In the mean-variance plan, this corner portfolio is the one for which the portfolio rating is I 0 .
To compute the portion of the efficient frontier for which the linear constraint on the portfolio
rating is binding, we apply the results of Best and Grauer (1990):
1
Pp J0 J1
O
1
V p2 J2 J1
O2
With
( P ' 1 I )(L ' 1 I ) (L ' 1 P )(I ' 1 I ) (L ' 1 P )(L ' 1 I ) (L ' 1 L )( P ' 1 I )
J0 I0
(L ' I ) (L ' L )(I ' I )
1 2 1 1
(L ' 1 I ) 2 (L ' 1 L )(I ' 1 I )
(L ' 1 P ) 2 (I ' 1 I ) ( P ' 1 I ) 2 (L ' 1 L ) 2(L ' 1 P )( P ' 1 I )(L ' 1 I )

J1 P ' 6 1 P
(L ' 1 I ) 2 (L ' 1 L )(I ' 1 I )
I ' 1 I 2I 0 (L ' 1 I ) (L ' 1 L )I 02

J2
(L ' 1 I ) 2 (L ' 1 L )(I ' 1 I )

Case G 1 0
In this case, the portfolio rating decreases linearly with the expected return in the traditional
L ' 1 I L ' 1 I
mean-variance optimization. The maximal portfolio rating is . If I 0 , the
L' L1
L ' 1 L
L ' 1 I
constraint on the portfolio rating is binding. If ! I 0 , the constraint on the portfolio
L ' 1 L
rating is binding at the top of the efficient frontier and inactive at the bottom. Indeed, the
constraint is binding for O O0 .
The equation of the portion of the efficient frontier for which the constraint is binding is the
same as for the case G 1 ! 0 .
Appendix 3
Following Best and Grauer (1990), in the traditional mean-variance case, the portfolio
solutions stand as:
1 1
Z ( P rL ) > @
O
1
Zr 1 >L ' 1
( P rL ) @
O
In the mean-variance plan, these portfolios are:

1
P p ZP Z r r r ( P rL )' 1 ( P rL )
> @
O
1
V p2 Z ' 6Z >(P rL )' 1
( P rL ) @
O2
1
The responsible rating of the efficient portfolios is a linear function of :
O
1
Ip I 'Z I * Z r I * >(I I *L )' 1
( P rL ) @
O
1
As the expected return P P is also a linear function of , it is possible to express the
O
responsible rating I P of the efficient portfolios as a linear function of the expected return P P :
Pp r
Ip I 'Z I * Z r I * >(I I *L )' 1
( P rL ) @
( P rL )' 1 ( P rL )
As a consequence, it possible to write I p G 0* G 1* P p with

(I I * L )' 1 ( P rL )
G 0* I * r
( P rL )' 1 ( P rL )
(I I *L )' 1 ( P rL )
G 1*
( P rL )' 1 ( P rL )
Appendix 4
The mean-variance optimization with the linear constraint on the portfolio rating has the same
portfolio solutions as the standard mean-variance optimization while the constraint is not
binding. This is verified if the portfolio rating of the efficient portfolio from the traditional
mean-variance optimization is above the threshold I 0 . Two cases have to be distinguished:
where G 1* ! 0 and where G 1* 0 .
Case G 1* ! 0
In this case, the portfolio rating increases linearly with the expected return in the traditional
mean-variance optimization. The minimal portfolio rating is I * . If I * ! I 0 , the constraint on
the portfolio rating is inactive and the solutions are the same as in the traditional mean-
variance optimization: the Capital Market Line is not modified. If I * I 0 , the constraint on
the portfolio rating is binding at the bottom of the efficient frontier and inactive at the top.
Indeed, the constraint is binding for O ! O*0 with:
(I I *L )' 1 ( P rL )
O*0
I0 I *
The risk aversion parameter O*0 corresponds to the portfolio with the expected return:
1
E 0* r ( P rL )' 1 ( P rL )
O*0
And the expected variance:
1
V0* 2
( P rL )' 1 ( P rL )
O*0
To compute the portion of the efficient frontier for which the linear constraint on the portfolio
rating is binding, we apply the results of Best and Grauer (1990):
(I I *L )' 1 ( P rL ) 1 ((I I *L )' 1 ( P rL )) 2
Pp r (I 0 I *) (( P rL )' 1 ( P rL ) )
(I I * L )' (I I *L )
1
O (I I *L )' 1 (I I * L )

(I 0 I *) 2 1 ((I I * L )' 1 ( P rL )) 2
V p2 2 (( P rL )' 1 ( P rL ) )
(I I *L )' (I I *L ) O
1
(I I *L )' 1 (I I *L )
Indeed, when the constraint on the portfolio rating is binding, the Capital Market Line is no
longer a linear function in the mean-standard deviation plan but a hyperbola.
Case G 1* 0
In this case, the portfolio rating I p decreases linearly with the expected return P p in the
traditional mean-variance optimization. The maximum portfolio rating is I * . If I * I 0 , the

constraint on the portfolio rating is always active: the entire Capital Market Line becomes a
hyperbola in the mean-standard deviation plan. If I * ! I 0 , the constraint on the portfolio
rating is binding at the top of the Capital Market Line. The modification of the Capital Market
Line occurs for the same risk aversion parameter O*0 as for the case G 1* ! 0 .
Appendix 5 Descriptive statistics for the EMBI + indices in US dollars, January 1994 to
October 2009
Ann. Std. Sharpe
Ann. Mean Skewness Kurtosis Maximum Minimum
Dev. Ratio
ARGENTINA 4.94% 28.84% 0.03 -1.07 9.37 33.80% -43.90%
BRAZIL 15.03% 21.40% 0.51 -0.62 8.22 26.47% -27.17%
BULGARIA 15.10% 20.26% 0.55 -0.98 13.96 25.77% -36.38%
ECUADOR 16.19% 33.51% 0.36 -1.47 10.10 28.29% -55.78%
MEXICO 9.83% 11.48% 0.50 -0.78 7.72 12.84% -14.59%
PANAMA 15.31% 20.73% 0.54 0.26 8.18 28.88% -22.63%
PERU 14.88% 22.17% 0.49 -0.58 10.50 34.50% -29.93%
PHILIPPINES 7.82% 11.91% 0.32 -2.27 13.41 7.75% -20.44%
RUSSIA 18.27% 34.05% 0.42 -1.92 20.27 35.63% -72.18%
VENEZUELA 13.79% 23.07% 0.42 -0.84 12.84 34.05% -39.13%

Appendix 6 Optimal portfolios for a representative investor in the absence of a risk-free asset
Threshold I0 Expected return Expected volatility
(%/year) (%/year)
No constraint 21.38 24.03
82 21.38 24.03
84 22.10 25.25
86 22.98 27.17
88 23.86 29.48

WP-018
The Management of Retirement

Savings revisited
Didier Maillard, Professor at
Conservatoire National des Arts et Mtiers
Senior Advisor on Research Amundi
December 2011, Revised: September 2012
Retirement is not the only motivation of saving but it is a prominent

one. Whether channelled through pension funds or individual
accounts, the question of how to allocate retirement savings, and in
particular which degree of risk to tolerate, is fundamental.
Investing in risky assets should not be viewed as a way to compensate
for insufficient savings during a life time, or a way to optimise the
likelihood of reaching a future consumption target, whatever the
consequences in bad circumstances.
However, as risk free assets tend to vanish, or yield negative returns,
investing in risky assets is a way to improve expected returns on
savings, and thus expected purchasing power at old age, provided
the cost of risk may be mitigated. One way of increasing the
tolerance to investment risk is the potential stream of future labour
income if there is some flexibility on the retirement departure age or
the possibility having a job (full or part-time) during the first years
of retirement.
With reasonable parameters, such flexibility provides a significant
incentive to increase investment in risky assets and provide
significant welfare gains. Finally, labour supply flexibility gives a
reason for the optimal share of risky assets to decline with age.

1. Introduction
The return on risk-free assets, if such assets truly exist, is very low today and will
undoubtedly remain so for a long period of time. Therefore, if investment is made only in
risk-free assets or moderately risky assets, savings for retirement will procure low purchasing
power when it comes time to retire and will be insufficient for reaching any consumption
targets (such as maintaining a certain lifestyle).
If wealth is invested in risky assets, the expectations of return are higher; however, there is a
risk factor: in some adverse configurations, purchasing power could be perceived as
catastrophically low.
Risky investing is not an appropriate way to make up for insufficient savings 1. When risk is
rewarded, risk-taking can increase the likelihood of a reaching a minimum purchasing power
target in retirement but at the price of an exposure to adverse circumstances, which risk
measures such as VaR or CVaR accurately capture.
In the specific case of saving for retirement, the individuals concerned have, however, a way
to mitigate the consequences of an unfavourable configuration for return on investment in
risky assets: supplementing inadequate purchasing power with labour income. In practice,
such income can be secured by putting off retirement or by getting a job, even if part-time,
during the first few years of retirement.
In this research paper, we will simultaneously model, with conventional utility and expected
utility functions, the labour supply at the time of retirement and the portfolio allocation choice
for retirement savings. We examine the importance of labour supply flexibility and the impact
of such flexibility on asset allocation. We find that that the portion allocated to risky assets
can be substantially increased.
This result is consistent with the work done on labour supply flexibility and portfolio choice
(Bodie, Merton and Samuelson, 1992) and generalizes it for different levels of risk tolerance
(or aversion).
In light of these results, we can at last give thought to the identification of pension fund
commitments, or objectives.
1
As Zvi Bodie has shown, expressions such as I cannot afford not to invest in risky assets should be banished.

2. Characteristics of building savings for retirement
Retirement can be defined as a period in life in which an individual no longer receives income
from his or her profession or labour. This individual's consumption needs will therefore have
to be covered by other funds: either transfers or labour income earned prior to retirement that
was not consumed immediately and was therefore saved. Saving is therefore a key factor in
having funds available and consuming during retirement.
Conversely, retirement is a motivation to amass savings during a person's working life. It is

not the sole motivation people also save as a precaution, to meet temporary interruptions in
labour income, and to satisfy the desire to transmit purchasing power to their heirs but it is
an important motivation.
One characteristic of saving for retirement is the length of the time horizon (delayed
purchasing power). Usually retirement lasts 20 years and follows forty years of work. On
average, a period of approximately thirty years elapses between the time savings are amassed
and when they are used for consumption.
This length varies depending on the age of the working individual of interest: for a young
person entering the workforce, this period is almost fifty years a half-century. For an
individual about to enter retirement, it would be ten years or so.
There are two principal schemes for saving for retirement: an individual format and an
institutional format, which we will refer to generically as pension funds. Under the individual
format, the saver (assisted by his or her advisers) has primary responsibility for asset
allocation. Under the institutional format, the fund itself is responsible for asset allocation.
This does not prevent that at the end of the day it is usually the saver who is impacted by the
consequences of the choices made, with one major exception: defined benefit retirement plans
guaranteed by a sponsor who is often the individual's employer. In this case, risk is ultimately
borne by the sponsor (except in the event of bankruptcy) and must be managed within the set
of risks to which he is exposed.
All methods combined, savings for retirement must be substantial. The targeted goal is often
defined as a ratio 50% to 70% of the benefit received to the labour income in the last years
of employment or sometimes the benefit received to the average of labour income over the
working life.

In fact, it is achievable consumption that is the aim, so that a certain lifestyle can be
maintained after retirement, at least partially. Arguments are made that the needs of
consumption are lower after retirement due to, among other reasons, children leaving the
household. But there are also arguments in favour of aiming at higher level of resources to
cover care and medical bills.
With an actual rate of return (after tax 2) on investment near zero, which is optimistic today for
risk-free savings, approximately one-third of all labour income must be set aside for savings,
with a desirable replacement rate of two-thirds.
Table 1
The savings effort required based on real return
T = 30 years, replacement rate = 2/3
Required Saving Rate

Real return
(a) (b)
-2% 0.611 0.379
0% 0.333 0.250
2% 0.184 0.155
4% 0.103 0.093
6% 0.058 0.055
(a) : Replacement rate as a proportion of working life income
(b) : Replacement rate as a proportion of working life income net of retirement savings
This assessment is made using a two-period model, assuming that savings is amassed mid-
career and its fruits spent midway through the retirement period. In practice, the model should
be fine-tuned to take account of the characteristics of the labour income time profile and
mortality tables. It is worth bearing in mind that the orders of magnitude obtained are
nonetheless significant.
Savings here should be understood in a sense widened to include contributions to mandatory

plans, in particular through pay-as-you-go regimes. Naturally, the savings effort is very
dependent on expected real return. This is obviously very important for risk-free investing,
where expected real return is very low and sometimes even negative. The effort can be
substantially reduced for higher real returns but they can only be achieved through risk-
taking.
2
Pension funds and pension accounts usually work under tax neutrality: contributions are deductible from the
income tax base and pensions when retrieved are added to it. If income tax rates do not differ between the two
periods, tax does not impact real return. With other saving channels, tax generally eats into real returns, the more
so inflation is high (Maillard, 2011b).

In case of low real returns, the saving effort is so huge that it is certainly fair to compute the
ratio on a net-of-saving basis, to express the target in terms of old-age potential consumption
as a fraction of working age potential consumption, or working age income net of retirement
savings (column (b) in Table 1). That reduces somewhat the savings ratio target but it remains
important for low returns.
3. The question of managing savings for retirement
3.1. Limitations on management in terms of target.
3.1.1. The issue
In this section, we assume that savings have been accumulated and that the individual has, at
moment in time 0, wealth or capital, W0 , that he or she must invest in one way or another (or
that must be invested on their behalf).
The time horizon to retirement is T. We assume that are efficient mechanisms to transform
wealth, WT , secured on this horizon, into annuities 3, and that this final capital reflects the
degree of achievement of this person's goals.
If there is a risk-free investment between dates 0 and T, and if the initial wealth is invested in
this risk-free asset, the capital secured is known with certainty. However, most often, the
initial wealth will be invested, as least in part, in risky assets and the capital secured will be
exposed to risk.
Pension funds have often obligations (in the case of defined benefit plans) or explicit or
implicit objectives of paying specified amounts, defined nominally or in purchasing power,
i.e. WT *. If savings is not managed by an institution, the individual may also think in terms of
a target.
Commitments will be kept, or the targets reached if:
WT t WT *
There are two possible cases. Risk-free investment (nominal or corrected for inflation) can
produce a return, r f , sufficient to comply with commitments or reach targets.
3
The hypothesis is therefore formulated that pensions are not exposed to risk once liquidated, which
undoubtedly satisfies the desires of most retirees.

t WT *
rf T
W0 e
The question of asset allocation is straightforward in that case. It can consist of investing a
portion of wealth, e WT * , in risk-free assets, and the remainder in high-risk assets. The
rf T
proceeds from investment in this portion could be disbursed as a pension bonus or used to
reimburse contributions by the retirement plan's sponsor.
The first example is perhaps not the most typical, especially in the early 2010s. Often it is
impossible to meet commitments or to reach targets by means of a risk-free asset.
WT *
rf T
W0 e
Complying with commitments or reaching targets is no longer certain. How then should be
question of asset allocation be formulated?
Should the probability of complying with commitments be maximised, or, by the same token,
should the chances of default of the fund due to its obligations be minimised?
Max((Pr(WT ! WT *))
Formalisation of this type could create more room for risky assets (the probability of
complying with obligations is zero if investments are made solely in risk-free assets). But if
we go deeper, it could lead to renouncing all upside in excess of the WT * limit and hence to
selling puts at this threshold on the portfolio of risky assets. Going beyond that, the optimum
solution would be found in investing in a binary option backed by a portfolio of risky assets,
paying 0 with a low probability, and exactly W T * with the highest possible probability.
The drawback of formalising by minimising the probability of default is clear at this stage:
missing targets is not punished in a manner specific to the degree of failure reflecting the
shortfall between the target and actual performance.
Progressive penalisation based on such a shortfall should be introduced and in a manner

consistent with the cost inflicted on the pension beneficiaries and which would take account
of the fact that these beneficiaries can mitigate this cost by virtue of increased labour supply.
We come back to the problem of optimising for the benefit of the ultimate investor as the
pension fund is transparent (but ensures, of course, the functions of allocating the lifetime risk
and handling the financial administration of the savings).

We therefore assume that the charge of the shortfall does not land on the shoulders of the plan
sponsor. We also do not take into account fund's possible regulatory constraints on asset
allocation.
3.1.2. Modelling
Wealth can be invested in risk-free assets, if one assumes they exist, or it can be invested in
an optimally-managed risk portfolio 4 (instantly optimising the Sharpe ratio) if adopting a
dynamic management style 5. Here is average annual expectation of return over the risk
portfolio period, is annualised volatility and t is the Sharpe ratio.
P rf
t
V
The eventual value of the wealth, if a portion, , is (continually) invested in the risk portfolio,
is:
1 2 2
r f D ( P r f ) 2 D V T DV T H
WT W0 e
where is a random variable with zero mean and unitary standard deviation. In the interests of
simplicity, we will use a Gaussian risk distribution, in particular because of the distant
horizon.
The target will be reached or exceeded if:
r * r f
1
H t H* t DV T
DV 2
Pr(WT ! W *T ) 1 ) (H *)
LVWKHODZRIFXPXODWive frequency distribution of the risk.
r* is the risk-free rate of return that must be secured in order to reach the target with a 100%
investment in risk-free assets.
4
The portfolio of risky assets is not necessarily the market portfolio. Room is left for active portfolio
management to improve the performance of the risk portfolio.
5
We are speaking of "lite" dynamic allocation in the sense that the parameters are deemed constant over the
period (no predictability). Its results will differ very little from that of static allocation, with sufficient risk
aversion (Maillard, 2011). The choice of dynamic allocation allows an analytic treatment of the optimization
problem.

If this threshold performance is less than the actual risk free rate, investing all wealth in a
risk-free asset is sufficient to achieve a probability of 1 in reaching the target (trivial, = 0,
= -
If it is greater, there is a funding gap at the effective risk free rate and the difference
represents a shortfall in annualized returns. The threshold therefore decreases with the portion
of risky assets. The theoretical optimum is found, for an infinite investment in the risk
portfolio, with leverage over the risk-free asset.
The quid pro quo is obviously risk, which can be assessed using conventional measurements,
the standard deviation of final wealth or Value-at-Risk (VaR) or Conditional Value-at-Risk
(CvaR, aka Expected Shortfall).
As an illustration, the chart below provides a representation of the parallel change in the
probability of reaching a target and VaR and CVaR at a threshold of 99% in proportion to the
investment, with an investment term of 20 years, a risk free rate of 2%, a credit spread of 4%,
a annualised volatility of the risk portfolio of 20% and a funding gap of 2% per year.
Chart 1
Probability of meeting the target and risk
1
0,8
0,6
0,4
0,2
0
0% 50% 100% 150% 200% 250% 300% 350% 400%
-0,2
-0,4
-0,6
Share of risky assets
Pr(WT>WT*) VaR 99% CVaR 99%
Given the limitations of reasoning in terms of targets, it is logical to place the asset allocation
question within a framework of optimisation.

3.2. Management in response to optimisation
3.2.1. No flexibility in labour supply after retirement
We use the classic Neumann-Morgenstern framework to optimise the expected utility

delivered by consumption derived from purchasing power of the value of the accumulated
savings at the time of retirement.
MaxE (U (WT ))
As to the form of the utility function, we have opted for a function with the feature of constant
relative risk aversion (CRRA). In the catalogue of standard utility functions, we have rejected
the quadratic function (with which utility decreases with consumption after a certain
threshold), and the constant absolute risk aversion (CARA) function. With this last class of
functions, the optimal proportion of risky assets decreases with the initial wealth of the savers,
which is counter-intuitive (See Annex 4). We do not retain a hyperbolic absolute risk aversion
(HARA) function due to the difficulty in identifying irreducible consumption and
distinguishing if from the "targets" described above. However, we will use the results related
to optimisation and describe how to adjust our results if a hyperbolic risk aversion function
were used (Annex 4).
The utility function eventually retained has the form:
C 1J
U (C )
1 J
The range usually deemed realistic for risk aversion is along the lines of 2 to 7/10.
The expected utility is maximised (see Annex 2) for:
P rf
D
JV 2
Using this value for the portion of risky assets:

(1J )( r f s )T
E (U (WT )) (1 J ) 1W0
1J
e
t2
with s
2J
Compared to a risk-free investment, the optimisation of asset allocation leads to a better

utility expectation equivalent to risk-free supplemental return equal to the square of the
Sharpe ratio divided by double the aversion coefficient related to risk.

3.2.2. Flexibility in the labour supply on retirement
To take into account labour supply flexibility, the utility function is augmented to:
C 1J ( L L)1J
U (C , L) b
1 J 1 J
C is the consumption for the period, L is the labour supplied, L is the maximum amount of
work that is possible to provide and L - L is therefore leisure.
The first term is the conventional utility function (CRRA) with constant relative aversion
equal to . The second term represents the contribution of leisure to total utility, with a
weighting dependent on coefficient b.
Using the same exponent for consumption and for leisure firstly cross-references a
conventional utility function of the constant elasticity of substitution (CES) type. The
elasticity of substitution between consumption and leisure is:
1 1
s
1 (1 J ) J
As relative aversion to risk is generally greater than 1, the elasticity of substitution is confined
in a range 0 to 1, which is reasonable.
Furthermore, using the same exponent provides an analytical solution to the problem of
optimising utility, at the moment of interest, under a budget constraint.
The funds available for spending are in fact made up of accumulated wealth, W, and labour
income. If w represents the labour compensation rate, optimization is written:
MaxU (W wL, L)
And leads to labour supply at retirement age equal to:

1 1

J J
b w
L 1 1
L 1 1
W
1 1
J J J J
b w b w
The supply of labour decreases linearly with accumulated wealth 6.
Entering this value for labour supply into the utility function gives (see Annex 3):
6
The labor supply derived from this formula can be negative, which is tolerable: if performance of the risky
assets is very strong, the individual can move forward (if rules permit) the age at which he or she can receive his
or her pension.

1 1 J
(W w L)1J 1
U (C , L) 1 b J w J V (W )
1 J

The function V(W) is a HARA-type function, apart from the fact that the term is constant,
representing minimum consumption, is normally negative in this type of function. Here the
constant term is positive and represents the maximum value of supplemental labour income
that the retiree can secure by sacrificing all his or her leisure.
The second phase entails maximising V(W). We can use the results of Bajeux, Jordan and
Portait (2003) when maximising HARA utility. In this case, the optimal allocation is the
following combination:
- a risk-free investment ultimately yielding exactly the minimum consumption,
- a dynamic portfolio optimally combining risky assets and risk-free assets, the same as
that resulting from optimising a CRRA.
For the question at issue, optimal allocation is composed of:
- a short position in the risk free asset corresponding to the present value, at a risk free
rate, of the maximum labour income that the retiree can receive, or w Le
rf T
- a long position in the portfolio combining the risk portfolio and the risk-free asset
using dynamic management.
The proportion of the risky assets will change over time with the value of the portfolio. At the
outset, it is:
rf T rf T
W0 w Le w Le
D D 1

W0 W0
The portion invested in risky assets is increased by the ratio of the present value of potential
earnings from future work to savings accumulated for retirement. As accumulated savings
usually grows with the age of the saver, while the earnings from potential future work are
independent, the optimal portion of the risky assets therefore decreases with age.
The impact of flexibility on the risk asset portion is not immaterial. To demonstrate this point,
let's take an individual mid-way through his or her working life and for whom accumulated
savings represents four times the labour income (20 years times 20% the savings rate),
assumed to be constant. If this person can envisage working the equivalent of two years

subject to the same pay conditions and assuming the risk free rate is zero, the optimal
proportion of risky assets is increased by half.
It is finally possible to compute the welfare gains of labour supply flexibility (see Annex 3)
by comparing the expected utility with flexibility to the expected utility without (L =0). Part
of those gains stem from the ability to invest more in risky assets that labour supply flexibility
provides.
4. Discussion and conclusions
We have shown, using a simple model without being unrealistic, that labour supply flexibility
at the time of retirement can improve economic welfare, both directly and indirectly, by
making it possible to invest in riskier assets and to capture the rewards of risk.
In fact, it happens that working during retirement or deferring retirement procures additional
wealth at no risk or, in any event, with little risk (but not without pain). This assumes that
there are good assurances against potential joblessness and, above all, against the inability to
work. The optimisation question ultimately translates into optimisation with the constraint of
a fixed investment in a given asset.
As to pension funds, and especially in the case where they manage most of the retirement
savings of their principals, we would recommend promoting flexibility on the liquidation age.
The ideal would be to offer la carte allocation, taking account of risk aversion on the one
hand and, on the other, the opportunity and willingness to get an old-age job or to defer
retirement.

References
Amenc, Nol, Felix Goltz, Lionel Martellini and Vincent Milhau, 2010, New Frontiers in
Benchmarking and Asset Liability Management, Working Paper, Edhec Risk Institute,
September 2010
Bajeux-Besnainou, Isabelle, James V. Jordan and Roland Portait, 2003, "Dynamic Asset
Allocation for Bonds, Stocks and Cash", Journal of Business, Apr 2003, Vol. 76, Issue 2
Bodie, Z., R. C. Merton and W. F. Samuelson, 1992, Labor Supply Flexibility and Portfolio
Choice in a Life Cycle Model, Journal of Economic Dynamics and Control, 16, 427-449
Bodie, Z., 2001, "Retirement Investing": A New Approach, Working Paper 2001-03, Boston
University School of Management
Maillard, Didier, 2011a, Dynamic versus Static Asset Allocation: From Theory (halfway) to
Practice, Working Papers Series N 1942901, SSRN, March 2011
Maillard, Didier, 2011b, Tax and Investment Return, Working Papers Series N 1968985,
SSRN, September 2011

Annex 1
Target is achieved iff:
1 2 2
WT W0 e
t WT *
1 2 2 WT *
r f D ( P r f ) 2 D V T DV T H t ln W
0
WT * 1 2 2
ln r f D ( P r f ) D V T
W0 2 r * r f 1
Ht t DV T H*
DV T DV 2
Pr(WT ! W *T ) 1 ) (H *)
Annex 2
If a portion of wealth is invested in a risk portfolio with an expected return of and

volatility on this return of , and a portion, 1- is invested in a risk-free asset whose return is
r, the relative variation in wealth over time can be represented by the following equation:
dW
(1 D )r f dt D ( Pdt Vdz ) >r f D ( P r f ) dt DVdz
@
W
where dz is standard Brownian motion.
Using ,W
VOHPPDand integrating leads to:
1 2 2
d ln W r D ( P r f ) 2 D V dt DVdz
Assuming the random process is Gaussian, and designing a zero mean unitary variance
random variable
1 2 2 1 2 2
r f D ( P r f ) 2 D V T DV T H r f D ( P r f ) 2 D V T DV T H
W (T ) W (0)e
WT W0 e
If we assume that the utility function is CRRA with parameter ,
U (W ) (1 J ) 1W 1J
(1J )( r f D ( P r ) D 2V 2 / 2 )T
U (WT ) (1 J ) 1W0
1J
e e (1J )DV TH
(1J )( r f D ( P r ) D 2V 2 / 2 )T
E (U (WT )) (1 J ) 1W0
1J 2
D 2V 2T / 2
e e (1J )
(1J )( r f D ( P r f ))T (1J )D 2V 2 / 2 )T (1J ) 2 D 2V 2 / 2 )T
E (U (WT )) (1 J ) 1W0
1J
e
(1J )( r f D ( P r f ) JD 2V 2 / 2 )T
E (U (WT )) (1 J ) 1W0
1J
e

In fact, expected utility is the same as that obtained in the case of a risk-free investment, for
which the return would be:
J
rc r f D ( P r f ) JD 2V 2 / 2 E (rP ) V (rP )
2
Expected utility is maximised for:
P rf
D
JV 2
Using this value for the portion of risky assets:
(P r f ) 2 (P r f ) 2 V 2 1 (P r f )
2
1 t2
r f D ( P r f ) JD 2V 2 / 2 r J rf rf
JV 2 J 2V 4 V2 2 JV 2 2J
2
(1J )( r f t / 2J )T 1J (1J ) r f T
E (U (WT )) (1 J ) 1W0 (1 J ) 1W0
1J
e e
1
e sT
W0
W0
>e (1J )( t 2 / 2J )T 1J
@ e (t
2
/ 2J )
t2
s
2J
Annex 3
1) Optimization of labour supply
(W wL)1J ( L L)1J
MaxU (W wL, L) b
1 J 1 J
dU
w(W wL) J b( L L) J 0
dL
1 1

w J (W wL) b J ( L L)
1 1
1
1

1
L b J w J b J
L w JW

1 1

J J
b w
L 1 1
L 1 1
W
1 1
J J J J
b w b w

2) Resources at old-age
1 1 1 1

b J w J b J
b J
W wL 1 1
w L 1 w 1 1 W 1 1
wL 1 1
W
1 1 1 1

J J
J J
J J

J J
b w b w b w b w
1

J
b
W wL 1 1
(W w L)
1
J J
b w
1 1 1

b J w J w J
LL 1 1 1 L 1 1
W 1 1
(W w L)
1 1 1

J J
J J

J J
b w b w b w
3) Utility function
1 1J 1 1J

1 b J w J
U (C , L) (W w L) 1J
b 1 (W w L) 1J

1 J J1 1
1
J J 1
1
J
b w b w
1J 1J
1J 1J
1 J 1 1
U (C , L) b bw J
(W w L)
1J
1 J 1
1
1

1
J 1
1

b J w J b w J

1J
1 J
(W w L)1J 1 1 1 (W w L)1J 1 1
1
U (C , L) b b J w J 1 b b J w J
1 J

1
b J w1 J

1 J

1 1 J
(W w L)1J 1
U (C , L) 1 b J w J V (W )
1 J

Optimally, final wealth is:
1 2 2
WT W 0 w Le
rf T
e
wL
As to the funds available for consumption, they are

1 1

1 2 2
J J
b b
WT wL

1
1
1
(WT w L)

1
1
1
W 0 w Le
rf T
e
J J J J
b w b w
The utility is:

1 1 J
(WT w L)1J 1
U (C , L) 1 b J w J
1 J

1 1 J 1
(1J ) r f D ( P r f ) D 2V 2 T DV T H

1 1J
(1 J ) 1 b J w J W0 w Le f

1

r T
e 2

1 1 J
1 1J
E (U ) (1 J ) 1 1 b J w J W0 w Le f

r T
e
(1J )( r f t 2 / 2J )T
If there is no labour flexibility (L = 0), the expected utility is
E (U ) (1 J ) 1 W0 > 1J
e
(1J )( r f t 2 / 2J )T
bL
1J
@
We can measure the gains in welfare associated with flexibility by comparing the savings
leading to expected utility with flexibility and the higher savings needed to reach the same
degree of utility without flexibility.
Expressed in monetary terms, the resulting gain in welfare due to flexibility is:
W 0 W0
1 1 J
(1 J ) 1 W 0 > 1J
e
(1J )( r f t 2 / 2J )T
bL
1J
@ 1
(1 J ) 1 1 b J w J W0 w Le f

r T
1J
e
(1J )( r f t 2 / 2J )T

1 1 J
t2 1
u rf A( w) 1 b J w J
2

1J 1J
W 0 e uT bL
1J

A( w) (W0 w Le
rf T
)e uT
1
1J
W0 e uT
0
A( w) (W w Le r f T )e uT bL
1J 1J


Annex 4
Thoughts on the choice of the utility function
The utility functions commonly used are the quadratic function, the constant absolute risk
aversion (CARA) function, the constant relative risk aversion function (CRRA) and the
hyperbolic absolute risk aversion (HARA) function.
Quadratic function
It has the form:
M
U (C ) C C2
2
U ' (C ) 1 MC
Consumption utility decreases for C ! 1 / M , which is in conflict with the commonly accepted
hypothesis that an abundance of wealth is not harmful.
CARA function
U (C ) e aC
Although it is used in the context of optimising a static portfolio (the result cannot be obtained
analytically in the context of dynamic optimisation, but should not differ significantly), you
obtain a result that is relatively counter-intuitive.
In fact, the final portfolio is worth, if is the portion invested in the risk portfolio (with the
notations in the body of the text):
WT
W0 (1 D )e
rf T
De PT V TH
W0 A BD A e
rf T
B e PT V TH
e
rf T
U (WT ) exp> a @ exp> aW0 A BD @ exp> aW0 A@exp> aW0 BD @

E (U (WT )) exp> aW0 A@E exp> aW0 BD @
W0
If maximises this expression, then D '* D * maximises
W '0
E (U (W 'T )) exp> aW ' 0 A@E exp> aW ' 0 BD @
This means that the optimal percentage allocated to risky assets is inversely proportional to
the initial wealth, which appears contrary to the perception of reality.

CARA function
It has the form:
C 1J
U (C )
1 J
HARA function
It has the form:
(C C )1J
U (C )
1 J
C is irreducible consumption, below which the question of utility is irrelevant. It should not
be confused with a target such as that discussed in the body of the article.
To maximise the expectation of a HARA function, on the one hand the risk-free asset must
exist and, on the other, placing in this asset exactly what is needed to attain C , the remainder
being placed in a dynamic portfolio identical to that which would result from maximising a
CRRA function with the same risk aversion parameter.
With labour flexibility, optimisation in the context of such a function would yield a solution
whereby it would be necessary to place the difference (algebraic) between what is necessary
to achieve C and the present value of the maximum of labour income in the risk-free asset.

WP-022
Incorporating Linkers
in a Global Government Bond
Risk Model
Marielle de Jong,
Head of Fixed Income Quantitative Research, Amundi
February 2012, Revised: June 2012
Since many governments over the world have started issuing

inflation-protected securities, or linkers, alongside the conventional
nominal bonds, the investment universe for fixed-income investors
has de facto been extended. It is not evident how to incorporate
the new securities into the investment processes, in particular,
how to build them into a bond risk model such that the overall
global price covariance is captured. We study this question taking
the International Capital Asset Pricing Model approach, where
in essence risk is decomposed in a global systematic- and a local
specific component.
The challenge is to fit a linear factor model onto the triangular
relationship that exists between the nominal bond yields, the
inflation-linked bond yields and their yield spreads called the
inflation breakeven rates. The model we develop provides a
consistent view on risk for internationally invested portfolios
containing nominal bonds, linkers and/or inflation swaps.

The historically stable fixed-income investment market has been elicited over the last ten to
twenty years by the issuing of inflation-linked (real) bonds alongside the conventional
nominal bonds in most of the major worlds markets. This market innovation does not merely
increase the choice in sovereign debt securities, it essentially adds a new dimension to the
scope of the bond investor. It creates the possibility to take a relative position on real debt
versus nominal debt, and as such place a directional bet on the inflation level in a country.
Naturally, this market innovation calls for a revision of the existing risk models serving
government bond investment strategies. Since the pricing of inflation has been made explicit
in a way via the bond yield differentials called the breakeven inflation rates, it would in
principle be possible to model inflation risk in an explicit way as well. This was not
conceivable before. Nominal bonds are, unlike real bonds, inherently exposed to inflation risk
since their payoff is typically not compensated for the inflation incurred over the time to
maturity, but the price influence of that cannot be made explicit in an easy way without a real
bond comparative.
We explore, in this article, how to build the new inflation dimension into a bond risk
modeling framework. We have not found studies in the literature making such attempt. We
choose to do this in an international setting focusing on the cross-border price dependency
between national bond markets as a whole, rather than between individual financial
instruments. We adopt the International Capital Asset Pricing Model introduced by
Solnik [1974]. This model is a generalization of the Sharpe-Lintner CAPM that was initially
developed for equity instruments within one country, towards an international scale including
other asset classes as well. The essence of the model is that price returns are being
decomposed into a systematic global market risk component and country-specific residual
components.
The model design typically suits an international bond investor who runs a country allocation
strategy. The dichotomous decomposition of risk facilitates the assessment of the tactical
positions split into global directional bets and individual country bets. The term structure in
the bond yields is not considered in this article. The currency risk inherent to purchasing
foreign assets is not considered either, this by making the assumption that all foreign positions
are fully hedged against exchange rate risk. The model is as such an abbreviated version of

the usually more general fixed-income models used by practitioners and discussed in the
literature; see for example Fabozzi [1998].
Within the modeling setting that we have chosen we uncover an intriguing price correlation
relationship between the bonds, which would not be easy to detect in a national context. In
next section we discuss this pricing phenomenon. In section 3 we formulate the challenge it
represents for modeling the risk. In section 4 we present which modeling approach would suit
best, and section 5 concludes.
The global price covariance between nominal- and inflation-linked bonds
We start by exploring the price covariance of the inflation-linked bonds with the nominal
bonds over the recent past. In Exhibit 1 the price correlation matrix is displayed between the
two types of bond traded in the major developed world markets. Correlations are measured
between the time-variation of the nominal bond yields (NBY), the inflation-linked (real) bond
yields (RBY), and the yield spreads, the breakeven inflation rates (BEIR). Six countries have
been retained in this study: Australia, Canada, the Euro Area, Great-Britain, Sweden and the
United States, which correspond to the members in the Barclays Developed World Inflation-
Linked Bond Index that have started issuing linkers before 2002.1
The correlations, as well as the annual volatilities which are displayed on the diagonal, are
measured on weekly bond returns in the seven-to-ten-year maturity range with an average
duration of 7.5 years as calculated by Barclays Capital, over an eight-year period from June
2002 to July 2010. The particularly volatile months between October 2008 and May 2009
have been discarded for reasons discussed separately below. In de Jong [2010] it is verified
that essentially the same numbers are obtained when measuring on monthly rather than on
weekly data over the same observation period.

EXHIBIT 1
Correlation/covariance matrix between the RBY, NBY and BEIR variation
RBY NBY BEIR
es
es
es
n
n
ai
ai
ai
at
at
at
ea
ea
ea
rit
rit
rit
St
St
St
lia
lia
lia
en
en
en
Ar
Ar
Ar
-B
-B
-B
da
da
da
ra
ra
ra
d
d
at
at
at
ed
ed
ed
ite
ite
ite
na
na
na
ro
ro
ro
st
st
st
re
re
re
Sw
Sw
Sw
Au
Un
Au
Un
Au
Un
Eu
Eu
Eu
Ca
Ca
Ca
G
G
RBY Australia 4,4%
Canada 0,27 3,1%
Euro Area 0,40 0,45 4,5%
Great-Britain 0,43 0,44 0,67 5,1%
Sweden 0,43 0,37 0,65 0,54 3,5%
United States 0,43 0,51 0,60 0,52 0,51 5,4%
NBY Australia 0,83 0,26 0,49 0,45 0,51 0,46 6,4%
Canada 0,49 0,59 0,59 0,54 0,52 0,66 0,55 4,3%
Euro Area 0,46 0,40 0,79 0,65 0,69 0,59 0,56 0,68 5,2%
Great-Britain 0,52 0,43 0,74 0,83 0,66 0,60 0,59 0,67 0,81 5,2%
Sweden 0,49 0,35 0,72 0,60 0,77 0,56 0,59 0,65 0,82 0,77 5,1%
United States 0,52 0,46 0,61 0,56 0,55 0,80 0,58 0,80 0,76 0,72 0,69 6,6%
BEIR Australia (0,24) 0,13 0,37 0,28 0,36 0,29 0,75 0,36 0,42 0,41 0,44 0,40 3,7%
Canada 0,36(-0,16) 0,32 0,27 0,31 0,35 0,43 0,70 0,48 0,44 0,47 0,57 0,32 3,5%
Euro Area 0,19 0,01(-0,11) 0,12 0,21 0,11 0,24 0,28 0,52 0,28 0,34 0,38 0,18 0,33 3,2%
Great-Britain 0,16 -0,01 0,14(-0,28) 0,21 0,15 0,25 0,24 0,30 0,31 0,30 0,29 0,23 0,30 0,29 3,0%
Sweden 0,31 0,16 0,43 0,37 (0,15) 0,33 0,39 0,45 0,56 0,51 0,74 0,49 0,31 0,41 0,31 0,24 3,2%
United States 0,27 0,06 0,19 0,21 0,22(-0,05) 0,33 0,42 0,45 0,37 0,38 0,57 0,25 0,46 0,48 0,28 0,36 4,0%
Period: June 2002 to July 2010, excluding the crisis period from October 2008 to May 2009.
Data frequency: weekly (Friday to Friday).
The numbers on the diagonal are annual return volatilities (bond duration 7.5 years).
Threshold of significance for the correlation numbers on a 5% confidence level with T=389 weeks is 0.10.
Data source: Barclays Capital. Calculations made by the author.
Note that the correlations are significantly positive, they are more accentuated within markets
between the NBY and RBY, and between the NBY and BEIR (shaded in grey), yet in contrast
they are lass accentuated between the RBY and BEIR (in brackets). The positive cross-border
correlation overall is usual for financial assets. It reveals a global common price movement,
indicating that there is a global systematic risk component in the bond price behavior. There
appears to be a country-specific component as well. The inflation-linked bonds are more
correlated with the nominal bonds within the same country than to those of other countries,
which is intuitive. The same holds for the breakeven rates with respect to nominal yields. The
observation concerning the RBY and BEIR seems odd though. Why would those variables be
systematically less correlated within countries than between?
Cette and de Jong [2008] give an explanation. They make evident that the country-specific
movements of the RBY and BEIR are systematically anti-correlated with each other and this
as a result of local market distortions. Issues like market liquidity and a time-varying aversion
to inflation risk have been raised in the literature as reasons for those distortions; see
Christensen, Dion and Reid [2004] for a survey. As soon as the inflation-linked bond price

moves due to such country-specific market-related events, which are typically not mirrored in
the nominal bond price, the breakeven rate, the third leg in the triangle as it were, moves
mechanically in opposite direction. The global systematic correlation between the RBY and
BEIR, which appears to be positive, is thus offset locally. This feature is shown to be a
stylized fact that is persistent in time since the first introduction of inflation-linked securities
on the capital markets.
The observations we make are schematized in Exhibit 2 below. The covariance between the
real bond yields and the breakeven inflation rates between countries is the net sum of positive
systematic covariance and negative residual covariance (second row in the Exhibit). This is
atypical compared to the more intuitive situation where the residual covariance within
countries is positive (first row). In next section it is discussed how this particular price
behavior can be take into account in an overall bond risk model.
EXHIBIT 2 Atypical covariance structure
typical covariance structure total

covariance
systematic covariance
residual covariance

applying to NBY with RBY 0
and to NBY with BEIR 0

atypical covariance structure total
covariance
systematic covariance
residual covariance

applying to RBY with BEIR 0 0
0 0

In their most recent paper, Cette and de Jong [2012] investigate the persistence of the global
correlation between the common trends in the RBY and BEIR. This correlation has been
positive during six years from June 2002 to September 2008, has turned significantly negative
over the particularly turbulent period from October 2008 to May 2009, and has then turned
positive again attaining the same correlation level as before. The explanation given by the
authors for the radical sign change during the crisis is that the market distress, and the market-
related issues related with that, had become a global phenomenon. Market distortions lead to
opposite price movements between the RBY and the BEIR, and if they are concerted over the
globe, the opposite movements are being perceived globally as well.

The violent events that shook the markets during the crisis period in 2008 have been discussed
since in the literature; see for example Campbell, Shiller and Viceira [2009], or Grkaynak,
Sack and Wright [2010]. The consensus is to say that the events were exceptional and that the
price swings they provoked were highly unusual. In the objective to build a risk model
serving bond portfolio investment in normal market conditions, supposing normally
distributed price variations, crisis periods should typically be discarded. This is what is done
in the remainder of this paper. The fact that the bond prices settle back into the same regime
as they were in before this crisis period, as shown by Cette and de Jong [2012], reinforces the
thought that this was a one-off exceptional period.
The risk modeling challenge
We aim to build a risk model that specifies the overall price covariance structure effectively in
the newly-extended bond universe, containing thus nominal- as well as inflation-linked
government bonds. The challenge is to encapsulate the atypical price covariance that has been
made apparent in previous section. In order to appreciate its practical implications, consider
the following situation. When inflation swaps, i.e. derivative instruments directly priced on
the breakeven inflation rates, are being added to a nominal bond portfolio, the total risk will
increase, due to the strongly positive price correlation between the two instruments. If the
swaps were added to a portfolio invested in inflation-linked bonds, the risk would not
necessarily increase and may even decrease, due to the negative correlation.
Let us quantify this example for the case of British Gilts, taking the relevant risk parameters
from Exhibit 1.2 We read in the Exhibit that a portfolio fully invested in nominal Gilts and
one invested in inflation-linked Gilts both have an annual volatility level of 5%. Adding 100
percent inflation swaps, with an annual volatility of 3%, increases the overall volatility of the
nominal bond portfolio to 7%, that is 7% # 5% 2 3% 2 2 5% 3% 0.3 , while it leaves the
volatility of the inflation-linked bond portfolio unchanged at 5%,
since 5% # 5% 3% 2 5% 3% 0.3 .
2 2
It is not directly obvious how to take account of this situation in a global risk model. We
describe how we proceed in the context of the International Capital Asset Pricing Model, the
I-CAPM of Solnik [1974]. We start by formulizing a global nominal bond risk model in an I-
CAPM framework. Let the nominal bond yield variations in countries i, denoted as 1%< i , be

decomposed into a global market factor, denoted as F1%<, and country-specific residual terms
i which are assumed to be independently distributed of each other and of the common factor:
'NBYi E i
'NBY
F 'NBY H i (1)
so that the covariance structure between the bond yields is specified as:
2
E i E j V 'NBY
'NBY 'NBY
if i z j
cov'NBYi , 'NBY j 'NBY 2 2
(2)
E i E j V 'NBY V H
'NBY
if i j
where E i'NBY are the market betas, the sensitivities of assets i to the global market factor,
V 2
'NBY
is the global market variance and V H2 the country-specific residual variances.
Question is how to incorporate the inflation-linked bonds into this modeling framework. One
approach would be to define a global inflation-linked market factor onto the real bond yields
in an analogous way, denoted as F5BY, and assume a linear relationship with the nominal
market factor, through a parameter . The global NBY variation that is not captured by the
real bond factor is denoted as F%(,5. All residual terms are assumed to be independent of each
other and of the factors:
'NBYi E i F H i
'NBY 'NBY

'RBY E 'RBY F 'RBY K
i i i i (3)
F 'NBY O F 'RBY F 'BEIR
so that the covariance structure between the two bond yields is specified as:
cov'NBYi , 'RBY j E i'NBY E j'RBY O V '2RBY (4)
Given this model, the covariance structure of the bond yields with the breakeven inflation
rates can be derived, by applying the definition BEIR = NBY RBY. This is done for the
RBY with the BEIR in equation (5), by which, interestingly, the country-specific anti-
correlation appears. For this aspect the model given in (3) fits the empirically observed price
behavior of the bonds. In next section we test by means of a statistical analysis to what extent
the model adapts to the data more generally.

cov'RBYi , 'BEIR j covE i'RBY F 'RBY K i , E j'NBY F 'NBY H j E j'RBY F 'RBY K j
E i E j O E i E j V 'RBY
'RBY 'NBY 'RBY 'RBY 2
if i z j (5)
'RBY
E i E j O E i E j V 'RBY V K
'NBY 'RBY 'RBY 2 2
if i j
It appears, and this is the interest of this paper, that a relatively small modification in the
model specification leads to much better results in terms of data fitting. The idea is to add a
global breakeven market factor to the original risk model (1) instead of a real bond factor as
was done in (3). This model is specified in (6) below. A breakeven factor, denoted as F%(,5,
LVGHILQHGRQWRWKHEUHDNHYHQLQIODWLRQUDWHYDULDWLRQV%(,5WKDWDUHGLUHFWO\GHULYHGIURP
the market data, and a linear relationship with the nominal market factor is specified through a
parameter . Again all residual terms are assumed to be independent of each other and of the
factors:
'NBYi E i F H i
'NBY 'NBY
'BEIR E 'BEIR F 'BEIR Z

i i i i (6)
F 'NBY P F 'BEIR F 'BRBY
The reader may verify that this model specification embeds the country-specific anti-
correlation between the RBY and BEIR just as well. The reader may verify also that the third
possible combination of bond variables, defined on the real bond yields with the breakeven
rates, does not have this quality. The residual terms of the RBY and of the BEIR would in this
combination be assumed independent of each other, in a standard CAPM approach, which
would exclude from the start the possibility of negative correlation.
The market data fit
In this section we test to what extent the two model specifications, given in (3) and in (6),
adapt to the market data. We do this by carrying out principal components analyses, by which
we retrieve the first two eigenvectors that explain at maximum the bond yield variance. The
results are displayed in the two Exhibits below. The sensitivities to the first two eigenvectors,
denoted as ev1 and ev2 on the axes, are plotted of the twelve yields, that is six NBY and six
RBY in Exhibit 3, and analogous for Exhibit 4.
We look if the eigenvectors we obtain resemble the global market factors that have been
defined in the two respective models. If they match, it tells that the model factors are the ones
with the highest statistical significance. Note that in both models the first factor represents a

global market factor that is common to all bonds. The second factor then makes a distinction
between the nominal- and real bond yields in model (3), and between the nominal yields and
the breakeven rates in model (6).
Let us first look at Exhibit 3. The fact that all sensitivities to the first eigenvector are positive
means that it indeed represents a worldwide market factor commonly shared by the nominal-
and real yields. This factor is statistically significant, explaining 62% of the variance. Yet the
second eigenvector doesnt match. As can be seen in the Exhibit it represents a factor that
somehow makes a distinction between Canadian, Australian and American bond yield
movements on the one hand versus Euro Area, Swedish and British yield movements on the
other, regardless of the bond type. This geographical division doesnt correspond with the
factor specified in model (3), which makes an explicit division between the two bond types.
Model (3) is for this matter statistically rejected.
EXHIBIT 3
Results of a principal component analysis on the NBY and RBY correlation matrix
2,00
ev2
1,50
RBY Canada
1,00
RBY United States NBY Canada
RBY Australia NBY United States
0,50
NBY Australia
0,00
-2,00 -1,50 -1,00 -0,50 0,00 0,50 1,00 1,50 ev1 2,00
-0,50 NBY Great-Britain

RBY Great-Britain
NBY Euro Area
RBY Euro Area
NBY Sweden
RBY Sweden
-1,00
-1,50
-2,00

EXHIBIT 4
Results of a principal component analysis on the NBY and BEIR correlation matrix
2,00
ev2 BEIR Euro Area
1,50 BEIR United States
1,00
BEIR Great-Britain
BEIR Canada
0,50
NBY United States
NBY Euro Area
BEIR Sweden
NBY Canada
0,00
-2,00 -1,50 -1,00 -0,50 0,00 0,50 1,00 1,50 2,00
ev1
NBY Sweden
-0,50 NBY Great-Britain
-1,00
NBY Australia
-1,50
BEIR Australia
-2,00
Looking at Exhibit 4, it can be seen that the first eigenvector again corresponds to a global
market factor. It is statistically significant as well explaining 52% of the variance. Note that in
this case the second eigenvector matches. The factor reflects a systematically different
movement between the nominal yields on the one hand versus the breakeven rates, with the
exception of Australia, on the other. This factor explains an additional 10% of the total
variance, which gives support for adopting the model (6).
So we find model (6) more in line with how bond prices behave. It is striking that the
relatively small modification we had made in the data presentation, which has no incidence on
its information content, has such a different outcome in the estimation results. We point at two
details in the model specification that are at the origin of this. Firstly, the two models dont
make exactly the same assumptions for the residual terms. In model (3) independence is
assumed between the residual NBY and RBY, and in model (6) between the residual NBY
and BEIR. This has a direct consequence for the volatility levels. The residual variance of the
BEIR is according to model (3):

residual variance 'BEIRi varH i K i V H2 V K2 (7)
which is strictly superior to that of the RBY. That doesnt strike with the empirical data, as
can be seen in Exhibit 1. The reader may verify that model (6) inverses this hierarchy: the
country-specific real bond yields are by construction more volatile than the breakeven
counterparts. The fact that this tends to be true in practice may be attributed to the before-
mentioned market distortions agitating the inflation-linked bond prices in particular.
The second seeming detail is that the data presentation indirectly imposes a hierarchy in the
two market factors. In model (3) the first factor defines a global common movement between
the nominal- and real yields, and in model (6) between the nominal yields and breakeven
rates. Those are essentially different factors. With reference to Fishers [1930] seminal
Interest Rate Theory and supposing that this theory applies to long-term sovereign bonds, we
can say that the first factor in (3) is driven by structural shocks that affect the global real
economy. Nominal- and real bonds have this risk source in common in principle. The first
factor in (6) is rather driven by inflation shocks. We may call this the global inflation factor,
with reference to Fisher.
It is an interesting question, yet beyond the scope of this article, whether the inflation shocks
have indeed more impact than structural shocks on sovereign bond prices on a worldwide
scale. The inflation-linked bond markets may, as of today, well be too inexperienced and
illiquid to assert this. What we have shown in this article is that defining a global inflation
factor in the way it is done in model (6) is an effective way to capture the new risk dimension
that the new inflation-linked bonds have introduced.
We point out that a bond investor, who would want to assess the risk of an internationally
invested portfolio using the enhanced modeling framework we propose, is not refined to stay
in the data presentation in which the model was estimated. In other words, although the model
factors are defined onto the nominal yields and the breakeven rates explicitly, the portfolio
may contain inflation-linked bonds as well. In order to assess the portfolio risk, it suffices to
adapt the presentation of the model. This can be done easily by means of a rotation operator
L3 given in equation (8) below. The covariance matrix that is spanned by model (6), denoted
as VNBY,BEIR, can be re-expressed in terms of nominal- and real bond yields, denoted as
VNBY,RBY, by applying the following matrix multiplication:

1 1
V NBY , RBY LT V NBY , BEIR L , where L = (8)
0 1

The original matrix given in (9a) transforms into the matrix in (9b). In order to verify the
overall coherence, the reader may check that if the rotation operator is applied three times
consecutively, the original covariance matrix reappears.
VNBY ,NBY VNBY ,BEIR NBY ,RBY VNBY ,NBY VNBY ,NBY VNBY ,BEIR
V NBY ,BEIR V (9)
V
NBY ,BEIR VBEIR ,BEIR
V
NBY , NBY
V NBY , BEIR
V NBY , NBY
2 V NBY , BEIR
VBEIR , BEIR

(a) (b)
Conclusion
The issuing of inflation-protected securities in countries over the globe has expanded the
investment universe of fixed-income investors and has brought an essentially new dimension
into their performance opportunity set. An investor traditionally holding nominal bonds can
seize this new opportunity by adding inflation swaps, or by adding linkers, or alternatively, by
replacing the entire portfolio by one that contains linkers and inflation swaps only. In
principle the three options are equivalent in terms of risk-and-return profile, however, it is not
trivial to assess the portfolio risk in this new situation in an effective and coherent way.
The bond literature is surprisingly silent on the question. In this article a modeling approach
has been presented which tackles the issue in an international bond allocation setting. The
choice of setting allows dealing with an intriguing price phenomenon that plays on an
international scale. A two-factor linear model has been developed that is set to capture the
international price covariance between the nominal- and inflation-linked bonds. The first
factor may be loosely interpreted as a global inflation factor priced by the bonds. The second
factor divides the nominal yields from the breakeven rates and may for that matter be
interpreted as a real economy factor driven by global structural macroeconomic shocks.

Endnotes
1
Japan, which is member of the Barclays index since 2004, has not been included in the study
for its relatively short data history and because serious price-efficiency issues persist due to
the fact that the inflation in Japan is close to zero. For all index data issues we refer to James
Global Inflation-Linked Products User Guide (2010).
2
We make a number of simplifying assumptions here, intentionally to focus on the main idea.
We assume that the inflation swaps have exactly the same price behavior as their cash
equivalent, i.e. a long position in a nominal- and a short position in an inflation-linked bond of
the same duration. In complementary tests we have verified that the small price divergence
doesnt alter the principle results. In addition, we assume that all instruments in the example
have a 7.5-year duration, that they are not exposed to any source of risk other than interest
rate movements, and that the risk parameters measured in the past are representative.
3
The same operator is used for currencies to convert a covariance matrix measured from one
base perspective into another, see Solnik [1974].

References
Campbell, J., R. Shiller, and L. Viceira. Understanding Inflation-Indexed Bond Markets.

NBER Working Paper Series, No. 15014, 2009.
Cette, G., and M. de Jong. The Rocky Ride of Breakeven Inflation Rates. Economics
Bulletin, Vol. 5, No. 30 (2008), pp. 1-8. A more complete version with the same title is
available as a Banque de France working paper, No. 230, 2009.
http://www.banque-france.fr/gb/publications/telechar/ner/dt230.pdf.
__________ Breakeven Inflation Rates and Their Puzzling Correlation Relationships. under
revision with Applied Economics.
Christensen, I., F. Dion, and C. Reid Real Return Bonds, Inflation Expectations and the
Break Even Inflation Rate. Bank of Canada Working Paper, No. 43, 2004.
Fabozzi, F. The Handbook of Fixed Income Securities. McGraw-Hill, 1998.
Fisher, I. Theory of Interest. Macmillan, New York, 1930.
James, A. Global Inflation-Linked Products A Users Guide. Barclays Capital Research,

2010.
de Jong, M. The Perception of Investment Risk. PhD dissertation, University of Aix-

Marseille II (DEFI), 2010.
Grkaynak, R., B. Sack, and J. Wright. The TIPS Yield Curve and Inflation Compensation.
American Economic Journal, Macroeconomics, Vol. 2, No.1 (2010), pp. 70-92.
Solnik, B. An Equilibrium Model of the International Capital Market. Journal of Economic

Theory, No. 8 (1974), pp. 500-524.

WP-030
Market-Implied Inflation
and Growth Rates Adversely
Affected by the Brent
Gilbert Cette,
Associated Professor in Economics, Universit dA ix-Marseille II
Marielle de Jong,
October 2012
The inf lation and the real yield component deduced from
inf lation-linked and nominal bond prices are adversely affected
by two market effects: price distortions due to certain market-
related events and oil price movements. Their underlying time-
correlation without those effects is stable and positive. Market
data analysis carried out on the worlds major bond markets
gives valuable new insight in the long-debated relationship
between inf lation and growth prospects.

1. Introduction
When inflation-linked bonds were introduced on the world bond markets one to two decades
ago, there was positive belief that their pricing would reveal the inflation and growth
expectations of the market participants. In a short version of Fishers (1930) interest rate
theory, the yield of these bonds, the real yield, reflects the economic growth forecast, while
the yield differential (nominal minus real) called the breakeven inflation rate, reflects the
inflation forecast. It has proven difficult though to make such assertions on the market data
that has become available since.
Furthermore, it has proven difficult to learn from the bond data how inflation and economic
growth mutually interact. In our previous articles (Cette and de Jong, 2008, 2013) we had
made an attempt, making apparent that the time-correlation between real-bond yield (RBY)
and breakeven inflation (BEIR) variations is continuously distorted within countries by
market-related events. Observations made within local markets, which is the standard in the
literature on inflation-linked bonds, may therefore be misleading. By taking an international
approach we had been able to separate out the correlation due to country market distortions to
a certain extent, so as to obtain a view on the more fundamentally-driven correlation. It
showed that the correlation measured on a global aggregate scale is positive between RBY
and BEIR, except during the heat of the financial crisis in 2008/2009.
What does this say about the interaction between inflation and growth prospects? We show in
this article that the oil price plays an important role. There is an apparent adverse relation
between breakeven inflation and real yield movements, the former being driven up by an oil
price rise while the latter is pushed down. When eliminating the effect of oil from the bond
prices, the net global correlation between BEIR and RBY rises. In the crisis sub-period in
2008/2009, the oil price was particularly turbulent provoking large adverse movements
Again, taking an international study approach is essential in making the observations. The
influence of oil is easier to detect in global aggregate bond yield variations where the country-
specific effects are diversified away and oil, a common denominator for all economies,
remains. The new test results contribute to the longstanding debate on the relation between
inflation and economic growth prospects.

Section 2 presents the database. Section 3 gives the correlation structure between RBY and
BEIR variation and Section 4 presents the role of the oil price within this. Section 5
concludes.
2. Data
The bond market data has been retrieved from Barclays Capital. The markets, member of the
World Government Inflation-Linked Bond Index (WGILB), which have been issuing
inflation-linked bonds since at least a decade, have been retained. 1 They constitute, in June
2012, of nineteen Inflation-Linked Gilts issued in the United Kingdom, thirty-three Treasury
Inflation-Protected Securities (TIPS) in the United States, twenty-six Obligations
Assimilables du Trsor indexes sur linflation (OATi) in the Euro Area, five Treasury
Indexed Bonds in Australia, six Index-Linked Treasury Bonds in Sweden and six Real Return
Bonds in Canada.
For calculating the breakeven inflation rates, the Barclays Breakeven Comparator indices
have been used, which are nominal bond indices purposely designed to match the
characteristics of the WGILB members. This is to avoid that the breakeven rates which are
simply calculated as the nominal minus the real yield, are being distorted by rotations in the
yield curves. For each market and each bond type, aggregate yields are calculated by Barclays
over all the maturities in the index. We refer to James (2010) for more details on Barclays
calculus.
The crude oil Brent FOB US dollar price series, available via Datastream, is used as the oil
price.
The observation period runs over ten years from July 2002 to June 2012, tests being done on
monthly data, which corresponds to 120 observations. In order to define the crisis months, we
have measured market turbulence by means of the standard deviation of the weekly variation
in the breakeven rates over four weeks over all countries in the dataset. If this measure
1
Japan has not been retained for this reason. This country started issuing inflation-linked bonds in 2004 and has
suspended its program in 2008 until further notice. See http://www.mof.go.jp for press releases by the Ministry
of Finance.

exceeds twice its historical average, the market is deemed in crisis. Through this method, the
eight months from October 2008 to May 2009 have been labelled as such.
3. Fishers hypothesis revisited
In his seminal book Theory of Interest, Fisher (1930) hypothesized that the two components
of the nominal interest rate, the real rate and the inflation expectation, should be unrelated to
one another, this since they are driven by independent economic factors. In Cette and de Jong
(2008, 2013) we find that the respective bond components, observable since the issuance of
inflation-linked bonds, are not univocal on the matter. Correlations between real yield and
breakeven inflation variations measured locally country by country are close to zero, giving
indication that Fishers hypothesis holds. However, their cross-border correlations are
systematically positive, which indicates that it doesnt.
The deadlock can be broken by separating local and international price effects. Through a
standard regression analysis, we estimate worldwide common bond yield- and country-
specific movements. 2 Ignoring the small cross-correlation terms, the complete correlation
matrix between the 'RBY and 'BEIR over the various countries, given in Figure 1 (a), is
decomposed into a common (b) and an idiosyncratic (c) correlation matrix. This exercise is
carried out over the entire observation period from July 2002 to June 2012, in I, and over the
period barred the crisis months, in II. The eight crisis months have been defined in the data
section.
2
Common movements are obtained, both for RBY and BEIR, by regressing on time-fixed effects. The
idiosyncratic components are the residuals of the regressions. Consequently, the common correlation, in
matrix (b), is identical for all countries and cross-combinations.

Figure 1
Decomposition of the correlation matrix between 5%<DQG%(,5
Period: July 2002 to June 2012 monthly frequency
Data source: Barclays Capital. Calculations made by the authors.
I Entire ten-year observation period
Total correlation Common correlation Idiosyncratic correlation

'RBY 'RBY
es
es
in
in
at
at
ita
ita
ea
ea
St
St
lia
lia
Br
Br
en
en
da
da
Ar
Ar
ra
ra
d
d
ed
ed
at
at
na
na
ite
ite
ro
ro
st
st
re
re
Sw
Sw
Ca
Ca
Un
Un
Au
Eu
Au
Eu
G
G
Australia -0.0 -0.1 0.0 -0.0 0.2** -0.1 -0.4*** 0.1 0.2** 0.1 -0.0 0.1
Canada 0.3*** -0.4*** -0.0 0.0 0.3*** -0.0 0.2** -0.6*** 0.1 0.1 -0.1 0.2**
'BEIR
Euro Area 0.1 -0.3*** -0.5*** -0.3*** 0.2* -0.3*** = -0.1 + 0.2** 0.2*** -0.5*** -0.1 0.2** -0.0
Great Britain 0.0 -0.2* -0.1 -0.3*** 0.3*** -0.2* -0.2*** 0.2*** 0.1 -0.4*** 0.1 0.1
Sweden 0.3*** 0.0 -0.0 0.1 0.0 -0.0 0.0 0.2** -0.0 0.3*** -0.6*** 0.2**
United States 0.1 -0.3*** -0.2* -0.2* 0.2*** -0.3*** 0.2** -0.1 0.1 -0.0 0.5*** -0.6***
*** **
, and *: significant at, respectively, the 1%, 5% and 10% level - Using an asymptotic T-test with T=120.
(a)
(b )
(c )

0

Schematically 0
II Ten-year observation period barred the crisis months from October 2008 to May 2009
Total correlation Common correlation Idiosyncratic correlation

'RBY 'RBY
es
es
in
in
at
at
ita
ita
ea
ea
St
St
lia
lia
Br
Br
en
en
da
da
Ar
Ar
ra
ra
d
ed
ed
at
at
na
na
ite
ite
ro
ro
st
st
re
re
Sw
Sw
Ca
Un
Ca
Un
Au
Eu
Au
Eu
G
Australia 0.2*** -0.0 0.2* 0.3*** 0.4*** 0.2** -0.4*** -0.0 0.2* -0.0 0.1 0.2**
Canada 0.5*** -0.4*** 0.1 0.2** 0.3*** 0.1 0.2 -0.6*** 0.1 0.1 0.1 0.0
'BEIR
Euro Area 0.5*** -0.0 -0.3*** 0.2* 0.2** 0.1 = 0.2** + 0.2* 0.4*** -0.6*** 0.1 -0.0 0.1
Great Britain 0.3*** -0.1 0.1 0.0 0.3*** 0.1 -0.3*** 0.3*** 0.2* -0.4*** 0.1 0.1
Sweden 0.4*** 0.0 -0.0 0.2*** -0.0 0.1 0.1 0.2** -0.0 0.1 -0.6*** 0.1
United States 0.4*** -0.3*** 0.0 0.1 0.3*** -0.1 0.3*** -0.3*** 0.1 0.1 0.4*** -0.5***
*** **
, and *: significant at, respectively, the 1%, 5% and 10% level - Using an asymptotic T-test with T=112.
(a)
(b )
(c )

0

Schematically 0

It can be noted that (i) country-specific correlation is systematically negative, and (ii) the
correlation between the global RBY and BEIR movements is usually positive, yet negative in
the crisis. In our previous articles we explained these stylised facts. The negative idiosyncratic
correlation can be directly related to market-related events that distort bond prices. As soon as
an inflation-linked bond price is being distorted whilst not the nominal, the (differential)
breakeven inflation rate moves in exactly opposite direction as the real yield. Those local
market events are recurrent, resulting in systematically negative correlation over time. The
global correlation turning negative in the crisis months can be explained by the fact that in
this period the market events were concerted over the globe and distorted prices on a global
scale.
The market distortions are discussed in the finance literature (see Christensen et al., 2004, for
a survey). They are recognised to lead to a price premium; less attention is paid to their
influence on the correlation structure between bonds. A series of articles mention that
liquidity problems on the inflation-linked bond market are the main cause of the price
distortions (see for example Sack and Elsasser, 2004, Shen, 2006, DAmico et al., 2010, and
Grkaynak et al., 2010). Another series of articles points rather at the behaviour of investors.
Hesitance in taking on inflation risk makes prices fluctuate (see for example Hrdahl and
Tristani, 2007, on Euro Area data, Ejsing et al., 2007, or Emmons, 2000, on US data). A few
recent articles recognise both causes and estimate the respective price premiums
simultaneously (see Pflueger and Viceira, 2011, Haubrich et al., 2011, and Christensen and
Gillan, 2012).
The global market distortion in 2008-2009 is discussed in the literature as well. James (2010)
and Campbell et al. (2009) report massive flights to liquidity. Hu and Worah (2009) as well as
Bekaert and Wang (2010) mention that the bankruptcy of Lehman Brothers has added to the
turmoil, for it was the world leader in inflation-secured investment instruments. Pond (2012)
actually mentions that in this period the usual price relations were inverted. Figure 2 below
makes the situation clear. It shows that in the developed countries, the breakeven inflation
level dropped dramatically in late 2008, to renormalize in early 2009, back to pre-crisis levels.
The abnormal BEIR levels stem from a simultaneous decrease in the nominal yields and
increase in the real yields.
In the same Figure, the oil price is displayed which is remarkably synchronised with the
breakeven levels.

Figure 2
BEIR in the developed world (in %, left scale) and Brent value (in $ per Barrel, right
scale)
5 140
4 120
3 100
2 80
1 60
0 40
-1 20
US UK Australia Canada EUR Sweden Brent
4. Impact of the Brent
Regarding Figure 2, it seems relevant to take account of the oil price, and decompose the
common correlation between RBY and BEIR (matrix b in Figure 1) further in a Brent-induced
component (b 1 ) and a residual component (b 2 ). The Brent-induced correlation should be
negative. Oil being an important factor of inflation, it should be positively correlated to
LQIODWLRQH[SHFWDWLRQVDQGWKXVWKH%(,5VHHIRUH[DPSOH&KHQDQG'H*UHJRULRet
al., 2007). Meanwhile, the oil price has an opposite impact on the economic activity
HQJHQGHULQJ QHJDWLYH FRUUHODWLRQ ZLWK JURZWK DQG WKXV WKH 5%< VHH %DUVN\ DQd Kilian,
2004, and Cuado and Prez de Gracia, 2003).
The negative impact of oil on the economy may pass through two channels: a production cost
effect (an increase in the production costs decreases the output equilibrium level) and a
Mundell-Tobin effect, which is a behavioural effect (in reaction to an oil price rise households
increase their savings which lowers the output equilibrium level). Ang et al. (2008) find
(weak) evidence of the Mundell-Tobin effect in American bond data.

To integrate the Brent in our tests, we augment the regression equations that were used in the
decomposition discussed in previous section by a term that captures the country-common
reaction to the Brent (x t ). Thus, we estimate:
(1) yit E y xt J yt . I t H yit

common component
where y it are the bond yield variations ('RBY and 'BEIR) in country i over month t, x t is a
function of the Brent price variation, E y measures the sensitivities of the two yield variations
to the Brent, It are time dummies, J yt measures the non-Brent common variation and H yit are
the residuals, which are assumed to be identically and independently distributed, and
represent the idiosyncratic variation component of y i .
)RU ERWK 5%< DQG %(,5 WKH PRGHO LV HVWLPDWHG WKURXJK 2UGLQDU\ /HDVW 6TXDUHV RQ WKH
whole period in two steps. First the common- and idiosyncratic variation is split, and then the
common component is split further into a Brent and non-Brent sub-component. Best results
are obtained with a non-linear impact of the oil price. To capture nonlinearity, we mount the
oil price (P) log-returns to the power three, i.e. x t = ln (P t /P t-1 )3. We have deliberately kept
the specification and estimation of the model simple.
The second-step estimation results are given in Table 3. The sensitivities to the oil price have
the intuitive signs as commented above. Interestingly, the total effect of the oil price on the
nominal yield, which is by construction the sum, E 'NBY E 'RBY E 'BEIR , is non-
significantly different from zero. It is perhaps for this reason that there is little discussion in
the literature on the effect of oil on bonds. The introduction of inflation-linked bonds on the
capital markets has made this observable.
Table 3
Estimation results
The common component of y is regressed on the Brent price changes
Explanations in the text
y='BEIR y='RBY
Ey 8.07 -4.95
T-statistic 6.95 -3.90
R2 0.11 0.04
N obs. 120 120

The regression results are inserted into the correlation decomposition given in Figure 1. The
split of the common correlation matrix (b) is displayed in Figure 4. It can be seen that the
Brent-induced correlation (b 1 ) is negative and the ex-Brent correlation (b 2 ) positive.
Figure 4
Schematic decomposition of the correlation matrix between 'RBY and 'BEIR
I Entire ten-year observation period
(a) b (c)

> 0.1@ 0

b1 b2
-
> 0.2@ >0.1@ 0
II Ten-year observation period barred the crisis months from October 2008 to May 2009
(a) b (c)

>0.2@ 0

b1
b2
-
> 0.1@ >0.3@ 0
We find the three components constituting the correlation between RBY and BEIR to have
stable signs, yet the total correlation (the sum) to be unstable over time. The net sum depends
on the share of each component, which is time-varying.
5. Conclusion
We have shown that the breakeven inflation and real yields deduced from the developed bond
markets are adversely affected by two factors: price distortions due to market-related events
and oil price movements. Without the influence of those, their correlation is positive. This
finding contributes, we reckon, to a better understanding of the long-debated complex
interrelationship between inflation and economic growth prospects. The effect of oil on bond
prices has become measurable thanks to the emergence of inflation-linked securities on the
markets. The results fit in with macroeconomic theory. An oil price rise drives up inflation
and slows down economic growth.

References
Ang, A., G. Bekaert and M. Wei (2008): The Term Structure of Real Rates and Interest
Inflation, Journal of Finance 63, n 2, 797-849.
Barsky, R. and L. Killian (2004): Oil and the Macroeconomy since the 1970s, NBER
Working Paper 10855.
Bekaert G. and X. Wang (2010): Inflation risk and the inflation risk premium, Economic
Policy, 755-806.
Campbell, J. Y., R. J. Shiller and L. M. Viceira (2009): Understanding Inflation-Indexed

Bond Markets, NBER Working Paper Series, n 15014, May.
Cette, G. and M. de Jong (2008): The Rocky Ride of Breakeven Inflation Rates, Economics
Bulletin, Vol; 5, n 30, 1-8.
___________________ (2013): Breakeven inflation rates and their puzzling correlation

relationships, Applied Economics, 45:18, 2579-2585.
Chen, S. (2009): Oil Price Pass-Through into Inflation, Energy Economics 31, 126-133.
Christensen, I., F. Dion and C. Reid (2004): Real Return Bonds, Inflation Expectations and
the Break Even Inflation Rate, Bank of Canada Working Paper, 2004-43.
Christensen, J. and J. Gillan (2012): Could the U.S. Treasury Benefit from Issuing More
TIPS?, Federal Reserve Bank of San Francisco Working Paper, 2011-16.
Cuado, J. and F. Prez de Gracia (2003): Do oil price shocks matter? Evidence for some
European countries , Energy Economics 25, 137-154.
DAmico S., D. H. Kim and M. Wei (2010): Tips from TIPS: The informational content of
Treasury Inflation-Protected Security prices, FEDs finance and economics discussion series,
2010-19.
De Gregorio, J., O. Landerretche and C. Neilson (2007): Another Pass-Through Bites the
Dust? Central Bank of Chile Working Paper n 417.
Ejsing, J., J. A. Garcia and T. Werner (2007): The Term Structure of Euro Area Break Even
Inflation Rates, European Central Bank, Working Paper Series, n 830, November.
Emmons, W. (2000): The Information Content of Treasury Inflation-Indexed Securities,

Federal Reserve Bank of St Louis Review, November/December.
Fisher, I. (1930): The Theory of Interest, Macmillan CO., New-York.
Grkaynak, R. S., B. Sack and J. H. Wright (2010): The TIPS Yield Curve and Inflation
Compensation, American Economic Journal, Macroeconomics, 2:1, 70-92.

Haubrich, J., G. Pennacchi and P. Ritchken(2011): Inflation, Real Rates, Expectations and
Risk Premia: Evidence from Inflation Swaps, Federal Reserve Bank of Cleveland Working
Paper n 1107.
Hrdahl, P., and O. Tristani (2007): Inflation risk premia in the term structure of interest
rates, BIS Working Papers, n 228, May.
Hu, G. and M. Worah (2009): Why Tips Real Yields moved significantly higher after the
Lehman Bankruptcy, PIMCO, Newport Beach, CA.
James, A. (2010): Global Inflation-Linked Products A Users Guide, Introduction, in

Barclays Capital Research, March, 1-2.
Pflueger, C. and L. Viceira (2011): An Empirical Decomposition of Risk and Liquidity in

Nominal and Inlfation-Indexed Government Bonds, NBER Working Paper n 16892.
Pond, M. (2012): Beta Calculations, Drivers and Uses, Barclays Capital Research
document.
Sack, B. and R. Elsasser (2004): Treasury Inflation-Indexed Debt: A Review of the US

Experience, Federal Reserve Bank of New York, Economic Policy Review, May, 47-63.
Shen, P. (2006): Liquidity Risk Premia and Breakeven Inflation Rates, Economic Review,
Federal Reserve Bank of Kansas City, Second Quarter, 29-53.

WP-017
Breakeven Inflation
Rates and their Puzzling
Correlation Relationships
Gilbert Cette,
Associated Professor in Economics, Universit dA ix-Marseille II
Marielle de Jong,
December 2011
It is generally assumed that the two Fisher components of the

interest rate -the real interest and the inflation- evolve
independently over time, considering that they are driven by
unrelated economical events. However, the market pricing of those
components deduced from newly-available bond data does not
provide conclusive evidence. While studying the price behaviour
of inflation-linked (real) bonds beside nominal bonds in the major
fixed-income markets, we observe that the real bond yields and the
yield differentials, the breakeven inflation rates, have the propensity
to be positively correlated between each other across the various
countries, yet are pushed into a negative correlation relationship
due to market-related price distortions. As long as those distortions
are local, the net result is near-zero correlation within countries;
when they become global, as in the heat of the current crisis, the
correlations turn negative worldwide. In this paper insight is gained
by taking an innovative worldwide study approach and thanks to
revealing crisis period events.

1. Introduction
The yield of an inflation-linked bond, informally called the real bond yield (RBY), reflects the
market pricing of the long-term real interest rate in the same manner as the nominal bond
yields (NBY) price the nominal interest rate. The relatively recent issuance of inflation-
protected securities by governments around the world has made it possible, for the first time 1,
to pair up the two bond types and observe the yield differentials, called the breakeven
inflation rates (BEIR). The longstanding idea of decomposing interest rates into two
components, introduced by Irving Fisher in his seminal book named Theory of Interest
(1930), can eventually be tried now the new bond markets are maturing and becoming more
liquid.
Fisher had hypothesized that the two components should be unrelated to one another: the
real interest rate is entirely determined by the real factors in an economy, i.e. the productivity
of capital and the investors time preference, and should thus be unrelated to the inflation
expectation. Many efforts have been undertaken to provide empirical evidence; see Cooray
(2002) for a literature review. In Cette and de Jong (2008), we had made a renewed attempt
with tests on recent bond market data, observing that the Fisher hypothesis seems to hold
country per country yet is definitely rejected in an international context. Tests were based on a
historical correlation matrix measured between the two interest rate components, BEIR and
RBY, across various countries, featuring near zeros (more exactly, unsystematically negative,
nil or positive numbers) on the diagonal, i.e. within countries, and strictly positive numbers
elsewhere, i.e. between countries.
By means of standard statistical analysis, by which worldwide common bond yield

movements are separated from idiosyncratic movements within countries, we had succeeded
to disentangle two effects in the correlation structure. The schema is given in Figure 1 below.
The complete matrix, displayed in (a), turns out to be the net sum of a positive common
global correlation (b) and a negative idiosyncratic country-specific correlation (c). The
diagonal terms in (a) are the sum of a positive term (from matrix (b)) and a negative term
(from matrix (c)), and can consequently be nil, negative or positive.
1
The scale in which such bonds are being issued is new, not the concept. According to Shiller (2003) the first
inflation indexed bonds were issued by the Commonwealth of Massachusetts in 1780 during the Revolutionary
war to deal with severe wartime inflation.

Figure 1
Schematic decomposition of the correlation matrix between BEIR and RBY variation
across countries
total correlation
common correlation
idiosyncratic correlation

0 0

0 0
(a) (b) (c)
The finding is puzzling. The correlation structure, untypical for bonds and casting doubt on
the Fisher hypothesis, pulls all analyses traditionally made in a national perspective into
scrutiny.
In our previous article we had shown the correlation structure to be stable over time up to
mid-2008. In this article we study what happened after, in the current financial crisis, up to
mid-2010, during which bonds have been in great turmoil, as documented by many such as
Campbell et al. (2009). Our main findings are the following. While the idiosyncratic
correlation structure remained unchanged in the turmoil, the common correlation became
negative, resulting in a negative correlation matrix overall with significantly negative
coefficients on the diagonal. We explain this change by a worldwide lack of liquidity in the
real bond markets affecting the prices, and we observe that, as soon as this simultaneous
liquidity problem resolved in 2009, the bond prices settle back into the same regime as given
in Figure 1. Our international study approach remains original in the literature to our
knowledge, many other articles focusing on national indexed bond markets only. It proves
essential in gaining insight in the effects behind the interest rate component movements.
Section 2 discusses the data issues, section 3 and 4 give an analysis of the bond correlations
respectively before and since the crisis, and section 5 concludes.

2. Data issues
The data has been retrieved from Barclays Capital. Developed countries issuing inflation-
linked bonds since at least a decade have been retained. 2 It covers the Inflation-Linked Gilts
issued in the United Kingdom, the Treasury Inflation-Protected Securities (TIPS) in the
United States, les Obligations Assimilables du Trsor indexes sur linflation (OATi) in the
Euro Area, the Treasury Indexed Bonds in Australia, the Index-Linked Treasury Bonds in
Sweden and the Real Return Bonds in Canada. Generic bond yields have been calculated by
Barclays Capital per interval of maturity dates. The seven-to-ten-year term-to-maturity
interval with a bond-duration close to 7.5 years has been selected for this study, since it is by
and large the most liquid category. We refer to Barclays Global Inflation-Linked Products
edited by James (2010) for more details on their calculus.
The observation period, from mid-2002 to mid-2010 has been divided in three in order to
separate out the period of great market turbulence in 2008-2009. There is a pre-crisis period,
from July 2002 to September 2008, a turbulent period, from October 2008 to May 2009, and a
post-turbulent period, from June 2009 to June 2010. In order to set the cut-off dates we have
measured market turbulence by means of the standard deviation of the weekly variation in the
breakeven inflation rates over four weeks and over all countries in the dataset. If this measure
exceeds two times its historical average, the market is deemed turbulent.
3. The bond correlation structure before the financial crisis
The correlation matrix measured in Cette and de Jong (2008) between the major bond markets
over a six-year period from 2002 to 2008 is reprinted in Table 1.
2
Japan has not been retained for this reason.

Table 1
Correlation matrix between breakeven inflation variation and real bond yield variation
across countries
Period: July 2002 to September 2008 Weekly data frequency. 3
Data source: Barclays Capital. Correlation measures made by the authors.
RBY
Australia Canada Euro Area Great Britain Sweden United States
Australia 0.25*** 0.16*** 0.36*** 0.30*** 0.40*** 0.31***
Canada 0.34*** -0.14*** 0.38*** 0.28*** 0.29*** 0.39***
Euro Area 0.18*** 0.04 -0.11** 0.05 0.15** 0.11**

BEIR
Great Britain 0.15*** 0.02 0.17*** -0.26*** 0.25*** 0.17***
Sweden 0.29*** 0.16*** 0.43*** 0.34*** 0.13** 0.35***
United States 0.21*** 0.02 0.10* 0.13*** 0.15*** -0.07
***
: significant at the 1% level (critical value at 0.13); **: significant at the 5% level (critical value at 0.11);
*
: significant at the 10% level (critical value at 0.09) - Using an asymptotic T-test with T=325.
A dual phenomenon can be observed. The correlations are nil, or more precisely they are
unsystematically negative, nil or positive, as discussed in the introduction, within countries
(on the diagonal), while in contrast they are strictly positive between countries (the cross
terms). This is puzzling. In practical terms, it means that the yield variation of an American
TIPS, to take an example, is uncorrelated with the breakeven inflation movements in the US
(bottom right number in the matrix). It indicates that the bond price is insensitive to inflation
concerns, which is in effect the raison dtre of the security. Yet why would its yield correlate
with breakeven movements registered in other countries (the numbers in the rightmost
column)? The correlations do not strike with theory either. The Fisher hypothesis seems to
hold within countries in effect a joint test of zero correlation is not rejected even on a 10 %
error level-, however it is rejected, on a 1 % significance level, across countries.
3
On a monthly data frequency results are very similar.

2. Data issues
The data has been retrieved from Barclays Capital. Developed countries issuing inflation-
linked bonds since at least a decade have been retained. 2 It covers the Inflation-Linked Gilts
issued in the United Kingdom, the Treasury Inflation-Protected Securities (TIPS) in the
United States, les Obligations Assimilables du Trsor indexes sur linflation (OATi) in the
Euro Area, the Treasury Indexed Bonds in Australia, the Index-Linked Treasury Bonds in
Sweden and the Real Return Bonds in Canada. Generic bond yields have been calculated by
Barclays Capital per interval of maturity dates. The seven-to-ten-year term-to-maturity
interval with a bond-duration close to 7.5 years has been selected for this study, since it is by
and large the most liquid category. We refer to Barclays Global Inflation-Linked Products
edited by James (2010) for more details on their calculus.
The observation period, from mid-2002 to mid-2010 has been divided in three in order to
separate out the period of great market turbulence in 2008-2009. There is a pre-crisis period,
from July 2002 to September 2008, a turbulent period, from October 2008 to May 2009, and a
post-turbulent period, from June 2009 to June 2010. In order to set the cut-off dates we have
measured market turbulence by means of the standard deviation of the weekly variation in the
breakeven inflation rates over four weeks and over all countries in the dataset. If this measure
exceeds two times its historical average, the market is deemed turbulent.
3. The bond correlation structure before the financial crisis
The correlation matrix measured in Cette and de Jong (2008) between the major bond markets
over a six-year period from 2002 to 2008 is reprinted in Table 1.
2
Japan has not been retained for this reason.

Table 1
across countries
Period: July 2002 to September 2008 Weekly data frequency. 3
Data source: Barclays Capital. Correlation measures made by the authors.
RBY
Australia 0.25*** 0.16*** 0.36*** 0.30*** 0.40*** 0.31***
Canada 0.34*** -0.14*** 0.38*** 0.28*** 0.29*** 0.39***
Euro Area 0.18*** 0.04 -0.11** 0.05 0.15** 0.11**

BEIR
Great Britain 0.15*** 0.02 0.17*** -0.26*** 0.25*** 0.17***
Sweden 0.29*** 0.16*** 0.43*** 0.34*** 0.13** 0.35***
United States 0.21*** 0.02 0.10* 0.13*** 0.15*** -0.07
***
*
A dual phenomenon can be observed. The correlations are nil, or more precisely they are
unsystematically negative, nil or positive, as discussed in the introduction, within countries
(on the diagonal), while in contrast they are strictly positive between countries (the cross
terms). This is puzzling. In practical terms, it means that the yield variation of an American
TIPS, to take an example, is uncorrelated with the breakeven inflation movements in the US
(bottom right number in the matrix). It indicates that the bond price is insensitive to inflation
concerns, which is in effect the raison dtre of the security. Yet why would its yield correlate
with breakeven movements registered in other countries (the numbers in the rightmost
column)? The correlations do not strike with theory either. The Fisher hypothesis seems to
hold within countries in effect a joint test of zero correlation is not rejected even on a 10 %
error level-, however it is rejected, on a 1 % significance level, across countries.
3
On a monthly data frequency results are very similar.

An explanation can be given on the basis of an elementary matrix decomposition analysis.
The movements of the breakeven inflation and real bond yields are decomposed into a
common- and an idiosyncratic component per country. The common component of each
variable is extracted through a regression across countries with time-fixed effects as
explanatory variables; the idiosyncratic component is for each variable the residual of this
regression. Ignoring the (small) cross terms and ignoring that correlations dont exactly add, it
results, as shown in the schema in Figure 1, that the total correlation, displayed in Table 1,
turns out to be the net sum of two distinct effects 4:
(i) The common global component of the BEIRs tends to move in the same direction as
the common global component of the RBYs, resulting in a positive common
correlation matrix;
(ii) The idiosyncratic country-specific component of the BEIRs tends to move against the
idiosyncratic components of the RBYs within the same country, yet have no statistical
relation with those of other countries, resulting in a diagonal negative correlation
matrix.
We interpret the two effects separately. The negative idiosyncratic correlation can be directly
associated with certain time-varying market distortions that are being mentioned in the
literature; see Christensen et al. (2004) for a survey. As soon as the real bond price moves due
to such country-specific market-related events, which are typically not mirrored in the
nominal bond price, the breakeven rate mechanically moves in opposite direction. Those
events appear to be sufficiently recurrent to provoke a systematic idiosyncratic negative
correlation over time. Our observation complements the literature that analyses a price
premium on real bonds relative to nominal bonds resulting from those market distortions.
Abrupt market liquidity problems are mentioned to be the main cause of the price distortions,
for example by Craig (2003), Sack and Elsasser (2004), Shen (2006), DAmico et al. (2009),
Campbell et al., (2009) as well as Grkaynak et al. (2010). Grkaynak et al. (2010) relate the
liquidity premium, which is being observed in the inflation-linked bond prices compared to
the nominal bond prices, to the particularly low trading volumes for inflation-linked bonds. 5
They explain in a regression analysis the time-variability of the liquidity premium on TIPS
4
The detailed decomposition is available upon request from the authors.
5
Grkaynak et al. (2010) report that in the US, the TIPS market expressed as a share of total Treasury trading
represented about 0.5% in 1999 and 2% in 2006.

directly by the variations in the TIPS trading volumes. These market-related events being
usually country specific, explain the idiosyncratic negative correlation between the BEIR and
RBY.
Others observe a risk premium rather, which varies depending on the aversion to inflation
uncertainty among the market participants; see Hrdahl and Tristani (2007) on Euro Area
data, Emmons (2000), on US data, Evans (1998), on US and UK data, and Ct et al. (1996),
on Canadian data. Ejsing et al. (2007) show that the seasonality in consumer prices over the
year adds to the price discrepancy as well. Note that those price influences are also mainly
idiosyncratic.
The positive correlation between the global BEIR and RBY is less commented in the
literature, since it is not easily observable on a national level. In fact, market practitioners do
recognise its existence indirectly when they mention the beta effect (see Pond, 2008). It is the
observation that the RBY tends to move in the same direction as the NBY yet in smaller
amplitude. Note that in that situation, the BEIR mechanically moves in the same direction as
well, resulting in positive correlation. The market phenomenon is illustrated in Figure 2 on a
global scale. The weekly global RBY variation is set out on the Y-axis against the global
NBY variation in the same week on the X-axis. The dots cover the off-crisis weeks and the
triangles the crisis-period weeks. Note the R-squared of the regression line through the off-
crisis observations being significantly high.
Does this market observation match with usual macroeconomic theory? In theory the nominal
and real interest rates should move together with the same amplitude after a shock that has no
significant impact on inflation expectations (for example, at the first order, a technological
shock). Or otherwise, the nominal interest rate should move alone after a pure inflation shock.
Notice in Figure 2 that those events rarely occur in practice. The fact that most of the time the
net result of the two scenarios is observed means that the two types of shock very often
coincide or that somehow bond prices absorb the two types of shock simultaneously.

Figure 2
The beta effect between global nominal- and real bond yield movements
Period: July 2002 to June 2010
x: Weekly NBY and RBY variations registered from 07/2002 to 09/2008 and from 06/2009 to 06/2010
': Weekly NBY and RBY variations registered from 10/2008 to 05/2009, in the heat of the crisis
0.3
'RBY
0.2
0.1
'NBY
0.0
-0.3 -0.2 -0.1 0.0 0.1 0.2 0.3
-0.1
-0.2 DRBY = 0.64xDNBY

R2 = 0.82
-0.3
4. Bond market experience since the financial crisis
We investigate what happened during the current crisis period. The correlation matrix
between breakeven inflation and real bonds measured between September 2008 and May
2009 is given in Table 2. Over this period, the correlations are (i) strongly negative within
countries (on the diagonal), and (ii) weakly negative between countries (the cross terms). The
test of zero correlation is integrally rejected on a 1% significance level. Despite the admittedly
reduced reliability of the tests in non-stationary times, it may be concluded that the Fisher
hypothesis does not hold, even within countries.

Table 2
across countries
Period: October 2008 to May 2009 Weekly data frequency.
RBY
Australia -0.21 -0.31* -0.08 -0.23 -0.21 -0.22
Canada 0.12 -0.62*** 0.19 -0.26 0.23 -0.34**
Euro Area 0.04 -0.04 -0.70*** -0.38** 0.07 0.03

BEIR
Great Britain -0.11 -0.40*** -0.28* -0.68*** 0.20 -0.33**
Sweden 0.22 -0.01 0.06 0.06 -0.32* 0.06
United States 0.26 -0.38** 0.13 -0.12 0.18 -0.59***
***
*
The same elementary matrix decomposition produces the schema as displayed in Figure 3. 6
Note that the sole difference with the pre-crisis period lies in (b): the correlation between the
common global BEIR and RBY movements turns negative. The sign change is informative. It
is, to us, caused by the same issues of liquidity and risk attitude mentioned above. The
worldwide market distortions have provoked a negative common correlation between the
BEIR and the RBY, in the same way as they do on a national scale in normal times. In the
heat of the crisis, the inflation-linked bond markets contracted in all countries simultaneously.
Considering the relatively small trading volumes on these markets, the price shocks the
inflation-linked bonds incurred provoked a negative correlation with the breakeven inflation
rates in all countries. Any global price shock in the real bonds that is typically not registered
in the nominal bond markets leads to a mechanic opposite movement in the global breakeven
inflation.
6
The complete decomposition is omitted but can be obtained from the authors upon request.

Figure 3
Schematic decomposition of the correlation matrix in Table 2
total
correlation
common correlation
idiosyncratic correlation

0

0
(a) (b) (c)
This interpretation fits in with the finance literature reporting on the crisis events: there was a
massive flight to liquidity. James (2010) wrote: the extreme deleveraging phase that
engulfed almost all financial markets included the majority of off-benchmark investors in
inflation-linked bonds being stopped out of their positions. The bankruptcy of Lehman
Brothers added to the turmoil for it was the world leader on inflation-secured instruments (see
Hu and Worah, 2009, or Bekaert and Wang, 2010). Simultaneously in the economic literature,
the near-meltdown of the financial sector was seen as the start of an extended low-growth
period, with lower inflation than expected before the crisis. The relevance of inflation
issuance by inflation-linked bonds decreased, reducing its demand. The price fall of
commodities -in particular petrol- was reinforcing this view.
The flight to the mature nominal bond markets, which penalized TIPS demand and increased
their risk premium in 2008, as show Grkaynak, Sack and Wright (2010), was worldwide, its
factors being themselves worldwide, as mentioned before. It is shown in Figure 4 that not
only in the US, but in all developed world countries, the BEIR decreased dramatically from in
the last quarter of 2008, to normalize after in the first half of 2009, back to pre-crisis levels.
The abnormal BEIR levels in 2008 and 2009 stem from a decrease in the nominal rates as
well as an increase in the real rates. These global co-movements, that are a good illustration of
the generalized flight-to-liquidity behaviour, explain the common negative co-movements of
the BEIR and the RBY, which explain on its turn why the common correlation matrix exhibits
negative numbers during this short period.

Figure 4
The 10-year breakeven inflation rates in the developed world
Period: July 2002 to June 2010.
Data source: Barclays Capital.
From June 2009 when markets calmed down, the correlation matrix settled back into the same
regime as before (see Table 3), and its decomposition matches with Figure 1. This shows, or
gives strong indication, that the untypical correlation structure between BEIR and RBYs is
nevertheless robust.

Table 3
Period: June 2009 to June 2010 Weekly data frequency.
RBY
Australia 0.24* -0.06 0.37*** 0.18 0.24** 0.22*
Canada 0.42*** -0.34*** 0.19 0.23* 0.49*** 0.19
Euro Area 0.20* -0.07 -0.14 0.27** 0.42*** 0.15

BEIR
Great Britain 0.25* -0.15 0.04 -0.31** 0.16 0.04
Sweden 0.36*** 0.08 0.44*** 0.50*** 0.24** 0.17
United States 0.40*** 0.10 0.48*** 0.42*** 0.54*** -0.02
***
*
5. Conclusion
The events on the capital markets during the present crisis provide new insight in the price
covariance structure of bonds. The two components of the nominal bond yield, the breakeven
inflation and the real bond yield, have the propensity to be positively correlated, yet are
pushed into a negative correlation relationship by certain market events. As long as those
market distortions are local, the net result is near-zero correlation within countries; when they
become global, as was the case in the heat of the crisis, the correlation between real bond
yields and breakeven rates turns negative worldwide. Those empirical findings have been
shown to be robust over the pre- and post-crisis period.
This untypical correlation behaviour has been left uncommented in the literature, the reason
for that being that studies on inflation-linked bonds are traditionally made in a national
context, while an international analysis is necessary to reveal the underlying effects. We
consider the demonstration of the correlation relationships an important contribution to the

finance literature; its understanding is essential in bond risk analysis, in particular in
establishing the risk profile of fixed income portfolios.
The results are a contribution to the economic literature as well, the more that they are not in
line with conventional theory of Fisher (1930) postulating zero correlation between the two
interest rate components. The discrepancy between macroeconomic theory and financial
practice has been indirectly reported by central bankers, e.g. in Bernankes (2004) speech
What Policymakers Can Learn from Asset Prices. The issuance of inflation-linked bonds had
in part been motivated by the expectation that the observed breakeven inflation would in some
way reflect the credibility granted to Central Banks regarding their control on inflation.
Bernanke reckons the volatility of the breakeven inflation too high, to the extent that the
market data remains as it stands of very limited use for policymaking purposes.

Acknowledgements
The authors thank Hubert Kempf and Jean-Paul Renne for their valuable comments.
References
Bernanke, B., (2004): What Policymakers Can Learn from Asset Prices, The Federal
Reserve Board.
Bekaert G. and X. Wang (2010): Inflation risk and the inflation risk premium, Economic
Policy, 2010, pp. 755-806.
Campbell, J. Y., R. J. Shiller and L. M. Viceira (2009): Understanding Inflation-Indexed

Bond Markets, NBER Working Paper Series, n 15014, May.
Cette, G. and M. de Jong (2008): The Rocky Ride of Breakeven Inflation Rates, Economics
Bulletin, Vol; 5, n 30, pp. 1-8.
Christensen, I., F. Dion and C. Reid (2004): Real Return Bonds, Inflation Expectations and
the Break Even Inflation Rate, Bank of Canada Working Paper, 2004-43.
Cooray, A., (2002): The Fisher Effect: a Review of the Literature, Research paper n 206,
department of economics, Macquarie University.
Ct, A., J. Jacob, J. Nelmes and M. Whittingham (1996): Inflation Expectations and Real
Return Bonds, Bank of Canada Review, Summer, pp. 41-53.
Craig, B. (2003): Why are TIIS Yields so High? The case of the Missing Inflation-Risk
Premium, Federal Reserve Bank of Cleveland Economic Commentary.
DAmico S., D. H. Kim and M. Wei (2009): Tips from TIPS: The informational content of
Treasury Inflation-Protected Security prices, mimeo, December, 29.
Ejsing, J., J. A. Garcia and T. Werner (2007): The Term Structure of Euro Area Break Even
Inflation Rates, European Central Bank, Working Paper Series, n 830, November.
Emmons, W. (2000): The Information Content of Treasury Inflation-Indexed Securities,

Federal Reserve Bank of St Louis Review, November/December.
Evans, M. (1998): Real Rates, Expected Inflation and Inflation Risk Premia, Journal of
Finance, pp. 187-218.
Grkaynak, R. S., B. Sack and J. H. Wright (2010): The TIPS Yield Curve and Inflation
Compensation, American Economic Journal, Macroeconomics, 2:1, pp. 70-92.

Hrdahl, P., and O. Tristani (2007): Inflation risk premia in the term structure of interest
rates, BIS Working Papers, n 228, May.
Hu, G. and M. Worah (2009): Why Tips Real Yields moved significantly higher after the
Lehman Bankruptcy, PIMCO, Newport Beach, CA.
Hunter, D. and D. Simon (2005): Are TIPS the real deal? A conditional assessment of their
role in a nominal portfolio, Journal of Banking and Finance, 29 (2005), pp. 347-368.
James, A. (2010): Global Inflation-Linked Products A Users Guide, Introduction, in

Barclays Capital Research, March, pp. 1-2.
Pond, M. (2008): Effective duration of linkers and beta, in Barclays Capital Research:
Global Inflation-Linked Products, February.
Sack, B. and R. Elsasser (2004): Treasury Inflation-Indexed Debt: A Review of the US

Experience, Federal Reserve Bank of New York, Economic Policy Review, May, pp. 47-63.
Shen, P. (2006): Liquidity Risk Premia and Breakeven Inflation Rates, Economic Review,
Federal Reserve Bank of Kansas City, Second Quarter, pp. 29-53.
Shiller, R. (2003): The invention of inflation-indexed bonds in early America, Cowles

Foundation Discussion Paper, n 1442, October.

About the authors
FLORIAN BERG, Research Analyst Fixed Income Quantitative Research
Amundi, PhD Candidate at Paris Dauphine University
Florian Berg divides his time between the fixed-income quantitative research team
at Amundi and Dauphine University in Paris where he conducts research for his
doctoral thesis. He obtained a BA in applied economics and an MSc in financial
engineering from Dauphine University. His current research focuses on the
implementation of socially responsible investment strategies in bond portfolios.
Moreover, he lectures macroeconomics at Paris Dauphine and Sciences-Po Paris.
DP-03, March 2014, p. 51
TEGWEN LE BERTHE, CFA, Research Analyst Equity Quantitative

Research, Amundi
Tegwen Le Berthe joined Amundi in 2008 as an SRI analyst within the SRI Analysis
team. Since 2011 he is working as a quantitative analyst within the equity quantitative
research team. Prior to joining the company he worked as an SRI analyst at Inspire
Invest (2007-2008).
He graduated from a French engineering school, ENST Bretagne, and has a
specialized master in management of sustainable development from HEC School of
Management. He is a CFA charterholder since 2013.
DP-03, March 2014, p. 51
ZVI BODIE is the Norman and Adele Barron Professor of Management

at Boston University
He holds a PhD from the Massachusetts Institute of Technology and has served
on the finance faculty at the Harvard Business School and MITs Sloan School
of Management. Professor Bodie has published widely on pension finance and
investment strategy in leading professional journals. His books include The Future
of Life Cycle Saving and Investing and Foundations of Pension Finance.
His textbook, Investments, coauthored by Alex Kane and Alan Marcus is the market
leader and is used in the certification programs of the CFA Institute and the Society
of Actuaries. His textbook Financial Economics is coauthored by Nobel Prize
winning economist, Robert C. Merton. His latest book is Worry Free Investing: A
Safe Approach to Achieving Your Lifetime Financial Goals. In 2007 the Retirement
Income Industry Association gave him their Lifetime Achievement in Applied
Retirement Research Award.
WP-039, October 2013, p. 159

LING-NI BOON, PhD Candidate at Tilburg University and University Paris
Dauphine; Research Analyst at the Investor Research Center
at Amundi
Ling-Ni Boon is a PhD in Finance candidate at both University Paris Dauphine and
Tilburg University. Prior to beginning her doctoral studies, she completed a Master in
Quantitative Finance at the University of Paris 1, Panthon-Sorbonne. She received
her undergraduate degree from McGill University in Canada. Her PhD is pursued in
collaboration with Amundi, where she works as a research analyst. Additionally, she is
a junior research fellow of the Network for Studies on Pensions, Aging and Retirement
(Netspar).
WP-035, April 2013, p. 241
MARIE BRIRE, PhD, is Head of Investor Research Center at Amundi in Paris.

She is also Associate Professor with Paris Dauphine University and Associate
Researcher with the Centre Emile Bernheim at Solvay Business School, Universit Libre
de Bruxelles. Since 2012, she is the new Editor of the scientific Journal Bankers Markets
and Investors. Marie Brire started her working career in 1998 as a quantitative
researcher at the proprietary trading desk at BNP Paribas. She joined Credit Lyonnais
Asset Management in 2002 as a fixed income strategist, then a Head of Fixed Income,
Forex and Volatility Strategy at Credit Agricole Asset Management. She has taught in
several universities and business schools. Since 2012 she teaches empirical finance,
portfolio management and quantitative investment strategies at Paris Dauphine
University. Marie Brire is the author of a book on anomalies in the formation of interest
rates, and a number of her scientific articles have been published in books and leading
academic journals including the Financial Analyst Journal, Journal of Banking and
Finance, Journal of Money, Credit and Banking, Journal of Portfolio Management, etc.
WP-039, October 2013, p. 159
GILBERT CETTE, Associated Professor in Economics Universit dAix-

Marseille II
Gilbert Cette is Director of the Microeconomic and Structural Analysis Department
at the Banque de France and Associate Professor in Economics at the University
of Aix-Marseille (GREQAM). He holds a PhD in Economics from the University of
Paris I (1991). His fields of research include Growth, Productivity, Employment,
Labour Market Equilibrium, and Macroeconomic Modelling, fields in which he
has extensive track of publications. He intervenes and is considered a reference
in French policy debates. He is author of the recent French bestseller Changer de
Modle co-written with Philippe Aghion and Elie Cohen.
WP-030, October 2012, p. 341
WP-017, December 2011, p. 353

BASTIEN DRUT, PhD, Fixed Income, Forex Strategist
Bastien Drut joined Amundi in 2008 as strategist within the Strategy and Economic
Research team.
He holds a degree from the Ecole Centrale de Lyon (2008), ENSAE (2008) and Paris
School of Economics (2008). He also holds a PhD in economics from the Universit
Libre de Bruxelles and Universit Paris Ouest Nanterre La Dfense (2011). He is
the author of several academic articles published in refereed journals ( Journal of
Business Ethics, Finance) and of two books.
WP-009, April 2010, Revised: January 2012, p. 273
FLORIAN IELPO, PhD in Econometrics, University of Paris 1,

Panthon-Sorbonne; Portfolio Manager
Florian Ielpo completed his PhD in econometrics at the University of Paris 1,
Panthon-Sorbonne while working as an economist in the banking industry. He
then occupied various positions: an econometrician for a fixed income department
in an asset management institution and a strategist for a global macro hedge fund.
He now works as a portfolio manager in Switzerland.
At the same time, he is an associate researcher at the Centre dEconomie de la
Sorbonne in Paris, France. He teaches various aspects of applied finance at the
Sorbonne and Dauphine Universities and at the Ecole Nationale des Techniques
Avances (ENSTA) in Paris, France. His peer-reviewed scientific publications can be
found in various journals such as Quantitative Finance, the Journal of Forecasting,
Finance Research Letters or the Journal of Investing.
WP-035, April 2013, p. 241
MARIELLE DE JONG, Head of Fixed Income Quantitative Research, Paris

Marielle de Jong is Head of the Fixed-Income Quant Research team at Amundi Asset
Management in Paris. She graduated in econometrics at the Erasmus University in
Rotterdam, did an MSc in operational research in Cambridge UK, and holds a PhD
in finance from the University of Aix-Marseille (2010).
She started her working career in London in 1994 as an equity fund manager, moved
to Paris in 1997 where she joined the quantitative research team at Sinopia (HSBC)
and then at Amundi in 2011. She has published a series of articles on equities, on
currencies, and recently on fixed-income risk modelling issues.
WP-022, June 2012, p. 327
WP-30, October 2012, p. 341
WP-017, December 2011, p. 353

SYLVIE DE LAGUICHE, Head of Quantitative Research, Amundi
Sylvie de Laguiche has been heading the quantitative research team at Amundi
(previously Credit Agricole Asset Management) since 2005. Prior to that, she started
her professional career as a consultant at Associs en Finance in Paris. She then
served as a quantitative analyst at Indosuez Asset Management between 1991
and 1997 and was then appointed Head of the fixed income and asset allocation
quantitative research team at Credit Agricole Asset Management in 1997.
Sylvie de Laguiche holds a PhD in Mathematics and Finance from Paris Dauphine
University. She was a student at the Ecole Normale Suprieure in Paris. She has
given some lectures at Paris Dauphine University and HEC and is a board member of
Inquire Europe. She is the author of several papers on fixed income and quantitative
research.
DP-03, March 2014, p. 51
DP-02, March 2014, p. 93
WP-032, November 2012, p. 177
DIDIER MAILLARD, Professor at CNAM, Senior Advisor on Research to Amundi

Didier MAILLARD is Senior Advisor to Amundi on Research. He his since 2001
Professor at Conservatoire national des arts et metiers (CNAM), where he holds
a Chair of Banking. Previously, he has been an economist at the French Ministry
of Finance and at the OECD (1980-1992) economic forecasts, economic policy,
public finance, tax studies, financial sector, and has occupied various positions at
Paribas (and then BNP Paribas) from 1992 to 2001: chief economist, head of asset
management, risk advisor. He is a graduate from Ecole polytechnique (Paris) and
Ecole nationale dadministration.
His main fields are portfolio optimization, asset management, wealth management
and tax incidence (in particular on investment return).
WP-018, September 2012, p. 307
GIANNI POLA, PhD, Balanced Quantitative Research, Amundi

Gianni Pola joined Amundi SGR (previously Credit Agricole Asset Management SGR)
in 2005 and the Balanced Quantitative Research team in June 2012 as a quantitative
analyst. He started his professional career in 2005 as quantitative analyst at
Nextra Investment Management (Banca Intesa, Milan). Previously, he collaborated
with the Universities of Manchester UMIST (UK), Newcastle upon Tyne (UK), and
LAquila (Italy) on research topics related to neural coding, econophysics, and
robust optimization. He currently lectures in Quantitative Finance at the School of
Management MIP at the Milan Polytechnic (since 2011).
He holds a PhD in Computational Neuroscience (2004) from the University of

Newcastle upon Tyne and a first class honours degree in Theoretical Physics (2000)
from the University of LAquila (Italy). In 2000 he was INFN (National Institute
for Nuclear Physics) fellow at LNGS (National Laboratory at Gran Sasso, Italy). He
has written several articles on peer review journals, book chapters, conference
proceedings and working papers on portfolio construction, diversification, optimal
dynamic asset allocation, expected returns, computational models for time.
WP-032, November 2012, p. 177
WP-034, May 2013, p. 109
ALESSANDRO RUSSO, CFA, Head of Equity Quantitative Research, Paris

Alessandro Russo has been heading the Equity Quantitative Research team at
Amundi since 2007.
He started his professional career in 2001, as an Equity Risk Manager at Nextra
Investment Management (Intesa Group), Milan. Still at Nextra SGR he has served as
a quantitative Fund Manager from 2004 to 2005, when he has been appointed Head
of the Quantitative Research team at the Italian branch of Credit Agricole Asset
Management SGR. Alessandro Russo has been a CFA Charterholder since August
2005 and holds a Degree in Economics of Financial Markets and Institutions from
Bocconi University, Milan.
DP-04, May 2014, p. 9
DP-03, March 2014, p. 51
WP-033, March 2013, p. 211
ANTOINE SORANGE, Head of SRI analysis of Amundi

Antoine Sorange joined Amundi in 2000 as a Quantitative analyst. He then joined the
SRI Expertise Department as an SRI Equity portfolio manager in 2004. He notably
managed Amundi Actions Euro ISR and launched two SRI thematic funds, Aqua
Global and Clean Planet, respectively in 2006 and 2008. He then became head of
sustainable research in October 2008. Composed of 10 extra-financial analysts, this
team is responsible for the SRI analysis and the ESG integration of the whole Amundi
Group that manages more than 70 billion euros SRI portfolio as of December 2013.
Antoine Sorange is a member of the Eurosif Lobby Advisory Group as well as the SRI
commission of the French Asset Management Association AFG
Antoine Sorange holds a Master Degree in Fundamental Physics as well as a Master
in Finance.
DP-03, March 2014, p. 51

About the editors

PASCAL BLANQU
Deputy Chief Executive Officer,

Head of Institutional Investors and Third Party Distributors,
Group Chief Investment Officer
Pascal Blanqu has been Global Chief Investment Officer at AMUNDI (formerly
known as Crdit Agricole Asset Management, CAAM), Head of Institutional
Investors and Third Party Distributors and member of the Executive Committee
since February 2005.
He was also appointed Chairman of the Board of CPR AM in January 2007. From
2000 to 2005 he was Head of Economic Research and Chief Economist of Crdit
Agricole. Before joining Crdit Agricole, Pascal Blanqu was Deputy Director of the
Economic Research Department at Paribas (1997-2000) following four years as a
strategist at Paribas Asset Management in London (1992- 1996). He began his career
in institutional and private asset management at Paribas in 1991.
As an economist and a financial historian Pascal Blanqu is the author of several
contributions. His research interests and his academic work focus on monetary
issues, the functioning of financial markets and the philosophical foundations of
economics. His books include Money, Memory and Asset Prices, The Social Economy of
Freedom, Philosophy in Economics and Essays on Positive Investment Management,
all published and distributed by Economica and Brookings Institution Press.
His publications in international Journals and specialized newspapers include
numerous articles concerning financial history, economics and policy making. He
taught, gave lectures or carried out research at Ecole Normale Suprieure, Ecole
Polytechnique and Paris-Dauphine University. He is a member of the French Socit
dEconomie Politique.
Pascal Blanqu was awarded European CIO of the Year 2013 by Funds Europe
Magazine.
Born in 1964, Pascal Blanqu studied at the Ecole Normale Suprieure. He is a
graduate of Pariss Institut dEtudes Politiques and holds a Phd in Finance from
Paris, Dauphine University.

PHILIPPE ITHURBIDE
Global Head of Research, Strategy and Analysis

Philippe Ithurbide joined Amundi as Global Head of Research, Analysis and Strategy
in July 2010.
From 2006 to 2010, he has worked at Caisse de Dpt et Placement du Qubec
(Montral). A member of the Executive Committee and of the asset allocation
committee he was, in turn, Executive Vice President in charge of fixed income,
currency and credit from 2007 to 2009, and of multi-asset class overlay strategy
subsequently. He was previously in charge of strategic investment in the fixed income
and credit division. Prior to that, he worked for Socit Gnrale (SGCIB) from 1991
to 2006 as head of research in foreign exchange, fixed income and commodities,
having joined as chief economist. Member of the board of SGs pension fund (2000-
2006), he was in charge of asset allocation. In 1989 he was appointed chief economist
at the Manufacturers Hanover Bank France after serving in the same role at the
Banque Franaise de Commerce Extrieur.
Philippe Ithurbide started his career as teacher/researcher at the University
of Bordeaux, Laboratoire dAnalyse et de Recherche Economiques (part of the
Centre National de la Recherche Scientifique, CNRS) in 1982. For more than
20 years he taught (or gave lectures) at several universities, in France (in Bordeaux
in particular), in Spain, in Colombia, in the US, and more recently, between 2000
and 2005, at HEC Paris. He published theoretical papers on international finance
(topics such as Rational bubbles and gold market, Voluntary exports restraints: a
general equilibrium model, Pension funds: the role for Indexed linked bonds in an
Asset-Liabilities Management perspective) and books such as The economic status
of gold, Protectionism and exchange rate regime, or French firms and foreign exchange
risk: dollar versus EMS.
Born in 1958, Philippe Ithurbide holds a PhD in International Economics and Finance
from the University of Bordeaux. He is member of the AMFs Scientific Advisory
Board (the French regulator), and member of the board of OEE (LObservatoire de
lEpargne Europenne) and of IODS (INSEAD OEE Data Services).

Research publications
At May 2014

Cross Asset Investment Strategy Articles
Subjects Contributor (s) Date
The bank stress test methodology should ITHURBIDE Philippe 2014-05

not be a game changer. In fact, we would
be tempted to say the opposite
Economic divergence vs financial ITHURBIDE Philippe 2014-05

defragmentation. Monetary policy issues
confronting the eurozone
Much ado about not much: the ECBs call DOISY Nicolas, 2014-05
for a negative rate DRUT Bastien
The Feds normalisation dilemma: what DOISY Nicolas 2014-05

does normal mean? Choosing between
two evils: inflation vs. unemployment
How vulnerable to deflation is the DRUT Bastien 2014-05

eurozone?
Eurozone: are nominal and real BOROWSKI Didier 2014-05

convergence compatible?
The premium offered by peripheral AINOUZ Valentine 2014-05

markets is increasingly low
The Indian rupee is attractive for several DRUT Bastien 2014-05

reasons
Stars in the first quarter still untapped BERTONCINI Sergio, 2014-05

potential in Italian equities MIJOT Eric, WANE Ibra
Central banks support the growth outlook ITHURBIDE Philippe 2014-04
The good, the bad and the ugly: DOISY Nicolas 2014-04
deflationary risks in America and Europe
Emerging economies: it is time to revert BEN ABDALLAH Marc-Ali, 2014-04

to simple segmentation principles MOUSSAVI Julien
China: the big banks are not defenceless BEN ABDALLAH Marc-Ali 2014-04
against rising credit risk
Peripheral bonds outperforming: DRUT Bastien 2014-04

yes, but
HY spreads have entered areas of BERTONCINI Sergio 2014-04

historically tight levels: how low can they
fall?

A strong euro: the root causes of the BOROWSKI Didier 2014-04
problem
Emerging market assets remain under WANE Ibra 2014-04

pressure
Automotive equipment suppliers: LABIA Frdric 2014-04

the new contenders
The Chinese central bank modifies its ITHURBIDE Philippe 2014-03

exchange-rate policy
What affect will this have?
2014 earnings: beware the depreciation WANE Ibra 2014-03

of emerging market currencies!
Emerging debt markets: time for reason BEN ABDALLAH Marc-Ali 2014-03
A recovery driven by domestic demand BOROWSKI Didier 2014-03

in the major advanced economies
United Kingdom : the recovery remains PERRIER Tristan 2014-03

strong while becoming more broad-based
Where is there still yield left in European BERTONCINI Sergio 2014-03

fixed income markets?
The euro hybrid non-financial debt market: AINOUZ Valentine 2014-03

an attractive alternative
Australia and Canada: rebalancing DRUT Bastien 2014-03

economies
Equity markets: should we fear a new MIJOT Eric 2014-03

bubble?
Cloud Computing, or how history VU NGOC Van 2014-03

repeats itself
Emerging Markets in Turmoil: what are the ITHURBIDE Philippe 2014-02

implications for asset allocation?
A protracted crisis in the emerging ITHURBIDE Philippe 2014-02

markets? What implications for advanced
economies?
Emerging markets: central banks are BEN ABDALLAH Marc-Ali 2014-02

opting for strong-arm tactics
Eastern Europes currencies on diverging DRUT Bastien, 2014-02

paths KWOK James

What could (will?) the ECB do in 2014? ITHURBIDE Philippe 2014-02
2014 HY corporate bond default outlook BERTONCINI Sergio 2014-02
Mergers & Acquisitions: their recovery WANE Ibra 2014-02

should bolster the equity market in 2014
Will 2014 be the year for more M&A LEOPOLD Marie-Hlne 2014-02
in the pharmaceutical sector?
The Japanese machinery sector: KOEDA Yoshinori 2014-02

the sun rises again
Scenarios, risk factors and asset allocation ITHURBIDE Philippe 2014-01

for 2014
United States: a cyclical recovery BOROWSKI Didier 2014-01

with traditional aspects
Eurozone: Towards positive, yet slow PERRIER Tristan 2014-01

and uneven growth in 2014
Japan at the crossroads BOROWSKI Didier 2014-01
European political risk in 2014: DOISY Nicolas, 2014-01

Eurosceptic barbarians at the gate? VICQUERY Roger
Emerging economies: individual stories LEBOUGRE Catherine 2014-01

now and in future
Monetary policies in the advanced ITHURBIDE Philippe 2014-01

countries: the main risk factor,
and wide divergences between countries
2014: another positive year for equities MIJOT Eric 2014-01
The rebound in equity flows looks WANE Ibra 2014-01

set to continue in 2014
Bond markets: DRUT Bastien 2014-01

rising rates continue in 2014
Credit markets: BERTONCINI Sergio, 2014-01

the search for high yield is going on AINOUZ Valentine
A stronger US dollar in 2014 DRUT Bastien 2014-01
Emerging debt: BEN ABDALLAH Marc-Ali 2014-01

a year of greater differentiation
ECB sending ever clearer messages ITHURBIDE Philippe 2013-12

Does the eurozone still hold some appeal? ITHURBIDE Philippe 2013-12
A few genuine strengths and somewell-
identified risk factors
Equity markets: a snapshot of long-term MIJOT Eric, 2013-12

prospects GEORGES Delphine
Germany, source of deflation? BOROWSKI Didier 2013-12
What determines the euro-dollar? DOISY Nicolas, 2013-12

Essentially the ECBs (deflationary) stance DRUT Bastien
US and European credit cycles AINOUZ Valentine 2013-12

are diverging
Refunding risk of US and European High BERTONCINI Sergio 2013-12

Yield companies in next two years
Refinancing mortgage loans in Denmark: JEANNIARD Rmi 2013-12

an envied but somewhat ambiguous model
The reinvention of electronic components VU NGOC Van 2013-12
US government shutdown and debt ceiling: ITHURBIDE Philippe 2013-11

another reason to favour European assets
Debt, deficits and governance. ITHURBIDE Philippe 2013-11

Diversifying outside of the United States:
simple common sense
Is the European banking sector still DE BRAY Yasmine, 2013-11

vulnerable? LAPEYRE Thomas
Italy and Spain: two different models BOROWSKI Didier 2013-11

but the same challenge of restoring growth
How far can public debt financing DRUT Bastien 2013-11

by banks in Spain and Italy go?
The residential housing market PERRIER Tristan 2013-11

and economic growth: where we stand
in the four largest eurozone countries
High Yield bonds, the major outperformers BERTONCINI Sergio, 2013-11

of European fixed-income markets AINOUZ Valentine
Defence: up in arms LENOIR Francine 2013-11
The Fed breathes new life into the markets ITHURBIDE Philippe 2013-10
European equity markets: a sleeping GEORGES Delphine, 2013-10

beauty waiting to awaken MIJOT Eric, WANE Ibra

Emerging markets: strong drawing power LEBOUGRE Catherine 2013-10
Spains brave new world: DOISY Nicolas 2013-10

anatomy of a lost decade in Europe
France: Market complacency or excessive ITHURBIDE Philippe 2013-10

bashing?
Germany: domestic demand should PERRIER Tristan 2013-10

support growth in 2013-2014, but significant
medium-term challenges remain
Fed will taper IF BOROWSKI Didier 2013-10
The US yield curve: bear flattening DRUT Bastien 2013-10

in early 2014?
What are the best opportunities still AINOUZ Valentine 2013-10

available on the European IG market?
Electronic cigarettes: MARTIN Corinne 2013-10

where theres no smoke, theres fire
Telecoms: M&A keeping them trim DETERCK Virginie 2013-10
Emerging markets: a gathering storm, ITHURBIDE Philippe, 2013-09

or merely a correction? BEN ABDALLAH Marc-Ali
Emerging debt: an overview of its main BEN ABDALLAH Marc-Ali 2013-09

characteristics and outlook
Germanys European strategy is unlikely DOISY Nicolas 2013-09

to soften much after the election
Eurozone growth rebound: A real recovery BOROWSKI Didier 2013-09

or a flash in the pan?
Japan: for whom the bell tolls? YOSHINO Akio 2013-09
Could unconventional monetary policy ITHURBIDE Philippe 2013-09

become conventional? The case of Japan
and Japanese debt
The link between credit spreads and bond BERTONCINI Sergio 2013-09
yields: is this time different?
Earnings growth likely to surprise WANE Ibra 2013-09

to the upside in 2014, particularly
in the eurozone and Japan
Norwegian krone and Swedish krona DRUT Bastien 2013-09

will appreciate against the euro

E-commerce: undeniably a commercial CHAZELLE Jacques 2013-09
success story, but not a financial success
yet, particularly in food segment
Excess liquidity, QE, deleveraging and ITHURBIDE Philippe 2013-07

debt deflation: the reality is very different
by economic zone: what are
the consequences?
Where will eurozone long rates come ITHURBIDE Philippe, 2013-07

to rest? DRUT Bastien
US, Eurozone, UK, Japan: what lessons BOROWSKI Didier 2013-07

can be drawn from the macro-economic
adjustments carried out since 2008 ?
Scandinavian countries: diversely PERRIER Tristan 2013-07

impacted by the crisis, but fundamentals
remain robust
A pyrrhic victory for the US dollar in H2 DRUT Bastien 2013-07

2013?
Commodities: a bad semester LAMBINET Rmy 2013-07

in perspective
Will a major change in the Feds monetary MIJOT Eric 2013-07

policy cast doubt on the equity cycle?
The US and European high-yield market BERTONCINI Sergio, 2013-07

AINOUZ Valentine
Is the party over for German auto LABIA Frdric 2013-07

manufacturers?
Inflation, disinflation, deflation, debt ITHURBIDE Philippe 2013-06

deflation, depression where exactly
do we stand?
Strained financial and economic situation ITHURBIDE Philippe 2013-06

balanced by awkward allocation
Equities: spring is in the air, but the winter MIJOT Eric 2013-06
blues still need shaking
Relaxing fiscal constraints in the eurozone BOROWSKI Didier 2013-06

will not be enough to solve the problem
Eurozone: the challenge of structural PERRIER Tristan 2013-06

reform in a recessionary climate
Banking Union: is a new reality taking LAPEYRE Thomas 2013-06

shape in Europe?

Americas Great Reflation (Year 3): DOISY Nicolas 2013-06
whats with QE-3? Back to Keynes!
The American economic recovery: LENOIR Francine 2013-06

myth or reality for the capital goods
sector?
The hunt for yield: where to find spreads? BERTONCINI Sergio, 2013-06
DRUT Bastien
Worsening of economic conditions and the ITHURBIDE Philippe 2013-05

positive outperformance of risky assets:
Whats wrong with this picture?
Deflation is knocking at Europes door: DOISY Nicolas 2013-05

yet another challenge for the Eurozones
policy mix
Taking stock of deflationary pressures PERRIER Tristan 2013-05

in the eurozone
What impact have the BoJ announcements DRUT Bastien 2013-05

had on European bond yields?
Are European banks in a Japan-style DE BRAY Yasmine, 2013-05

deleveraging process? LAPEYRE Thomas,
WATANABE Kazuo
Will France remain attractive to foreign BOROWSKI Didier 2013-05

investors?
United States: how has non-residential ROGER Florian 2013-05

investment performed?
European High Yield default rates remain BERTONCINI Sergio 2013-05

low. Why? And what trends to expect in
the next quarters?
After its records, is Wall-Street still MIJOT Eric, 2013-05

in pole position WANE Ibra
The rise of US shale gas: a real threat PARIS Laurent 2013-05

to the European chemical industry,
but in the medium term!
When a small country like Cyprus ITHURBIDE Philippe 2013-04

threatens to bring down Europe
What does the future hold for the pound DRUT Bastien 2013-04
sterling?
Real estate recovery powers US growth PERRIER Tristan 2013-04

High Yield: an increasingly significant BERTONCINI Sergio, 2013-04
segment for investors AINOUZ Valentine
Brazil: finding the right balance between LEBOUGRE Catherine 2013-04

demand and supply-side policies
The new Mexican revolution VARTANESYAN Sosi 2013-04
Equity markets peak: what comes next? MIJOT Eric, WANE Ibra 2013-04
The correlation between currency GEORGES Delphine 2013-04

and stock returns is moderating
Japan: auto sector revival KATU Masho, LABIA Frdric 2013-04
Italys elections: a striking display ITHURBIDE Philippe 2013-03

of vulnerability and tension but not
necessarily a reversal in trends
United States: BOROWSKI Didier 2013-03

gauging the sequesters impact
When does the single currency ROGER Florian 2013-03

appreciation start penalizing European
countries?
Dutch mortgage debt: what are the risks? PERRIER Tristan 2013-03
How high should US bond yields be? DRUT Bastien 2013-03
Is the Feds role in the supply/demand BOROWSKI Didier 2013-03

balance of Treasuries overestimated?
Financial issuances remain attractive AINOUZ Valentine 2013-03

versus non-financial
Corporate profits: bottomed-out? MIJOT Eric, WANE Ibra 2013-03
TV is not dead! LABIA Frdric 2013-03
What are the risk factors for portfolios ITHURBIDE Philippe 2013-02
in a risk on mode?
Spain: domestic demand to remain ROGER Florian 2013-02

depressed in 2013
Divergences in exports: PERRIER Tristan 2013-02

the external adjustment of major EMU
countries remains limited
Weakness of the yen. Will it boost exports BOROWSKI Didier 2013-02

and the stock exchange?
Goodwill: the moment of truth WANE Ibra 2013-02

Equity markets: soaring too fast? MIJOT Eric 2013-02
Another look at eurozone equity valuations GEORGES Delphine 2013-02
Is there any value left in Eur IG credit? BERTONCINI Sergio, 2013-02

AINOUZ Valentine
Towards a further re-rating LEOPOLD Marie-Hlne 2013-02

of the European Pharmaceutical sector?
2013 and beyond: what are the likely ITHURBIDE Philippe 2013-01
scenarios and which allocations to make?
2013: Global economy still under fiscal ROGER Florian 2013-01

constraint but with less financial stress
In most euro zone countries, reaching their ROGER Florian 2013-01

2013 public deficit target is likely to be
postponed
Loose monetary policies for the long term BOROWSKI Didier 2013-01
Fixed-income markets: DRUT Bastien 2013-01

rates increase will be very gradual in 2013
Foreign exchange market: DRUT Bastien 2013-01

yen headed for sharper depreciation in 2013
Credit in 2013: BERTONCINI Sergio, 2013-01

from recovery to normalization? AINOUZ Valentine
European banks: earnings recovery is still LAPEYRE Thomas 2013-01

not on the horizon
Strong rebound of inflows in 2012 WANE Ibra 2013-01
The equity rally that started in October MIJOT Eric 2013-01

2011 will continue in 2013
Commodities: a pause in the super-cycle? LAMBINET Rmy 2013-01
It may have lost its AAA rating ITHURBIDE Philippe 2012-12

but is France really a time bomb?
United States: what if the fiscal cliff ROGER Florian 2012-12

turned out to be more of a fiscal trail?
The banking union would separate bank AINOUZ Valentine 2012-12

debt from sovereign debt
Ireland vs. Portugal: some key lessons BOROWSKI Didier 2012-12

from a living experience
US, eurozone diverging on bank lending PERRIER Tristan 2012-12

Can the short end of the bond yield DRUT Bastien 2012-12
curve stay so flat?
US equities: impending tax hikes and WANE Ibra 2012-12

Apple are disrupting sector performances
Local debt attractiveness should not be only BEN ABDALLAH Marc-Ali 2012-12
assessed through currency appreciation
The auto sector: the death of the mid-price LABIA Frdric 2012-12
range
Eurozone: what are the equilibrium ITHURBIDE Philippe, 2012-11

sovereign bond yields? DRUT Bastien
Economic dynamics: couples break-up ROGER Florian 2012-11

in the eurozone!
When fiscal multipliers determine BOROWSKI Didier 2012-11

economic policy
Spanish banks: what is underpinning the GOT Xavier 2012-11

credit performance?
Spains bad bank: an important step in the LAPEYRE Thomas 2012-11

restructuration of the Spanish banking system
The roadmap for central banks is looking MIJOT Eric 2012-11

favourable for equities
Opportunity knocks in the commodities ABADIE Pauline 2012-11

sector
The trend for corporate credit metrics BERTONCINI Sergio, 2012-11

in US and Europe AINOUZ Valentine
Italy and Spain: a tale of two situations and ITHURBIDE Philippe 2012-10
two trajectories
What do the fundamentals tell us about DRUT Bastien 2012-10

the fair hierarchy of bond yields in the
eurozone?
Eurozone country allocation: a closer look GEORGES Delphine 2012-10

at equity market valuations
The inevitable japanisation of European ITHURBIDE Philippe 2012-10

debt holdings?
How do unconventional central banks BOROWSKI Didier 2012-10

policies impact activity and asset prices?
Flows: markets more reckless than investors MIJOT Eric, WANE Ibra 2012-10

Corporate domestic spreads: a new BERTONCINI Sergio 2012-10
dimension of Eur IG credit valuations
Forecasting European HY default rates: BERTONCINI Sergio 2012-10

our tools, baseline forecasts and stressed
scenarios
Emerging assets and quantitative easing BEN ABDALLAH Marc-Ali 2012-10

by the Fed: an empirical analysis
Emerging Markets represent a long term LEOPOLD Marie-Hlne 2012-10

growth driver, but not for every MedTech
segment
Euro debt crisis: brief respite or lasting ITHURBIDE Philippe 2012-09

relief?
Can the ECB become like the Fed or the BOROWSKI Didier 2012-09
BoE?
Competitiveness in EMU peripheral ROGER Florian 2012-09

countries: outcome and remaining efforts
required?
The Spanish model is under scrutiny BOROWSKI Didier 2012-09
Spanish regional debt on the rise DRUT Bastien 2012-09
United Kingdom: the double dip puts PERRIER Tristan 2012-09

austerity and quantitative easing to the
test
HY spreads implied default rates vs our BERTONCINI Sergio 2012-09

scenarios. Are spreads cheap by historical
standards?
Interim results plagued by the economic WANE Ibra 2012-09

situation and the sovereign debt crisis
Media: the Digital Opportunity LABIA Frdric 2012-09
Emerging market risky assets faced BEN ABDALLAH Marc-Ali 2012-09

with decreased economic flexibility
What is the state of speculation in LAMBINET Rmy 2012-09

commodities? The teachings of an index
Federalism, banking union or debt pooling ITHURBIDE Philippe 2012-07

which way out for EMU?
A European banking union ITHURBIDE Philippe 2012-07
A fiscal (and political) union ITHURBIDE Philippe 2012-07

Projects bonds ITHURBIDE Philippe 2012-07
Eurobonds ITHURBIDE Philippe 2012-07
Eurobills ITHURBIDE Philippe 2012-07
Blue bonds, Red bonds ITHURBIDE Philippe 2012-07
Union bonds ITHURBIDE Philippe 2012-07
The European Stability Mechanism ITHURBIDE Philippe 2012-07
European Debt Redemption Fund ITHURBIDE Philippe 2012-07
June 28-29 2012, European summit ITHURBIDE Philippe 2012-07
EU Bank resolution Regime GOT Xavier 2012-07

Another one bites the dust
The financial repression is required DRUT Bastien 2012-07

to liquidate public debt
Asset reallocation by US investors: BOROWSKI Didier 2012-07

a form of financial repression?
Spain: when excessive debt makes bank AINOUZ Valentine, 2012-07

recapitalisation inevitable LAPEYRE Thomas,
PHAM Dung Anh
US fiscal cliff and European sovereign ROGER Florian 2012-07

crisis: beware of the reverberation!
Eurozone equity markets: lessons gleaned LAMBINET Rmy, 2012-07

from a regime-switching model MALONGO ELOUAI Hassan
The luxury sector experiences greater CHAZELLE Jacques 2012-07

short-term cyclical effects
Spanish utilities: the government is DEDISE Sophie 2012-07

introducing a major reform of the tariff
deficit
Equity markets: European equity exposed MIJOT Eric 2012-07

to emerging-market growth: a mature but
still valid theme
EMU: an initiative in favour of growth ITHURBIDE Philippe 2012-06

is legitimate and inevitable
A Greek exit from the eurozone ITHURBIDE Philippe 2012-06

The wrong right solution?
Greece: would devaluation ultimately BOROWSKI Didier 2012-06

stimulate real GDP growth?

Renewed tensions in Europe: impact on ROGER Florian 2012-06
the global economy
A promising future for value-added tax PERRIER Tristan 2012-06
Caution is the watchword AINOUZ Valentine 2012-06
Beyond default rates stands the recovery GABORIEAU Corinne, 2012-06

rate DE LESTRANGE Hien,
MAUDUIT-LE-CLERCQ Claire
Equity markets: an assessment of Q1 2012 WANE Ibra 2012-06

earnings season
Equity markets: what is happening with MIJOT Eric 2012-06

growth stocks?
Basic materials hit by China slowdown ABADIE Pauline 2012-06

fears
Geopolitical uncertainties are keeping MARTIN Emmanuel 2012-06

a short-term premium on oil prices
French Economic Policy: shifts in policy ITHURBIDE Philippe 2012-05

but no revolution
Political change and equity markets: WANE Ibra 2012-05

Frances story
France: growth vs. austerity. Which ITHURBIDE Philippe 2012-05

direction should it take in the coming
months?
Deleveraging vs. growth impact on ITHURBIDE Philippe 2012-05

business, employment, inflation and
interest rates
France: what if we put the company back ROGER Florian 2012-05

at the heart of the debates?
The end of the Great Moderation: BOROWSKI Didier 2012-05

implications for the equity markets
LTROs: sharp contrasts from country to BOROWSKI Didier 2012-05

country in terms of their impact on bank
lending
The dilemma around the funding of the LAPEYRE Thomas 2012-05

Spanish banking system restructuring
Strong investments flows into US spread BERTONCINI Sergio, 2012-05

products while European credit face new AINOUZ Valentine
headwinds from wider sovereign spreads

Pharmaceutical sector, much healthier CHASTENET Emmanuelle 2012-05
than perceived, brighter prospect ahead
Euro zone: is the market rally sustainable? ITHURBIDE Philippe 2012-04

An evaluation of risk factors
The Feds top priority: keeping long-term BOROWSKI Didier 2012-04

interest rates low
Commodities: gold, oil and food ITHURBIDE Philippe 2012-04

what are the concerns?
Global economy: what are the risks and ROGER Florian 2012-04
dynamics of inflation?
The renminbi grinds to a halt DRUT Bastien 2012-04
The high-yield market: strong growth AINOUZ Valentine 2012-04

potential in Europe
Russian banks: now better able to absorb MARTIN Pol-Louis 2012-04

shocks
Insurers show resilience to the difficult DIJKMAN DULKES Esther 2012-04

environment
Autos in Europe: emergency measures LABIA Frdric 2012-04

needed
China opportunity sets a new normal level LABIA Frdric 2012-04

for premium car sales
Equity markets: towards a transition phase MIJOT Eric 2012-04
Equity markets: earnings revisions will play WANE Ibra 2012-04

a key role going forward
Emerging market equities: a bit more BEN ABDALLAH Marc-Ali 2012-04

patience is needed
Year starts in risk on mode will it last ITHURBIDE Philippe 2012-03

as far as the euro zone is concerned?
Budgetary discipline in Europe: ITHURBIDE Philippe 2012-03

the road to credibility will be a long one
Hard times in the euro zone: imbalances BOROWSKI Didier 2012-03

and competitiveness at the heart of the
problem
Spain and Italy: Forecasts and growth ROGER Florian 2012-03

trajectories increasingly diverged between
countries

Portugal: more trouble ahead VARTANESYAN Sosi 2012-03
Japan: Will the trade deficit significantly YOSHINO Akio 2012-03

affect JGBs and the yen?
The LTROs: buying valuable time for DE BRAY Yasmine 2012-03

banks deleveraging
Credit recovery led by technical BERTONCINI Sergio 2012-03

factors: demand and supply trends look
increasingly supportive
Metal and Mining sector, a medium term ABADIE Pauline 2012-03

winner
The risk premiums justify a gradual MIJOT Eric 2012-03

reallocation to equities
Toward moderate risk deployment ITHURBIDE Philippe 2012-02

in portfolios
Public debt in Europe: ITHURBIDE Philippe 2012-02

the Sword of Damocles that is Greece
Sovereign debt ratings in the eurozone: ITHURBIDE Philippe 2012-02

what is the outlook for 2012 and 2013?
European banks: sovereign debt holdings, ITHURBIDE Philippe 2012-02

recapitalisation needs, liquidity
Where do we stand?
The ECBs balance sheet: hit by the crisis AINOUZ Valentine 2012-02
LTRO, lending and growth: causality and ROGER Florian 2012-02

economic outlook for the EMU
Governments vulnerability and sovereign BOURGEOT Rmi 2012-02

spreads: an old relationship
Unconventional monetary policies and BOROWSKI Didier 2012-02

asset prices
The depreciation of emerging currencies is DRUT Bastien 2012-02

partly related to the sovereign debt crisis
Covered bonds, the last refuge GUERMEUR Christophe 2012-02
Equity markets: monetary policies vs. MIJOT Eric 2012-02

corporate profits, which will win?
Emerging Equities: time to take stock GEORGES Delphine 2012-02
Another year of record outflows in 2011 WANE Ibra 2012-02

Debt crisis: a positive outcome in the ITHURBIDE Philippe 2012-01
eurozone is still possible
Asset allocation in 2012: ITHURBIDE Philippe 2012-01

bringing risk back into portfolios?
2012 economic scenario: inflexion, ROGER Florian 2012-01

elections, disinflation (and reflation):
key words
Monetary policy: maintaining low rates ITHURBIDE Philippe 2012-01

for at least two years
Fixed-income markets: sovereigns being ITHURBIDE Philippe, 2012-01

pulled both ways! VARTANESYAN Sosi
Eurozone countries financing needs ITHURBIDE Philippe 2012-01

in 2012: troubles in sight
Credit market: the European sovereign AINOUZ Valentine 2012-01

crisis is still the most important driver
Investment Grade Corporate bonds: LY Lai 2012-01

where to from 2011?
Forex markets: the euros fate will play BOROWSKI Didier 2012-01
out in 2012
Equity markets: asymmetric risks in 2012 MIJOT Eric 2012-01
After the full defensive play of 2011, be ready WANE Ibra 2012-01
to add a touch of emerging play for 2012
Emerging equities to make a comeback BEN ABDALLAH Marc-Ali, 2012-01

in 2012 GEORGES Delphine
Commodities in 2012: one year, LAMBINET Rmy 2012-01

two semesters!
European debt crisis: what can be done ITHURBIDE Philippe 2011-12

and how do we get out of it?
European debt crisis: one crisis after ITHURBIDE Philippe 2011-12

another
The euro is stabilising amidst a full-blown BOROWSKI Didier 2011-12

sovereign debt crisis: the return of a paradox
Deleveraging, but no European credit LAPEYRE Thomas 2011-12

crunch yet!
Country Effect: an analysis using risk GEORGES Delphine 2011-12

measures

China: a choppy but controlled domestic ROGER Florian 2011-12
slowdown
Energy: Improving fundamentals MARTIN Emmanuel 2011-12

despite skepticism
Is third round of QE needed for oil? LAMBINET Rmy 2011-12
Hope for equity is based more MIJOT Eric 2011-12

on reflation than on profits
Q3 2011 Earnings Season WANE Ibra 2011-12
European debt crisis: is the light ITHURBIDE Philippe 2011-11

at the end of the tunnel receding?
What can we learn from the economic ITHURBIDE Philippe 2011-11

and financial history of monetary unions?
Euro zone is looking for the last resort investor BOROWSKI Didier 2011-11
France now on the radar of the rating ITHURBIDE Philippe 2011-11

agencies
Is Italys debt still sustainable? ROGER Florian 2011-11
No fear Japan might Japanize further KONO Masanaga , 2011-11

YOSHINO Akio
Euro zone corporates: Watch out for AINOUZ Valentine 2011-11

access to financing!
Credit: despite recent recovery, attractive BERTONCINI Sergio 2011-11

valuations are still offered by HY, BBBs
and senior financials
How have Asset-Backed Securities gone JEANNIARD Rmi 2011-11

from one crisis to another, from a plague
to a safe haven security?
Equity markets: economic downturn WANE Ibra 2011-11

and sector allocation
China: dont we have to worry about banks? BEN ABDALLAH Marc-Ali 2011-11
What is really happening with recession ITHURBIDE Philippe 2011-10

risks and the euro debt crisis?
Germany: slowdown or recession in 2012? ROGER Florian 2011-10
When deleveraging threatens BOROWSKI Didier 2011-10

to sabotage growth
A moderate recession is no longer BEN ABDALLAH Marc-Ali 2011-10

a risk factor for European risky assets

Growth or recession: what will be TAILLEPIED Stphane 2011-10
the impact on corporate earnings in 2012?
What about the Eureca plan? ITHURBIDE Philippe 2011-10
For inflation-linked bonds, crises are DRUT Bastien 2011-10

coming one after another but every
one is different
How should CDS premiums be valued? ROUPHAEL Raymond 2011-10

The example of the Greek CDS yield curve
Will French banks be able to overcome DE BRAY Yasmine, 2011-10

their recent refinancing difficulties? LAPEYRE Thomas
Equities: which European markets MIJOT Eric 2011-10

will benefit most from fresh momentum
in emerging markets
Transmission of a sovereign debt crisis ALLEMAND Ingrid 2011-10

to equities: the example of OTE
Financial crisis, economic environment ITHURBIDE Philippe 2011-09

and investment advice
Financial crisis: consequences for the ROGER Florian 2011-09

global economy
Sovereign debt: the US leaves the (very ITHURBIDE Philippe 2011-09

exclusive) AAA club
Federalism, eurobonds, expanded EFSF ITHURBIDE Philippe 2011-09

role what path are we truly taking?
What do financial stress indicators teach BOROWSKI Didier 2011-09

us about the euro zone and the US?
Liquidity and monetary policy: BOROWSKI Didier 2011-09

what might be the impact on yield curves?
Credit has once again been weakened by BERTONCINI Sergio 2011-09

fears over the economy and sovereign debt
Emerging economies are truly attractive ITHURBIDE Philippe 2011-09

but are they truly emerging?
Equity markets: TAILLEPIED Stphane 2011-09

the drawback of being born European
Mediocre quarterly results in Europe WANE Ibra 2011-09
European firms and EMG countries: WANE Ibra 2011-09

fear of the void

The yen and the Swiss franc: DRUT Bastien 2011-09
safe haven status has a cost
Greece and Portugal are on rating ITHURBIDE Philippe 2011-07

agencies radars, bailout plan threatened
by selective default, initial signs of
contagion in Italy and Spain
How to solve a debt problem? ITHURBIDE Philippe 2011-07

Euro area specifics
Spain: the euro area situation is not enabling VARTANESYAN Sosi 2011-07
the country to fully get over the crisis
US job market: never has a recovery been BOROWSKI Didier 2011-07

so weak in terms of jobs (created)
United States real estate market: ROGER Florian 2011-07

still suffering major imbalances
Implied growth expectations: BEN ABDALLAH Marc-Ali, 2011-07

an inexcessive adjustment GEORGES Delphine
Credit: there is still value in High Yield bonds BERTONCINI Sergio 2011-07
Equity markets: how to emerge from MIJOT Eric 2011-07

current state of uncertainty?
Equity markets: volatility must not make WANE Ibra 2011-07

us lose sight of the dividends essential role
When the Chinese currency wakes up DRUT Bastien 2011-07
The ECB is strongly opposed to the Greek BOROWSKI Didier 2011-06

debt rescheduling proposed
by the German government
Euro debt: contagion and systemic ITHURBIDE Philippe 2011-06

crisis vs. investment opportunities
Banking sector rerating: the contagion LAPEYRE Thomas 2011-06

in Eurozone periphery has to be avoided
EMU financial debt: BERTONCINI Sergio 2011-06

current value vs. sovereign risk
Emerging countries: towards a peak ROGER Florian 2011-06

in inflation this summer
Are commodities to become just another LAMBINET Rmy 2011-06

financial asset?
Equities: are high margins a cause MIJOT Eric, WANE Ibra 2011-06
for concern?

Japan: are you cheering the short-term YOSHINO Akio 2011-06
recovery or cringing over the medium-term
stagnation?
Are Nordic currencies still attractive? DRUT Bastien 2011-06
American sovereign debt no better than ITHURBIDE Philippe 2011-05

certain eurozone peripheral countries?
An unavoidable Greek debt restructuring? ITHURBIDE Philippe, 2011-05

VARTANESYAN Sosi
The end of quantitative easing (QE) ITHURBIDE Philippe 2011-05

leads to greater prudence
Could the ECB derail growth in the Eurozone? ITHURBIDE Philippe 2011-05
Why is the euro rising and how far can it go? BOROWSKI Didier 2011-05
What is the outlook on global growth? ROGER Florian 2011-05
Chinese inflation: some good news ITHURBIDE Philippe 2011-05

but dangers remain
Commodities: a temporary surge in silver BOROWSKI Didier 2011-05

and gold prices?
Credit: Beta makes the difference in credit BERTONCINI Sergio 2011-05

markets
Equity markets: the initial trends emerging WANE Ibra 2011-05

from Q1 results are encouraging
What to do with emerging equities? MIJOT Eric 2011-05
Indian population: some results from the WAKANKAR Aashish 2011-05

2011 Census
Euro sovereign debt: A multi-track Europe ITHURBIDE Philippe, 2011-04

VARTANESYAN Sosi
Japan: The economic and market outlook HEGARTY James, KONO 2011-04
following the earthquake Masanaga, OHNI Shizuko,
YOSHINO Akio
How will the Japanese disaster affect ROGER Florian 2011-04

the global economy?
The Japanese crisis: the market has TAILLEPIED Stphane 2011-04

already discounted the short-term effects
Impact of the global geopolitical crises: TAILLEPIED Stphane 2011-04

despite being geographically close to
Africa and the Middle East, the direct
impact is limited for European companies

Currencies: what can movements in assets BOROWSKI Didier 2011-04
held abroad tell us about currency trends?
Credit: Corporate America doesnt stop BERTONCINI Sergio 2011-04

on the road towards improved credit
metrics, as shown by the Feds latest figures
Equity markets: The increase in exogenous MIJOT Eric 2011-04

risks, linked to factors such as geopolitics,
climate and radioactivity, does not dim the
appeal of equity investments, but it does
call for diversification
Bank stress tests redux BUTLER Richard, 2011-04

GOT Xavier
The commodities shock: what impact ROGER Florian 2011-03

will it have on growth?
Is a new spike in commodity prices ABADIE Pauline, 2011-03

in store? MARTIN Emmanuel
Commodity currencies and the Dutch disease DRUT Bastien 2011-03
The ECB on the verge to hike rates, the BOROWSKI Didier, 2011-03
Fed No! What impact on the yield curve ITHURBIDE Philippe
and on the euro?
End of quantitative easing and tensions ITHURBIDE Philippe 2011-03

in the Middle East the same
consequences for risky assets?
Balance sheet normalisation by US BOROWSKI Didier 2011-03

residents will curb the increase in bond
yields
Earnings season: solid reports, but signs WANE Ibra 2011-03

that margins are topping out
Emerging equities: how will they withstand WANE Ibra 2011-03

the shock?
Sovereign debt in advanced economies: ITHURBIDE Philippe 2011-03

what impact will it have on emerging markets?
Brisk demand bonds supports strong BERTONCINI Sergio 2011-03

HY bonds supply: as a result, refinancing
risk eased considerably over the next few
years
Japan: is change in sovereign outlook YOSHINO Akio 2011-03

a dangerous tipping point?
Government debt trajectories for some ITHURBIDE Philippe 2011-02

eurozone countries
The rebound in the US economy and the BOROWSKI Didier 2011-02
dollar
Inflation-linked bonds remain attractive BOROWSKI Didier, 2011-02

DRUT Bastien
Convertible bonds propelled by tailwinds BOROWSKI Didier, 2011-02

DRABOWICZ Alexandre
Excellent start to 2011 for the credit market BERTONCINI Sergio 2011-02
European financials were bullish at the MIJOT Eric 2011-02

beginning of the year
Why Japan? Why now? KONO Masanaga , 2011-02

YOSHINO Akio,
OHMI Shizuko
30% of profits from European companies WANE Ibra 2011-02

now come from the emerging markets
An analysis of flows by asset class WANE Ibra 2011-02

confirms that interest in risky assets has
picked up since the announcement of QE2
Sovereign debt in Europe: a lasting ITHURBIDE Philippe 2011-01

problem
The global economy is flooded with excess ROGER Florian 2011-01

liquidity
Discrepancies and dispersion in Europe in ITHURBIDE Philippe 2011-01

2011: a problematic situation
Monetary policies will remain BOROWSKI Didier, 2011-01

accommodating in 2011 ITHURBIDE Philippe
H2 promises to be challenging for the bond BOROWSKI Didier, 2011-01

markets ITHURBIDE Philippe
2011: a test year for emerging debt ITHURBIDE Philippe, 2011-01

VARTANESYAN Sosi
The bullish cycle on the equity markets is MIJOT Eric, WANE Ibra 2011-01
not over yet
Business sectors will set themselves apart TAILLEPIED Stphane 2011-01

by their ability to pass on commodities
price rises
The credit market is doing okay but BERTONCINI Sergio 2011-01

impacted by tensions due to sovereign risk

N Subjects Contributor (s) Date
WP-001-2008 Bond Market Conundrum: New Marie BRIRE, Jan-08

Factors to Explain Long-Term Interest Ombretta SIGNORI,
Rates? Kokou TOPEGLO
WP-002-2008 Do Inflation-linked Bonds Still Marie BRIRE, Apr-08

Diversify? Ombretta SIGNORI
WP-003-2008 Crisis-Robust Bond Portfolios Marie BRIRE, Apr-08

Ariane SZAFARZ
WP-004-2009 Volatility Exposure for Strategic Asset Marie BRIRE, Jan-09

Allocation A. BURGUES,
Ombretta SIGNORI
WP-005-2009 The Revenge of Purchasing Power Marie BRIRE, Jun-09

Parity on Carry Trades during Crises Bastien DRUT
WP-006-2010 Transmission of financial shocks Florian ROGER, Jan-10

to the real economy: the impact Cathy DOLIGNON
of the financial accelerator
WP-007-2010 Inflation-hedging portfolios: Economic Marie BRIRE, Feb-10

Regimes Matter Ombretta SIGNORI
WP-008-2010 Sovereign bonds and socially Bastien DRUT Apr-10

responsible investment
WP-009-2010 Social responsibility and mean- Bastien DRUT Jan-10

variance portfolio selection
WP-010-2010 The perception of investment risk Marielle de JONG Jun-10
WP-011-2011 Sovereign Wealth and Risk Zvi BODIE, Jun-13

Management. A New Framework for Marie BRIRE
Optimal Asset Allocation of Sovereign
Wealth
WP-012-2011 Hedging Inflation Risk in a Developing Marie BRIRE, May-11

Economy: the case of Brazil Ombretta SIGNORI
WP-013-2011 An adequate measure for exchange Marielle de JONG May-11

rate returns
WP-014-2011 Volatility Strategies for Global and Marie BRIRE, Jun-11

Country Specific European Investors Jean David
FERMANIAN,
Hassan MALONGO,
Ombretta SIGNORI
WP-016-2011 A Dynamic Inflation Hedging Trading Nicolas FULLI- Nov-11

Strategy using a CPPI LEMAIRE
WP-017-2011 Breakeven Inflation Rates and their Gilbert CETTE, Dec-11

Puzzling Correlation Relationships Marielle de JONG

WP-018-2011 The Managing of Retirement Savings Didier MAILLARD Sep-12
revisited
WP-019-2012 Is the Market Portfolio Efficient ? A Marie BRIRE, Jan-12

New Test of Mean Variance Efficiency Bastien DRUT,
when all Assets are Risky Valrie MIGNON,
Ariane SZAFARZ
WP-020-2012 Inflation and Individual Equities Andrew ANG, Jan-12

Marie BRIRE,
Ombretta SIGNORI
WP-022-2012 Incorporating Linkers in a Global Marielle de JONG Jun-12

Government Bond Risk Model
WP-023-2012 No Contagion, Only Globalization Marie BRIRE, Mar-12

and Flight to Quality Ariane CHAPELLE,
Ariane SZAFARZ
WP-024-2012 Capital Access Bonds: contingent Patrick BOLTON, Sep-10

capital with an option to convert Frdric SAMAMA
WP-025-2012 An Inflation Hedging Strategy with Nicolas FULLI- Jan-13

Commodities: A Core Driven Global LEMAIRE
Macro
WP-026-2012 Rehabiliting the Role of Active Michel AGLIETT, May-12

Management for Pension Funds Marie BRIRE,
Sandra RIGOT,
Ombretta SIGNORI
WP-028-2012 Do corporate bond markets value Florian BERG Jul-12

environmental, social or corporate
governance events?
WP-029-2012 Swapping Headline for Core Inflation: Nicolas FULLI- Aug-12

An Asset Liability Management LEMAIRE
Approach Ernesto PALIDDA
WP-030-2012 Market-implied inflation and growth Gilbert CETTE, Oct-12

rates adversely affected by the Brent Marielle de JONG
WP-031-2012 Alternative Inflation Hedging Portfolio Nicolas FULLI- Nov-12

Strategies: Going Forward under LEMAIRE
Immoderate Macroeconomics
WP-032-2012 Unexpected Returns. Methodological Sylvie de LAGUICHE Nov-12

considerations on Expected Returns Gianni POLA
in Uncertainty
WP-033-2013 Low Risk Equity Investments. Empirical Alessandro RUSSO Mar-13

evidence, theories, and the Amundi
experience
WP-034-2013 Managing Uncertainty with Dams. Gianni POLA May-13

Asset segmentation in response
to macroeconomic changes

WP-035-2013 Determining the Maximum Number Ling-Ni BOON Apr-13
of Uncorrelated Strategies in a Global Florian IELPO
Portfolio
WP-036-2013 Fundamental indexation for bond Marielle de JONG, Apr-13

markets Hongwen WU
WP-037-2013 A Tale of Two Eurozones: Banks Nicolas FULLI- Aug-13

Funding, Sovereign Risk & LEMAIRE
Unconventional Monetary Policies
WP-038-2013 Cross-Hedging of Inflation Derivatives Nicolas FULLI- Jun-13

on Commodities: The Informational LEMAIRE
Content of Futures Markets Ernesto PALIDDA
WP-039-2013 Optimal Asset Allocation for Sovereign Zvi BODIE Oct-13

Wealth Funds: Marie BRIRE
Theory and Practice
Forthcoming Do Social Responsibility Screens Marie BRIRE,

Really Matter? A Comparison with Jonathan PEILLEX,
Conventional Sources of Performance Loredana URECHE
Forthcoming Pension Regulation and Investment Ling-Ni BOON,

Performance: Rule-Based vs. Risk- Marie BRIRE,
Based Carole GRESSE,
Bas J.M. WERKER
Forthcoming Does Regulation Matter? Riskiness Ling-Ni BOON,

and Procyclicality in Pension Asset Marie BRIRE
Allocation Sandra RIGOT
Forthcoming Exploring the investment opportunity Marielle de JONG

of European asset-backed securities Vinh Cam Anh NGUYEN,
Hubert Vannier
Forthcoming Is your portfolio efficiently diversified? Gianni POLA

N Subjects Contributor (s) Date
DP-04-2014 Understanding Smart Beta: beyond RUSSO Alessandro May-14

diversification and low risk investing
DP-03-2014 SRI and performance: BERG Florian, Mar-14

impact of ESG criteria in equity De LAGUICHE Sylvie,
and bond management processes LE BERTHE Tegwen,
RUSSO Alessandro,
SORANGE Antoine
DP-02-2014 Risk-Free Assets: De LAGUICHE Sylvie Mar-14

What Long-Term Normalized Return?
DP-01-2014 Will the Real Janet Yellen Stand Up? ITHURBIDE Philippe Mar-14
Amundi Special Focus
Subjects Contributor (s) Date
Entropy, Diversification and the Inefficient Frontier POLA Gianni May-14
Ireland: the good news is piling up despite PERRIER Tristan Jan-14

very high public and private debt levels
The housing bubble in AAA-rated countries DRUT Bastien Dec-13
European equity markets: GEORGES Delphine Sept-13

a sleeping beauty ready to awaken MIJOT Eric
WANE Ibra
Bubbles and Regimes, LAMBINET Remy Jun-13

Two Complementary Approaches MALONGO ELAOUI
Hassan
MOREL Thierry
Rethinking Strategic Asset Allocation in Terms of POLA Gianni May-13

Diversification Across Macroeconomic Scenarios

Dynamic Asset Allocation Rebalancing Strategies De LAGUICHE Sylvie, Apr-13
in the Recent Crisis LEZMI Edmond,
ZHAO Shan
Cyprus Bailout A unique Burden-sharing Scheme BOROWSKI Didier Mar-13
Asset Class Pro-Cyclicality under Solvency II De LAGUICHE Sylvie Feb-13

and the impact on asset allocation
2013 and beyond what are the likely scenarios ITHURBIDE Philippe Jan-13
and which allocations to make
Sector Equity Allocation: WANE Ibra Jan-13

Methodology Note
Country Equity Allocation: MIJOT Eric Jan-13

Methodology Note
Forecasting Returns on Assets De LAGUICHE Sylvie, Jan-13

in an Environment of Uncertainty POLA Gianni
Euro Investment Grade Corporate AINOUZ Valentine, Dec-12

Bond Supply- Evolution and Prospects BERTONCINI Sergio
Insurance, Solvency II and Currency Risk De LAGUICHE Sylvie Dec-12
French Presidential Elections ITHURBIDE Philippe May-12
Everything you Want to Know about Sukuk CHAN Wai-Mei May-12

but are Afraid to Ask
Futures on Dividends how do they Behave De LAGUICHE Sylvie, Jan-12

and what they Can Tell us During an Equity Crisis RUSSO Alessandro
Credit Corporate Default Rate Forecasting BERTONCINI Sergio Jan-12
Resolving the Greek Crisis BOROWSKI Didier, Jul-11

and Stemming Contagion ITHURBIDE Philippe
Investment Grade Credit Financials De LAGUICHE Sylvie, Mar-11

Versus Industrials, as seen by European Insurers BERTONCINI Sergio

Chief Editors:
Pascal BLANQU
Deputy Chief Executive Officer
Head of Institutional Investors and Third Party Distributors
Group Chief Investment Officer
Philippe ITHURBIDE
Global Head of Research, Strategy and Analysis
Pia BERGER, Assistant Editor, Research, Strategy and Analysis

Benoit PONCET, Graphic designer, Research, Strategy and Analysis

In the European Union, this document is only for the attention of Professional investors as defined
in Directive 2004/39/EC dated 21 April 2004 on markets in financial instruments (MIFID), to
investment services providers and any other professional of the financial industry, and as the case
may be in each local regulations and, as far as the offering in Switzerland is concerned, a Qualified
Investor within the meaning of the provisions of the Swiss Collective Investment Schemes Act of 23
June 2006 (CISA), the Swiss Collective Investment Schemes Ordinance of 22 November 2006 (CISO)
and the FINMAs Circular 08/8 on Public Advertising under the Collective Investment Schemes
legislation of 20 November 2008. Under no circumstances may this material be distributed in the
European Union to non Professional investors as defined in the MIFID or in each local regulation,
or in Switzerland to investors who do not comply with the definition of qualified investors as
defined in the applicable legislation and regulation.
This document neither constitutes an offer to buy nor a solicitation to sell a product, and shall not
be considered as an unlawful solicitation or an investment advice.
Amundi accepts no liability whatsoever, whether direct or indirect, that may arise from the use of
information contained in this material. Amundi can in no way be held responsible for any decision or
investment made on the basis of information contained in this material. The information contained
in this document is disclosed to you on a confidential basis and shall not be copied, reproduced,
modified, translated or distributed without the prior written approval of Amundi, to any third
person or entity in any country or jurisdiction which would subject Amundi or any of the Funds,
to any registration requirements within these jurisdictions or where it might be considered as
unlawful. Accordingly, this material is for distribution solely in jurisdictions where permitted and
to persons who may receive it without breaching applicable legal or regulatory requirements.
The information contained in this document is deemed accurate as at the date of publication set out
on the first page of this document. Data, opinions and estimates may be changed without notice.
Document issued by Amundi, a socit anonyme with a share capital of 596,262,615 - Portfolio
manager regulated by the AMF under number GP04000036 Head office: 90 boulevard Pasteur
75015 Paris France 437 574 452 RCS Paris www.amundi.com
Photo credit: Thinkstock by Getty Images, Alexandre Guirkinger

Investment Strategy
Collected Research Papers
2014 Edition
About Amundi
Amundi ranks first in Europe1 and ninth worldwide1 in the
asset management industry with AUM of over 800 billion
worldwide2.
Located at the heart of the main investment regions in more

than 30 countries, Amundi offers a comprehensive range of
products covering all asset classes and major currencies.
Amundi has developed savings solutions to meet the needs

of more than 100 million retail clients worldwide and designs
innovative, high-performing products for institutional clients
which are tailored specifically to their requirements and risk
profile.
The group contributes to funding the economy by orienting
savings towards company development.
Amundi has become a leading European player in asset

management, recognised for:
- Product performance and transparency;
- Quality of client relationships based on a long-term
advisory approach;
- Efficiency in its organisation and teams promise to
serving its clients;
- Commitment to sustainable development and socially
responsible investment policies.
1. Source IPE Top 400 asset managers active in the European marketplace
published in June 2013, based on figures as at December 2012. Interviews
with asset management companies on their assets as at end-December
2012 (open-end funds, dedicated funds, mandates).
2. Amundi Group figures as of 31 March 2014.
research-center.amundi.com

Amundi Investment Strategy Collected Research Papers

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Amundi Investment Strategy Collected Research Papers

Uploaded by

Copyright:

Available Formats

Investment Strategy

Collected Research Papers

Editors: Pascal BLANQU & Philippe ITHURBIDE

For professional investors only

Amundi Discussion Papers Series

Amundi Working Papers

About the authors and editors p. 371

Research publications p. 381

Amundi Investment Strategy Collected Research Papers 1

2 Amundi Investment Strategy Collected Research Papers

Amundi Investment Strategy Collected Research Papers 3

International Investment Center Office dedicated to networkpartners Office dedicated to institutional

4 Amundi Investment Strategy Collected Research Papers

published in academic reviews

in the past 4 years (Applied 5

Economics, Bankers Markets

& Investors, Economic Policy,

European Financial Management,

Finance, Financial Analysts

Journal, Journal of Alternative

Investments, Journal of Asset Ass

Management, Journal of Banking

Amundi Research Center

Amundi research available on our website: www.research-center.amundi.com

Quant Emerging Markets

High Yield Sovereign Bonds

Corporate Bonds Forecasts

Amundi Investment Strategy Collected Research Papers 7

Understanding Smart Beta:

Smart Beta is the asset management industrys answer to some well-

Amundi Investment Strategy Collected Research Papers 9

Weight of Information Technology and Telecom

10 Amundi Investment Strategy Collected Research Papers

I - Smart Beta Strategies

Amundi claims several years of experience in minimum variance equity management,

where w is the vector of the optimal portfolio weights, V is the variance-covariance

Amundi Investment Strategy Collected Research Papers 11

1.2 The portfolio maximizing the diversification ratio

Several reasonable diversification measures exist, and maximizing each of them

12 Amundi Investment Strategy Collected Research Papers

Very often the portfolio maximizing the diversification ratio is presented as a

Maximizing diversification is an intuitive and transparent process, but as for the

For these reasons, when dealing with diversification-based strategies, we believe

1.3 The risk parity portfolio

Amundi Investment Strategy Collected Research Papers 13

Actually, marginal contributions to risk are both function of volatilities and

14 Amundi Investment Strategy Collected Research Papers

Percentage Risk Contributions by Sectors

HEALTH CARE MATERIALS

Amundi Investment Strategy Collected Research Papers 15

Percentage Risk Contributions by Sectors

CONS. DISCRET INF. TECH

HEALTH CARE MATERIALS

16 Amundi Investment Strategy Collected Research Papers

In addition, risk parity strategies have an embedded mean reverting mechanism,

II - Low Risk Anomaly and Diversification

Financial theory assumes that higher risk is remunerated on average by higher

In our previous work, we showed how researchers have been documenting

Amundi Investment Strategy Collected Research Papers 17

60% High Risk

Weight Distribution: Common Factor (Systematic) Risk

60% High Common Factor Risk

18 Amundi Investment Strategy Collected Research Papers

Weight Distribution: Specific Risk