18 views

Uploaded by vitaxe

Financial crises, recession, banking crises, debt crises, def lation, and so forth – all these have been at the sides of asset allocators and risk managers, more closely intertwined and requ…Full description

save

- Defense Industry Uncertainty
- The Mutual Fund Industry in India Started in 1963 With the Formation of Unit Trust of India
- PPT Linku Bisnis
- Hayward Lumber Company Case
- Literature Review
- Topic 10 - Direct Foreign Investment
- Class 1
- XIME Finitiative- 15AUG
- Aboite and About - June 2011
- 924309_634371650465575000
- Reduction of Portfolio Through Diversification
- The 15-Stock Diversification Myth.pdf
- Lecture 2 Uncertainty 2
- ch07
- Portfolio Selection
- ritterSilb_796031_tb07
- lecture7 (2) smu
- Chapter6v6
- What is a Mutual Fund
- 5631_Session 1-Introduction_mutual Fund Concept
- Brochure 2
- Ch06 SM
- Moody'SKMVproject Studentsample2
- casestudyflare-savvy-140109002021-phpapp02 (1).pdf
- Investment Diversification and Portfolio Analysis
- Chapter 9 - HW Solutions
- Singapore Property Weekly Issue 15
- Advantages of Mutual Fund
- Financial Stability
- Investment Management Index
- Beijing Lectures in Harmonic Analysis-Stein-0691084181
- Lebesgue
- Lebesgue.pdf
- FT and FFT.pdf
- Transient Stability Analysis with PowerWorld Simulator
- A First Course in Complex Analysis-Beck_Marchesi_Pixton_Sabalka-2016.pdf
- On the Use of Real-Time Simulation Technology in Smart Grid Research and Development
- Transient Stability Analysis with PowerWorld Simulator

**Collected Research Papers
**

2014 Edition

Editors: Pascal BLANQUÉ & 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 BLANQUÉ

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 Amundi’s visibility worldwide. It is important to

provide guidelines that explain and publicise Amundi’s 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 group’s 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 Amundi’s 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 Investment Strategy Collected Research Papers 3

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

4 Amundi Investment Strategy Collected Research Papers

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 , Fré

ro fe

Ris PPart dé

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 oﬁ 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

é, 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 é (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

rél

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 Étienne 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

çais rshi

e de p wit

ﬁna

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 Française

d’Economie…)

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

Amundi Investment Strategy Collected Research Papers 5

6 Amundi Investment Strategy Collected Research Papers

Amundi Discussion Papers Series Amundi Investment Strategy Collected Research Papers 7 .

8 Amundi Investment Strategy Collected Research Papers .

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 industry’s 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 investor’s 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.

Amundi Investment Strategy Collected Research Papers 9

**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

10 Amundi Investment Strategy Collected Research Papers

**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.

Amundi Investment Strategy Collected Research Papers 11

**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 value–related 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 Amundi’s 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 (Ω) of average

stocks’ volatility on portfolio volatility, as it was originally introduced by Choueifaty

and Coignard in 2008.

wɐ

ȳ=

, ww ɐɐɏ

**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

12 Amundi Investment Strategy Collected Research Papers

**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 doesn’t 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 (Ω), 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.

Amundi Investment Strategy Collected Research Papers 13

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 naïve 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

14 Amundi Investment Strategy Collected Research Papers

**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

Amundi Investment Strategy Collected Research Papers 15

**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

Source: Amundi Research

16 Amundi Investment Strategy Collected Research Papers

“average risk” and “high risk” stocks. In any case. II . risk parity strategies have an embedded mean reverting mechanism. In our previous work. We then group stocks into three equally populated families. due to its documented performance coupled with the unprecedented volatility experienced during the last two global financial crises. for US minimum variance portfolios. as we will show hereafter. low risk investing has recently gained a remarkable interest. for each of the three portfolios as well as for the MSCI World. as stocks and sectors with positive performance and increasing weights will be reduced in order to be aligned back to a risk parity weight. 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. some of them state that the low risk anomaly holds regardless of which dimension of risk –systematic or total– is used for stock selection. “Reversal” at a sector level is a successful risk control strategy.1 The Low Risk Anomaly Financial theory assumes that higher risk is remunerated on average by higher returns. at the model date of 12/31/2012). This is the main reason why risk parity portfolios are generally exposed to the low risk anomaly. Amundi Investment Strategy Collected Research Papers 17 . and the deviations from a 10% target become negligible. We build three portfolios (in Barra One. high risk stocks must have lower weight relative to stocks with lower risk. 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. restricting the investment universe to the constituents of the MSCI World Index. However. and Baker and Haugen (1991) find significant reduction in volatility with no reduction in returns. In addition. Only few exceptions instead (Ang et al. and a two-step approach accounts for it more effectively. according to their risk: “low risk”. the outperformance of low volatility stocks during the last 50 years has been among the most puzzling anomalies in equity markets. Finally we observe the percentage allocated to each family of stocks.Taking into account correlations at least across sectors reduces the dispersion of risk contributions.Low Risk Anomaly and Diversification 2. 2006) rather refer to idiosyncratic volatility. At the same time. in order to contribute the same to portfolio risk. We find that most of the relevant empirical studies focus on systematic risk. We impose that no stock can exceed a 5% weight. whatever the precision of our risk parity (with the exception of the equally weighted approximation).

the risk parity portfolio has only a slight tilt toward low risk stocks. the systematic risk is the most relevant measure when addressing the low risk anomaly and. a slight but intuitive exposure for the risk parity portfolio. compared to the standard index. while it clearly underweights average risk stocks. we see a clear and intuitive exposure to low risk anomaly for the minimum variance. using total risk as a grouping criterion. 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% Diversification Minvar Risk Parity Msci Source: Amundi Research The portfolio maximizing the diversification ratio now exhibits a much more significant percentage invested in low risk stocks. as we have documented. if we restrict our analysis to this component only. The low risk feature of the risk 18 Amundi Investment Strategy Collected Research Papers . 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 Source: Amundi Research In the chart above we see that while the minimum variance portfolio is exclusively invested in stocks with below average risk. This distinction is needed because. However we traditionally distinguish two components of risk: the systematic component (or common factor component according to Barra One terminology) and the specific component. and no exposure at all but rather a barbell allocation for the portfolio maximizing the diversification ratio. As a conclusion. The portfolio maximizing the diversification ratio is apparently well balanced in absolute terms toward low or high risk stocks.

the portfolio maximizing the diversification ratio may show an average total volatility that is not statistically different from that of a standard market index. the portfolio maximizing the diversification ratio exhibits almost 50% of the weight invested in high specific risk stocks. but will necessarily result in below average systematic risk stocks (the denominator effect).parity portfolio is somehow more significant as well. we expect the minimum variance optimisation to exploit uncorrelated stocks as well as low risk stocks. In order to check for diversification we compute the diversification ratio first. On the other hand. the minimum variance portfolio is still exclusively invested in low risk stocks. In other words we are supposed to find some diversification evidence in the minimum variance and in the risk parity portfolios as well. Weight Distribution: Specific Risk 100% 90% 80% 70% 60% High Specific Risk 50% Average Specific Risk 40% Low Specific Risk 30% 20% 10% 0% Diversification Minvar Risk Parity Msci Source: Amundi Research This time. 2. at the numerator. and above average specific risk stocks (the numerator effect). In addition to the portfolio maximizing the diversification ratio. while unsurprisingly. Interestingly we observe some different effects while investigating specific risk. the optimisation results in the selection of high specific risk stocks since the latter increase the numerator. Amundi Investment Strategy Collected Research Papers 19 . in the same way. while having little impact on the denominator: specific risk doesn’t matter at the denominator as it is easily diversified away. we have seen that the two-step risk parity process also somewhat relies on low correlations (at least across sectors) and volatilities.2 Diversification Diversification according to the risk model We now move to investigate how the three investment processes behave in terms of diversification. As a result. 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. and only marginal weight in low specific risk stocks.

4 Msci 1. while finding the same hierarchy. we confirm that the specific risk inflates diversification measures. 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.wi2m i2 l avg i .8 1.4 2. while the risk parity portfolio also provides some diversification improvement.6 Risk Parity 1.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 Source: Amundi Research 20 Amundi Investment Strategy Collected Research Papers .8 2. according to the CBOE methodology: N n .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.2 Diversification 2 Minvar 1. j 1 N <1 N i 1 2-.wi w jm im j lij < .6 2. We than compute the average correlation of stocks. and better explains why a process maximizing diversification is tilted toward high specific risk stocks. Diversification Ratio 3 2.

if those assets were equally weighted. Again. even investing in the same number of stocks. where a relevant risk factor is neglected. In order to address the capital diversification of the three portfolios.Average correlations do not change the picture: the portfolio maximizing the diversification ratio is the best diversified across risk factors. As shown in the chart below. low-correlation stocks and investors are comfortable with such a portfolio when the confidence in the risk model is very high. optimized portfolios are better diversified than risk parity portfolios. optimisation-based portfolios that typically invest in 70-80 assets have an entropy measure of roughly 40-45. investors may correctly believe that the ultimate insurance against unexpected risks is capital diversification. meaning that they have a capital diversification equivalent to an equally weighted portfolio of 40-45 assets. regardless of their true diversifying properties. Entropy 10000 1000 LOG Scale 100 1 276 740 10 47 39 1 Diversification Minvar Risk Parity Msci Source: Amundi Research Amundi Investment Strategy Collected Research Papers 21 . but once again we find some evidence of diversification in the minimum variance and in the risk parity portfolios. investors may be concerned by the effect of using a risk model that is not properly specified. Capital diversification Risk-based measures of diversification like diversification ratio and average correlation show that smart beta are better diversified than a standard index. out of a maximum possible of 1600 (the number of investment universe constituents. the specific risk component reduces measured correlation. In contrast. with an entropy measure of roughly 1300. or where the optimisation relies on incorrectly estimated correlations. In these cases. The risk parity portfolio is obviously much better diversified in terms of capital allocation. if equally weighted). 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. The entropy of a portfolio may be read as the equivalent number of assets held. we employ the entropy measure on the weights of their constituents. Also the risk parity portfolio has almost double the entropy of the standard market index. while within smart beta. Optimisation- based portfolios are thus quite concentrated on a few low-risk.

or whether they should rather be used by investors seeking absolute returns. while volatility and drawdowns are systematically lower. Is liquidity enough to allow for such a radical switch from traditional equities to smart beta? 3. Their returns over the last decade are at least equal to and very often higher than those of standard indices. investors should monitor liquidity as any equity strategy deviating from free float adjusted market cap is by definition less liquid than the latter. or whether they should constitute a new satellite. As for the diversification family. 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. we need to investigate if smart beta equities exhibit some evidence of different and hopefully more favorable correlations with bonds. and the FTSE Edhec Risk Efficient – 22 Amundi Investment Strategy Collected Research Papers . However. We try to explain the sources of these favorable deviations from market cap indices. in order to improve capital diversification of the optimisation-based portfolios. Finally. and thus replace the traditional benchmark. Another point of growing interest is whether smart beta should replace traditional equity as an alternative “equity core”. A similar issue is whether smart beta equities should be used to improve active returns relative to a traditional benchmark. Crucial to all these questions is the detection of the drivers behind the risk-return profiles of smart beta equities. we have analyzed two well-known indices– the FTSE Tobam Maximum Diversification. for some well-known global equity smart beta benchmarks. Also. III . thus favoring capital diversification over risk-model diversification.1 Performance drivers Smart beta strategies have proven to be more efficient than market cap indices from a risk-return standpoint. some more prudent constraints may be used on the maximum weight of any holdings (compared to the 5% that we use in this example). 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?). as well as for some Amundi smart investment processes.Smart Beta in asset allocation Since their introduction into the industries.We believe that risk parity is more suitable for investors that are not completely confident about the estimation of the full variance covariance matrix.

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).D. and not that close to an unconstrained portfolio maximizing diversification. FTSE FTSE MSCI Amundi MSCI Amundi Amundi Amundi CORRELATION EDHEC. Though these constraints are sound.E. and two Amundi processes: the first is a minimum variance with some liquidity constraints.4% Amundi MinVar . Interestingly we notice that the diversification bloc is highly correlated with the minimum variance block.4% 95. In the Table below we show the correlation matrix of active returns relative to the corresponding benchmark for each strategy.E.2% Amundi MinVar 14% 84% 85% 59% 48% 91. Not surprisingly.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).4% 91. 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. RW Parity MinVol Piot FTSE EDHEC-R. this index is well correlated to the MSCI World Risk Weighted index that applies similar constraints. In any case. neither more than 3 times such a quantity. they make this index half way between a market weighted and an equally weighted portfolio. As for the risk parity family. as explained in section 1 (the biggest difference with MSCI being the two-step sector-company approach for the Amundi process).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. Amundi Investment Strategy Collected Research Papers 23 . M. 14% 29% 61% 35% 20% 14% 10% Amundi Diversification 14% 78% 62% 61% 83% 84% 86% FTSE TOBAM M. 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. MinVar R. summarized in the chart below. while the second is a very similar process applied to a restricted list of high quality stocks according to the Piotroski score. TOBAM World Risk World MinVar - Diversif.Piot 10% 86% 84% 54% 48% 91. In the minimum variance family we study the MSCI Minimum Volatility.2% 95. we investigate the MSCI World Risk Weighted together with an Amundi risk parity process. while the risk parity block stays somewhere in the middle.D.

One can argue that we have such a high percentage explained as we use redundant information. For this reason we run a simplified PCA on a restricted sample of one strategy per family (FTSE Tobam Maximum Diversification. and by the 91% variance explained by the first two factors alone.5% 65% 1.5% 55% 50% 0.5% 90% 3.0% 95% 3. over the explanatory variables listed below.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. In order to detect those drivers. we regress the first two principal components of the complete sample of eight strategies.0% 85% 80% 2. Explained Variance 100% 4. since many strategies in our analysis (almost all the strategies within each family bloc) are very similar to each other. Amundi Risk Parity. 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. 24 Amundi Investment Strategy Collected Research Papers .0% 60% 0. 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. thus resulting in overlapping behaviors. 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. Amundi Minimum Variance).0% 70% 1.5% 75% 2.

5 1. a strong low risk anomaly effect (except during the rebound of 2009).2 1. As for the “low risk anomaly”.8 0. we have built a long basket of stocks belonging to the lowest quintile according to systematic risk (cf. and small caps. to delete positive correlation between value and dividend.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.3 1 1 0. common factor risk.4 1. 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. Cumulative Factors Performance 2 2.7 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. positive sector reversal (the latter is interesting as it exhibits very low volatility).6 1.8 2. The dividend yield factor has been simultaneously regressed over the market index and the value factor. Small Cap Value Dividend Msci World Source: Amundi Research The sample period has been characterized by strong equity markets despite the massive drawdown of 2008.6 1. 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. Amundi Investment Strategy Collected Research Papers 25 .9 1. Value and dividend yield have been flat. estimated by Barra One).2 1. positive momentum.

but with small caps only explaining a non-negligible portion of variance (5%).8% 9. small caps. both of them significant.0% 0.0% 0. The variance explained is 60% for market beta. while other correlations are sufficiently low to exclude muticollinearity problems.0% 0.8% Small Cap 0.1% -36. Sector CORRELATIONS Mkt Beta Moment. 5% for the dividend factor.0% 39.0 0. and about 1% for value.1% 9.9% Moment.0% -1.0 50.4% -5.0% 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.1% -19.0% 0.9% 8. As mentioned. However we recognize that each strategy may have different exposure to these 26 Amundi Investment Strategy Collected Research Papers .0% 1.8% 8. Risk 0.0% 0.4% -18. low risk anomaly.5% -5. The second component has small cap and sector reversal exposure. 0. over the full set of explanatory variables.5% -1. we regress the first two principal components of smart beta strategies.0% 39.0% 0.0% 0.0% -18.5 0. small caps and sector reversal.0% 0. Exposures Variance Explained (Log Scale) MKT (negative) MKT (negative) 2.0% Dividend 0.2% 0.1% LMHbeta 1. 0.0% -36.0% 32.0% 0.0% 0.0% .0% 0. and sector reversal. and significant exposures to all the other explanatory variables with the exception of momentum (positive but not significant). Small Cap Value Dividend Risk Reversal Mkt Beta 0. and we analyze their exposures and their explained variance.0% 0.0% 32.8% 0.0% 0.0% Value 0.0% 0.5% Dividend LMHbeta Dividend 3.5 12.0% 0. Overall we would argue that the active performance of smart beta strategies is finally due to low market beta. Low Syst.1% 0.0% 0. 15% for the low risk anomaly.8% -19.0% Sector Reversal 0.0% Low Syst.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.

explanatory variables. and once regression parameters are estimated. + Interact.2013 250% 200% 150% 100% 50% 0% -50% -100% -150% FTSE EDHEC-R. Interestingly. those strategies exhibiting the lowest market beta offset much of this negative effect with a positive contribution by the low risk anomaly. MKT Low Syst. and we need to estimate them separately.E. Amundi Investment Strategy Collected Research Papers 27 . Cumulative Active Returns vs. Amundi Diversif.00% 8. Explained Variability of Active Returns by Components 12. Standard Index: 2003 .00% 6. FTSE TOBAM MD MSCI World RW Amundi Risk Parity MSCI World MinVol Amundi MinVar Amundi MinVar - Piot Unexpl. We show cumulative effect over the period from the end of June 2003 to the end of December 2013 in the following Chart. We thus run seven multiple linear regressions. Risk Sector Reversal Momentum Small Cap value Dividend Source: Amundi Research With the exception of the FTSE Edhec Risk Efficient.00% 4.E. + Interact.00% 2.00% FTSE EDHEC-R. Intuitively the low market beta has a negative effect during upward markets. Amundi Diversif. FTSE TOBAM MD MSCI World RW Amundi Risk Parity MSCI World MinVol Amundi MinVar Amundi MinVar - Piot Unexpl. the regression model explains 80% to 90% of the variance of active returns and its F-test is significant for all the strategies investigated. and it explains a big percentage of the variance of active returns. 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.00% 0. The model is thus overall well specified. Risk Sector Reversal Momentum Small Cap value Dividend Total Source: Amundi Research 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.00% 10. MKT Low Syst.

and equity allocation by the equity department thereafter. In an investment process that is based on top-down strategic asset allocation by the investment board. 3. 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. a new equity core? The choice whether smart beta should be used in an absolute or in an active risk- return framework.may find it difficult to massively move toward smart beta equities. In the same way. While a fund manager with the objective of maximizing information ratio -under a limited tracking error constraint. and realized tracking error.2 Smart Beta for active or absolute returns. while it is basically absent on minimum variance portfolios. could use smart beta equity to make up the bulk (or the “new equity core”) of its equity investments. On the other 28 Amundi Investment Strategy Collected Research Papers . section 1 about the two-step company-sector methodology). but has little impact on returns. the value factor is basically absent in the performance chart and is also negligible in terms of explained variability. The dividend factor explains some variance. Unexplained component of returns is positive in the case of Amundi minimum variance. The quality filter delivers additional value. if the board allocates wealth based on traditional benchmarks allowing limited tracking error deviations. In this case. its deviations from a market weighted index are quite low in terms of cumulative active returns. depends on the utility function of the investor (or the mandate of the fund manager in the case of delegated asset management). with the Amundi Risk Parity benefiting the most. especially when the construction process is less constrained than the MSCI World Minimum Volatility. and the governance of the investment process. because small caps bring some additional volatility. an institutional investor aiming to maximize wealth under some absolute risk constraint. the variance explained is particularly high as the construction process of this portfolio is based on a systematic sector rebalancing (cf. 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.All strategies benefit from sector reversal. and because Amundi portfolios apply some liquidity filters as well. We have said about the constraints applied in its construction process and. Small cap effect explains both performance and variance for diversification-based portfolios and risk parity portfolios. unsurprisingly. we confirm the outlier behavior for the FTSE Edhec Risk Efficient Index. Finally. as it is quite flat over the period. since smart beta equities bring high tracking error relative to a standard market index. The only visible source of active return is the small cap exposure.

using up to 20% of the daily average volumes. Program Trade . Trade Source: Amundi Research Amundi Investment Strategy Collected Research Papers 29 . but cannot quickly absorb the program trades resulting from strategic asset allocation or tactical asset allocation decisions. the lower the likelihood that the investor will be willing to allocate a relevant portion of its equity to low beta stocks. 25 and 50 billion respectively. investors may switch their core equity allocation to smart beta. As we have seen in performance attribution. size. buying (selling) a relevant amount of global equity. if the investment board accepts to change its strategic benchmark into a smart beta benchmark. Trade 25 Bln Usd Prog. Trade 50 Bln Usd Prog. investing in low beta stocks is not efficient in a classical Markowitz framework either (cf. the investor should rather prefer to allocate smart beta equities to the satellite bucket of the portfolio. since smart beta equities bring high tracking error relative to a standard market index. but should be ready to change the equity benchmark. low beta itself penalizes profitability as long as it is not offset by some other positive effects.hand. sector reversal and so on. the MSCI World Minimum Volatility. if liquidity is not an issue. The lower this confidence. smart beta equities can effectively become the new equity core. We test three program trades of USD 10. Let’s consider the case of a big sovereign investor that is going to implement a big change in its strategic asset allocation. First. the investor must be confident that the performance drivers identified above are going to deliver positive performance. the MSCI World Risk Weighted. next section).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. However such a radical choice implies that several conditions are met. On the other hand. times four equity index hypotheses: the MSCI World. With no positive contribution from low risk anomaly. If the market can absorb the volumes needed for monthly or quarterly rebalancing. each day (the daily average volumes are estimated over the last three months as of end of December 2013). just as we have seen in the recent past. Second. according to a long-term estimation of volatility). Let’s assume the investor wants to complete the program trade in 10 days. the investor should be sure that smart beta strategies provide sufficient liquidity. and a risk weighted allocation of the latter two indices (43% MSCI World Risk Weighted and 57% MSCI World Minimum Volatility.

Program Trade . Trade Source: Amundi Research The most liquid index is unsurprisingly the maker weighted index: a huge program trade of 50 billion may be completed in five days. Trade 25 Bln Usd Prog. In the table below. in order to complete each program trade in 10 days. only a USD 10 billion program trade allows a relevant. completion after 10 days.W. Minimum Risk Weighted 43%R. than the Risk Weighted Index (1600). Trade 50 Bln Usd Prog.MSCI Minimum Vol Program Trade .V.V. not really because it is more exposed to small caps. Minimum Risk Weighted 43%R. Percentage Completion of Program Trade after 10 days 100% 95% 90% 85% 80% 75% 70% 65% 60% 55% 50% Risk Weighted 43%R. 30 Amundi Investment Strategy Collected Research Papers . but rather because it is concentrated over a lower number of stocks (248).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.W. the investor cannot hold 100% of equity in smart beta and should dilute his holding with traditional and more liquid equity investments. Minimum 57%M.V. In detail. The Minimum Volatility index is the least liquid.W. Volatility 57%M. Volatility 10 BLN USD 25 BLN USD 50 BLN USD Source: Amundi Research In order to effectively complete the program trades in 10 days. we show the maximum allocation in smart beta that the investor can afford. Volatility 57%M. As for smart beta in general. though not exhaustive. this is the percentage completion after 10 days.

smart beta allocation should be kept residual with respect to market weighted equity. Minimum Risk Weighted 43%R.W. Volatility 10 BLN USD 25 BLN USD 50 BLN USD Smart Beta weight MSCI World weight Source: Amundi Research If the investor is not likely to incur program trades bigger than USD 10 billion. Maximum Allocation in Smart Beta 100% 95% 90% 85% 80% 75% 70% 65% 60% 55% 50% Risk Weighted 43%R. Tracking Error Relative to Standard BenchParks 5. Volatility 57%M. Volatility 57%M. and the tracking error of an allocation with 40% in the smart beta above and 60% in global bonds. thus smart beta would be more suited to being a satellite bucket of the portfolio.V. risk weighted.W.0% 4. the easier the move toward smart beta equities as a new equity core. minimum volatility (to a lesser extent).0% 1. Volatility 10 BLN USD 25 BLN USD 50 BLN USD tracking error vs CW Equity tracking error vs 60-40 Source: Amundi Research The lower the impact of liquidity issues.W. if comfortable with the USD 10 billion hypothesis. Minimum Risk Weighted 43%R. Volatility 57%M. we show the tracking error relative to the MSCI World Index of all equity allocations from the example above. For higher sizes of program trades. But massive investments in smart beta equities Amundi Investment Strategy Collected Research Papers 31 . the investor that goes for smart beta as a new equity core.5% 3. as the investor can hold up to 100% of total equity in smart beta.0% 3. and 60% JPM Global Bond Index). Volatility 57%M.W.V. In the next chart. Minimum 57%M.W.W. Minimum 57%M.5% 4.V. should seriously consider changing its strategic benchmark. Minimum Risk Weighted 43%R.0% 2.V. relative to a classic balanced benchmark (40% MSCI World Index. Minimum Risk Weighted 43%R.V. and a mix of the two indices may all become a new equity core.5% 2.5% 0% Risk Weighted 43%R.0% 0.V. However.5% 1.

the improvement in the efficient frontier is straightforward: Efficient Frontiers . 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. Historical Historical Correlations Returns Volatility with bonds MSCI World 9. MSCI Min Vol: historical data Source: Amundi Research We can state that.3 Bond-Equity Allocation As historical returns may suggest. The chart is based on historical data only (returns.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. we can increase the relative weight of smart beta equities in the allocation (as smart beta equities are more conservative than traditional equities).90% JPM GBI 4.09% MSCI World MinVol 9. As a consequence. for the same level of risk of a traditional bond-equity allocation.41% 31. via both the higher percentage of equity and the higher return of 32 Amundi Investment Strategy Collected Research Papers .75% 6.86% Despite a slightly higher correlation with bonds. thus improving performance. and thanks to the far better risk return profile of the MSCI Minimum Volatility Index. and 5% relative to a traditional equity benchmark. we trace two simplified efficient frontiers using the JP Morgan Global Bond index for fixed income. as far as the risk is lower while returns are higher. MSCI Min Vol: historical data ------. 3. the risk-return profile of some traditional bond-equity allocation is improved by simply switching from market weighted equities to smart beta.Historical Data 12% 11% 10% 9% 8% 7% 6% 5% 4% 4% 6% 8% 10% 12% 14% 16% 18% ------.48% 15.72% 20. In the chart below. variance and covariance).58% 11. and the MSCI World Index or the MSCI World Minimum Volatility for equities.

.smart beta. as it was a combination of cash and a traditional index. low risk anomaly. On the contrary. To formalize this unfavorable scenario into our simplified efficient frontier.60 JPM GBI 4. thus making the frontier less attractive than using unconstrained equity..MSCI Min Vol: historical data ------. Historical Historical Risk Adjusted Returns Volatility returns MSCI World 9. introducing smart beta while keeping the weight of equity unchanged. have proven to hold up over the last decade.48% 15.41% 0. and all other residual factors are not going to deliver any additional return. Thus. Efficient Frontiers .88% 11.75% 6. if we apply an alternative and less favorable scenario where sector reversal. In the same way. 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. we improve the expected return of the portfolio. small caps.69 This is equivalent to the hypothesis that the MSCI Minimum Volatility Index is simply a low beta index.86% 0.MSCI Min Vol: historical data . and may be confirmed in the future. These statements are formally correct.Historical Data and Risk Adjusted Returns 12% 11% 10% 9% 8% 7% 6% 5% 4% 4% 6% 8% 10% 12% 14% 16% 18% ------.60 MSCI World MinVol 6.72% 0.. investing in smart beta would be simply equivalent to investing in low beta equity. 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. while reducing its risk.MSCI Min Vol: same risk adjusted return and correlation of MSCI World Source: Amundi Research Amundi Investment Strategy Collected Research Papers 33 . 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. 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.

with the exception of the FTSE EDHC Risk Efficient Index that is close to traditional equity. or correlation with bonds is higher for any smart beta. 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.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..MSCI Min Vol: historical data ------. Taking true correlation into account we have an even less interesting profile: Efficient Frontiers . Though we may identify some significant common behavior across them. we still have to investigate if some additional benefit could come from lower correlation with bonds: if so... in a balanced asset allocation with smart beta. even from a correlation standpoint. there are several factors behind the performance of smart beta equities. TOBAM MD World RW Parity MinVol MinVar MinVar .MSCI Min Vol: same risk adjusted return and correlation of MSCI World .Piot Smart Beta MSCI World Source: Amundi Research 3. Actually this is rather generalized evidence. smart beta equities would be more diversifying.. the relevant factors explaining such performance deviations from a standard index are not the same for every smart beta strategy. while its only 20% with the MSCI Index (cf. 34 Amundi Investment Strategy Collected Research Papers ..MSCI Min Vol: historical data .4 Diversifying and timing smart beta strategies As we have seen in previous sections. previous table).Historical Data and Risk Adjusted Returns 12% 11% 10% 9% 8% 7% 6% 5% 4% 4% 6% 8% 10% 12% 14% 16% 18% ------.Actually.. Unfortunately this is not the case as bonds’ correlation with the MSCI Minimum Volatility Index is 32%.

high or low average correlation).4. We first consider the case of an investor within an absolute risk-return framework. From an active risk perspective. with no surprise as we are dealing with equity portfolios. 10% of active variance is unexplained.0% 60% 77% 40% 43% 50. using the estimated parameters of linear regressions and their covariance matrix. the market factor explains roughly 90% of the absolute variance.Furthermore we observe that smart beta strategies perform differently according to the conditions of the equity market as a whole (bear or bull market. We can translate the analysis on performance drivers. risk is rather concentrated over low beta (obvious as minimum variance has very low beta). Equal Active Risk Contribution Ey Strategy 100% 80% 57% 50. 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. and MSCI Risk Weighted). and dividend yield. We provide several cases of multi smart beta allocation and for illustrative purposes we stay within the MSCI family (MSCI World Minimum Volatility.1 Diversifying across smart beta strategies There are several ways to diversify across smart beta strategies. Diversifying across smart beta may also be useful in addressing the issue of building a reasonable multi-strategy smart beta benchmark. he would allocate 57% of his assets to the minimum variance and 43% to the risk weighted indices respectively. according to a long term (10 years) covariance matrix. If the investor wants to achieve diversification by equalizing the two indices’ contribution to absolute risk. from section 3. high or low volatility. and not holding just one of them. These are all strong arguments for diversifying across smart beta strategies. From an absolute risk perspective.1. and diversification is possible even if the number of strategies involved is very limited. low risk anomaly. Amundi Investment Strategy Collected Research Papers 35 . trading range.0% 20% 23% 0% Weights Total Risk Contribution Active Risk Contribution MSCI World RW MSCI World MinVol Source: Amundi Research As requested. 3.

Contributions Wo Absolute Risk Contributions Wo Active Risk 100% 100% Unexplained 10. We thus maximize the measure of entropy as defined in section 2. 36 Amundi Investment Strategy Collected Research Papers . He still diversifies on the two indices. In this third example. and does not yet control for factor exposures directly.96% Sector Reversal 60% 60% 15. In this case he would rather invest 73% in the risk-weighted index and 27% in the minimum variance index respectively. while the percentage of active risk with an unknown source would be reduced.61% Momentum 80% 12.69% 67% 60% Value 40% 50% 17.7% 50% 5.37% Sector Reversal 60% 73. the investor might be much more sensitive to diversification across the factors than across the two indices themselves.08% 40% Dividend 40% Low Syst. as a side effect. Risk 0% 20% 37. we assume that the investor is willing to maximize the diversification of the sources of active risk. However. or performance drivers. diversification across the performance drivers would improve as well.43% Value 92.06% 80% Momentum 80% 12.2. Equal Active Risk Contribution Contributions Ey Strategy Wo Active Risk 100% 100% 8.19% Dividend 40% 20% Low Syst. computed over the active risk contributions by factors. active risk would be balanced across the two strategies and. Risk 48.3% 16. we assume that another investor prefers to diversify from an active risk perspective.67% 20% 20% Small Cap 0% MKT Beta 0% Factor Contributions Factor Contributions Source: Amundi Research In a second example.47% Small Cap Weights Total Risk Active Risk Contribution Contribution MKT Beta 0% Factor Contributions MSCI World RW MSCI World MinVol Source: Amundi Research By doing so.13% Unexplained 80% 33% 26.

9% 68% 27% Dividend 40% 20% Low Syst. Maximum Entropy Over Contributions Factor Contributions to Active Risk to Active Risk 100% 100% 7. small cap.3% 32% 7% 80% 4% Momentum 42.1% 80% 32% 6% Momentum 80% 11% 60% Sector Reversal 17% 60% Value 40% 79% 83. but that is willing to introduce into his allocation some exposure to the sector reversal factor. is mainly due to a base effect: portfolio exposure to the low risk factor decreases. relative to the standard index). The exposure to this factor is negligible in the three previous allocations. adding Amundi Risk Parity to the set of available strategies. Contrarily. its risk contribution as a percentage marginally increases. Risk 20% 0% 32% Small Cap Weights Total Risk Active Risk Contribution Contribution MKT Beta 0% Factor Contributions Amundi Risk Parity MSCI World RW MSCI World MinVol Source: Amundi Research Amundi Investment Strategy Collected Research Papers 37 . and low risk anomaly contributions are increased. however. because of the two-step stock-sector construction process. The smart beta portfolio that is most exposed to sector reversal is the Amundi Risk Parity. the allocation in the risk-weighted index would increase.36% Unexplained 21% 16.9% 80% 7% 60% 41% 16% Sector Reversal 34% 60% Value 40% 40. unexplained active variance is further reduced. Maximum Entropy Over Contributions Factor Contributions to Active Risk to Active Risk 100% 100% 5% Unexplained 21% 16. The change in the latter. in order to reduce the still dominant contribution of low market beta.9% 28% Dividend 37% 40% 20% 34% Low Syst. dividend yield. We thus repeat the last case study. but as the active risk decreases as well (risk weighted index has a much lower tracking error than minimum volatility. Risk 0% 20% Small Cap Weights Total Risk Active Risk 29% Contribution Contribution MKT Beta 0% Factor Contributions MSCI World RW MSCI World MinVol Source: Amundi Research In this case. because of its regular and low-volatility historical contribution to performance. As a fourth and last case. As a side effect. we now consider an investor that is comfortable with an objective of diversification across factors in an active management framework.

05 4 5 2.5 0 -0.2 2 0.5 0.2 -0. while the entropy measure on the risk factor would increase to 5.25 1. from June 2003 to December 2013. When the impressive rebound of March 2009 starts.1 0. should be to time the underlying factors and consistently allocate strategies. Results are interesting and often intuitive as well. the most straightforward way to implement some timing over the different indices or strategies. the risk parity portfolio often exhibits the best returns.5 -0. Unexplained risk would be further reduced to 5%. Rolling 12-Month Outperformance (LS) Relative to MSCI World (RS) 0. The following chart exhibits the 12-month cumulative outperformance of each of the three Amundi strategies relative to the standard index. and 34% of active risk.2 Timing smart beta strategies As several factors drive smart beta performance. Another way to time smart beta strategies might be to investigate their behavior according to different market conditions. 41% of absolute risk.25 3. while diversification and especially risk parity react well since they keep on delivering some positive outperformance.15 1.0 1 0. 38 Amundi Investment Strategy Collected Research Papers . rather than maximizing some diversification measure as we have done in previous case studies.89. but split into the MSCI Index and the Amundi process.02 from 4. minimum variance starts lagging the two other strategies. with a quarterly frequency. while during market crashes minimum variance is by far the winning strategy. minimum variance is the winning strategy with some nice resistance by the diversification strategy as well. Sector reversal would be introduced as a source of active risk with a 7% contribution. When the market is suffering some higher volatility without exhibiting a clear trend as in the period between mid-2011 and mid-2012.1 -0. Investors may develop a reliable style rotation model. and may apply some allocation where risk contributions match return expectations.The overall allocation to the risk parity strategies is basically unchanged.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 Source: Amundi Research In long and steady bull markets as was the case from 2003 to mid-2007. This latter now accounts for 37% of the assets.4.15 3 3. 3.0 -0.05 2.

correlation and turbulence. has an equally weighted target allocation on the three smart beta. in order to reduce turnover and avoid false signals.0% -20. as well as on the GICS industry group indices of the MSCI World. and assigns an overweight and an underweight of 5% to the most and least profitable strategy. on single country indices. and actually this is not our goal.0% 0. smart beta strategies are slightly lagging overall (with better risk adjusted returns than the market index.0% -15.0% -5. In this section.0% -10. The chart below exhibits the next-month average annualized returns.0% 5. however). we describe the dynamic allocation model that Amundi implements on a real money multi-smart beta fund on Eurozone equities. we overweight the minimum variance portfolio and we underweight risk parity. On the other hand.0% 20. but the risk parity strategy still captures the trend fairly well.0% Risk Parity Diversif. is higher than the average computed over the last 25 days (increasing VIX). Amundi Investment Strategy Collected Research Papers 39 . as estimated from beginning 2003 to mid-2012 (our sample period).0% 25. Minvar MSCI Source: Amundi Research The second model is based on average market correlation. When the average level of the VIX Index.During the recent low volatility bull market period. according to the market signal. We have computed average correlation according to the CBOE methodology. Minvar MSCI Risk Parity Diversif. The first model is based on market implied volatility. The model is based on three stand-alone dynamic strategies: each of them is based on a market signal. computed over the last 10 days. We are conscious that forecasting the market conditions of the future is not an easy task. we recognize that volatility. Average Returns Increasing Vix Average Returns Decreasing Vix 0. may be behind each of the five “states of the world” described above. the weight of the diversification-based smart beta is always neutral. According to the VIX model.0% 15. When the 10-day average is lower than the 25-day average (decreasing VIX) we overweight risk parity and we underweight minimum variance.0% 10. conditional to the VIX configuration. according to the conditional next-month average returns that we have computed historically. and works as a typical risk off – risk on model. 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.

and months following an average two- week correlation higher than the 104-week average (high correlation). returns on the right- hand chart are lower.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 . Minvar MSCI Risk Parity Diversif.2 0.06/2012 0. Average Returns High Correlation Average Returns Low Correlation 15.0% Risk Parity Diversif.0% -5. as the low correlation across sectors and countries includes a typical pre-crisis situation (October 2008 and July 2011).0% 10. the 5% threshold for neutral signals applies. as we want to capture the turbulence that is not already explained by the market implied volatility. As this indicator is closely correlated with the VIX index.1 0.05 -0. When average correlation is lower (right chart).0% 0. strategies based on diversification across risk factors benefit more than those diversified on stocks directly. computed over the GICS industry group indices. while searching for diversification across assets directly is probably more effective.As for the case of the VIX.0% 10. we normalize by the VIX itself. Again.15 0. Overall. there is less benefit in searching for diversification over risk factors.0% 5.05 0 -0.0% -10.0% 5. Average Correlation and Returns in the Eurozone Estimation Period: 01/2003 .2Y Av Corr Source: Amundi Research Our third indicator is a turbulence index. and the sample period ends in mid-2012. since the former favor those stocks exposed to uncorrelated factors.15 -0.1 -0. 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).0% 0. Minvar MSCI Source: Amundi Research The intuition behind this is that when correlation is very high (left chart). We define market turbulence as the cross- section dispersion of returns. 40 Amundi Investment Strategy Collected Research Papers .

0% 5. Starting mid-September 2012 data are out-of sample. the dynamic model outperforms by more than 40 basis points per year. and we finally compute next-month annualized average returns.0% 0. we compute rolling averages on two weeks and 104 weeks.2% Compared to an equally weighted composite of smart beta strategies.9% Conditional Var (5%) -12.29 0.1% -49.08 Return 6.4% 9. but the turbulence indicator is uncorrelated with the implied volatility and average correlation signals. There is no significant impact on the absolute volatility.0% 0.0% 4.0% 10.01).0% 8. The chart below shows the gross total return performance of the market-weighted index. and the dynamic strategy described so far.64 .1% 9. our final dynamic strategy consists in the portfolio that averages the three model portfolios based on the three market signals above. As mentioned.01 1. Each model portfolio overweights the best performing strategy according to the observed signal at the end of the previous month. and the ratio of returns to Conditional Var improves too (1.As for the correlation indicator.3% Volatility 17. Average Returns Average Returns High Turbulence Low Turbulence 15. even though the drawdown of 2008 is reduced.2% -50. an equally weighted basket of the three smart beta strategies.0% 6.53 1.06% IR 1. The Sharpe ratio rises to 0.2% 13.64 0. we apply the neutrality threshold at 5%.67 from 0. Amundi Investment Strategy Collected Research Papers 41 .1% Drawdowns -56.0% Risk Parity Diversif.67 Return / Conditional Var (5%) 0. Minvar MSCI Risk Parity Diversif.27 Mkt Wght Constant Mix Dynamic Strategy Sharpe Ratio 0. Minvar MSCI Source: Amundi Research Results are less intuitive than in the two previous cases. while starting from June 2013 the strategy feeds a real money portfolio. Turnover 72.5% 13.3% -9.6% -9.08 versus 1.0% 2.

50 1. transparent and investable.50 1.00 1. Mix (LS) Mkt W Index (LS) Cum.50 0. Benchmark characteristics: widely recognized (generally accepted).Are smart beta passive or active strategies? Defining an investment strategy as “active” or “passive” is usually not an easy task. there are some frequent intermediate situations where we can distinguish different levels or intensities in being active or passive. According to our definition.00 0. A classic condition for a “passive” strategy is that we clearly identify the systematic replication of a widely recognized. we need to clarify whether the market cap index should be considered as the benchmark of a smart beta strategy.05 3.02 1. Of course there are some extreme circumstances where such a definition is obvious: let’s 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.99 2014 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 Strategy with Timing (LS) Const. 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. 2.00 1. On the other hand. nor is their impressive tracking error a valuable argument for defining them as “active”.03 2. We believe this should not be the case. we must recognize both the elements in order to match the “passive” condition: 1. 42 Amundi Investment Strategy Collected Research Papers .01 1. transparent and investable benchmark. smart beta strategies should not be benchmarked to market weighted indices. Performance 3. Systematic replication. However. As a consequence.50 Returns 1. Active (RS) Source: Amundi Research IV .04 Active Returns 2. as the main argument for smart beta investing is the well-documented inefficiency of market capitalization benchmarks.00 1.

free float. subject to some constraints (minimum stocks threshold.However. or on some diversification benchmarks (such as the FTSE EDHEC Risk Efficient Index. or fundamental factor models) is usually needed. than among portfolios optimized by different optimizers: little differences in risk and correlation estimates may determine huge differences in optimal stock weights. Smart beta benchmarks require the estimation of a variance covariance matrix. An optimizer. market weighted indices offer an interesting reference point for comparison to our discussion. and its numerical algorithms 3. To some extent. This point is rather critical as we often observe some better consistency among risk forecasts provided by different risk models. and size). stocks’ upper bounds. The case for smart beta benchmark is different. Such benchmarks are thus dependent on some necessarily complex risk model. as well as the rules for index weighting.) All the points mentioned above contain provider-specific features (risk model. 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. at the cost of a negligible margin of error. Together with their 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 Amundi Investment Strategy Collected Research Papers 43 . Some risk measure estimation (the variance covariance matrix of stocks) 2. and differences in numerical algorithms across optimizers magnify those discrepancies in portfolio composition. investors could combine construction rules with publicly available information. all of them depend on: 1. or the FTSE TOBAM Maximum Diversification). etc. All of them rely on continuously and publicly available information as market capitalization. sector concentration constraints. A simple historical data approach is almost impossible: a statistical and parsimonious method (such as principal component analysis. as in the case for market capitalization: rather. index providers disclose the rules applied for companies’ inclusion or exclusion from the index (typically: geographical belonging. The objective function that is maximized: portfolio variance is minimized (or some diversification measure maximized). as a covariance matrix may ideally contain millions of parameters. smart beta composition is definitely influenced by the numerical algorithm (often very complex too) of the optimizer. market cap weighted benchmarks may be replicated even without knowing their composition directly: in order to derive it. optimizer). These indices do not rely on some objective and easily measurable metrics. Let’s have an example focusing on some minimum variance benchmarks (such as the MSCI Minimum Volatility index). Furthermore. sector classification.

or diversification benchmarks exist. If several and different minimum variance. We might argue that. Although reasonable. may be packaged in an ETF. the minimum variance indices so far available in the market exhibit reciprocal (historical) tracking errors ranging from 4% to 6%. We think this conclusion is too radical. which are usually lower than 1%. As a consequence. – general upper bound on each stock (in absolute terms or as a function of the daily liquidity). as far as there is no universally representative smart beta benchmark for any of the three categories. for instance. and prudent. Nowadays passive management is experiencing spectacular growth thanks to a highly comprehensive product offer. and the concrete possibility for the fund manager to replicate its payoff. or risk parity. – the index may be prevented to have major exposure to some style factors. but all mimicking the payoff of their own benchmark. 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. the dependency on different risk models and optimizers. In contrast.and sometimes illiquid. each with a very different payoff. while the disclosure of parameters and models may satisfy the transparency condition. it is highly influenced by a traditional definition of benchmark. passive management is basically precluded for smart beta: many alternative benchmarks exist and all are very different from each other. as it would be required instead. and the common practice of applying various and sometimes heterogeneous constraints prevent the benchmark to be easily recognized and universally representative. every asset class or strategy. and it does not recognize some significant trends in the passive asset management industry.stocks. In conclusion. 44 Amundi Investment Strategy Collected Research Papers . its transparency and its investability. realistic. These products are not limited to those asset classes with a universally recognized benchmark available. – sector and country holdings may be constrained in a range around their weights in a traditional index. Definitely more than the tracking errors among market weighted indices. we may potentially have several passive mandates replicating them. even the more exotic or customized. these rules are discretionary and lead to a benchmark that is provider-specific. rather than universally recognized. involuntary exposure toward styles such as small caps or momentum) some prudent constraints may be needed: – minimum holding thresholds. The only requirements are the mere existence of a benchmark (and very often we would better say the existence of an underlying asset).

stock picking is not contemplated. and also implies a formal commitment by the fund manager of never deviating from the benchmark itself. regardless of the decision of declaring the true benchmark. is transparent. We think that disclosing a benchmark is a not a pure formality.Of course. rebalancing at a pre- specified frequency with no room left for incorporating (time varying) views on market and stocks. 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. As a conclusion. The latter depend on time-varying forecasts of the future profitability of stocks. any risk parity portfolio that systematically Amundi Investment Strategy Collected Research Papers 45 . once we have verified that a pertinent benchmark exists. neither is it a trivial decision by the fund manager. The easiest example of an active strategy is some optimisation-based smart beta strategy (minimum variance or diversification). is whether or not the fund manager implements a systematic replication of it. Systematic replication Once the benchmark is built and made available. a smart beta portfolio is actively managed where the investment universe is filtered by some quantitative or systematic criteria. Similarly. In other words. Furthermore having an official benchmark would be a costly decision as smart beta benchmarks are far more expensive than traditional ones. based on some qualitative and judgmental criteria (the best investment ideas of the buy-side analysts. smart beta investments may fall in both the active and the passive category. and generates some non-negligible tracking error relative to any of the existing benchmarks available. On the other hand. investable and computed by a third party (the index provider). According to this criterion. for a strategy to be defined as “active”. deviation from the benchmark must be relevant. portfolios where risk factors’ exposures are managed according to market views are active portfolios. Finally. would concretely be a passive manager if he actually tracks an existing smart beta benchmark. On the other hand. and tracking error relative to the benchmark or relative to the reference strategy must be ideally equal to zero. and the main drivers of such deviations are investment decisions. for instance). if this filtering is specific to the fund manager. A formal benchmark implies the involvement of an index provider that provides objective calculation and transparency. the only criterion that we need to apply in order to categorize a smart beta product as active or passive. a smart beta fund manager that is formally benchmarked to a traditional market cap index. In the same way. where the process is applied to a restricted list of stocks. the goal of a passive fund manager is to systematically apply and comply with it.

smart beta might still become a new equity core. diversification. any replication of an index is passive. 46 Amundi Investment Strategy Collected Research Papers . ”Sector reversal” as a source of outperformance is more relevant for risk parity than for any other smart beta. Yet the unexplained variability corresponds to some non-negligible positive contribution to performance (thus further investigation is needed). however. while filtering the universe for some quality criteria proves to provide additional value. and that the fund manager assures very low tracking error. etc. the liquidity of those strategies must be consistent with the amount of assets the investor holds. at least when the investment horizon is shorter than 8-10 years. if those weighting rules differ from those applied by the available benchmarks. but we notice that the “low beta” and “low risk anomaly” are less explanatory than “small cap” and “sector reversal”. In this case. “low market beta” and “low risk anomaly” are still the most significant factors. Smart beta may become a “new equity core” if the investor’s relevant risk measure is absolute risk. absence of an auto-corrective mean- reversion mechanism. 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.). through a sophisticated numerical algorithm. Performance drivers behind the risk parity strategies are basically the same. As for the diversification-based strategies (portfolio maximizing the diversification ratio. As an opposite example. and investment universe restrictions. overrepresentation of large caps. even if the index is built by maximizing a highly innovative utility function. thus potentially compromising the risk-return payoff of traditional equities. The requirements are that the index is produced by an independent index provider that discloses calculation methodology. If the investor’s relevant risk measure is relative risk. employing highly specific sector and country constraints. risk efficient portfolio. with the addition of “small cap” and “sector reversal”. Conclusion Smart Beta equities are the asset management industry’s answer to some well- known drawbacks of market capitalization-based equity indices such as price noise. Some of these features may result in high volatility and massive drawdowns. especially where (as is the case in Amundi’s process) risk parity is achieved through a two-step stock- sector construction process. with some variance explained by “sector reversal” and “dividend yield”. and risk parity.applies some unique weighting schemes (as the two-step approach discussed in previous sections) is actively managed. In this study we provide a formal description of three popular risk-based smart beta strategies – minimum variance.

In addition to diversification. Amundi Investment Strategy Collected Research Papers 47 . because smart beta investing generates high tracking error relative to standard indices. 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. and thus there is room for diversification. the answer is obvious – if he designs his own smart beta process.but some more pertinent benchmarks should be designed. A clever benchmark. 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. most smart beta strategies should fall into the passive category. Another argument for diversifying across smart beta is the different behavior they exhibit in some typical market conditions. Finally. 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. 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. According to the pure replication criterion. investors can translate these different behaviors into some profitable timing strategies. If the fund manager replicates an existing smart beta benchmark. we discuss whether smart beta should be considered as passive or rather active strategies.

We have thus applied some constraints at these levels. In our case. Furthermore we also rebalance our portfolio quarterly. Sector. turnover in the investment universe is limited as the Piotroski score is based on balance sheet data that varies very little during one quarter. at least in the long run. D is the number of days that we accept to liquidate the fund. Turnover and liquidity High turnover is a critical issue in many systematic investment strategies like Minimum Variance and other optimisation-based strategies. At the same time we do not want to renounce an optimisation process which is completely independent from expected returns. 48 Amundi Investment Strategy Collected Research Papers . without using explicit expected returns. as suggested by Baker and Haugen (1991). country. we limited the amount held in any stock to the following percentage: where UB i is the upper bound on the ith stock. 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. we apply a qualitative filter to our investment universe. 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. countries or single stocks. more than turnover itself. the optimizer is left with a high degree of freedom and it tilts the optimal portfolio toward good quality stocks. Basically. 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. Minimum Variance and Diversification portfolios provide excellent diversification across risk factors. ADV i is the average daily volume over the last quarter. Keeping 60% of constituents available for investments. 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. To address this requirement. excluding the lowest quality stocks from the optimisation. Expected returns are very noisy in forecast and thus responsible for well known “error maximization” problems. and stock concentration As mentioned above. without incurring significant market impact costs. Nevertheless. For this reason. but may tend to be poorly diversified across sectors. without preventing the optimizer from choosing solutions that are far enough from a market index.

leverage…). we have decided not to manage them systematically as –again– we do not want to excessively restrict the optimisation process. momentum. we are conscious that optimisation-based smart beta portfolios may be systematically or incidentally exposed to fundamental factors such as size. while for single stocks we apply a general upper bound (GUB). we regularly monitor the behavior of all the risk factors of the BARRA model (size. Amundi Investment Strategy Collected Research Papers 49 . thus modifying the actual upper bound as follows: Management of asymmetries in factor returns Furthermore. value or momentum. On the other hand. We observe that much of our size exposure is corrected away by the liquidity constraints.On countries and sectors we accept deviations from the market index of 5% to 10%. from time to time the fund manager hedges the risk of an exploding bubble. imposing a neutral exposure to the suspected factor. 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. As for other factor exposures. growth. value.

Risk Parity.. The Journal of Portfolio Management.. No. Portfolio Selection. 38. De Silva H.L. CFA. Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk. Moulin P. 2008. R. Vol. The Journal of Portfolio Management.. 1952. 67. 2006. The Journal of Portfolio Management. Behavioral Portfolio Theory.. 1991. no. 39.. The Cross Section of Volatility and Expected Returns. The Journal of Portfolio Management. Social Science Research Network. Lu X. Financial Analysts Journal. The Journal of Finance. Vol.. 2009. The Journal of Finance. Vol. No. No. Vol. Teiletche J.(Spring 2013) Carvalho R. Demystifying Equity Risk-Based Strategies: A Simple Alpha Plus Beta Description. 2009. CFA. The Journal of Portfolio Management. Value Investing: The Use of Historical Financial Statement Information to Separate Winners from Losers. Bradley B. State Street Global Advisors 50 Amundi Investment Strategy Collected Research Papers . 2011... Vol.. Coignard.. CFA. No... 38. 1 (February) Baker M. Statman M.. Low Risk Equity Investments: Empirical Evidence. 17. LXI. Hodrick R. Wurgler J. Y.. Journal of Finance. 2000. 2011.J. Minimum Variance Portfolio Composition. 1 (Fall) Markowitz H. 2012. No. WP-033-2013 (March) Sharpe W. Working Paper Baker M. and comments that improved the quality of the manuscript. Wurgler J. 3. and the Amundi Experience. 1964. A. Zhang. CFA. no. 2 (June) Thomas R. 1. Roncalli T. Vol. 1991. H. Bibliography Ang A. A Behavioral Finance Explanation for the Success of Low Volatility Portfolio. Vol... Y. suggestions. S. New York University. Managed Volatility: A New Approach to Equity Investing. CFA. Bradley B.. Vol. No. Vol. De Silva H. L. CFA. 35. 1 (Fall) Clarke. Vol.. 3 (September) Shefrin H.F.. De Silva. 2011. CFA Institute Baker N. Benchmarks as Limits to Arbitrage: Understanding the Low Volatility Anomaly. Vol. VII. Working Papers Series (September) Piotroski J. The Efficient Market Inefficiency of Capitalization-Weighted Stock Portfolios. 2007. No. X. Shapiro R. and Minimum Variance: An Analytic Perspective.1 (March) Maillard S. Thorley. Supplement 2000 Russo A. Acknowledgements I would like to thank Sylvie de Laguiche for discussions. D.. Journal of Financial and Quantitative Finance. Xing. Towards Maximum Diversification. 33.35. Thorely S. 37. Y. 2000. On the Properties of Equally-Weighted Risk Contributions Portfolios. Theories. Minimum Variance Portfolios in the US Equity Market. Journal of Portfolio Management. 2 (Winter) Clarke R.. XIX. No. 2013. Thorely S. Maximum Diversification. Journal of Accounting Research. 3 (Spring) Choueifaty.. Vol. Amundi Working Papers. No. 3 (March) Clarke R.. Haugen R..

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. Amundi uses a best-in-class approach on a global investment universe. in absolute and relative terms compared to its benchmark. Tegwen LE BERTHE. there is no significant cost on either the European or global investment universes over the observed period. as well as non-monetary advantages in terms of reputation and responsibilities to future generations. What makes Amundi’s extra-financial rating system unique is that it complements the agency consensus with a subsequent internal analysis. Extra-financial Analysis March 2014 In this study. Amundi Investment Strategy Collected Research Papers 51 . The study revealed that after correcting for geographical and style bias. there is no significant underperformance or outperformance over the assessed period. In other words. Quantitative Research Antoine SORANGE. we examine whether an ESG signal adds performance when incorporated into a portfolio management strategy. DP-03 SRI and Performance: Impact of ESG Criteria in Equity and Bond Management Processes Florian BERG. In addition from an incompressible risk standpoint. Sylvie de LAGUICHE. SRI portfolio management is based partly on exclusion criteria for the worst-rated securities and partly on the portfolio’s overall ESG rating. Alessandro RUSSO.

a firm cannot simply maximize one objective function in order to deal with all potential contingencies. the stakeholder theory asks for the purpose of a firm and shared values with all stakeholders. management and finance. Both theories defend different views on the role CSR (Corporate Social Responsibility) should play in the definition of a firm’s objectives. customers. natural environment or communities. In a modern pluralistic society. the two main theses in play could be described as the “shareholder theory” and the “stakeholder theory”. The responsibility towards shareholders should always be considered as more important than the responsibility towards non-shareholding stakeholders such as employees. SRI AND PERFORMANCE: IMPACT OF ESG CRITERIA IN EQUITY AND BOND MANAGEMENT PROCESSES Introduction and Previous Literature The relationship between firms’ environmental. Jensen and Meckling (1976) and Fama and Jensen (1983). A stakeholder as defined by Freeman (1984) is “any group or individual who can affect or is affected by the achievement of an organization’s purpose”. According to Freeman (2004). This controversy has been fuelled by arguments from economics. (2007)) states that corporations should consider the interests of each stakeholder in their decision making. The “stakeholder theory” (Freeman (1984) and Freeman et al. Furthermore. Efficient corporate governance for instance may foster financial performance and facilitate debt financing. This thesis is notably upheld by Friedman (1970). As reminded by Kacperczyk (2009). 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 52 Amundi Investment Strategy Collected Research Papers . According to the “shareholder theory” corporate managers should focus solely on increasing their shareholders’ wealth. governance and social practices and their financial performance has attracted much debate in recent years. no stakeholder should have a prima facie obligation over another (Kacperczyk 2008).

which contradicts their theoretical analysis. Taking into account environmental. for instance. On the contrary. Moreover. Governance) news on stock returns. indicates that a firm is less likely to cause an environmental disaster that may induce unexpected expenses being financed by issuing debt. empirical results from the stock market cannot be applied directly on the corporate bond market. social and governance factors might have a positive impact on the firm’s financial performance. In the position of residual claimants. for instance. Amundi Investment Strategy Collected Research Papers 53 . may also affect the firm’s financial performance and thereby increase its default risk. Bauer and Hann (2010) find a negative relationship between the strength of the environmental profile of US public firms and their credit spread. conducted by Renneboog et al. shareholders are sensitive to upside and downside potential. (2003) and Margolis et al. A sound environmental profile. Inadequate corporate governance. for instance. some individual studies at the portfolio level. On the whole.company during hard times. they do not find evidence that a high CSP reduces the cost of debt. Bondholders are fixed claimants and have an asymmetric exposure to the downside risk of their securities relative to the upside potential. (2008) and Amenc et al. However. (2008) report statistically neutral and sometimes even slightly negative results. A positive event has no significant effect. To our knowledge. The subsequent deterioration of the firm’s balance sheet may induce a higher cost of debt and raise its default risk. Bondholders and shareholders do not have the same loss functions. Goss and Roberts (2011) show firms with high CSP benefit from slightly lower interest rates on the bank loan market. Social. (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. in a recent paper Krüger (2009) assesses the impact of ESG (Environment. Although the general picture seems to show a positive link between corporate social performance and corporate financial performance. He shows that a significant negative abnormal return is observed after the release of a negative event. albeit not significant. However. socially-compliant mutual funds exhibit a statistical neutral outperformance according to the literature. Literature surveys by Orlitzky et al. most studies on the link between CSP and financial securities focus on the stock market. For instance. (2007) study the relationship between different aspects of corporate social responsibility and the cost of debt financing. These studies tend to confirm the stakeholder theory. Sharfman and Fernando (2008). Stocks and bonds are not affected by news through the same channel. Menz (2010) and Chen et al. Good CSR could also be seen as a risk mitigating policy by preventing the risk of extreme negative events.

I . an analysis phase and a post-analysis phase: 54 Amundi Investment Strategy Collected Research Papers . 1. Extra-financial analysis is a so-called “best-in-class” analysis. etc. and the value of the business. pre-analysis) and ESG ratings (cf. t Environmental Dimension: Energy consumption and CO2 emissions. as it compares the securities of a single sector among themselves.Even though mutual funds are neutral in terms of performance. analysis) are based on the analysis of links among ESG criteria. social and governance factors into his portfolio since his utility function might comprise a socially-responsible component in addition to its financial component.1 Philosophy Responsible Investment is the financial translation of sustainable development. 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. t Governance Dimension: Independence of the Board of Directors.Amundi’s extra-financial analysis process 1. The calculations for weighting ESG criteria (cf. 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. shareholders’ rights. The assessment of securities from the extra-financial angle includes three phases: a pre-analysis phase. an investor might have a preference for integrating environmental. etc. water. protection of biodiversity. Therefore. Social. human resources. anti-corruption actions. This analysis incorporates the intangible risks tied to the business’ activity. Governance) rating based on a set of criteria.2 Analysis Process Analysts score businesses on a seven-step scale from A to G. t Social Dimension: Human rights. link with local communities. etc.

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. t Weighting of criteria: The ESG rating is a weighted average of E. 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. The choice of criteria and Amundi Investment Strategy Collected Research Papers 55 . S and G ratings. Using more than one supplier has various advantages: 1. 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). 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. The weighting varies according to the sector to which the security belongs.2.

motivate and retain qualified.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. Operational efficiency: this vector identifies the business’ capacity to improve technologies. Therefore. the impact on the business’ value remains low (Impact of 2 in 5). the leakage rate is also an environmental aberration. 3. The “Water” criterion may have an impact on the business’ value via the three performance vectors: 1. 2.” in the “Utilities” sector. 56 Amundi Investment Strategy Collected Research Papers . the extra-financial analysis must answer the following two questions: 1. ensuring the long-term development of the business and maintaining its know-how. The «reputation» vector: the issue of water leaks in routing circuits is unknown by the populations. it can be as high as 40%. particularly in areas suffering from water stress but also financial stress. The «Operational Efficiency» vector: The issue of leakage rate is a fundamental problem in the utilities sector. Reputation: it is tied to the business’ image with consumers and investors. competent staff. Thus. This vector is also tied to the business’ capacity to select. 2. We estimate more than $20 billion is lost each year by utility companies. for a given sector and criterion. 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. which exposes businesses to a slight risk to their reputation (probability of 1 in 5). In cities like London. thanks to more efficient management of human and energy resources. 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). extra-financial analysts consider that the impact of these criteria can be made via three performance vectors: 1. 3. If there is a proven controversy. regulatory obligations and fines. The «regulation» vector: European directives and other regulations around the world impose strict standards on utility companies in terms of water quality. processes and behaviors that reduce the costs of production or services. 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. Yet water lost in pipelines is water that has been previously treated by businesses and has therefore incurred a cost. Regulation: this concerns activities that are subject or potentially subject to a system of laws.

In the utilities sector. E. After discussion with businesses in the sector and with the relevant financial analyst. we get a score per criterion. on the criteria of utilities’ E dimension. pollution & waste management 6 t The weight of the dimensions E. S and G: Criteria Weight of criteria Energy consumption & GHG emissions 11% Water 21% Biodiversity. If new regulation were put in place. 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. For instance. 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: E S G Weights 38% 30% 32% Amundi Investment Strategy Collected Research Papers 57 . 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. the extra-financial analyst may believe that new significant regulation is likely to be introduced (probability of 3 in 5). we get the following scores: Criteria Score Energy consumption & GHG emissions 18 Water 36 Biodiversity. the cost generated for businesses would potentially be high (potential impact on the business’ value of 3 in 5).

t Analyses by Crédit Agricole Group.. in compliance with our desire to have a pragmatic approach. in-depth analysis is done on more than 250 securities. t Electric utilities. t Generate alert signals for extra-financial analysts in case of insufficient. an active. t NGOs.2 Analysis Phase A. i. 58 Amundi Investment Strategy Collected Research Papers . 1. To enrich their analyses. t Include the ESG evaluation of businesses done by the extra-financial analysis team and return it to the management teams. t Media and public documents. t Calculate the ratings. based on the most tangible risks and opportunities. A complementary qualitative approach In addition to the automatic rating calculation. For example.e. B. the more potential impact a criterion has on the value of the business. t Networks. each security is compared to securities belonging to a sector whose ESG issues are homogeneous. 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 Spread the ESG rating of an issuer to all underlying issues. To perform the analysis. the more it will be weighted in the analysis model. the utilities sector is subdivided into three homogeneous sectors: t Water utilities. obsolete or contradictory data..2. t Brokers. t Scientific reports. extra-financial analysts rely on several sources of extra-financial data: t Meetings with businesses and their sustainable development ratio.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. who are producing a greater quantity of increasingly refined studies on the topic of SRI and sustainable development.

On a given criterion. Amundi Investment Strategy Collected Research Papers 59 . particularly in certain geographic areas. water utilities are shown in blue. the benchmark’s “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 Source: Amundi Research Thus for each criteria in the benchmark. Indeed. in the water utilities sector. 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.8/10.thus it is on the left in the graph (4/10 risk). while the other businesses. 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. X-axis: The business operating in the United Kingdom is not exposed to much of a water risk . get a lower score. in our example of the water criterion. which are down on this criterion. while the American business operates in water stress areas and is therefore given a risk score of 8. the businesses get a risk exposure/risk management score of between 0 and 10. businesses that properly manage their extra-financial risks are those who succeed in managing their financial risk as well. Therefore there is a close link between financial and extra-financial risk. Finally. Y-axis: The business operating in the United Kingdom has the lowest leakage rate and thus gets a risk management score of 9/10.Next. each criterion is broken down as a risk and risk management indicator: For example.

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).00 8.00 10. This allows the analyst to recognize ESG criteria that are performance vectors 60 Amundi Investment Strategy Collected Research Papers .00 9. Amundi’s extra-financial rating is based on a consensus of extra-financial agencies.3 Post-Analysis Phase The post-analysis phase is based on the quantitative research team’s expertise.00 7.00 6.00 D E F G 0.2.00 American B Company 6.or underperformed most. in tandem with the management teams.00 7.00 4.00 English company 9. and includes: t Refining the algorithm for calculating ratings and ensuring the calculation’s traceability t Identifying and understanding the causes of the most marked scoring dynamics on securities t Identifying.00 2.00 0.00 Low exposure High exposure Risk exposure Water Utilities Electric Utilities Grid operators Ultimately. the securities that out.00 1. plus an internal analysis (weighting of criteria and specific ratings of 250 businesses).00 Global Company 1 5. for any extra-financial reasons. Example: the water issue in the Utilities sector Risk Management 10.00 2.00 A 8.00 C 3.00 3.00 1.00 Global Company 2 4.00 5. 1.

5 B 0.1 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.SRI for investors what added value? 2. However. it is reasonable to expect that portfolios that pick their stocks according to ESG criteria may outperform.5 F -2. the demonstration of this potential outperformance comes up against difficulties with the rating method.) and possibly correcting them.5 (E. One of the most traditional .5 C -0. as they have greater resources to set up a sustainable development policy. The choice was made to partially correct the stock-market capitalization bias inherent in the ESG rating. F and G rated stocks) II .5 2. In order to issue a signal that can be used by management.5 E -2. t Analyzing biases (capitalization. in order to be more demanding with biggest stock- market capitalizations.5 0. ESG scores are successively transformed into Z-scores.5 A 1.5 1.ways to test the added value of a Amundi Investment Strategy Collected Research Papers 61 .and simple . SRI managers must abide by these constraints: t the overall portfolio ESG rating has to be over 0.5 -0. ESG scores follow a normal distribution.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. etc. t Reconsidering the weightings of criteria by sector and therefore enriching the pre-analysis phase.5 D -1. This type of analysis is done by sector. country. Ultimately.5 -1. as shown in the table below: Min Z-score Max Z-score Rating 2.5 G On the basis of these ratings. criterion or geographic area. then ratings.

the graph below shows the performance of an SRI long short global equity portfolio.total and systematic . It is hard to justify attributing the relative out. you cannot establish a hierarchy between two companies belonging to different sectors using an ESG rating. There are too many economic.or underperformance of the geographic areas to differences in ESG ratings. The ESG rating of companies is done with a best-in-class approach within the sectors. it is thus foreseeable that a long short portfolio significantly reflects the geographic biases present in the ratings. which alerts us to the importance of these geographic biases in this long short’s performance. Thus we have also carried out an ex ante risk analysis of the long short portfolio. From a financial performance standpoint. excuse a company that is less attentive to those criteria. cyclical and political factors influencing such out. Then there is the issue of how these geographic biases are handled in the performance analysis of the long short portfolio. the sector biases resulting from the construction of a portfolio and must be neutralized when the long short is created. By way of illustration.are closely correlated. Amundi has chosen to keep geographic biases in the ESG ratings. Therefore. its systematic performance and its specific performance. within a sector.or underperformance. 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.signal is to prepare a long short portfolio in which you are buying the best-rated stocks and selling the worst-rated. 62 Amundi Investment Strategy Collected Research Papers . As a result.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 Source: Amundi Research The two curves . LS portfolio .

has every chance of showing strong systemic biases that are unstable over time. ESG . this effect is quite dominant. is very likely to be exposed to interest rate risk. As such. Thus a long short portfolio. It is suitable if the bulk of the systematic risk comes from Amundi Investment Strategy Collected Research Papers 63 . Create long short portfolios by geographic area. but only removes the biases that are specified straight away. if we restrict to a geographic area. and varies in a much broader spectrum than on equities. additional challenges emerge in the construction of a long short. at the level of security picking. Remove the systematic biases of the long short by performance attribution This method is preferable to a lack of treatment. for equities This is not a solution within Europe. Moreover. Furthermore. B. A long short portfolio. then the question arises of choosing and weighting the bonds. An alternative is to use fixed-maturity CDs instead of bonds. Depending on the maturity of the bonds. created without precautions. Several methods can be used to measure this added value: A. sensitivity to rate fluctuations is highly variable. this solution cannot be transposed to systematic biases other than geographic. because we see significant disparities in ESG ratings between European countries (Northern Europe vs. For bond portfolios. even if built with CDs. which makes it more difficult to establish neutral portfolios on a sector level. Southern Europe). many issuers have several bonds with different characteristics. moreover. with an unstable magnitude over time. but the credit risk is very different depending on the issuer’s rating.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 Source: Amundi Research The graph above shows that the geographic allocation effect explains between 50% and 90% of the ex-ante risk. the ratings are no longer centered within the sectors. and there- fore totally masks the effect of the SRI in the long short’s gross performance. We think the potential added value of ESG criteria is more likely to be detected in the performance after stripping out the systematic effects. In any case.

since we can specify that systematic risk must not exceed a certain % of the total risk. This method cannot be transposed in the case of bond portfolios.3. and it must at least be ensured that their contribution to long short risk is minimal. Any systematic biases other than geographic may persist. in order to see how much they affect results. Such information represents the cost of SRI implementation from a risk perspective. 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. To determine the minimum level of TE induced by transforming some well-known 64 Amundi Investment Strategy Collected Research Papers .not initially detected . where systematic biases are too numerous and unpredictable to be managed in a performance attribution. For instance. 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. C. However. This can be achieved by building portfolios that explicitly ensure systematic risk control via the objective function of optimization or via the constraints. it is possible. This method has its own drawbacks as the tested signal may be affected by the constraints introduced during the optimization process.4 on transfer coefficients. The test focuses on the portion of the signal that can actually be used. The second is that the test recreates the conditions in which the signal is actually applied in management. their potential impact on a portfolio’s outperformance will be low with regard to pure stock picking. 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). there is little risk turning a classic index into an SRI-compliant index while integrating ESG convictions. Europe. as we will see in Section 2.clearly identifiable geographic biases. Moreover.2 shows how the presence of a size bias . The analysis done on the governance criterion in Section 2.1 Minimum TE: what is the threshold for different SRI indexes (World. this method does have two advantages: the first is to control systematic risk factors.2 Does the constraint of being SRI penalize management? 2.gives a misleading measure of that signal’s added value. Pacific)? Measuring the minimum tracking error (TE) induced by an SRI process is critical information regarding the development of SRI. 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.2. to introduce turnover constraints or transaction costs into the simulation. If this minimum TE is low. North America. This is the most general method and it can be applied for both equities and bonds. whatever they are. 2. which is more relevant for our clients.

35% In general the implied TE of an SRI compliant portfolio is relatively low. But more attention has to be paid to the American and Amundi Investment Strategy Collected Research Papers 65 . above 1%.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 -0.00 1.55% on average. and even reach 2% TE levels in some specific conditions. The only constraints imposed to the optimizer are those to be respected to make a portfolio compliant with Amundi’s SRI rules.00 0. meaning that at each and every rebalancing date: t the overall portfolio ESG rating has to be over 0.00 1. BarraOne.00 0. However.21% Europe 0. TE ex-post ESG World 0. the minimum ex-ante TE for an SRI compliant process is low on average.indexes (MSCI World. minimum TE optimizations are performed through a risk-model.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.30% and 0.50 2. MSCI Europe. Ex-post levels of TE are globally in line with the ex-ante observations.50 2.60% North America 1. over the January 2005 – June 2013 period. but unsurprisingly it depends on the geographical area: ESG compliant portfolios ex-ante minimum TE (%) 2.5 As the following chart indicates.50 0. MSCI North America and MSCI Pacific) into portfolios respecting Amundi’s SRI rules. in particular for Europe and World.50 1. this is no longer the case for the North American and Pacific zones where minimum TE levels are higher. with respectively 0.50 0.07% Pacific 1.00 2.50 1.

38% t new exclusions. they are not necessarily stable over time.1 0.05% 2. it is interesting to focus on some specific situations.00% -0.92% 0. World North America Pacific Europe Source: Amundi Research Looking at the previous chart.61% Company 5 0.00% 0.Pacific areas where risk implications cannot be ignored.93% 0.15 -0.70 43.21% 66 Amundi Investment Strategy Collected Research Papers .1 0.00% 2.05 -0.09% 5.03 -1.2 0.1 37.54% to the TE increase over the period.02% 5. the large increase in ex-ante TE (+1. 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 2008.22% 0. most of which were large capitalizations making benchmark replication more complicated (in January 2009.00% 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 -------. during the crisis.57 37.21% 0.26%) was due to the combination of different events: Date TE 08/31/2008 1. such as the MSCI North America optimization which suffers from the biggest hikes and declines.12% Company 2 1.18 43.36% Company 3 1.5 2.0 0. %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 1.6 101.83% 0.4 0. 39% of companies under coverage were excluded versus 34% in August 2008).61% Company 4 0.12% 01/31/2009 2. Specific market conditions or large capitalization exclusions may lead to uncomfortable TE levels that we need to monitor. Even though the different TE levels are relatively low.8 0.00% 1.53 36.5 1.9 67.5 0.0 1.38 -0.06% 5. ESG compliance portfolios ex-ante minimum TE (%) 2.4 59.32 -0. The exclusion of the following companies contributed 0.60 23.1 65.06% 4.

60 -0.76% 26.18 -0.04 Company 5 0.04% 33. with a sudden reduction of the TE level. 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.38% 01/31/2010 0. 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.60 Company 7 -0.51% (2) What happened in 2010.57 Company 2 0.53 Company 3 1.38 -0.81% 23.85 Company 6 -0.67 -0.03 -0.6 29.2 7. t some already excluded companies whose volatility sharply increased along with their benchmark weights due to a strong relative performance compared to the index. Amundi Investment Strategy Collected Research Papers 67 .53 -0. Such combination led to an additional TE increase of 0.37% 1.97 -0.53 t conversely.96 0.26% 4. To limit such TE fluctuations over time and avoid such ins and outs by large frontier stocks. in particular for large capitalizations.04 Company 4 2. t higher market volatility made this rise even bigger. bringing back the exclusion level to the 2009 situation. t in-depth understanding of the ESG fundamentals for any of the excluded stocks.88 0. Company ESG score 08/2008 ESG score 01/2009 ESG score 01/2010 Company 1 -0. is almost the opposite of the previous situation: Date TE 01/31/2009 2.05 -0.67 -0. 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 1.17 43.57% Company 7 3.57 0.57% t the overall market volatility decreased leading to lower volatility of individual stocks.55%. in late 2010 the minimum TE partially rebounded as some of the previously mentioned companies were excluded again through qualitative over-writing.83 39.32 -1. t specific follow-up for problematic issuers.15 -0.11% 6.

12 North America 0. ESG IR 2005-2012 t-stat World -0.51 -2.93 -1. 2.10 1.47 0.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 -------.75 -0.42 -0.10 0. As a consequence the only cost of an SRI compliant investment process with a minimized TE is the level of TE itself.38 -1.73 -0.07 0.2 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.56 0.00 -2.30 -0.2.86 -0.31 1.55 0.56 -0.47 -0.70 1.06 Therefore the relative performance of a minimum TE SRI portfolio cannot reasonably be proven as different from 0. It is also worth measuring the impact from a performance point of view of creating such a minimum TE portfolio using ESG scores.36 Computing similar statistic tests on the full sample period IRs does not prove more significant.48 ESG Pacific -0.50 -0.38 -1.91 Europe -0.39 North America -0.00 0. World North America Pacific Europe Source: Amundi Research 68 Amundi Investment Strategy Collected Research Papers .52 0.41 0.94 0.14 -0.04 -0.24 -0. IR 1 Y 2005 2006 2007 2008 2009 2010 2011 2012 t-stat World 0. None of them turned out to be significantly different from 0 at an acceptable level of confidence.19 1.36 -0.79 Europe 0.32 0. Minimum TE .26 -0.73 Pacific 0.23 0. 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.55 -1. as the added value is neutral.54 0.35 -1.

3%. While in the previous exercise we tried to quantify the impact of ESG constraints on performance. 2. The risk aversion parameters have been calibrated in order to obtain average TE levels of roughly 2%. 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.5.For a minimum TE SRI compliant portfolio the impact on performance is neutral. we maximize our utility function applying a risk aversion of 9 for the low tracking error portfolio. Every month of our sample period from January 2005 to June 2013. respectively. with a monthly frequency. 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. we continue to impose Amundi’s SRI rules: t the overall portfolio ESG rating has to be higher than or equal to 0. the higher the relative risk that the investor should accept.5. t the overall portfolio ESG rating has to be higher than benchmark rating. we are now focusing on a more direct exploitation of ESG rating. we use a risk budgeting constraint: the common factors component of risk is limited to 10%. the lower its ambition should be in terms of ESG rating. whatever the geographic area. and ı is the tracking error of the portfolio relative to its benchmark. Ȝ is the risk aversion coefficient. we maximize a classical risk-return utility function as: Utility=ESG . and 4. we run several simulations allowing the risk aversion parameter to vary. leaving the remaining 90% of the risk budget available for specific risk. t the portfolio cannot be invested in stocks with an ESG rating below -0. As we want to capture the specific component of the ESG signal rather than some implicit (and involuntary) allocation effect. We are aware that the ESG signal may be biased toward some styles and some geographical areas. From January 2005 to June 2013. as in the previous simulation on tracking error minimization. the higher the portfolio’s required ESG rating. and a risk aversion of 3 and 1 for the average and high tracking error portfolios. In order to assess the impact on performance (as well as on portfolio characteristics) of a different target tracking error level.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. Amundi Investment Strategy Collected Research Papers 69 .Ȝı2 Where ESG is the weighted average ESG score of the portfolio.5% respectively. Obviously.

45 1. while those in the table below have been computed over eight annual information periods.72 3.31 -2.26 0.14 -0. 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.33 0.99% AV = 3 0.11 1.60% AV = 9 -0.08 -1. 2. at least with a tracking error ranging from 2.54 2.24 -0. for the full sample of our simulation and on an annual basis.15 We can infer from this analysis that whatever risk aversion we use.0% .41 0.55 0.We have computed relative returns.04 -0. we should expect an ESG portfolio to deliver performance in line with that of a classic portfolio and its benchmark.29 -0. calibrated through the choice of the risk aversion parameter.82 -1.39 0.43 -0.70 -0. ESG IR 2005-2012 t-stat TE ex-post min TE -0.70 -0.12 AV = 9 0.70 1.42 -0.60 0. IR 1 Y 2005 2006 2007 2008 2009 2010 2011 2012 t-stat min TE 0.07% AV = 1 0.33 -3. Thus. or that none of the stand-alone ratings or the aggregate ESG rating is neutral in any specific geographical area. As shown in the table below.35 -0.1 Various SRI criteria (ESG / E / S / G .14 AV = 1 -0.26 0.72 4. realized tracking error and information ratios.37 1.5%.51 -2.40 -0. the added value of maximizing the ESG rating is neutral and does not depend on the level of tracking error.00 0.74 0.03 0.World) However. 70 Amundi Investment Strategy Collected Research Papers .27 0.32% Although some observations may differ substantially from zero on an annual basis (table below).39 -0.3.98 ESG AV = 3 0.13 -0.06 -0. none of the results turns out to be significant at an acceptable level of confidence.36 1. Interestingly we notice that realized tracking error are in line with target levels. The t-stats in the previous table have been computed over 100 monthly returns.97 3.42 -0. full sample information ratios are very close to zero and t-stats are not statistically significant.73 -0.4.

as well as on the main four geographical areas. This time we maximize the E/S/G scores respectively instead of the overall ESG score. we set the risk aversion parameter to 3. AV3 . we run 12 additional back tests (on specific geographical areas and on the World developed markets separately). corresponding to an average target tracking error of 3%. with a risk aversion of 3.For this reason. Amundi Investment Strategy Collected Research Papers 71 . Most of the information ratios are positive. Coherently with a back test on World developed markets. 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). Europe. similarly we impose exclusions based on any stand-alone component each time. Again. the cumulative value added of the ESG score is very close to neutral.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. Euro zone and Pacific regions separately. maximizing the same utility function and applying the same constraints as in the previous case. We than repeat the test with stand-alone components in geographical areas and in the World developed markets as a whole. we first test the ESG signal on the North America.

26 -0.52 0.50 0.02 1.76 -0.65 -1.44 Europe 0.64 1.40 2.95 1.40 S Pacific 0.72 -0.71 North America 1.74 0. none of them turns out to be statistically significant at an acceptable level of confidence.24 -1.60 1.25 -0.08 -0.35 -0.83 North America 0.24 -0.20 -0.09 World 1.15 0.70 On an annual basis.51 Pacific 0.47 0.76 0.93 -0.37 1.01 North America 1.25 -0.31 1. the added value of the SRI signal is neutral.26 Europe 0.73 0.35 -0.33 Pacific 0.69 G IR 2005-2012 t-stat World 0.57 0.35 -0.88 0.79 -2. IR 1 Y 2005 2006 2007 2008 2009 2010 2011 2012 t-stat World 0.56 0.32 -2.20 1.00 -2.02 -0.42 0.25 North America -0.53 -1.33 -0.10 -0.66 0.73 -0.22 -0.16 S IR 2005-2012 t-stat World 0.56 1.03 0.16 0.15 -0.21 0.97 1.08 -0.02 -0.29 0.49 -0.95 North America -0.47 0.24 1.10 0.28 0.33 1.62 -2.22 -0.33 1.67 0.40 Europe 0.34 0.27 Europe -0.53 1.19 -0.81 -1.12 -0.37 G Pacific -0.10 0.41 1.39 Europe 0.09 0.25 0.59 -0.04 0.14 0.18 -0.59 -1.34 -2.14 0.42 -1.57 World 2.18 Europe 0.28 0. we can infer that investors may achieve full compliance with SRI rules at the costs of reasonable tracking error (especially 72 Amundi Investment Strategy Collected Research Papers .20 0.45 -0. some information ratios may be substantially different from zero.88 1.22 -0.69 -0.18 -0.04 Pacific 0.06 1.28 0.48 -0. we can state that whatever the extra-financial criteria and the geographic areas we investigate.95 0.36 0.04 -1.E IR 2005-2012 t-stat World 0.43 1.62 North America 3.51 -0.51 0.44 0.37 1. Combining our finding on ESG value added with those on typical tracking error of an ESG equity portfolio construction.70 As a conclusion.79 -0.57 -0.88 1.91 0.54 E Pacific -1.21 0.59 1.79 -1.04 0. however.66 -0.29 0.75 1.

2 Focus on Governance (World developed markets) While addressing the ESG investment process. since the governance rating is somehow addressed as the “most economic” among the ESG components.3. in the Pacific. many academic studies focused on governance as it often appeared to be the ESG component delivering the best returns. 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 Amundi Investment Strategy Collected Research Papers 73 . 2. without any risk budgeting constraint (in line with back tests presented in previous studies). even though it is not statistically significant in any of these areas. as well as in the global strategy. As shown in some previous tables. our latest empirical evidence only partially confirms these findings. On the other hand. while being confident that relative performance versus a standard market benchmark will likely be neutral. AV3 – G relative performance. and that governance is linked to a general effective functioning of the corporate structure. 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). industry. and more precisely by the risk budget constraint that we have introduced: contribution from regional. and style allocation must not exceed 10% of total tracking error. These findings have usually been considered intuitive at some instance. 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. in Europe. In the chart below we present a back-test on the world developed markets. The intuition being that good governance helps to avoid operational and reputational risk. In previous studies we have also documented a persistent and significantly positive impact of the governance filter on performance. and may even tilt the portfolio towards SRI criteria.in Europe and in the World developed markets).

The table below reports that all the common factor groups have a positive impact on performance but.86 2.86 and a t-stat of 2. sectorial.67** Currency 0.96% 0.27 0.0% e y th e ue ity n s e y ge m 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 E on M n s gn ng io N at ei ni e im r ar Fo iz st E S Source: Amundi Research -E on N 74 Amundi Investment Strategy Collected Research Papers . the only significant component is style with an average annual active return of 0.80%).67% 0.50** Specific 0. country/industry and currency effects explain 0. Although not significant.24 0. but rather weak (less than one fourth of the total active return is specific) and quite volatile.92 2. up to 70%-80% of the active performance and its variability is explained by common factors.0% 8.In this case.0% 10. or style allocation.0% 2. most of the cumulative performance is not purely ESG specific but comes from an implicit (and most of the time involuntary). Consequently. without risk budgeting 12. Component Ann.0% 0. Styles contributions to cumulative active performance over the 2005-2013 period.96% (out of 2.50 (significant at a 1% confidence level). in the table and in the chart below we rank stand-alone styles according to their contribution to performance.71 Style 0.21 Industry 0.0% 4.79 ** indicate return is significant at the 1% level For more details.90% of annual outperformance while the pure specific Governance component is positive.80% 0. thus not statistically significant.0% 6. a full sample IR of 0. AR IR t-stat Total Active 2.42 1.37% 0. among them. regional.52% 0.

95 Non-Estimation Universe 0.29 Others 0.28** Liquidity 0.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 Amundi Investment Strategy Collected Research Papers 75 .96% 0.33 0.93* Value 1.08 0. the best contributor to performance is the “size” factor (the portfolio is exposed to small caps.07% 0.ct an.51% 0. the cumulative active return of the Governance portfolio is much lower.Common Factor Cumulative AR IR t-stat Size 5.pr Jul.38 -1. If we impose a risk budget constraint on common factors.ct Jan Apr Jul Oct an.10 Momentum -0.02 Size Non-Linearity -0.22 Foreign Exposure -0.pr Jul.32% 0. the picture changes dramatically. Contribution from “value” is also significant (especially in the US and Australia).86 Earnings Variation 0.21 Growth 1.78 2.52% 0. G .Jul.ct an.pr Jul. and growth (negative exposure. Among the worst performers.ct an.04 * and ** indicate return is significant at the 5% and 1% levels respectively By far.23% 0.ct an.45% -0.01 -0.67 1. “yield” and “momentum” have limited impact and are not statistically significant. with small cap risk factor performing extremely well over the last 8 years).44 1.64 1.71% -0.33** Volatilty 1.ct an.pr Jul. and interestingly it fits the unconstrained portfolio’s specific component very well.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.72 Leverage -0. significant contribution).44% 0.80 2.27% -0.pr Jul.36 -1.59 -1. Some additional positive contribution comes from low risk stocks (negative exposure to volatility).01% -0.06% 0.42 1.15 Yield -1.20 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.ct an.40 -1. Limiting the common factors to 10% of the total ex ante risk.96% -0.pr Jul.

00 0.50 2. ESG score (max min and median) and risk aversion 2. especially in Europe and in the global developed markets. thanks to both a favorable timeframe up to 2009-2010.50 1.00 2.47 -0.99 World -0.1 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. However.14 Europe 0.00 1. and without negative impact on performance.49 -0.20 0.50 2.4.53 1. empirical evidence may be time varying as well.15 Pacific 0. Here we consider the two sub-periods 2005-2009 and 2010-2012: the signal “G” is not statistically significant.50 1.06 -0.50 0. G IR 2005-2009 t-stat G IR 2010-2012 t-stat World 0. we can come out with a higher average portfolio ESG score. an investment process that is compliant with governance criteria may be achieved at a cost of low tracking error. 2.00 1.00 0. 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.19 Europe -0.45 Pacific -0.4 Signal utilization within optimized portfolios 2.82 North America -0. and thanks to some bias that should be taken into account properly. In addition.85 We believe that the governance filter has gained popularity in academic studies as a well performing signal.00 minTE AV9 AV3 AV1 Median Source: Amundi Research 76 Amundi Investment Strategy Collected Research Papers . but it works well up to 2009 and quite poorly thereafter.Empirical evidence on “governance” is thus actually weaker than its consensus in the literature would suggest. All in all. as far as we allow higher tracking error.08 -0. but we cannot state that the process is likely to benefit from superior performance.50 0.44 0.67 -1.14 North America -0.

thus leading to some short positions in the portfolio. Also. When short positions are forbidden.4. We first have to keep in mind that higher tracking error is an explicit cost. risk budget constraints. etc.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. all the other ESG constraints (exclusions. and as far as the relationship between TE and average portfolio ESG rating is positive. and the transfer coefficient decreases. as it necessarily implies a higher variability of active returns themselves. the marginal benefit of any asset (its alpha. In addition to that. under-weights are limited to the weight of the stocks in the benchmark. it should provide an average ESG score that is higher than the standard market index. as far as the impact on performance is null. yet we believe that some more clever indicators may be used for this purpose. thus a higher probability of extremely negative (though unlikely) outcomes. we may agree that the average ESG score is a quick and transparent criterion to address the ESG quality of a portfolio. the process is thus not exploiting the full potential of the signal. minimum average ESG score. or its ESG score in our example) should be equal to its marginal cost (the marginal contribution to risk). Theoretically. its expected return. 2.We might conclude that. In our view. in an unconstrained portfolio maximizing a classical mean-variance utility function. In a well optimized ESG portfolio. but should also assure a generalized coherence between the ESG score and the portfolio composition. The correlation between marginal cost and marginal benefit should thus be perfect. 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. With no short constraints. but in practice this is rarely the case because mean-variance optimizations bear many constraints. the higher the TE the better the portfolio’s ESG profile. Amundi Investment Strategy Collected Research Papers 77 . 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). a good ESG portfolio should respect some compliance criteria (such as not being invested in very negatively ESG rated stocks). the contribution to risk of any constituent should be related – and ideally proportional – to its contribution to the portfolio’s average score. The long-only constraint alone often prevents the transfer coefficient from exceeding 75%.) further reduce the transfer coefficient.

00 0. and just one of the ESG maximizing portfolio thereafter.50 0. 78 Amundi Investment Strategy Collected Research Papers . portfolios optimized with an ESG maximization objective and a risk aversion of 9.90 0.80 0.70 0.00 minTE AV=9 AV=3 AV=1 Median Source: Amundi Research While an investment process aiming to be ESG compliant while minimizing TE has an average transfer coefficient of 0.90 0.70 0. B & C).20 0. the impact of those constraints on portfolio compositions is not the same for any utility function (minimizing tracking error or maximizing average score).60 0.00 0.30 0. the average transfer coefficient is lower and its variability is higher.52). have an average transfer coefficient of 0.45 (and ranging from a minimum of 0.00 1. Transfer coefficient for different risk aversions/TE 1.50 0. F & G). “average” (D) and “bad” (E. and we compute an average transfer coefficient over the three groups separately. We then plot the minimum tracking error portfolio first.As we can see in the following chart. nor for any level of tracking error.60 0.37 to a maximum of 0.40 0. as we will show that the three of them have similar profiles.30 0. In order to better understand how differently the three optimization methods exploit the signal.80 0.10 0.40 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.20 0.10 0. But this is true only up to a point: when risk aversion is equal to 3 and then to 1. we split the ESG signal into three categories: “good” (ESG notes A.

00 Good 0% R² = 0.00 ESG score Source: Amundi Research As for the minimum tracking error portfolio.00 -2. Low transfer coefficients among any groups of stocks. may result in a (artificially) positive transfer coefficient overall.00 Bad -5% -10% -4. Amundi Investment Strategy Collected Research Papers 79 .00 4. Minimum TE . thus with negative contribution to risk). The picture changes completely if we investigate one of the ESG maximizing portfolios.00 0. however. the average transfer coefficient is very low among the three groups. Below we show the case for risk aversion equal to 3.5 (thus excluded.00 Average R² = 0.00 2. gives the charts an overall positive slope (positive information coefficient). This depends on the highly asymmetrical distribution of the scatter plot of the marginal contribution to risk and the ESG score. thus with negative active weights.Transfer coefficient breakdown into ESG score levels 10% Marginal Contribution to Active Total Risk 5% R² = 0. The negative contribution to risk of stocks with an ESG rating lower than -0.

AV3 -Transfer coefficient breakdown into ESG score levels 10% Marginal Contribution to Active Total Risk 5% R² = 0.32 R² = 0. Generally speaking. transfer coefficient is a correlation measure. We plot their median in the following chart.00 R² = 0. When leading with asymmetrical distribution (as the marginal contribution to risk of stocks in a long only portfolio). 80 Amundi Investment Strategy Collected Research Papers . we should better investigate the transfer coefficient among several partitions of the scatter plot. for any portfolio. 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).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. and among any group of stocks. and it is very explicative when dealing with two normal distributions. In order to allow for some generalization we compute transfer coefficients at any date of our sample period.

ESG compliance would be respected by simply investing in the benchmark without any further ESG improvement needed.50 0. that is periods where the average ESG score is naturally above the threshold. 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.5). and if many stocks are excluded because of their negative rating. as it forces the optimizer to choose well rated stocks to reach the minimum threshold. These latter are quite similar to each other. the lower the tracking error. if the average ESG rating of our benchmark is much lower than our minimum threshold. targeting ESG compliance implies a significant ESG improvement. since all of them better exploit the signals relative to the minimum tracking error. On the opposite hand. especially among well-rated stocks. In order to distinguish between periods where ESG compliance is straightforward and periods where ESG compliance is harder to accomplish. If our investment universe has an average ESG rating above our minimum threshold (0. In the same way. an ESG process should not be limited to pure compliance. as “easy” periods. and if in our investment universe there are only a few small companies that must be excluded because of their negative rating. Amundi Investment Strategy Collected Research Papers 81 . the validity of this assumption may depend on how ambitious the ESG compliance constraints are. we have monitored the average ESG score of the minimum tracking error portfolio. However. Among them. the better the transfer coefficient is. We consider periods where the average ESG rating is higher than our threshold. even though it should ideally remain in a low tracking error range. Transfer coefficients for different risk aversions / TE 1.50 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.00 -0.00 0. In order to gain coherence between portfolio composition and stocks’ ESG rating. but should rather allow for some further ESG rating improvement.

20 0.27 R² = 0.40 0. Transfer coefficient in "difficult" periods for different risk aversions / TE 1.00 Good 0% Average Bad -5% -10% -6 -4 -2 0 2 4 6 ESG score Source: Amundi Research 82 Amundi Investment Strategy Collected Research Papers . Minimum TE portfolio .95 R² = 0. 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.60 0.80 0. and compliance is an ambitious goal to accomplish.00 0.00 Good Average Bad minTE AV9 AV3 AV1 Source: Amundi Research When the ESG profile of the investment universe is poor.We compute the transfer coefficient in “difficult” periods only and we compare the different portfolios.Transfer coefficient breakdown in "difficult" periods 10% Marginal Contribution to Active Total Risk 5% R² = 0.

Indeed. we could favor minimum tracking portfolios. 2. Where is a vector containing the weights of the bonds. we set a constraint of the weight of each currency of +-5 basis points. the latter better exploit the ESG signals (at least from a transfer coefficient perspective). If compliance is hard to accomplish. This index only contains investment grade bonds. By keeping the sector and regional weights neutral. is the risk aversion and the specific risk multiplier. Put another way.As a general rule. However.5. region or any other systematic asset allocation. Our bond picking process uses a maximization of the following utility function. This supposes that these criteria have an intrinsic value leading to a source of alpha. In a socially responsible fund it comes down to integrating ESG ratings as expected returns. we use the Merrill Lynch Industry Level 3 that classifies all issues according to 16 different sectors. We track currency risk closely and Amundi Investment Strategy Collected Research Papers 83 . is a matrix containing the specific variances on the diagonal. more than 90% of the active risk is idiosyncratic. we should design a low tracking error process with some ESG maximization target.1 Bond portfolios At Amundi the usual framework of a best-in-class fund within the fixed income universe is a bond picking process.5 Performance of SRI signals in corporate bond markets 2. r is the vector of expected returns. since in these circumstances. A bond picking process consists of choosing bonds for their intrinsic value and not for any bet on style. such a process is in line with the best-in-class idea of comparing each company to its direct competitors. We set the same constraint on the entire portfolio. rather than a pure compliance goal. The specific risk multiplier is set to a very small value so that we have an almost free floating active specific risk. 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. VCV is the covariance matrix of the systematic risk. 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 should be aware of how far from being compliant our investment universe is. Furthermore. The asset selection process is a maximization of the forecast intrinsic value of each company. expo is a vector containing the exposures of all bonds to the risk factors. at any given point in time in our portfolio simulations. Our benchmark is the Merrill Lynch Large Cap corporate bond index. Furthermore. We optimize each portfolio on a monthly basis between January 2010 and July 2013. in order to exploit the ESG score of well rated stocks. the main active risk source is the bond’s specific part of the risk. Thus such a process supposes a minimization of the portfolio’s active systemic risk compared to its market proxied by a benchmark. coupled with a straight tracking error minimization.

29 T-stat 0. The portfolio optimized with G as the expected return exhibits a significant outperformance at the 10% level of 0. The portfolio optimized with E as the expected return shows a non-significant outperformance of 0.00% Asia/Pacific IR 0.56 0. Zone ESG E S G Ann Act Ret 0.25 0. we restrict the portfolio to 300 assets.41 0. All optimizations are carried out with BarraOne’s optimizer.77) 0.4%. On a global scale.13 1. Indeed.60% Europe IR 0.10) (-0.23 (-0.20% 0. No issue and no issuer may exceed one and two per cent in terms of weight of the portfolio.05 -0.47 -0. respectively.11 By looking at the graph.9 (-0.10% 0.5 99 98.57) (-0.10% US IR -0.30% 0.notice that it is negligible at all times (less than 5% of the very low active systematic risk).5 101 100.88 T-stat 1.00% -0.67 Ann Act Ret 0.5 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 -----.13) Ann Act Ret 0.40% -0.58) (-0.59) 1.10% 0. the outperformance of the European portfolio is mainly driven by the factors E and G. of all portfolios optimized with the factor G as the expected return.31 0. not a significant outperformance.00% 0.68) (-0.10% -0.36 -0.ESG Europe ESG US ESG World ESG Asia/Pacific Source: Amundi Research As one can see in the table below. ESG – Cumulative relative performance 102.30% -0.40% 0.07 T-stat (-0.06 T-stat 0. 84 Amundi Investment Strategy Collected Research Papers .4%. Furthermore.23 0.44 0.30% -0. we see that the governance factor has an impact on the performance only in Europe.5 102 101.13 0.65 -0.6 0.3 -0. the World portfolio exhibits a statistically non- significant outperformance of 0. the active performance of the North America and Asia/Pacific portfolios is almost zero. By looking at the graph we see that the portfolio containing European bonds outperforms its counterparts by an annualized 0.20% -0.3 0.3% (annualized).31 -0.55 Ann Act Ret 0.20% World IR 0.6%.5 100 99.

the market as a whole moves according to macroeconomic news. The Leverage is defined as total liabilities over total assets. The control variables specific to the bond issue are Amundi Investment Strategy Collected Research Papers 85 .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 -------.2 Bond spreads We run panel regressions to identify a potential link between our ratings and the bond spread. some unobserved variables such as management capability are likely to affect the spread.5. It represents the profitability of the firm and thus the ability to pay back its debt. The ROA is the accounting return on assets. On the one hand. All issuer specific data stems from FactSet. We opt for fixed effects on the industry level and time effects on the monthly level. G . Since the fixed assets could be claimed by a creditor in case of a default. The higher the leverage ratio. NB: The panel is unbalanced because of some bonds entering and exiting the index. On the other hand. A high leverage ratio should increase the cost of debt. Europe US World Asia/Pacific Source: Amundi Research The optimized portfolios incorporating the ESG rating do not exhibit any significant outperformance. It indicates how much debt a firm has. The panel is built according to the following function. We use the same period and the same data as for our back tests on bonds. ݃ܮ൫ܵ݀ܽ݁ݎǡ௧ ൯ ൌ ݂ሺܩܵܧǡ௧ିଵ ǡ ݏܿ݅ݐݏ݅ݎ݁ݐܿܽݎ݄ܽܿݎ݁ݑݏݏܫǡ௧ ǡ ݏܿ݅ݐݏ݅ݎ݁ݐܿܽݎ݄ܽܿ݁ݑݏݏܫǡ௧ Ϳ We use control variables specific to the issue and to the issuer. Large firms are widely perceived as less risky and thus benefit from a lower cost of debt. The Capital Intensity is the ratio of fixed assets to total assets. Only the portfolio containing European bonds incorporating the governance factor has a significant positive outperformance. 2. We calculate cluster robust standard errors on the industry level. It is thus important to control both dimensions. 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. a high capital intensity should decrease the level of the spread. Loss is a dummy variable that equals 1 if the firm’s net income before extraordinary items is negative in the current and prior fiscal year. the riskier the firm.

Our control variables seem to explain the biggest part of the spread and are in line with the academic literature.003) (0.012 0.376) (1.019 -0. Moreover.008** -0.019) (0.061*** -0.035) (0. The table below shows the results of a panel regression for all regions of the spread’s natural logarithm on the ESG rating and the control variables.034 -0.017) Loss Indicator 0.7.003) (0.76 0.042) (0.003) (0.006 -0.125*** 0.298*** 0. we integrate dummy variables as a control for financial ratings.029** (0.166** (0.004) (0.002) (0.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.006*** 0.005 -0. Moreover.052*** 0.004 0.033*** 0.072) Capital Intensity -0.046) (0.110*** 0.01 -0.026) (0.57.026) (0.108 (0.005** 0.004 0.005) (0.021) Modified Duration 0.038) (0.010*** (0.74 86 Amundi Investment Strategy Collected Research Papers .465) (1.542 0.021) (0.014) (0.046) ROA -0.046*** 0.558 -0. Large issues tend to be more liquid than small ones.002) (0.173*** 0. 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. Almost all control variables are statistically significant and have the expected sign.042) (0.004) log(Nominal) -0.294 0.024 0.57 0. We use the Nominal as a proxy for liquidity.015) Industry Fixed Effects Yes Yes Yes Yes Period Fixed Effects Yes Yes Yes Yes Adjusted R-Squared 0.01 (0. the adjusted R squared is above 0.377 (0. 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 -0.005) (0.064) (0.019) (0.098) log(Size) 0. except for Europe where it is 0.72 0.01 (0.397*** (0.174) (0.091** -0. Larger issues should have a lower spread than small issues.005) (0.345) Leverage 0.

009 0. ** and *** indicate that the regression coefficient is significant at the 10%. In North America no statistic is significant. The standard deviation is in parentheses.039 0.037 G 0.021 S 0.015 S -0.006 0. Both are significant at the one per cent confidence level. By looking only at Europe.019 G 0.026 Asia/Pacific E -0. we see a positive link between the spread and the social rating as well as the governance rating.054*** 0.022 G -0.017 G -0.018 0.024 S -0.029 0. In the table below.058*** 0.015 0.032 0.*.019 Europe E 0.059 Industry Fixed Effects Yes Period Fixed Effects Yes Amundi Investment Strategy Collected Research Papers 87 .59 and the panel is quite large. but since we only observe a t-stat of 1. 5% and 1% levels respectively. Dependent Variable log(Spread) Independent Variable Coefficient Standard Deviation World E 0. There is no evident link between the spread and the ESG ratings in Asia/Pacific.068 0.018 North America E 0. We observe that the governmental performance is negatively linked to the spread.005 0. we cannot conclude that it is significant. we separate the environmental.013 S 0 0. social and governance performance and we still do not observe any impact on the world level.

*. it is not significant according to our statistics. ** and *** indicate that the regression coefficient is significant at the 10%. It seems that the spreads of well rated firms in Europe decrease during the 3 years of our backtests. Only in Europe. 5% and 1% levels respectively. 88 Amundi Investment Strategy Collected Research Papers . we find a link between the cost of debt of a firm and the environmental. social and governance ratings. Portfolios using the factor governance as expected return outperform their benchmark in Europe. This is mainly driven by the factor governance. Interestingly. the cost of debt increases with a better ESG performance. Event though the overperformance portfolios optimized using the ESG ratings is positive in Europe.

but not in countries or geographic areas. depending on the type of management being done . a careful analysis of the optimized portfolios must be done. Finally. Amundi set up qualitative tracking of these borderline securities at a Rating Committee meeting. Conclusion This paper presents our viewpoint on the impact of SRI on management performance. the US. In terms of outperformance. To obtain the greatest consistency between the securities’ ESG ratings and the portfolios’ over. Depending on the periods. but they are not statistically significant. Therefore.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. The measurements are taken using the SRI approach as seen by Amundi. Being SRI also has no significant cost in terms of risk. the geographic areas and the components of the signal. in our historical backtests. but Amundi makes a choice about the weighting of the most appropriate ESG criteria for each sector. if these securities fall on either side of the exclusion limit. 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. In addition. On the company’s ESG ratings. the portfolios must be built limiting every component of the systematic risk (geographic and size mainly). in absolute and relative terms compared to its benchmark and its universe. from an incompressible risk standpoint the inclusion of SRI Amundi Investment Strategy Collected Research Papers 89 . and for Europe or the world. Too many constraints in terms of average ESG rating mean that the signal is only used on certain categories of ratings. The good outcomes we have seen in the past on governance are largely due to inadequate correction of systematic biases (size effect). excluding the worst-rated securities may lead to significant as well as unstable active risk. For large capitalizations. to a lesser extent. However. as well as on the portfolio’s overall ESG rating. This means that while performing the tests. the cost of SRI in terms of incompressible tracking error is higher. SRI portfolio management is based on exclusion criteria for the worst-rated securities.or underweights. the Amundi approach is a best-in-class approach in the sectors. SRI portfolios do not underperform or outperform significantly over the assessed period. Next. The ratings depend on data by criteria coming from several outside providers. neither positive nor negative. we do not find any significant added value. we can see positive results. From this viewpoint. the portfolio with the highest ESG rating is not necessarily the most SRI. for the Pacific and.

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. Choosing to invest in SRI products also involves non-monetary considerations in terms of reputation and investors’ responsibilities to future generations. 90 Amundi Investment Strategy Collected Research Papers . SRI management can therefore be a relatively cost-free way to benefit from this evolution.criteria. with differences in corporate practices having considerable impacts on profitability.

V. C.C. New Haven CT. Theory of the firm : managerial behavior. G. Business Ethics Quarterly. September 13. Vol. M. and Success. E. J.C. 2003..S. Harindra de Silva.. agency costs and ownership structure. Fernando. The impact of corporate social responsibility on the cost of bank loans. L... stakeholder theory. H.. Corporate social responsibility : Is it rewarded by the corporate bond market? A critical note. Wicks. 2007. Schmidt. Marshfield. M. MA. Journal of Financial Economics 3. 29. 2010. 30. Environmental risk management and the cost of capital. Strategic management : a stakeholder approach (Pitman Series in Business and Public Policy).. Friedman. Steven Thorley. M. Organization Studies. R. University of British Columbia... Fama. vol 35. 2002. 2008.University of Michigan. 2011.E.. Financial Analyst Journal. R. 1970. 2009. Harrison. Journal of banking & finance.. 2011. Margolis. Yale University Press. Kacperczyk. 14 : 302-322. and the corporate objective function. A. Zhang.E.C. Working Paper. et Walsh. 1984. Does it pay to be good? A meta-analysis and redirection of research on the relationship between corporate social and financial performance. Journal of Corporate Finance.. M. C. F. R.. M.. Managing for stakeholders : Survival.. Reputation. Value maximization. S. 48-66. Corporate Environmental Management and Credit Risk.. The price of ethics and stakeholder governance : the performance of socially responsible mutual funds. Clarke. J. 1983. Do non-financial stakeholders affect agency costs of debt? Evidence from unionized workers. The Social Responsibility of Business is to Increase its Profits. Freeman..M.C. Jensen. 261-285.. Bibliography Amenc. 117-134. H. 301-325. 403-441.. Freeman... D. M. 2010. Jensen. Document de travail. Chen. Ross School of Business . 2007.. Pittman. et Hann. n°5 (September/ October). With greater power comes greater responsibility? Takeover protection and corporate attention to stakeholders. 2002. Renneboog. Les performances de l’investissement socialement responsable en France. Strategic Management Journal. Maastricht University. K.. A. Portfolio Constraints and the Fundamental Low of Active Management. 569-592 Amundi Investment Strategy Collected Research Papers 91 . Edhec Working Paper. J. Goss. Elfenbein.. issue 7. The New York Times Magazine. 305-360. Ortiz-Molina.. et Le Sourd. J. 235-256. 2007. Sharfman. N. Strategic Management Journal. 58. et Rynes.. 12.. European Centre for Corporate Management. 2008. 1976.S. R.. Orlitzky. Separation of ownership and control. Journal of Business Ethics 96. A. Bauer. Menz. Corporate social and financial performance : A meta-Analysis. M. Roberts. Kacperczyk. 24. H. Ter Horst. 2008. Jensen. Journal of Law and Economics 26.

92 Amundi Investment Strategy Collected Research Papers .

Amundi Investment Strategy Collected Research Papers 93 . The simple operational solution derived from these findings consists in a compromise between: . 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. there is in fact no such thing over the long term. .inflation. 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. which avoids endless extrapolation from either very restrictive or extremely accommodating monetary policies linked to a specific backdrop (25%). even if regularly renewed short-term investment does present some relevant characteristics. It first examines the notion of “risk-free” assets in themselves in order to demonstrate that.the current short rate given the strong autocorrelation between short rates (25%). DP-02 “Risk-Free” Assets: What Long-Term Normalized Return? Sylvie de LAGUICHE.long rates over the full horizon as the yield curve provides some information on future short rates (50%). .

This is not. Normalized returns are taken to mean returns that are coherent but that are not based on any macroeconomic forecasts. We use historical data from the US to ascertain how accurate these different approaches are. II . This report focuses on long-term horizons of at least 5 to 7 years. 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. For investors who choose not to spend in the immediate term but to invest over the long term. however. “RISK-FREE” ASSETS: WHAT LONG-TERM NORMALIZED RETURN? I . First we analyse the notions of cash and risk-free rates.Cash. the risk is perceived as the uncertainty linked to what the money they 94 Amundi Investment Strategy Collected Research Papers . and offer a compromise that is easy to implement. One could.Why this question? When building a strategic allocation over the long term. and should prompt us to question the notion of risk itself which can take a number of forms over a long investment period. To do so. investors assess an asset’s appeal according to its risk-return trade-off. they often reason in terms of its expected excess return in excess of a “risk-free” investment. 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. therefore. The aim is to establish a set of rules that use market information that is readily available when determining normalized returns. 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. then we review the most commonly-used approaches to establish the normalized returns on cash.

The first thing we naturally think of is the volatility of the markets. for which there is a certain adjustment in short rates in relation to inflation) is therefore less risky in real terms. 10 and 30 years. Amundi Investment Strategy Collected Research Papers 95 . we determine risk in the form of an equivalent normal volatility which offers the same dispersion (i. Over longer horizons. dispersion is similar to that of an asset with a volatility of 12% (namely around 60-80% of equities). we see that equivalent volatility increases. the uncertainty linked to the future level of rates means there is a risk of opportunity at the time of the reinvestment of coupons. 3. the risk of default.e. If. over the long term. This means that. cash is not without its risks.have invested will be worth in the future. although it is true that nominal returns are known. As such. Over the long term. the hierarchy of risks can also differ from over the short term. to avoid this risk of reinvestment. 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 Source: Amundi Research It is clear that. a regularly renewed short-term investment (i. over long and forward-looking timeframes. A zero-coupon investment over the long term is seriously exposed to the risk of erosion by inflation. The graph below shows the standard deviation in the annualised return on cash over rolling periods of 1. the risk of reinvestment is far from negligible when it comes to cash. In fact. is therefore to define zero-coupon investments with a nominal rate over the horizon in question as risk-free. However. for each horizon. systematic risk and. the risk of reinvestment linked to cash is high. dividends. by multiplying it by the root of the horizon). volatility is not the only source of risk. While the natural response. or repayments. This uncertainty can have several causes.e. First. even in the absence of these elements. 7. The calculations apply to the US between 1929-2013. There is also the risk of erosion via inflation. this is no longer the case if we reason in real terms. over a 30-year horizon. this definition poses several problems.

5% 2% 1. depending on the issuer.5% 0 1929-2013 1959-2013 1989-2013 Cash ZC buy and hold Source: Amundi Research Second. much in line with the drop in ratings for peripheral countries. and at least for strong currencies. 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. the crisis in confidence in the banking system and the increase in Libor rates led to a distortion in swap rates.5% 1% 0. the tendency was to consider government bond rates as risk-free. linked not to the risk of these instruments. Libor and OIS 10y swap rate: Euro 6 0. One remedy was to replace the Libor 3-month rate by a daily rate that was therefore less distorted.US 3. a common reference curve is a fundamental requirement. the sovereign debt crisis led to a serious discrepancy in government borrowing rates depending on the country. such an investment may carry a credit risk. but rather to the very strong distortion of the Libor 3-month rates used as a reference for variable rate swaps.3 2 0. Within the eurozone.2 1 0.6 5 0.5 4 0. 96 Amundi Investment Strategy Collected Research Papers . 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.5% 3% 2. Unless we accept the highly debatable view that returns on assets in euros would be different according to country. Swap rates are of course an option as their collateralisation means they incur very little credit risk.4 3 0. Having said that. Dispersion of annualised real return .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) Source: Amundi Research Lastly. Before the crisis.

in any case. in nominal terms at least. Wherever possible. but it is only justified if investors anticipate that rates will remain stable which will mean a flat yield curve. 3. Furthermore. We have seen that this point of view is debatable and. This approach is based on two arguments. the rate at which the economy finances itself is a mix of short and long rates. The virtue of this solution is its simplicity. Finally.These findings suggest that there are in fact no universally risk-free assets over the long term.Approaches 3. 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. In practise. 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. it is not entirely risk-free. 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. the return on a buy and hold bond investment is known in advance and can therefore be considered as risk-free. much too variable over time to make it an acceptable hypothesis. proxies can be used as long as they do not present any substantial risk in relation to a day-to-day investment. in no way guarantees the link between the initial long rate and the return Amundi Investment Strategy Collected Research Papers 97 . It is nonetheless prudent to remember that. this method also requires an estimation for growth and inflation at equilibrium. 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. 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.1 – Macroeconomic equilibrium model In a balanced economy. except over short-term horizons. The slope between long and short rates is. the interest rate at which the economy finances itself is equal to the sum of the equilibrium values for (real) GDP growth and inflation.Yield curve The great merit of methods that use the yield curve is that they are based on observable data and not on forecasts. An investment decision must factor in the known or unknown horizon and the outlook in nominal or real terms.2 . 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. while this asset may bear the least risk. Another approach is to use the zero-coupon rate over the investment horizon as a normative forecast for the return on cash. The first is to say that. A first solution is a pure and simple extrapolation of the current short rate. III . however. and they correspond more to what would be achieved via a regularly renewed short-term investment.

particularly for long horizons. for long or forward rates. The second is that short rates are essentially administrated and do not factor in agents’ future forecasts. Morton [1989] and Hull and White [1997]. Over the last 30 years.50% -0. For our purposes. The first is that it implicitly supposes that short rates evolve in a deterministic manner and fails to factor in their random nature.60% -0. As illustrated in the graph below. countless variations have been applied to different asset classes. this theory is used to establish that future short rates are lower than forward rates.00% DIFFERENCE MODEL .70% -0. It also provides a framework which allows for the clear separation. after which adjustments over the entire yield curve were introduced by Heath. Despite its simplicity. when it comes to current standard levels. Correction in absolute terms is the more significant so as volatility is high and risk aversion is strong.on cash as the investment is not made on the same asset.90% -1. Expected annualised cash return .10% -0.20% -0.SPOT -0. The first to apply the theory to interest rates was Vasicek [1977] for short rates.correction to be applied to spot ZC rate 1 6 11 16 21 26 31 36 41 46 0. i.e. this model is criticised for various reasons. between what is linked to rates forecasts and what constitutes a risk premium. to say that the best forecast for the behaviour of short rates is that they will evolve towards forwards.3 Arbitrage pricing theory This theory was presented for the first time by Black and Scholes to value options on stocks. there is a risk premium between long and short rates which explains why the slope of the yield curve is most often positive. regulatory changes which affect the appetite of buyer pension funds for very long bonds).80% -0. 3. 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. 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. on the other hand. Long rates. In actual fact. Jarrow. 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]). This theory is used as the basis for the valuation of option derivatives and explicitly takes into account the uncertainty linked to future rates.40% -0. the size of the correction is significant. This also enables us to define a coherent trajectory for short rates.00% Horizon (yrs) HJM Source: Amundi Research 98 Amundi Investment Strategy Collected Research Papers .30% -0.

spot ZC -0. 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 . It is nonetheless useful to keep in mind the implications of this theory. Today. over a long period of time.Historic analysis of long series in the US We have examined. The following graph shows the difference between dispersion figures. and whose long and short rates are not too distorted by credit risk and the government administration of foreign exchange rates. the level of volatility affects the risk premium.the difference constituting the forecast error. Over a 7-year horizon. the effect becomes really significant when volatility of long term rates is high.20% -0.60% -0. For the 7-year horizon.00% -1. Volatility Source: Amundi Research While these approaches are interesting. we compared the dispersion of the forecast errors with the actual unconditional dispersion of the annualised returns on cash over the same horizon. except over short horizons. Correction to be applied to initial spot ZC rate in order to forecast annualised cash return as a function of volatility 0. To ascertain whether the use of the initial short rate is actually relevant. This is more delicate when it comes to the eurozone where. short rates were affected by German reunification and the currency crises in the European monetary system. in the 1980s and 1990s. Forecasts based on the initial short rate alone are not very credible. for a volatility of around 0.5%. the relationship between long and short rates in the US which is the country for which we have the longest historical series.00% Difference model . Amundi Investment Strategy Collected Research Papers 99 . IV .20% % % 1% % 2% 25 50 50 0. they require complex calculations and the calibration of a substantial number of parameters.80% -1.40% -0. 0. the correction is around 35 basis points. All other parameters were kept constant for these calculations. 1. We looked at several horizons ranging from 3 months to 30 years. for example.Over a given horizon. The graph be- low shows how the level of volatility on long rates affects the correction.

However. We examined whether it is more useful to know the initial long rate. Our aim is to identify the relationships between the forecast errors using the short rate or the initial long rate and other variables.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. 100 Amundi Investment Strategy Collected Research Papers .5 % 2% 1. This is not verified over long horizons for which it is therefore much less important to know the initial short rate. we examined the cumulative returns on a short-rate investment over 7-year periods.5 % Annualised dispersion 3% 2. The dispersion is a little lower than with the initial short rate but is still strong.5 % 1% 0. 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. In fact. and analysed the forecast errors made using the initial long rate for each sub-period. the dispersion on forecast errors is much lower than the dispersion on the actual unconditional returns for cash. Dispersion of cash performance 3. because a coupon bond with a 10-year maturity has a shorter duration (around 7 years). over short horizons. it is virtually nil which is how cash earned its reputation as a “risk-free” asset. and the long rate for which we have very long series is the 10-year bond yield.

00 -4.In the first instance.00 -10.75% -3.00% FORECASTING ERROR 5. this forecast error is linked to the slope.00 Foracasting error using initial long term rate Forecasting error using initial short term rate Linéaire Linéaire (forecasting error using initial short term rate) (foracasting error using initial long term rate) Source: Amundi Research What we find is that there is a positive relationship between the forecast error using the initial short rate and the slope.00% -2.00 -8.00% 3.00 0 -8. we limited our analysis to periods where initial conditions are not fundamentally different (initial inflation between 0% and 5%).00 6.00 -6.25% SLOPE -7.00 -2.50% 1. This means that the initial long rate globally overestimates the future return on cash. it is also variable over time with a sensitivity of around 50% on average. 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.00 2.75% 0% -1.00 0 2.00% 4.00 -4.00 SLOPE -12. More specifically.00% VForecasting error using initial short term rate –––––Linéaire (forecasting error using initial short term rate) Source : Amund Research While this relationship is significant and robust. Forecasting error on cash return and initial yield curve slope: 1930-2013 7.00% 2. Amundi Investment Strategy Collected Research Papers 101 . The graph below shows that there is also a link with the volatility of rates.25% 3.00 -2.00% -1. all the more so when the slope is steep.00 4.00% 0% 1. the graph below shows that when we use the initial short rate alone.00 4.00 FORECASTING ERROR 6.00 -6.50% -5. Using the initial long rate on its own creates an error with a negative and often downward bias in relation to the slope. Forecasting error on cash return and initial slope: 1929-2013 with filter 8.

25% -6.50% 3. The scope of the phenomenon is also significant since.3. The second is that the increase in risk aversion during periods of strong volatility appears to accentuate the phenomenon.8%.0% 8.50 2.75 1.75% 0% -1.25% 3. We can also see that the forecast error is linked to the real short rate.0% 4. ForFcBsting error on cash return and initial volatility: 1929-2013 with filter 6. There are two possible explanations here.0% -4. distorted due to a very restrictive or 102 Amundi Investment Strategy Collected Research Papers .50% INITIAL VOLATILITY OF LONG TERM RATE -6. The first is that the decline in rates during the period in question meant that the Sharpe ratio was higher. While this is consistent with the arbitrage pricing theory.50% INITIAL REAL RATE -5.00 VForacasting error using initial long term rate –––––Linéaire (foracasting error using initial long term rate) Source: Amundi Research Using the initial long rate means we overestimate future returns. which involves complex calibration.75% -3.00% 1.50% -3. The correction linked to volatility. for 1% volatility.50% 1. using the initial long rate overestimates the return on cash by an average 0. all the more so when volatility is high.00% 0 0.25 3. Forecasting error on cash return and initial real rate: 1929-2013 with filter 7.0% 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.50% 0% -1.0% 0% 2.00% FORECASTING ERROR 5. using short and long rates together therefore seems more effective than using either one or the other on their own.00% FORCASTING ERROR 4. is less useful.00% -4. Given today’s extremely positive curve and the low volatility in rates. it is even higher than what is predicted by the theory with a Sharpe ratio of 0.0% 6.0% -2.

where available.50% 0% -1.5% 5. Today.0% VForecasting error using mixture –––––Linéaire (forecasting error using mixture) Source: Amundi Research The preferred alternative. is forward inflation. We have found no link between the forecast error and growth. This enables us to eliminate a substantial part of the dependency between the forecast error and the slope and real initial rate.8% 0% 1. The initial long rate segment (50%) factors in the data on the future trend in rates extracted from the yield curve.50% -5. Forecasting error on cash return and change in inflation: 1929-2013 with filter 6.accommodating monetary policy. failing that. The initial short rate segment (25%) reflects the fact that. inflation (25%) and zero-coupon long rates (50%) over the horizon as a normative forecast for returns on short-term investment. V . The graph shows that there is a strong relationship with the variation between initial and cumulative inflation. we recommend using the OIS curve rates or. is not consistent with the fundamentals. government bond rates where governments have an acceptable rating or are in control of their currency. on average.Operational implications In order to marry operational simplicity and coherence with theory and observation.8% 3.3% 7. which.50% 3.00% 1.50% -3.5% -1.00% FORECASTING ERROR 4. short rates are lower than their forecast using long rates and present a strong level of autocorrelation. even if it does remain highly sensitive to a change in inflation regime. 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.3% -3. we recommend using a combination of short rates (25%). Our forecast model mixes the short and long rates and initial inflation.INITIAL INFLATION -4. in the long term.00% OBSERVED . Amundi Investment Strategy Collected Research Papers 103 .

t an inversion in the yield curve. 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 Source: Amundi Research 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. 104 Amundi Investment Strategy Collected Research Papers . t a significant increase in the volatility of long rates.

Gianni Pola and Eric Tazé-Bernard at Amundi for their valuable input in improving this paper. White A [1997] “Options. Vol 5. Amundi Investment Strategy Collected Research Papers 105 . Shiller R [1973] “Inflation. 3rd Edition. Vol 40. 27. Journal of Finance. Hull JC. Morton A. Economica. Prentice-Hall. Vol. Rational Expectations and the Term Structure of Interest Rates”. Musiela M. REFERENCES Black F. Journal of Financial Economics. 2nd Edition. Ruthowski M [2004] “Martingale Methods in Financial Modelling”. Modigliani F. 60. Futures and Other Derivatives”. Journal of Economics. Econometrica. Scholes M [1972] “The Valuation of Option Contracts and a Test of Market Efficiency”. Lûtz FA [1940] “The Structure of Interest Rates”. Heath DC. Springer. Vol 55. Vasicek O [1977] “An Equilibrium Characterization of the Term Structure”. Acknowledgements I would like to thank my colleagues Jean Gabriel Morineau. Jarrow BA. [1992] “Bond Pricing and the Term Structure of Interest Rates: A New Methodology for Contingent Claims Valuation”. Vol.

106 Amundi Investment Strategy Collected Research Papers .

Amundi Working Papers Amundi Investment Strategy Collected Research Papers 107 .

108 Amundi Investment Strategy Collected Research Papers

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.

Amundi Investment Strategy Collected Research Papers 109

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.

110 Amundi Investment Strategy Collected Research Papers

**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

Amundi Investment Strategy Collected Research Papers 111

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 Attié and Roache (2008), Amenc et al.

(2009) in developed countries and Brière 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).

112 Amundi Investment Strategy Collected Research Papers

**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 Pearson’s 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

Amundi Investment Strategy Collected Research Papers 113

**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.

114 Amundi Investment Strategy Collected Research Papers

**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

Amundi Investment Strategy Collected Research Papers 115

moreover it is difficult to find a very deep historical database. Stocks move not because of low or high growth but mainly because growth is above or below expectations. The comparison between assets and factors is synchronous: we compared them in the same time period (month or quarter). Despite the CPI index rising sharply from 23. falling inflation. hence we preferred to study the polarization of asset returns towards historical variations of macroeconomic variables. 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). Chart 3 plots the alternating regimes between rising and falling scenarios for US inflation since 1950. -1). rising growth. dependency between fundamental factors and asset returns is assessed comparing asset returns with fundamental factors’ levels (e. In order to represent factors’ changes. the evaluation is ex-post and we do not need to forecast market scenarios. rising and falling scenarios have been about equally likely: rising and falling scenarios express (respectively) acceleration and deceleration of fundamental factors. 116 Amundi Investment Strategy Collected Research Papers . 13 It is worth mentioning that our point here is simply to name a specific time interval as rising or falling. we preferred to follow a parsimonious approach: for each fundamental factor we define rising and falling scenarios. and falling stress. More explicitly we related to each factor’s time-series a Boolean time sequence (+1. rising stress. Measuring expectations in macroeconomic variables like inflation and growth 12 is tricky. falling growth.75 in October 2012. and do not allow one to perform a robust statistical analysis. In this document we follow a different approach. 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. indicating in each time period a rising or falling scenario 13 (see Annex C for more details). in Annex F we addressed the issue of introducing a temporal-lag between the factor measurement and asset performance. 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.61 in December 1949 to 231.Traditionally. The aim of the present work is to segment asset classes with respect to the six scenarios. Each asset is evaluated since inception: our aim here is to individuate a long term relationship.g. According to this procedure we identify six scenarios: rising inflation. Unfortunately time-series are very short. we question whether high inflation or low/negative inflation hurt asset returns). in Pola and de Laguiche 2012.

PMI. it cannot be applied to discrete random variables. f2. Rising and falling inflation scenarios in the US 2. hence the standard Pearson’s correlation coefficient is not applicable 14. and hence the dependency among factors. real GDP. We quantify the dependency between fundamental factors as the distance between the joint probability P(f1. Annex B for a brief review). We remind the reader that fundamental factors are expressed by Boolean time-series of +1 and -1. f4. x a non-zero distance signals a statistical relationship among some factors. f2. The Kullback- Leibler distance (KL distance in the following) permits one to quantify the distance between two probability measures (Cover and Thomas 2006. 14 The standard Pearson’s correlation coefficient measures the relationship between two continuous variables. For the sake of simplicity let us name the four factors (CPI. pair-wise relations are more relevant than higher order ones.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. 12 Amundi Investment Strategy Collected Research Papers 117 . f4) and the product of marginal probabilities P(f1)P(f2)P(f3)P(f4): x if the distance is zero.Chart 3. moreover it is not very appropriate for our sample because time-series inceptions are not homogeneous across factors. But usually. For this reason we preferred to investigate pair-wise orthogonality 15. This approach does require large amount of data samples (see Annex B for a discussion on data sampling). 15 Indeed pair-wise orthogonality does not guarantee full independence among factors. f3. factors are independent of each other. f3. and market stress) as f1.

x CPI vs. The equilibrium is slightly moved towards off-diagonal elements: this means that rising (resp. The off-diagonal figures are more likely than the diagonal ones. 118 Amundi Investment Strategy Collected Research Papers . indicating a strong (positive) relationship between them. Among the computations. STRESS. real GDP. STRESS. x Real GDP vs. The main findings are: x CPI vs. Chart 4 summarizes the result: The scenarios are well distributed. rising) growth ones. 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. x CPI vs. In order to have perfect orthogonality between factors.First of all we computed the probability of each scenario. The most likely joint scenario is falling real GDP and falling stress. PMI. The equilibrium is clearly moved towards the diagonal elements. we expect to find a joint probability close to 25%. PMI. Probabilities are estimated from the frequency table. x Real GDP vs. this corresponding to rising and falling scenarios of the two factors being completely unrelated each other. falling) inflation scenarios are more likely linked to falling (resp. those involving the stress scenario are less significant because of the short data sample. Almost uniformly distributed.

PMI. we computed the historical joint probability for CPI and real GDP scenarios in rolling windows (10-year trailing windows. a deeper historical analysis is advisable. at least. and computed the KL distance.x STRESS vs. In order to test significance.000 times. As we showed in Pola and de Laguiche 2012. This outcome is confirmed by the lower panels which report the marginal probabilities for CPI and real GDP. 40 quarterly observations). 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. part of the anomaly can be explained by the sample specificity of the analysed period: in order to assess the significance of the result. Chart 6 reports the main findings. and market stress.000 computations of the KL distance. 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. in order to get 10. Test of significance has been achieved through a non-parametric approach 16. In order to compute the KL distance we only need the two Boolean sequences indicating rising or falling scenarios. 14 Amundi Investment Strategy Collected Research Papers 119 . the joint probability is identical to the product of marginal probabilities). 16 In order to clarify the approach. and x orthogonality among real GDP. The analysis shows that: x there is clear redundancy between real GDP and PMI. real GDP. PMI. any possible relationships are certainly spurious and originate from the data sample. real GDP. we firstly randomly reshuffled the Boolean sequence of GDP. It is worth stressing that computations from reshuffled pairs should not signal any relationship between CPI and real GDP.e. investigating the likelihood of inflationary and deflationary scenarios in a deeper historical analysis. This way allows one to evaluate if the KL distance is significant or spurious. it emerges that the last century itself represented an anomaly. The upper panel plots the joint probability of CPI and real GDP: the 70s exhibited a marked increase of scenarios characterized by rising inflation. the last century has been characterized by inflationary scenarios. CPI and market stress holds only as a first approximation. We believe that. 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. let us consider the first pair: CPI vs. however. and greater than zero otherwise (see Annex B for more details). We performed this procedure 10. In order to further investigate the non-negligible dependency observed among factors.

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 . Chart 6.3 0.9 0. as second-order correction to the main frame. than perfectly orthogonal factors (from statistical tools) which are difficult to interpret and unstable over time. +) Marginal Probabilities of rising-falling scenarios on CPI .2 0.2 0.4 0.Indeed it is worth stressing again that we prefer to work with quasi-orthogonal axes that are easily related to macroeconomic dynamics.5 0.6 0.7 0.10 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 (.3 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 . Historical joint and marginal probabilities (focus on CPI and GDP) Joint Probability of rising-falling scenarios on CPI and real GDP .6 0..10 yr rolling analysis in US 1 0.7 0. 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.10 yr rolling analysis in US 1 0.8 0.9 0.5 0. CPI + Marginal Probabilities of rising-falling scenarios on real GDP .8 0. real GDP + 120 Amundi Investment Strategy Collected Research Papers . -) (.. +) (+ .4 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. Chart 7 (left panel) reports the (conditional) average of excess-returns over 3-month T-bills in different scenarios 17. x nominal bonds exhibit clear characteristics in terms of factors scenarios: they are good hedge against falling inflation. keeping the dependency on growth and market stress fixed).60% 1. 18 The average figure for the stress indicator is significantly different from those of CPI and growth because of a shorter data sample.3. This is not possible because in financial markets all factors act synchronously to determine asset prices. inflation in 70s.20% 1. moreover factors are orthogonal only as a first approximation. Dispersion of Excess Return across factors' scenario 1. it might be the case that some factors are more important than other (e. Conditional analysis of US Treasury across macroeconomic scenarios US Treasury. In small data sample. A longer data sample enables this problem to be alleviated. Chart 7. studying the dependency of nominal bonds versus movements of inflation.00% 5% 0. falling growth and rising stress scenarios. the chart includes also the unconditional average 18.20% -5% 0.00% -0. Average Excess Return across factors' scenarios US Treasury. We start with a naïve approach: for each asset and factor we divide asset return in correspondence of rising and falling scenarios. 16 Amundi Investment Strategy Collected Research Papers 121 . Observations are quarterly. Ideally it is desirable to investigate the relationship between assets and each factor separately (e. market stress in 2008). leading potentially to misleading results. Two main findings emerge: x nominal bonds present positive risk premia.g. growth in the Great Depression.40% 0. In Chart 7 we perform this exercise for the US treasury. 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. whereas asset excess return over cash responds mainly to surprises in expectations of macroeconomic variables and market stress. Growth in this chart refers to real GDP.40% 10% 1.60% 0% 0.80% 15% 1.g.80% 0.

x abs(P) corresponds to the probability to polarize to specific rising/falling scenarios (obtained by a t-test for the parametric computation. x 0. We define significance bands. (We name with P param and P non-param respectively the parametric and the non-parametric indicators. as follows: x 0. as well as figures within +1 standard deviation and -1 standard deviation. 19 We report the maximum and minimum observations. First of all we consider two statistical tests: a t-test on the difference of means between rising and falling samples (parametric test).99P<1 preference for rising scenarios with a high confidence level. and a bootstrap technique for the non-parametric one). P shares with ȡ to be in between -1 and 1.90P<0. see Annex D for all the technical details regarding the derivation of the indicators. x 0. it presents two limitations: x hypothesis on gaussianity of returns is certainty violated in many asset classes. Chart 7 (right panel) reports the dispersion 19: results are noisy.) The main properties of P are: x sign of P indicates a preference of the asset class for scenario +1 (rising) or -1 (falling). hence we need to quantify how robust the polarization is towards a specific factor scenario.70 weak tendency to prefer rising scenarios. 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). We believe that the coupled application of the two tests may help us to conduct a more robust analysis. 122 Amundi Investment Strategy Collected Research Papers .99 preference for rising scenarios with a good confidence level. While the parametric test is the standard in the industry.70P<0. The parametric and non-parametric tests allow one to define the polarization coefficient P 20. and a non-parametric test based on the historical probability that the rising sample “outperform” the falling sample (see Annex D for more details).90 preference for rising scenarios with a moderate confidence level. and x it does infer about the mean of the two samples.60P<0. and some further investigation is needed.However. whereas we are much more interested in the entire distributions. whether ȡ compares two continuous variables. a more detailed analysis shows that the picture is not very neat. x 0. 20 We label the polarization measure with capital rho P: this is to indicate the analogy with the standard Pearson’s correlation coefficient ȡ.

alpha equity indices (long equity sectors and styles. domestic equity styles (value.88% . x -0. Asset segmentation: empirical results In this section we compute the polarization coefficient P for a large investment universe from a US investor’s perspective.60P<0. and a global risk parity balanced strategy. x -0.70P<-0. large.99 preference for falling scenarios with a high confidence level. long-only trend follower strategies on a diversified investment universe).53%. corporate bonds). commodities. Moreover we considered zero-duration credit indices (in order to disentangle credit dynamics from interest rate ones).90P<-0.70 preference for falling scenarios with a moderate confidence level.65% . 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 country’s 18 Amundi Investment Strategy Collected Research Papers 123 . P non-param = -99. x -0. x -1P<-0. then we move towards a more comprehensive three-dimensional chart that expresses polarization of assets to the three fundamental factors (the polarization cube). international emerging market bonds (local currency and hard currency). x Rising Stress: P param = 99.85%.x -0. domestic equity.60 weak tendency to prefer falling scenarios.85%. x Falling Growth: P param = -99. long-short trend follower strategies on long term nominal bonds and fx-rates. growth. hedge fund strategies (including CTAs. falling growth and rising stress with a high confidence level. 4. short S&P 500 to neutralize beta to S&P500). We firstly introduce a two-dimensional representation (the polarization plan). Indeed the parametric and non- parametric measures are coherent.90 preference for falling scenarios with a good 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. P non-param = 99. volatility.60 blend response (uncertainty). small) and sectors. The figures above demonstrate that nominal bonds are sensitive to falling inflation.99P<-0. inflation-linked bonds.67%. short S&P 500).1 Asset class universe We consider the point of view of a US investor investing in domestic bonds (nominal bonds. P non-param = -99. 4. and beta-neutral equity indices (long equity sectors and styles.

x FX-rates: 19 indices. x Dynamic strategy: two strategies (including a Risk Parity strategy26. Germany. Canada. long-short trend follower on fx-rates24. Japan. Long-short trend follower on fx follows the same investment scheme. Oil WTI). Canada). Australia. USD. Copper. AUD. and 25 beta-neutral indices 23 (including sectors and styles). Commodity (Gold. CAD. Japan). China. BRL. Italy. x Trend-follower and CTAs: six indices/strategies (including long-short trend follower on nominal bonds 24. below) the signal is long (resp. MXN. and a dynamic strategy27 selecting securities according to their historical sharpe ratio). ZAR (carry returns indices). NOK. Emerging Market Debt in Hard currency and Local currency) and three zero-duration indices 21 (Investment Grade. High Yield. the portfolio then is leveraged in order to match a target volatility of 8% (estimated on the previous year historical observations). US. Australia. UK. Portfolio construction is given by asset historical volatility for the previous year (risk budget is the same for all the markets). Risk allocation is rebalanced once a month (the last Thursday of the month). Each month the portfolio is determined according to asset historical volatility for the previous year. 21 Indices obtained taking the excess returns over government bond of same maturity (Merrill Lynch computation). UK. short). GBP. Japan. Germany. and USD vs. NZD. Equity (Euro. volatility targeting is on a daily basis. We considered transaction costs equal to 15 bps for 100% portfolio turn-over. JPY. and invests in 7-10-yr bonds from US. 124 Amundi Investment Strategy Collected Research Papers . Agriculture. 24 alpha indices 22 (including sectors and styles). x Volatility: three indices. TWD. two long-only trend follower strategies on a diversified investment universe 25). Beta estimation is on the whole data sample. we analyzed 139 assets/strategies: x Nominal bond: four indices with different maturities. Risk budget is equally divided between nominal 7-10-yr bonds (US. x Inflation-linked bond: three indices (two US. Going into more details. 22 Synthetic indices obtained going long on the sector or style and going short 100% on the S&P 500. AUD. Portfolio is not leveraged. High Yield. calculation on a monthly basis. UK. x Credit markets: eight indices (including Investment Grade. x Hedge Funds: 11 indices.macroeconomic dynamics and market stress. KRW. calculation on a monthly basis. Long-short signals are given comparing the index with the moving average for the last 100 days: if the price is above (resp. 24 Long-short strategy is designed on a monthly basis (the last Thursday of the month). one global emerging). x Commodity markets: six indices. Investment universe is EUR vs. Our hypothesis is that strategies are implemented through derivatives instruments. Emerging Market Debt in Hard currency). 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. JPY. 26 Risk-parity allocation is derived within a monthly rebalancing. x Equity market: 25 (including sectors and styles). 25 They are labelled in the following as riskparity8 trend and cstra95. CAD.

the closer the spots are to the border. whereas real estate stocks are very sensitive to growth. they are rather insensitive to inflation changes. in the next section we present a more comprehensive analysis. 27 The strategy invests in a large. On the other hand. We remind the reader that. and to a lesser extent to falling inflation scenarios. commodities). Let us summarize the main findings: x Nominal bonds are excellent hedges in case of falling growth and falling inflation. to a lesser extent in case of falling growth. x The 1-5 segment of Inflation-linked bonds is an excellent hedge for rising inflation scenarios (and rather insensitive to growth). 20 Amundi Investment Strategy Collected Research Papers 125 . x Equity is an excellent hedge in case of rising growth. Emerging Market Debt in Local currency (EMDL) behaves like commodity CRB. preferring rising growth and rising inflation. equities. Chart 8 shows the behaviour of a selection of asset classes versus CPI and real GDP as a proxy of growth (here. The Inflation-linked all maturity index responds well to falling growth scenarios. according to the definition of P. Emerging Market Debt in Hard currency (EMDH) results almost blend versus changes in inflation and growth. Explicit constraints allow us to control portfolio turnover. This section is devoted to presenting the main ideas. High-yield bonds offer a good preference for rising growth scenario. The strategy is long-only on all assets. but weakly related to inflation changes. x Corporate bonds do not present a homogeneous response. with the exception of fx-rates where we take long-short positions.2 Getting started: the polarization plan. The equity energy sector tracks commodity investments. The strategy is designed to invest on the best ten indices in terms of realized historical Sharpe ratio on a given time interval. and global investment universe including more than 150 indices (fx-rates. nominal bonds. Investment grade bonds are an excellent hedge in case of falling inflation. diversified. An effective way to represent polarization of asset classes to two factors is to scatter plot their polarization coefficients P. description) of the indices.Annex A reports the details (Bloomberg code. performing better in rising growth and rising inflation. 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). inception. we report P param for quarterly observations). the more significant the asset response to macroeconomic changes. and. 4. surprisingly.

x proxy rising inflation: portfolio of commodities and inflation-linked bonds. x The VIX index is similar to nominal bonds in terms of growth. Agriculture is very sensitive to inflation. 126 Amundi Investment Strategy Collected Research Papers . because it is strongly polarized to growth as well. but rather independent from growth. as follows: x proxy rising growth: portfolio of equities and commodities. indeed they are sensitive to growth as well). x proxy falling growth: portfolio of nominal bonds and inflation-linked bonds. Industrial Metals. 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. x proxy falling inflation: portfolio of nominal bonds and equities. it might be wiser to consider combinations of assets. In order to individuate diagonal proxies 28 for the rising/falling scenarios. Gold presents a tendency to rising growth and rising inflation. commodity is not a proper proxy of inflation. 28 With this term we mean proxies that exhibit a clear relationship versus a factor. to a less extent with respect to inflation. even if the result is less significant. but are rather insensitive to other factors: e. Chart 8. commodities are commonly reported as proxy of inflation.g.x Commodities CRB.g. and Crude Oil WTI are excellent hedges for rising inflation and growth.

inflation-linked bonds (magenta). top panel). slightly changing the definition of rising and falling scenarios. The above scheme for proxies can be interpreted. Different colours refer to different asset classes. equity beta-neutral indices (yellow). respectively. and in some cases in reverting the relationship in a counterintuitive way. In order to express the polarization of asset returns to the three factors. In Annex F1 we investigated whether a temporal lag may lead to more robust indications or not. Results are robust and coherent. equity alpha indices (light orange). CTAs and dynamic strategies (black). Chart 9 lower panel reports the two-dimensional projection along the stress axis. we scatter plot the polarization coefficients P in a three-dimensional space. the plot refers to the parametric evaluation of the polarization coefficient P for quarterly data. as follows: nominal bond (blue). hedge funds.3 The polarization cube. ask what is the position of a portfolio of those assets. from a geometrical point of view. trend-follower. commodity (green).These combinations allow one to get well polarized proxies versus growth and inflation without any relevant biases towards inflation and growth. The effect of lagging is to reduce the significance. 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. zero-duration credit markets (cyan). In the next section we present the results in a more exhaustive way. We conclude this section with three remarks: x Comparison between assets and factors is synchronous. We introduced the blend scenario in case of weak signals. the result will generally not lie within the linear interpolation of the two points. moreover we consider the real GDP as proxy of growth. credit markets (light blue). 22 Amundi Investment Strategy Collected Research Papers 127 . Coherently with Chart 8. fx-rates (brown). equity (orange). x In Annex F2 we verified the robustness of the analysis. volatility (red). 4. By construction all assets will be represented by spots in the interior of a cube (see Chart 9. x The mathematics behind the polarization coefficient P is not linear: this means that if we take two assets in the polarization plan. as a rotation in the polarization chart of about half of a right angle.

128 Amundi Investment Strategy Collected Research Papers .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). Telecommunication Services equity- sector (alpha and beta-neutral indices). to a lesser extent. Russell 1000 Growth and S&P 500 Growth (alpha and beta-neutral indices). sorting asset classes according to their joint polarization to inflation and growth. x Falling-inflation AND falling-growth corner.70. For the sake of simplicity we plot the three slices integrating away the dependency on the market stress factor (see Charts 10. In the following we describe each slice in detail. the zero-duration indices are definitely placed in the rising-growth rising-inflation corner.70. alpha and beta-neutral indices permits us to populate the falling-growth falling- inflation corner. The best hedge is given by Nominal Bonds (intermediate. but it is less significant in terms of polarization to growth) and. x Rising-inflation AND falling-growth corner: this class is empty. x Falling-inflation AND rising-growth corner: this class is empty. Less significant assets in this scenario are USDNOK fx-rate. 11. and 12). all and long maturity). x the falling stress slice is defined for assets exhibiting P stress between -1 and -0. Trend-follower strategy on Nominal Bonds (dynGOVY). x despite the behaviour of credit markets is not homogeneous. the Health-Care and Consumer- 24 Amundi Investment Strategy Collected Research Papers 129 . x the blend stress slice includes assets with P stress between -0.70 and 0. The best hedge is given by the Information Technology equity-sector alpha index (the beta-neutral index is placed in the same corner. VIX Futures Enhanced Roll index. USDSEK fx-rate. Chart 10 indicates assets that perform well in case of rising-stress: x Rising-inflation AND rising-growth corner. and USDCAD fx-rate. and the equity spread Growth vs Value. 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. 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.

and Agriculture. HFRI Short bias. x Rising-inflation AND falling-growth corner: Inflation-linked all maturity and USDJPY fx-rate. x Falling-inflation AND falling-growth corner. The best hedge is given by Consumer Staples. Long-only Trend follower on a diversified basket (riskparity8 trend). and to a lesser extent Health Care equity-sector and Emerging Market Inflation- linked bond. Less significant are the Corporate Investment Grade (all 130 Amundi Investment Strategy Collected Research Papers . Less significant are Inflation-linked 1-5 years (high confidence level on inflation axis. x Falling-inflation AND rising-growth corner. polarizations are weak in this case. Long maturity Corporate Investment Grade. Staples equity-sectors (alpha and beta-neutral indices). The best hedge is given by the S&P 500 High Dividend alpha index. VIX. Chart 10 Chart 11 includes assets that are rather independent from market stress factor: x Rising-inflation AND rising-growth corner. but poor polarization to growth direction). The best hedge is given by the Information Technology equity-sector. EURUSD fx-rate. CTA long term.

High Yield zero-duration index. EURUSD fx-rate. Russell 2000 Value (alpha and beta-neutral indices). The best hedge is given by the S&P 500. Industrials equity-sector (alpha and beta-neutral indices). KRWUSD fx-rate. ZARUSD fx-rate. Consumer Staples and Financials equity-sectors. x Rising-inflation AND falling-growth corner: RUBUSD fx-rate. The best hedge is given by the TWDUSD fx-rate. Crude Oil WTI. See Annex E for more details. Emerging Market Debt in Local currency. Asian dollar index. and Small Cap 600 growth (alpha and beta-neutral index). S&P500 Energy equity-sector (pure and beta-neutral indices). 26 Amundi Investment Strategy Collected Research Papers 131 . Industrial Metals. the polarization is very weak in this case. Gold Miners stocks. x Falling-inflation AND rising-growth corner. USDDKK fx-rates. HFRI Fixed Income Convertible arbitrage. Chart 11 Chart 12 reports assets that polarize to falling-stress scenarios: x Rising-inflation AND rising-growth corner. S&P 5000 High Dividend. Corporate Investment grade zero-duration index. Gold miners stocks beta-neutral index. and to a lesser extent. Commodity CRB. See Annex E for more details. and Utility equity-sector alpha index. HFRI Equity Hedge. maturity and intermediate maturity).

the same for equity and inflation-linked 132 Amundi Investment Strategy Collected Research Papers . 5. growth and market stress. to a lesser extent. Chart 12 We remind the reader to refer to Annex E for more details. 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. equity. The best hedges are Financial equity- sector beta-neutral index and Russell 1000 Value alpha index. inflation-linked bond.x Falling-inflation AND falling-growth corner. growth and market stress changes. S&P 500 High Dividend and Russell 1000 Value (beta-neutral indices). and. Implications of DAMS segmentation: duality and synthetic assets From the previous section. similarly inflation-linked bonds and equity might be combined together to neutralize their environmental biases. it emerges that nominal bond. we will say that nominal bonds and commodity are dual with respect to inflation. On this basis.

pension funds in some countries) are not allowed to invest in commodities. 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). 28 Amundi Investment Strategy Collected Research Papers 133 . P param (growth)=89. in designing balanced funds which are less sensitive to specific macroeconomic scenarios. at least. (3): P param (inflation)=94. relationships in Eqs (1) and (2) may help us to design synthetic asset classes.g. Let us consider for example the case of Commodities. even if in this case the relationship is less robust. Indeed it might be wiser to combine Eqs (1) and (2): Commodity CRB § Inflation-linked bonds + Equity – Nominal bonds. (1).61%. We then computed P param for the synthetic series (quarterly data) computed according to Eq. hence it might be wiser to design allocations that are more balanced across macroeconomic scenarios. Firstly. (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. 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. The same calculations on monthly time-series read: P param (inflation)=99.21%. The relationships above offer a possible solution. Many funds (e. however this position would be also structural short of the risk premia of nominal bonds.bonds. According to Eq. Macroeconomic trends may last for many decades. for two purposes. the macroeconomic behaviour of commodities might be represented by a structural short position in nominal bonds.41%. P param (stress)=-99. These relationships can be useful. x and the portfolio leverage has been set in order to target the long term volatility of CRB.80%. Secondly.

134 Amundi Investment Strategy Collected Research Papers . Nevertheless recent portfolio construction schemes.g. inflation in the seventies. This statement suggests a new way to segment asset classes according to their similarity in responding to changes in expectations.35%. and stress in 2008). challenging. 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. P param (growth)=97. equity.g. growth in the Great Depression. make of diversification the kernel of asset allocation. The factors approach provides a new. the new challenge in asset allocation suggests diversifying on fundamental factors that are believed to be the main drivers of asset price dynamics. with the differences coming from market specificities that are not captured by the factors dynamic. While most approaches diversify on asset class level. Even non-stationary patterns of correlation between asset classes might be brought back to the dominance of one factor over the others (e. 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. we found that each asset class presents mixed relationships with fundamental factors. like risk parity and maximum diversification. high yield and real estate stocks). This new way forces us to rethink asset segmentation. In this work. and powerful way to interpret financial markets. P param (stress)=-99.99%.g. 6. 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. commodity). (ii) assets at the edges present high polarization to two factors. and rather blend the response to the third factor (e. We conclude that the basket accurately represents the response to macroeconomic changes of CRB. nominal bond.

(iii) constructing asset allocations better adapted to face perspective macroeconomic scenarios like inflationary ones 29. 30 Amundi Investment Strategy Collected Research Papers 135 . (viii) generating Monte Carlo synthetic assets that are sensitive-or-blend to specific macroeconomic environments. 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. dynamic strategies like trend follower and risk parity). (vii) determining expected returns better adapted to the current uncertainty in financial markets. and we will illustrate the DAMS investment process). (iv) investigating the underlying macroeconomic bias in already managed balanced portfolios. G Pola. it is desirable for portfolios to be placed in the centre of the cube especially in uncertain macroeconomic and financial market conditions. (vi) building a macro equivalent synthetic asset allocation which tracks precise macro behaviour. (ii) providing new insights in crisis management to protect portfolios against extreme financial events and stress. correlation among asset classes. Amundi Special Focus. (iii) assets in the centre of the cube’s faces respond strongly to one factor only (e.g.g. Whether assets with specific polarization to factors are particularly appealing for asset allocation. 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 ”. and the relation of equity sectors and styles to macroeconomic dynamics29. (iv) assets in the centre of the cube are rather insensitive to factors dynamics (e. (v) deepening our understanding of asset prices dynamics. inflation-linked 1-5 yr bond).

rather than on standard mean-variance optimization packages that need forecasting returns.mostly rephrased in terms of asset environmental biases towards macroeconomic and stress factors. 136 Amundi Investment Strategy Collected Research Papers .

Moreover I would like to thank Ph. DAMS is in place at Amundi Milan within a range of flexible funds (diversified DAMS). Facchinato and C.Acknowledgement I would like to thank S. absolute return (income DAMS) and equity funds (equity DAMS) since December 2011. 32 Amundi Investment Strategy Collected Research Papers 137 . The ideas investigated in this paper constitute the foundation of the DAMS investment process. Ithurbide and S. which has been developed by Amundi Milan Investment Management and G. Casadei for very illuminating discussions and suggestions regarding this research topic. Pola (Amundi Paris Quant Research). de Laguiche for their contribution which improved the research study and quality of the manuscript.

138 Amundi Investment Strategy Collected Research Papers . which represent the number of standard deviations that current financial conditions lie above or below the average of the January 1994-June 2008 period. Financial Conditions Index combines yield spreads and indices from U.S. Time-series are up to December 2012. Time-series from September 1948. x Stress Index. Bloomberg US Financial Conditions Index 30 (Bloomberg code is BFCIUS Index). US Real GDP Seasonally Adjusted (Bloomberg code GDP CHWG Index). and Bond Markets into a normalized index. ISM Manufacturing PMI Seasonally Adjusted (Bloomberg code is NAPMPMI Index).S. We investigated 139 assets/strategies as detailed in tables A1. The values of this index are Z-scores. Equity Markets. Time-series from September 1996. A2 and A3. Time-series from December 1948. x PMI. US CPI Seasonally Adjusted (Bloomberg code CPI INDX Index).Annex A – Database description We considered four fundamental factors: x Inflation. Time-series from December 1953. 30 The Bloomberg U. x Real Growth. Money Markets.

34 Amundi Investment Strategy Collected Research Papers 139 .

140 Amundi Investment Strategy Collected Research Papers .

36 Amundi Investment Strategy Collected Research Papers 141 .

y).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.y) and P(x)P(y). The KL distance permits one to measure the distance between P(x.y) be the joint probability of outcomes (x. The KL distance is defined as follows: P ( x. 2006). hence to quantify the degree of dependence. Let us consider two random variables x and y taking values from a discrete set 31.y)=P(x)P(y). Let P(x. y ) DP ( x. x and y will be independent if and only if P(x. Let P(x) and P(y) be (respectively) their probability distributions. This notion of independence is stronger than that one given by the standard Pearson’s correlation coefficient 32. we do not need the set to be of finite dimension. y ) || P( x) P( y ) .

(B1) x. ¦ P( x. y) log P( x) P( y) . y ZKHUHWKHVXPLVH[WHQGHGWRDOO[DQG\VXFKWKDW3[.

DQG3\.

and equal to zero if and only if P(x.y)=P(x)P(y).WFDQEHSURYHQWKDW the KL distance is always greater than zero. This property justifies why this measure is called distance 33.. y ) || P ( x) P ( y ) . Equation (B1) can be rearranged as DP ( x. hence any dependence between x and y is certainly to increase the KL distance.

and the joint variable (x. 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.y). y ). This representation allows one to write down the following inequality: DP ( x. H(y) and H(x.y). y ) || P( x) P( y ) . (B2) where H(x). H ( x) H ( y ) H ( x.y) stand respectively for the Shannon Entropy 34 of x.

xy2. 142 Amundi Investment Strategy Collected Research Papers . as we will see in this section. x2y2. There are several ways to get a symmetric KL measure. Finally the Pearson’s correlation coefficient is a linear measure. The KL distance enables us to alleviate many of these drawbacks. d max( H ( x). 31 We present here the KL distance for discrete random variables. It can be proven easily that the Pearson’s correlation coefficient is zero in this case. 33 However it should be stressed that this measure is not symmetric. The only con. We prefer here to work with the simplest version of the KL distance. 32 In fact the standard Pearson’s correlation coefficient can even wrongly interpret two deterministic related random variables: let x be uniformly drawn in the interval [-1. is that computing a reliable estimate of KL distance requires large data samples. and let y=x2. 1]. 34 The Shannon entropy for random variable x is defined as H(x)=-Ȉx P(x) log P(x). 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. it completely misses non-linear terms like x2y. H ( y )). as shown clearly in Lhabitant 2011. 2006). but the formalism is more general and can be applied to continuous random variables (Cover and Thomas. decreases) as well” is wrong. etc.

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 ) .

market stress). If we require at least 30 observations (on average) for each state we come out with 30Â24=480. x N ) D¨¨ P ( x1 . index i running from 1 to N. This simple evaluation clarifies why measuring the KL distance on all the factors might be difficult in practice. Let us come back to the practical example with four factors (CPI. 30Â22=120). where the KL distance was extensively used to study neural coding of sensory information.e... which is equivalent to 120 years of quarterly observations. 35 Given an entropy measure H=-Ȉi Pi log Pi .. N y and N xy are respectively the average number of relevant states 35 of P(x). (B3) © i 1 ¹ x. real GDP. 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. x N ) || P ( xi ) ¸¸ ¦ P( x . 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. 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. i. 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. A brief review can be found in Pola et al 2002. PMI... ¨ N ¸ © xy ¹ where N x .. x 1 N ) log N ..y). (B3). and P(x... log¨ x y ¸. and why we prefer to evaluate pair-wise dependency (in this case we can lower 120 years of quarterly observations to 30 years.. P(y). the average number of relevant states DUH GHILQHG DV Ș H[S+. The generalization of the KL distance to more random variables is straightforward: § N · P ( x1 .

W FDQ EH SURYHQ HDVLO\ WKDW LI 3i= 1/M for each i= 1. . …. M1 Ș 0 7KLV DQDO\WLFDO result justifies the definition. 38 Amundi Investment Strategy Collected Research Papers 143 . Meucci (2009) introduced an analogous definition.

x PMI. 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. 36 Recently Kritzman et al 2012 investigated this issue in the context of Markov switching models. Take the quarterly variations. At each point compare the current month’s variation to the average of the last twelve variations (including the current figure). otherwise as a falling-growth (symbol -1) quarter. 37 In order to simplify notation. 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. If the last observation is above the average. In case of monthly data. 144 Amundi Investment Strategy Collected Research Papers . PMI. real GDP. market stress). We will follow a heuristic approach 36. x GDP. At each point compare the current quarter’s figure (average of the three monthly observations) to the average of the last twelve observations (including the current quarter’s figure). If the last observation is above the average. x STRESS 37. At each point compare the current quarter variation to the average of the last four variations (including the current figure). we will name the quarter as a rising-stress quarter (symbol +1). we will name the month as a rising-inflation month (symbol +1). we will name the quarter as a rising-inflation quarter (symbol +1). In case of quarter data. Take the quarterly variations. the approach consists in: x CPI. If the last observation is above the average. otherwise as a falling-inflation (symbol -1) month. otherwise as a falling-stress (symbol -1) quarter. the approach consists in: x CPI. Take the monthly variations. we will name the quarter as a rising-PMI quarter (symbol +1). otherwise as a falling-inflation (symbol -1) quarter. If the last observation is above the average. we will name the quarter as a rising-growth quarter (symbol +1).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. At each point compare the current quarter’s figure (average of the three monthly observations) to the average of the last twelve observations (including the current quarter’s figure). otherwise as a falling-PMI (symbol -1) quarter.

x STRESS35. If the last observation is above the average. At each point compare the current month’s figure to the average of the last twelve observations (including the current figure).x GDP. 40 Amundi Investment Strategy Collected Research Papers 145 . Not available with monthly frequency. If the last observation is above the average. otherwise as a falling-stress (symbol -1) month. we will name the month as a rising-stress month (symbol +1). x PMI. otherwise as a falling-PMI (symbol -1) month. we will name the month as a rising-PMI month (symbol +1). At each point compare the current month’s figure to the average of the last twelve observations (including the current figure).

are independent random samples from normal distributions with equal means and different and unknown variances (Behrens-Fisher problem). 38 We considered the general case of different variances for sample A+ and A-. For each reshuffling we estimated P(i) (a+>a-). x Non-Parametric.. under the null hypothesis.(0) the true sets. quantifying how much an asset class responds to rising or falling scenarios. against the alternative that the means are not equal. Secondly we divided the performance sample of A in correspondence of symbol +1 and -1. and one-tailed distribution.Annex D – Technical note: the polarization coefficient P The aim here is to define a polarization indicator. We estimated from the frequency tables the probability that a+>a-. We question whether there is an ordering between sets A + and A . We performed 10000 reshufflings. The p-value 38 is reported as p param. Let us consider that we want to verify if there is a relationship between asset A and fundamental variable V. We indicate by A + (0) and A . we randomly bootstrapped the order of the symbol +1 and -1 to get A + (i) and A . The t-statistics. falling) scenario of fundamental factor V. p param <0.. 146 Amundi Investment Strategy Collected Research Papers .be respectively two elements of A + and A . -1). Let a+ and a. we will name them respectively as A + and A . we will name this probability P(a+>a-). Firstly we define symbol +1 and -1 in correspondence of a rising or falling scenario for indicator V. the same holds for p non-param .. In order to quantify how reliable the estimate is. In order to answer to this question we perform a conditional historical analysis. Looking at the entire estimates of P(i) (a+>a-) we will be able to extract the distribution of P(a+>a-). and by P(0)(a+>a-) the true probability. then we conclude that asset A is a good candidate to hedge against rising (resp. We considered two statistical tests: x Parametric. lower) than A . If we prove that A + is on average significantly greater (resp.50 (resp.(i) for each reshuffling (i).50) indicates a preference of asset A for scenario +1 (resp. 2002). t-test of the null hypothesis that data in the samples A + and A . is distributed according to an approximate Student’s t distribution with a number of degrees of freedom given by Satterthwaite's approximation (see Timm. It is worth stressing that p param >0. 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 .

Amundi Investment Strategy Collected Research Papers 147 .50 sign(p param -0.50)-0.Finally we define the polarization coefficients as: P param = p param +0.50 sign(p non-param -0. P non-param = p non-param +0. 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.50.50)-0.50.

Annex E – Empirical Results This section includes eight tables corresponding to the following scenarios: rising inflation. and relevant non-gaussian features in the return distributions. parametric and monthly data. falling inflation. and falling market stress. falling growth (real GDP). rising growth (PMI). rising growth (real GDP). Differences between the parametric and non-parametric evaluations are small. In general we found coherence among the four metrics (parametric and quarterly data. For lack of space we did not investigate this issue further in this work. non-parametric and quarterly data. 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 each scenario we report only assets which present a moderate polarization coefficient. non-parametric and monthly data). falling growth (PMI). Asset classes are sorted according to the average P. rising market stress. In each table we report the parametric and non-parametric estimations of P for quarterly and monthly data. 148 Amundi Investment Strategy Collected Research Papers . and may come from small data-sample effects.

Amundi Investment Strategy Collected Research Papers 149 .

150 Amundi Investment Strategy Collected Research Papers .

Amundi Investment Strategy Collected Research Papers 151 .

152 Amundi Investment Strategy Collected Research Papers .

Amundi Investment Strategy Collected Research Papers 153 .

154 Amundi Investment Strategy Collected Research Papers .

Amundi Investment Strategy Collected Research Papers 155 .

So the new rule reads as: x If the ratio is greater than 1+¨ then rising scenario.+¨] then blend scenario. It is evident that P is rather stable for different ¨. Commodity CRB). Results are reported in Chart 11. whereas macroeconomic scenarios are evaluated at time k. F1. 39 Asset performance is measured at time k+1. Table F1 permits one to compare the synchronous measurements to the temporal lagged ones: lag 1 indicates three months lag for quarterly series. Effect of temporal lag on polarization analysis. rising and falling scenarios are defined comparing the value of an index to a given moving average. Polarization analysis with rising. if the ratio is greater than 1 we name the scenario as rising. F2. We performed the polarization analysis introducing a temporal lag of one month for monthly data. and in some cases reversing the relationship in a counterintuitive way. Data analysis has been developed on the main asset classes (US Treasury. falling otherwise. We investigated the polarization indicator P for the main asset classes as a function of different threshold ¨. US Inflation-linked bond. We slightly changed this rule introducing a band around 1. x If the ratio is less than or equal to 1-¨ then falling scenario. As described in Annex C. 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. and three months for quarterly data 39. S&P500. x If the ratio is in (-¨. The effect of temporal lag is in weakening some relationships. and one month lag for monthly series. 156 Amundi Investment Strategy Collected Research Papers . falling and blend scenarios.Annex F – Additional Analysis In this section we investigate some side issues related to the polarization analysis.

Amundi Investment Strategy Collected Research Papers 157 .

Timm. G. 158 Amundi Investment Strategy Collected Research Papers . S. EDHEC-Risk Institute Working Paper. Attié. “Managing Diversification”. Working paper. 2012. R. and de Laguiche. 2002..... 2002. Kluwer. The Journal of Portfolio Management. Cover. 22-39. L. Pola... 2009. G.. p. G.. S. Amundi Working Paper WP-012-2011. April. Pola. S.. A.com. “Applied Multivariate Analysis”. M. A. Young.References Amenc. S. Panzeri S. Panzeri. 2013. “A practical guide to information analysis of spike-trains”. Ziemann. 1-44.. “Correlation vs. M. “Data-robust tight lower bounds to the information carried by spikes times of a neuronal population”... P. S. S. “Unexpected Returns.. 35(4). M. “Inflation Hedging for Long-Term Investors”. Pola. Methodological considerations on Expected Returns in Uncertainty”.. Balancing macroeconomic scenarios”. and Facchinato.. IMF Working Paper 09-90. Pola. “Regime Shifts: Implications for Dynamic Strategies”. Thiele. Amundi Special Focus. Trends: A Common Misinterpretation”. Brière. 2009. Page. M. 2011. A. “Alternative Investments for Institutional Investors.K.. Pola. S. Thomas. D. Risk Budgeting Techniques in Asset Management and Asset-Liability Management”.. Neural Computation. 2005. 2009. 2011. V. Roache. Schultz.. Book chapter in “A practical guide to neuroscience databases and associated tools”... Petersen. and Turkington. Martellini.P. 2006.. N.. Meucci. S. Amundi Working Paper WP-032-2012. 94-110. Signori. symmys. London. 2012.. “Hedging Inflation Risk in a Developing Economy: the case of Brazil”. Wiley series in Telecommunications and Signal Processing. 17. Petersen. A. “Managing uncertainty with dams. G. “Rethinking strategic asset allocation in terms of diversification across macroeconomic scenarios”. T. Lhabitant F. Kritzman. Springer. N. G. J. “Elements of Information Theory”. Financial Analysts Journal. in preparation 2013.... H. S. 2013. O. R. 68 (3).. R.

Paris Dauphine University. Amundi Investment Strategy Collected Research Papers 159 . Boston University Marie Brière. In this framework. which compromises the coordination of sovereign wealth management with fiscal policy. Real-life SWFs asset allocations differ strongly from theoretical ones. WP-039 Optimal Asset Allocation for Sovereign Wealth Funds: Theory and Practice Zvi Bodie. we frame sovereign fund management as an asset-liability management (ALM) problem. Amundi. we suggest institutional arrangements that could overcome this obstacle and enable efficient coordination. Financial management of the sovereign balance sheet is hampered by a lack of aggregate data. Starting with the sovereign’s balance sheet. covering all public entities and taking explicit account of all sources of risks affecting government resources and expenditures. Université Libre de Bruxelles October 2013 This paper addresses management of sovereign wealth from the perspective of the theory of contingent claims. monetary policy and public debt management.

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. Kazakhstan and Qatar used SWFs or public pension 160 Amundi Investment Strategy Collected Research Papers . SWFs can be managed by different institutional structures. or supporting the domestic economy (Avendano and Santiso. with low diversification and poor medium-term performance (Bernstein et al. Moreover. Chhaochharia and Laeven. mainly equity stakes in listed firms. such as controlling politically sensitive industries. Although this takes into account only a fraction of SWF investments. intervening to support their domestic financial markets (Clark and Monk. Brown et al. 2012). A large body of empirical research has analysed the public investment strategies of sovereign wealth funds and their performance. Bortolotti et al.. They may also pursue political strategies. when a government is short of liquidity to meet its debt payments. 2010. for example. 2013. Introduction Interest in sovereign wealth funds (SWFs) as key players in financial markets has grown rapidly over the last years. etc. such as budget stabilization. Ireland. in the wake of the subprime crisis. In 2010. it shows that SWFs tend to invest in large foreign firms. saving for future generations. Russia. 2013). Ang. 2011.. But the example of the recent crisis clearly shows that other sovereign liabilities have to be taken into account: debt. contingent liabilities. 2009. Raymond. 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. from central banks to independent financial corporations. Dyck and Morse. Research on optimal sovereign wealth management is scarcer. the SWF’s assets are often available to substitute for the funds initially earmarked for this purpose. 2010). SWFs serve various economic objectives.1. (2010). diversification from commodities. SWFs also served as “investors of last resort” during the last crises. Scherer (2009a and b). often in the finance and energy sectors. 2009.

central bank. and the residual saved in the SWF. 2009) or a foundation (Merton. In our sovereign case. How much should be saved and how it should be invested is a classic ALM problem. 2008). Merton (1993) solved a similar problem for a university endowment fund. 2014). traditional macroeconomic data lack a significant dimension. 1992. 1993). including all the related institutions (budgetary government. not stocks. Bodie et al. and are unsuitable for valuing intangible assets such as human and natural capital (Aglietta. Most of the macroeconomic variables monitored at present describe flows. a pension fund (Bodie et al. This lack of aggregate data makes it difficult to coordinate sovereign wealth management with fiscal policy. Amundi Investment Strategy Collected Research Papers 161 . 2007). namely risk (Gray et al. The optimal allocation and expenditures of the sovereign will crucially depend on the nature and size of its assets and liabilities. 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. The central government receives tax revenues each year. 2010). In a recent paper (Bodie and Brière. Financial management of government resources and expenditures raises difficult issues in practice.. central bank reserves. and the sources of their uncertainty. Part of this income can be spent. The “sovereign” is considered in the broad sense. SWFs. pension funds and public entities placed under the sovereign's authority). or the public pension fund.. 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..fund assets to invest in banks or shore up equity markets. Standard macroeconomic tools are ill-suited to estimating sovereign economic balance sheets. the optimal sovereign allocation differs slightly.. 1969. Bodie et al. Moreover. monetary policy and public debt management. Managing the wealth of a sovereign is not very different from managing the wealth of an individual (Merton.

Sovereign Wealth Fund Investment While there is an abundant literature on the allocation of foreign-exchange reserves. 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”). whereas the euro is a better hedge in Emerging Europe. The central bank uses its reserves to repay the short-term foreign debt and minimize the variance of its portfolio in real terms. since the funds invested in SWFs often come from foreign exchange reserves. 2. We present our conceptual framework for optimal sovereign wealth management (Section 3). In their empirical investigation.In this paper. The two topics are nevertheless interlinked. We start this section with a state of the art review for these two topics. we review the literature on SWF investment. Beck and Rhababi (2008). and we show how real-life SWFs asset allocations differ from theoretical ones (Section 2). We finally conclude (Section 5). giving country examples and suggesting possible institutional arrangements that would enable efficient coordination (Section 4). optimal portfolio weights depend. providing efficient protection against sudden stops in emerging markets. In this framework. in addition to the standard minimum variance demand term. often linked to global liquidity crises. both from a theoretical and an empirical point of view. on the extent to which the assets can be used to hedge against sudden stops. 162 Amundi Investment Strategy Collected Research Papers . 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. Caballero and Panageas (2005b) suggest the use of assets based on the S&P 500 implied volatility index. there are only a few papers devoted to SWF optimal asset allocation. We then discuss its practical implementation. Caballero and Panageas (2005a and b).

In a recent study (Bodie and Briere. (2013) show that SWF portfolios tend to be insufficiently 1 Das et al. taking explicit account of all sources of risk affecting the sovereign’s balance sheet. and most authors concentrate on SWFs’ equity interests in listed companies. 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. and shows that in this case the optimal asset allocation of the SWF should include a hedging demand against commodity price variations.The authors do not have a uniform view of SWFs’ objectives. Aizenman and Glick (2010) compare the optimal allocations of foreign-exchange reserves by the central bank and by an SWF. Comparing theory on optimal SWF asset management with real-life data could provide interesting insights. 2013). In this framework. and (2) maximizing the expected utility of a domestic representative agent for the SWF. This reflects the different roles that governments assign to SWFs in practice. the authors show that the SWF must hold a riskier foreign-asset allocation than the central bank. Dyck and Morse (2011) and Bernstein et al. (2012) offer a literature review on the use of ALM techniques applied to sovereign fund management. 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. a large portion of SWF investments remains private. Brown et al. Unfortunately. We used Merton’s approach (1974) to estimate the process of the country's assets. Martellini and Milhau (2010) propose a dynamic asset allocation framework for SWFs having liabilities exhibiting inflation indexation. (2010) propose an allocation model for different types of SWFs. and then we optimized the balance sheet using the ALM approach. Amundi Investment Strategy Collected Research Papers 163 . we proposed a framework for optimal asset allocation of sovereign wealth.1 This framework expanded previous results on SWFs’ optimal asset allocations by introducing three additional sources of risk affecting the sovereign balance sheet. which have different objectives: (1) reducing the probability of sudden stops for the central bank.

Kazakhstan and Qatar). SWFs and state-owned enterprises. Bortolotti et al. Conceptual Framework We consider the concept of “sovereign” in the broad sense. exposing them to new risks. etc. Finally. Hong Kong. 164 Amundi Investment Strategy Collected Research Papers . the state tweaked the funds’ regulations to allow them to buy a larger share of the sovereign debt. Kuwait. Qatar. states modified their funds’ investment rules on a discretionary basis. defence. rescuing major Western banks or recapitalizing their home equity markets. 2009). with a strong home bias. Russia. such as investment in key sectors or industries for future growth.geographically diversified. They also tend to take large stakes in companies facing financial difficulties. 3. contradicting the principles of sound diversification.2 finance and telecommunications (Bertoni and Lugo. including pensions and financing of social services (infrastructure such as hospitals. Chhaochharia and Laeven. particularly future 2 Even when the country is producing commodities 3 For example in China. roads. 4 Distinctions among various state entities are less and less meaningful. both abroad and domestically (Raymond. Others are social.). Others used the resources to directly recapitalise ailing banks (Ireland. To achieve its objectives. Some are purely financial. SWFs tend to have significant holdings in large companies in politically sensitive industries. but also the other institutions related to it. 2013). education. The performance of those investments is generally poor in the long run. some SWFs3 played the role of “investor of last resort”. 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). 2010). as recent crises have shown. including not just the state’s budgetary institutions and monetary authorities (central bank). such as debt repayment and setting aside foreign exchange reserves to cope with liquidity crises. the sovereign has a variety of resources. 2012. in some countries with greater borrowing capacities. Saudi Arabia and Singapore. Still others are economic. During the subprime crisis. For this purpose. 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. In 2010 several countries turned to public institutions for assistance in coping with the crisis-related credit crunch. like energy.4 The sovereign has a multitude of objectives. 2013. such as pension funds. even if the announcement of SWF investments yields positive abnormal stock-price returns in the very short run (Bortolotti et al.

2007). fees. The idea is to estimate all the state’s assets and liabilities at market price. and to measure the risks (volatility and sensitivity to economic shocks) associated with each balance sheet item. security. Table 1: Simplified Presentation of a Sovereign Balance Sheet ASSETS LIABILITIES Foreign reserves. 2008). 1974 and 1977.tax revenues. SWF assets. gold. Just as a company’s balance sheet is regularly used to assess the risk of bankruptcy (Merton. KMV.. and possibly a stock of financial assets (foreign exchange reserves. real estate) Contingent claims: implicit guarantees (to banks. Gray and Malone.. 2007.) Present value of future taxes. but more generally. seigniorage. 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. etc. Defining the Sovereign Economic Balance Sheet The sovereign’s global economic balance sheet is key to a full understanding of its situation and risks (Gray et al. as we shall see. government administration. which is obviously a minimal objective. as well as income from other sources such as state-owned enterprises. public pension funds. fees. This is useful not only with regard to the state’s debt repayment capacity (Gray et al. 2002). Table 1 gives a simplified example of a sovereign balance sheet.). seigniorage Present value of expenditures on economic and social development. with regard to its ability to meet its long-term social and economic objectives. the same analytical framework may be applied to a state. benefits to other sectors Present value of target wealth to be left to future generations Amundi Investment Strategy Collected Research Papers 165 . etc.

Part of these revenues are spent. it is necessary to rearrange the balance sheet entries and adopt an integrated presentation. The two liabilities can then be valued as contingent claims on sovereign assets. 1993). Determining how much should be saved and how it should be invested is a standard ALM problem. subtracting the present value of expenses from the present value of income.. Bodie et al.5 depending on the allocation of the 5 Measured as sovereign assets minus sovereign liabilities. 2007. (2007).. Optimal Sovereign Wealth Management From a theoretical standpoint. To do this. The value of the sovereign’s assets and their volatility can then be estimated as a function of the default barrier (promised payments in foreign currencies). 1977) and Gray et al. a pension fund (Bodie et al. to do this. (Gray et al. which can be modelled as a call option on the total value of the sovereign's assets. 1992. 1969. An alternative method is to estimate the market’s valuation of the balance sheet. as described by Merton (1974. 2008). 2009) or a foundation (Merton.An initial approach to measuring a sovereign’s economic balance sheet is to estimate the market price and volatility of all its component assets and liabilities separately. which is a function of its Global Sovereign Surplus (GSS). 166 Amundi Investment Strategy Collected Research Papers . central bank reserves.. The sovereign receives tax revenues each year. and subtracting the value of contingent liabilities from assets. We assume that the sovereign’s objective is to maximise its expected utility. Bodie et al. the present value of future income and expense flows has to be estimated. managing the wealth of a sovereign is similar to managing the wealth of an individual (Merton. The foreign currency debt is considered as a “senior claim”. 2014). and the local currency debt plus base money as a “junior claim”. However. Bodie and Brière. An implied value for the sovereign's assets can be estimated from the observed prices of liabilities. and the residual is saved in SWFs. or public pension funds.

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.sovereign’s assets. FS Ω FA. three hedging demand terms are added to the speculative portfolio.). FS + Ω −FA1 Ω FA. etc. FL . taking explicit account of all sources of risk affecting the sovereign balance sheet. DL (5) ( ρ − 1)α α α α with μ FA the vector of annualized expected returns of the n financial assets in the portfolio over the investment horizon. in order to concentrate on the asset allocation choice. Ω FA their covariance matrix. α the fraction of total sovereign assets dedicated to financial wealth (the remainder is the fiscal surplus). 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. FL + Ω FA Ω FA. Amundi Investment Strategy Collected Research Papers 167 . Ω FA. considered as constant. β the fraction of total sovereign liabilities dedicated to foreign debt (the remainder is domestic debt).t − Ω FA Ω FA. These results shed new light on the optimal allocation of the sovereign’s wealth. and the sources of their uncertainty. DL the covariance of the financial asset returns with the fiscal surplus. 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 SWF’s preference in real terms and observe a hedging demand against realized inflation. Bodie and Brière (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 − α ) −1 β (1 − β ) −1 w* = Ω −FA1 μ FA . foreign liabilities and domestic liabilities respectively. In a more general framework. Ω FA. which both influence 6 We disregard other potential macroeconomic decision variables (tax rate. We recommend taking into account not only the risks from inflation and fluctuations in natural resource prices.

the fiscal surplus variability is influenced not only by commodity prices and inflation volatility. and from foreign and domestic liabilities.). This has important implications. Traditional Public Finance Data and their limitations To implement sovereign ALM. Practical Implementation The practical implementation of sovereign ALM raises several difficulties. as we will see in our estimation for Chile in Section 3. It does not include public enterprises. and so on. with a 7 This leads to another important difference from the previous literature. but all the risks stemming from the fiscal surplus. In our framework. as the fiscal surplus may not have a sensitivity of one to natural resource prices. but also by the sovereign’s policies on natural resource extraction.7 4. 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. The definition of the “government” entity differs between countries8 and may not correspond exactly to our broad definition of the sovereign. 2009b)). The IMF's GFS database. Moreover. 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 variability of the fiscal surplus. and the accumulation phenomena that lead to systemic risks. We discuss institutional arrangements that could enable efficient coordination. which are included in the corporate or financial sector and cannot be disentangled from it. 8 In the US. The general government sector comprises only central. the European Central Bank and Eurostat “Euro Area Accounts” have a more restrictive definition. the federal government (including government-owned corporations and agencies that issue securities individually) and the monetary authority. what really needs to be measured is the actual nature of macroeconomic and financial risks. taxation. In Europe. 168 Amundi Investment Strategy Collected Research Papers . with their non-linear features (contingent liabilities modelled as options. not the fluctuations in commodity prices themselves (Scherer (2009a. etc. created in 2001. remedies these differences with a unified base of 153 countries’ data on government balance sheets. the variability of the flow of revenues from the sale of natural resources needs to be hedged. state (regional) and local government and the social security or pension funds belonging to it. the “Flows of Funds” statistics consider state and local governments (excluding employee retirement funds). monetary policy and public debt management.

These data are a very useful supplement to the existing figures because they provide an estimate of stocks11 of natural resources and intangible assets. and on the other hand to inventory changes (natural resource depletion. calculated as a residual. wage costs. the difference between total wealth and the sum of produced and natural capital. exports.particularly broad scope for the sovereign. World Bank estimates of natural and human capital can be used to estimate the present value of the fiscal surplus. 2006 and 2011). In 2000 the World Bank took the unprecedented step of measuring the wealth of nations (World Bank. SWFs. pastureland and protected areas) and (3) intangible capital (human. given a certain level of desired 9 It comprises not just the central government budgetary authority but also the central bank. risks are related on the one hand to market price fluctuations (for commodities. these data. which are purely accounting-based and generally available on an annual basis. subnational governments and other government agencies. population growth. deposit insurance funds. mineral resources.). state-owned enterprises. Finally. domestic net investment. (2) natural capital (energy resources.) that cause the government’s income and expenditures to fluctuate. investment in education. In the case of sovereign balance sheets.10 Total wealth is broken down into: (1) produced capital (machinery. etc.9 The IMF’s GFS data nonetheless have significant limitations. pension funds. there are no estimates of contingent liabilities.). The total wealth of each nation is estimated as the present value of future flows of consumption. such as too-big-to-fail guarantees to the financial sector and implicit guarantees to provide social benefits when various needs arise. non-timber forest resources. or expenditures. 10 For years when adjusted net savings are negative. timber resources. the actual consumption rate is added to adjusted net savings. are not sufficient to measure the risks associated with each item. etc. Amundi Investment Strategy Collected Research Papers 169 . structures and urban land). There is no evaluation of the present value of future tax revenues. etc. Consumption levels are based on past historical data but are adjusted to be “sustainable”. Moreover. 11 Flow variables are also available: depletion of natural resources. cropland.

New Zealand. there needs to be a high level of coordination between institutions that control sovereign assets and sovereign liabilities (at least the central bank. possibly conflicting. including setting the risk-free yield curve and providing highly liquid securities. the mandate of the debt management office is to keep the net foreign currency position close to zero. South Africa and Turkey are the handful of countries that have made significant steps in the direction of developing an ALM framework. This is not without drawbacks since the issuance of government debt might also respond to other. the debt management office. the ALM exercise has been performed by the debt management office. What the most efficient institutional arrangement would be is still an open question. the treasury and the ministry of finance). already responsible for cash management and treasury services. because of the 170 Amundi Investment Strategy Collected Research Papers . explicitly matching foreign currency assets and liabilities and hedging exchange-rate movements. In Canada. In New Zealand and South Africa. The need for central coordination To implement sovereign ALM in practice. In New Zealand. with the goal of minimizing currency and interest-rate risks by matching the assets to the liabilities funding them. these data were estimated in 2000 and 2006 for the World Bank’s 2006 and 2011 reports and are not available as a historical series. there is a specialized asset-liability management unit that analyses the sovereign’s balance sheet. Britain.taxation. the government decided to continue debt issuance even though there was no need for government borrowing. debt management is also defined in an ALM framework. for example. objectives. In Turkey. and the few country examples show that very different organizations are possible. Canada. Denmark. In Australia and Norway. Government debt has public good characteristics. ALM was introduced for the tactical management of foreign reserves in 1997. Unfortunately. In most of the example countries cited (Canada being an exception). in close cooperation with the reserve management office.

Conclusion This paper presents an analytical framework for sovereign wealth and risk management. The example of Canada. extending the theory of contingent claims analysis. but the optimal institutional arrangement may in the end depend on the political organization of each country. The sovereign wealth fund would actually be an excellent candidate for the job of implementing the sovereign ALM. is a good example of this. On the other hand. and discusses its practical implementation. 5. which gave the central bank tactical reserves management office an ALM mandate. South Africa has organized such a framework with a common committee bringing together the South African reserve bank and the treasury. this may be facilitated by the fact that the finance ministry is responsible both for debt issuance and fiscal policy and for determining the SWF’s strategic asset allocation. and there could be potential conflicts of interest with monetary policy. including. a coordinated approach to the management of the national balance sheet would necessitate central responsibility. Amundi Investment Strategy Collected Research Papers 171 .importance of sustaining functioning capital markets. a strategy was developed jointly by the reserve bank and the treasury to bring down this exposure. In any case. In many countries. This supposes the broadest possible definition of the sovereign. Probably the most realistic scenario would be to encourage more links and consultation between the different agencies. The finance ministry might be the best candidate to lead this coordination. in particular. entities subordinated to the state. When South Africa had a net negative forward currency position in the late 1990s. with detailed instructions from the ministry of finance. assigning ALM to the asset management offices would also make sense. But responsibility for the wider government balance sheet would sit uneasily with central bank independence. 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.

a relatively precise understanding of the government’s economic objectives and an accurate estimate of contingent liabilities are also needed. to be measured precisely. SWFs. 2006). it is necessary to measure not only the sovereign’s financial wealth. the practical implementation of our ALM framework requires reliable macroeconomic data on a regular basis. pension funds. both assets and liabilities. government agencies and state-owned enterprises. including taxation level. However. to concentrate on asset allocation. The reason is that the funds. A general equilibrium model endogenizing all of the state’s decision variables would be more realistic.such as the central bank. strong coordination is needed between the sovereign entities. as well as their risks. but also its human and natural capital. we consider macroeconomic variables as exogenous. First. the sovereign benefits from many more policy instruments. Our ambitious approach has limitations. Second. This approach also requires all balance sheet items. and the debt management office on the liability side. it can nevertheless be thought of as providing useful guidelines for efficient 172 Amundi Investment Strategy Collected Research Papers . To do this. One significant application of this analytical framework is the management of financial wealth under direct state control. but also much more complex. become fungible if a crisis arises. Similarly. It can also inflate or repudiate its debt (Landon-Lane and Oosterlinck. far removed from current practice. A sovereign ALM strategy can thus be developed for managing asset risks in a way that is consistent with the sovereign entity’s liabilities. In practice however. 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. even if located in different 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. even if the implementation of the ALM framework for SWF asset allocation is an unfeasible first-best solution for many countries. Moreover. The ministry of finance is particularly well positioned as a central institution to facilitate this coordination.

In particular. it should help to improve the diversification of sovereign assets and the hedging of important risk factors affecting the sovereign balance sheet. Amundi Investment Strategy Collected Research Papers 173 .management of sovereign wealth.

“The Four Benchmarks of Sovereign Wealth Funds”. 174 Amundi Investment Strategy Collected Research Papers . “Optimal Reserve Composition in the Presence of Sudden Stops. "Contingent Claims Analysis and Life-Cycle Finance... 77(1). References Alfaro. (2008). Journal of Economic Dynamics and Control. Lerner J. (2008). Journal of Investment Management. April.. and Treussard J. European Central Bank Publication. in Bolton P. Optimal reserve management and sovereign debt. Bernardell C. (2004). (2012).." American Economic Review. Bodie Z.. Forthcoming. 2009. No 916. 219-238. Diebold F.X. “Sovereign Wealth and Risk Management: A Framework for Optimal Asset Allocation of Sovereign Wealth”.. Aglietta M. Beck R. 14(3).. Avendano R. “Pension Funds. “Labor Supply Flexibility and Portfolio Choice in a Life Cycle Model”. European Central Bank Working Paper. N° 283.. Merton R.. “Should Larger Reserve Holdings Be More Diversified?”. eds. 427-449. (2014). 415-444. and Brière M. Cardon P. Sovereign Wealth Funds and Long-Term Investing. and Rahbari E. (2010). 27(2).F. Bodie Z. May.. mimeo. Ruffino D. Univesita di Torino Working Paper. Fugazza C.. N°16392. July. Journal of International Economics.. p 94-105. NBER Working Paper. Life-Cycle Asset Allocation and Performance Evaluation”. Bodie Z. and Nicodano G. p. C.. L. (2009). Bagliano F. Kanczuk. “The Investment Strategies of Sovereign Wealth Funds“. Bernstein. “Are Sovereign Wealth Funds’ Investments Politicall Biased? A Comparison with Mutual Funds”. F. and Manganelli S. and Stiglitz J. 23-36.. 291–96. and Samuelson W. “Sustainable Growth: Do We Really Measure the Challenge?”. (1992). the Euro and the Dollar as Safe Haven Currencies”.. Ang A. 16. March.Acknowledgment The authors are grateful to Ariane Szafarz for her comments on a previous version of this paper. p. OECD Development Center Working Paper. 98(2). 2009. Journal of Economic Perspectives. Columbia University Press. “Risk Management for Central Bank Foreign Reserves”. and Glick R. (2011). and Santiso J. (2013). “Asset Class Diversification and Delegation of Responsibilities between Central Banks and Sovereign Wealth Funds”. Beck R. International Finance. and Weber S. S. p.. Samama F. Coche J. Aizenman J. p. and Schoar A.

10.J. NBER Working Paper. p. (2010). 1-24. Bortolotti B.. UK. “Macrofinancial Linkages of the Strategic Asset Allocation of Commodity-Based Sovereign Wealth Funds”. “Contingent Reserves Management: an Applied Framework” Journal Economía Chilena (The Chilean Economy). and Panageas S. Annual Review of Financial Economics. Journal of Finance. University of Oklahoma Working Paper. Review of Economics and Statistics. KMV (2002).S. (2009).. p. Merton. No 10/9. 247-257. Landon-Lane J. (2007).J. (2005b). and Malone S.. “Hope Springs Eternal: French Bondholders and the Soviet Repudiation (1915-1919)”. Gray D. Lu Y. October. (2009). R. “Lifetime portfolio selection under uncertainty: The continuous-time case”. 5(4). SSRN Working Paper. “Sovereign Wealth Fund Portfolios”. January.. (2013).cfm?abstract_id=1262383 Clark G. (1974).ssrn.C.com/sol3/papers. Merton. Review of Finance. No 12/241. Amundi Investment Strategy Collected Research Papers 175 . Gray D.. 429-451. and Laeven L. 8(2). “Contingent Claim Approach to Measuring and Managing Sovereign Credit Risk”. 449-470. Pacific Review. Chicago Booth Research Paper.F. Megginson W. p.. (2008). and Panageas S. 1..C. Papaioannou M. 249-286. 23(4). Chhaochharia V. Oosterlinck (2006).J. N°11-15.Bodie Z. p. p. Das U. and Bodie Z.. “Macrofinancial Risk Analysis”. Brown A. Journal of Investment Management. “Government of Singapore Investment Corporation: Insurer of Last Resort and Bulwark of Nation State Legitimacy”. IMF Working Paper. IMF Working Paper. Central Bank of Chile. Merton R. Papaioannou M. G. Wiley Finance. Fotak V. 51... Edhec-Risk Institute Publication. 45-56. R.H. Caballero R. p. “Asset-Liability Management Decisions for Sovereign Wealth Funds”. 507-535. “A Quantitative Model of Sudden Stops and External Liquidity Management”. October. (2012). N° 11293. Dyck I. “Life-Cycle Finance and the Design of Pension Plans”. available at: http://papers. 29.F. and Rindisbacher M.L. K.L. N°1262383. (2010). (2005 a). (1969). "Modelling Default Risk". and Petrova I. and Petrova I. and Morse (2011). Detemple J.C. (2010). “The Investment Allocation of Sovereign Wealth Funds. “The Sovereign Discount in Sovereign Wealth Fund Investments”. and Milhau V. Moodys KMV. p. Caballero R. Martellini L. “Sovereign Risk and Asset and Liability Management—Conceptual Issues”. and Monk A.B. “On the Pricing of Corporate Debt: The Risk Structure of Interest Rates”.

The International Bank for Reconstruction and Development/The World Bank. R. Studies of Supply and Demand in Higher Education. (2009b). World Bank (2006). p. “Portfolio Choice for Oil Based Sovereign Wealth Funds”. 1. “Optimal Investment Strategies for University Endowment Funds”. The Changing Wealth of Nations. (1993). Clotfelter C. Journal of Alternative Investments. 24-34. Scherer B. 211-242. 121-159. 123. University of Chicago Press..C. 13(3).T. “A Note on Portfolio Choice For Sovereign Wealth Funds”. 176 Amundi Investment Strategy Collected Research Papers . and Rothschild M. R. (1977). p. p. Journal of Banking and Finance. World Bank (2011). Scherer B. Raymond H. Where is the Wealth of Nations? Measuring Capital for the 21s Century. (2010). (2009a). 3-11. International Economics.C.Merton. “An Analytic Derivation of the Cost of Loan Guarantees and Deposit Insurance: an Application of Modern Option Pricing Theory”. ed. 23(3). 315-327. The International Bank for Reconstruction and Development/The World Bank. Financial Markets and Portfolio Management. Merton. “Sovereign Wealth Funds as domestic investors of last resort during crises”.

Balanced Quantitative Research. a more careful approach would be required. equal for all asset classes. • Complementary approaches. the traditional Sharpe ratio approach. • As for establishing a forward looking expected return which can serve as a reference for discounting liabilities. despite these major changes.unpredictable effects of non-conventional monetary policies on macro-economic variables. • For strategic allocation of diversified portfolios. may still be of interest because it provides good risk diversification. especially inflation. 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. Methodological Considerations on Expected Returns in Uncertainty Sylvie de Laguiche. Amundi Gianni Pola. WP-032 Unexpected Returns.the fact that some market variables (interest rates) are in uncharted territory. . Amundi Investment Strategy Collected Research Papers 177 . 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. . econometrics and quantitative finance may be required now in order to overcome the challenge. Head of Quantitative Research. mixing skills in macroeconomics.

making them effectively unobservable. Denmark. 1999) allow us to relate a given (optimal) portfolio to a set of expected returns.0. In this document. and. in the latter the level of returns itself is crucial to determine discount rates. once agreed on the market risk model and the investor’s risk aversion. while volatility is somewhat predictable as "large changes tend to be followed by large changes. Netherlands.at least ten years . Switzerland. and that prevailing market conditions can strongly affect returns and cause them to deviate significantly from the long- term average. x the two-year yield synchronously negative in many countries in 2012 (Germany. While the finance industry and academia are aware of the inadequacy of the Markowitz model. see chart 2. A pension fund. conversely. returns are un-correlated over time and probably unobservable 1. While in the former. right panel). reverse optimisation techniques (Cantaluppi. This is particularly true today: the recent crisis in the Eurozone and the previous sub-prime crisis in 2008. and small changes tend to be followed by small changes" (Mandelbrot. and x the decoupling of the Eurozone: the two-year yield divergence (see chart 2. left panel). never reached going back to many decades. and x assessing returns on assets and setting an appropriate discount rate for liabilities. Introduction Expected returns are closely related to the portfolio allocation: according to the Markowitz portfolio selection (1952). 178 Amundi Investment Strategy Collected Research Papers . risk-adjusted hierarchy between assets is more relevant. of either sign. we refer to long horizons . 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. 1963). expected returns unambiguously determine the portfolio. remind us that returns dynamics are not stationary in time. Unfortunately. Many financial variables are in a peculiar territory. it is clear that estimation of expected returns is a key issue for strategic asset allocation. needs expected returns on asset classes for: x building a strategic allocation. for example. 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).somewhat in line with the average liabilities in pension funds.

Chart 1 10-year US Treasury yield 16 12 8 4 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 p 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. investment styles (value. Amundi Investment Strategy Collected Research Papers 179 . They modelled equity expected returns as a time-varying baseline given by cash or bonds. the yield for a bond.g. Recently. and Fama & French (1989). the long-term equity risk premium. expected returns were considered to be time-dependent. Ilmanen (2011) faced the problem of estimating expected returns on major asset-classes broadening the investments to non- traditional assets (commodities. In the next two decades. In the eighties.). the financial community came back to the origin: risk premia were believed to be time-varying quantities themselves. carry. etc. inflation. 1988b) and continued with the works of Asness (2000). They were estimated from asset fundamentals (e. plus a constant term. Arnott (2002). trend. illiquidity. the dividend discount model for stocks. and tail risks). volatilities). This counterrevolution started with Campbell and Shiller (1988a. thinking on the subject changed according to the work of Ibbotson and Sinquefield (1976a and 1976b). and underlying factors (growth. real estates).

Today. x incorporate time-varying risk premia. 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. Time-varying risk premia can be profitable if and only if investors are able to predict them. Recently more extreme approaches emerged in the financial arena. leading to Bayesian approaches (Black & Litterman model. while the needs for expected returns might be questionable in an optimisation process. there are always lessons that can be learned from history. and x model market inefficiencies due to investor irrationalities and market frictions. 2011). skewness and liquidity preferences. our aim here is to investigate this issue. Rather than providing a specific recipe for estimating expected returns and computing return figures. and few side empirical analyses to sustain our ideas. considerations. 1990) and robust asset allocation models (Meucci. This manuscript is about expected returns. 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. 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. We refer to Ilmanen (2011) as a comprehensive review on the subject. stimulating the reader with relevant questions. Modern approaches: x derive asset-prices from asset returns co-variation with “bad times”. nor can equilibrium-based models be easily estimated given the rise in macroeconomic volatility. and then questioning the possibility today of 180 Amundi Investment Strategy Collected Research Papers .Current thinking today is more complex and closer to experimental evidences. minimum variance. risk parity approaches). aiming to construct portfolios without any specific views on expected returns (e. x take into account supply-demand effects on asset prices. The paper is organised as follows. Nevertheless. Indeed. In section 1 we will start investigating how the crisis changed the relationship between risk and return. x are based on multiple risk-factor models. researchers and practitioners prefer to incorporate uncertainty and estimation errors in expected returns. maximum diversification.g.

reverse optimisation techniques to get implied returns from equilibria allocation. Hence in section 2 we will introduce the two main classes of estimation methods (Statistical and Equilibrium-based approaches) highlighting the pros and cons. A conclusive section delineates the main take-home messages that we think are relevant to address expected return estimates in uncertainty. 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”. The relationship between risk and return Past average returns are rarely related to future performance. The annex includes all the details about the charts and data-analysis presented in sections 1 and 3. Even if the relationship is not clearly linear. factors-model focusing mainly on the consequence of deflationary or inflationary scenarios on asset-returns. This intuition is the key idea of modern theories on asset-pricing which. In section 7 we will then illustrate how portfolio construction can help to handle uncertainty in expected returns to deliver robust strategic asset allocation. rather than deriving prices from future discounted cash flows.1. see chart 3b in annex A1 for more details. and stressing the peculiarity of the estimation problem in this specific era for financial markets: in sections 3. 1. we measured the volatilities. the volatility is a more general measure of returns dispersion (global measure). 1981. and performance of some asset classes in two historical periods: pre-crisis sample (1990-2007) and the full sample (1990-2012).calibrating a model on past history. Amundi Investment Strategy Collected Research Papers 181 . The financial crisis questions: 1. 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. maximum draw downs (MDD) 2. and finally the equilibrium-based models. 5 and 6 we respectively critically evaluate the Sharpe ratio approach. build expected returns from asset returns’ co-variation with “bad times”. 2008) for financial markets and global economy. 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). it is evident that assets that performed poorly in crisis periods delivered higher returns in the long-run. 4. Given this argument.

Euro credit markets suffered less with respect to the pre-crisis period. Australia and Emerging markets. 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. 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. Euro peripherals. MDD for US investment grade more than doubled. The worst equity markets in the crisis period were US. World high-yield doubled its MDD. 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. 2008) 182 Amundi Investment Strategy Collected Research Papers . 1981. Commodity indices registered new MDDs in the crisis sample. 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.crisis sample: higher volatility and worst MDD should suggest a higher premium for future returns due to uncertainty. gold performed better during the crisis (the average compound return increased). The crisis period increased long-term volatility in most of the asset classes. The 2008 crisis was dramatic for spread markets: all volatilities increased markedly. volatilities increased. the volatilities of equity markets increased but only marginally with respect to the pre-crisis period.

We calculated the new millennium performance (2000-2012) of US 10-year bonds and US equity. the sample period is crucial: long time windows reduce sample specificity and allow for more robust statistical inference.45% and 8. but may miss structural changes happening in the market and may not be able to catch up non-stationary risk premia dynamics. and recently from convergence back to divergence again with the debt crisis. and the emergence of structural changes in the world economy. In chart 5b we rearranged the data in a scatter plot of the real performance of bonds and equity. and give an incomplete picture of the financial dynamics.14%). Chart 4 reports the rolling historical volatilities of German. and Spanish 7-10 year bonds (upper panel). On the other hand. Equity performance in the new millennium was very poor compared to the post-war period 1950-2000. and uncertainty prevents us from tracing a clear picture. it is rather dangerous to extrapolate over multi-year time windows when reversals often take place. if compared with the full sample starting from 1871. Amundi Investment Strategy Collected Research Papers 183 . 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 compared them with the past performance since 1871. In chart 5a we computed the annualised compound returns for each decade.1. The most relevant sample to estimates markets’ returns in this case is not obvious. Given the level of many financial indicators today.48% and 6. however we should admit that the post-war period was misleadingly favourable to equity.2. Financial markets are moving from pre-crisis equilibrium to a new one: at the moment we are in between. In historical estimates. Italian.19% (real figures are respectively +2. it is rather difficult to select an appropriate sample to estimate econometric models: recent history may or may not be relevant. 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). The average nominal figures for bonds and equity are respectively 4. small samples may lead to non–significant results.

4 -0.2 -0. 1 yr rolling) 1 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 184 Amundi Investment Strategy Collected Research Papers .6 0.8 0.4 0.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.2 0 -0.6 -0.

on a ten-year horizon. they were on average positively correlated.05%.95% to 12.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. performance in the new millennium was higher than the long-term average: bonds benefited from the marked fall in interest rates (see chart 1). This analysis highlights the importance of extending the historical sample. models from behavioural theories. in order to get more robust estimates it might be wiser to complement model-free historical figures with models from financial theories. when bonds suffered from the increase in 10-year interest rates from 1. This paradigm was not verified only from the beginning of 1941 to the end of 1980. delivering a negative real performance. On the bond side. Amundi Investment Strategy Collected Research Papers 185 . A final remark on correlation between performance of bond and equity: the scatter plot shows that. forward- 3 In the 1970-1980 decade the annual inflation rate was +8.84%. In addition.

g. equilibrium-based models forecast future returns incorporating forward looking macro variables. 2. The separation between them is not very neat. The above approaches: x lead to unbiased and. discretionary views may help as well in case of strong structural changes. or on more fundamental methods looking at asset-classes as vehicles of more systematic factors (e. principal component analysis. 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. hence.g. 2. CPI. 1999) to get expected returns from an equilibrium portfolio. We intend with these approaches: x a normative method based on Sharpe ratio hypothesis. independent component analysis). Should expected returns reflect an unbiased view about the market.looking market indicators (e. and both of them can be built on macroeconomic relationships among variables. Risk Aversion). Illiquidity. and are not equipped with valuation tools. more robust expected returns. Both of them are certainly statistical as most of the models are derived according to econometric relations. 186 Amundi Investment Strategy Collected Research Papers . We would say that while statistical models estimate returns mainly looking at the past history. x factors-model approaches based either on statistical analysis tools (e. valuation ratios). and x reverse optimisation techniques (Cantaluppi.g. The main differences between them are that the former do not assume any specific macroeconomic views.1 Statistical methods. GDP.

This way may lead to more robust estimations. 2. Amundi Investment Strategy Collected Research Papers 187 . 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. and Barberis and Huang 2001). and it is usually used to assess the efficiency of an asset-class. They are: x good candidates for computing discount rate. Sharpe ratio is usually computed as the average excess return of a risky asset over the risk-free rate and normalised for historical volatility. 2008). and might be useful to diversify statistical and equilibrium models. 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. x superior to forecast future returns. We mean models based on macroeconomic assumptions (e.2 Equilibrium-based methods. x sometimes they lead to corner solutions in an optimiser. x are good candidates for portfolio optimisation leading to well-posed solutions and diversified portfolios 4. investment strategy or fund. is a simple constant figure still relevant? Sharpe ratio is a risk-adjusted performance measure. x but are not very appropriate for computing discount rate for pension funds and insurance companies. These approaches enable us to explain many observed anomalies in risk premia 5. long-term expectations on inflation and GDP) and valuation arguments. Sharpe ratio.g. In this document we do not face the class of Behavioural models. Behavioural models provide an explanation (see Bernatzi and Thaler 1995. 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. 3. 5 The equity premium puzzle indicates the difficulty of explaining the observed equity risk premium within standard macroeconomic models.

Table 1 in annex A2 reports some descriptive statistics: the median and the average figures are reported for each country and three-. Computations have been performed on rolling windows of 10.Usually it corresponds to a short-term or long-term government yield. 0. but which one now? Government debt may no longer be risk-free. x Instability over time Chart 7 plots the Sharpe ratio for the US. For the equity markets. The longer the time window.25 is a reasonable estimate for the 10-year horizon. the more stable the Sharpe ratio. Chart 6 reports the 10-year CDS for Italy and Germany from 2005.and 10-year time horizons.and three-year horizons. Japanese. UK. German and French equity markets. 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 188 Amundi Investment Strategy Collected Research Papers . five.

5 -2 juil.-67 juil.-81 juil.5 1 0.-65 juil.-03 juil.-09 juil.-69 juil.-87 juil.-97 juil.-67 juil.-75 juil.-95 juil.-79 juil.-05 juil.5 -1 -1.5 0 -0. practitioners and researchers (see Frazzini & Petersen 2010) documented some anomalies regarding the Sharpe ratio.-65 juil. Amundi Investment Strategy Collected Research Papers 189 . We report in the following the most relevant: x On historical basis low volatility asset-classes delivered higher Sharpe ratio6.-11 USA JAPAN GERMANY FRANCE UK x Differences among asset classes. This evidence inspired minimum-variance investments in the equity markets.-73 juil.-01 juil. We tested this hypothesis in different baskets.-87 juil.-85 juil. and high-beta assets lower risk-adjusted returns.-71 juil. if investors are leverage averse.-91 juil.-95 juil.Chart 7 Historical Sharpe ratio (10yrs rolling) 3 2.-07 juil.-91 juil.-93 juil.-89 juil.-93 juil. Leverage aversion breaks the standard CAPM. (2011). Does it make sense to make a normative assumption of equal Sharpe ratios for all asset classes? Recently.-83 juil.-89 juil.5 2 1.-05 juil.-81 juil.-73 juil.-11 USA JAPAN GERMANY FRANCE UK Historical Sharpe ratio (3yrs rolling) 3 2.-77 juil.-99 juil.-83 juil.5 -1 -1.-85 juil.-79 juil.-03 juil.-07 juil.-71 juil.5 0 -0.5 2 1.-69 juil.-09 juil.5 -2 juil.-75 juil.-77 juil.5 1 0. as follows: 6 As reported by Asness et al.-97 juil.-99 juil. low-beta assets will offer higher risk- adjusted returns.-01 juil.

10 0.30 0. Equity markets. We considered a diversified investment universe in the US market since 1970 (see annex for details). R2 is 0.60 0. Chart 8 (top left panel) reports the Sharpe ratio as a function of the volatility.2210).9401). this relation is particularly sample specific: restricting the analysis within each macro asset-class can alleviate this problem.10 0. Diversified basket.00 0.50 0.40 sharpe (ann) sharpe (ann) 0. 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. In chart 9 (lower panels) we report two pictures taken from Baker et al. Each spot stands for a specific asset-class. The plot confirms the anomaly. The plot shows clearly a reverse relation of the Sharpe ratio to the volatility (linear regression. R2 is 0.20 0% 5% 10% 15% 20% 25% 0% 10% 20% 30% 40% vol (ann) vol (ann) 190 Amundi Investment Strategy Collected Research Papers .20 0.40 0.20 0. Chart 8 US Diversified (1971-2012) US MSCI Sectors (1995-2012) 0. even if the relation is noisy (linear regression. When dealing with diversified investment universe.30 0.10 -0. They studied the US equity market from January 1968 to December 2008. 2011. and plotting them against the volatilities.00 -0.50 0.

illiquid assets tend to exhibit smoothed prices thus leading to underestimation of the true risk.00 1. They complement the analysis replacing volatility as a risk measure with beta. this estimation error is certain to artificially increase the Sharpe ratio.20 0. and they found similar results.30 which is greater than most of each country’s ones. They did the analysis twice: left panel reports the analysis performed on all the stocks. The result confirms the benefit of diversification. Chart 9 Sharpe ratio EMU area 2002-2007 2.19. Each month they sorted stocks into five groups according to their historical volatilities calculated on five-year trailing windows. less liquid assets offer a compensation for the higher trading costs and lower flexibility to rebalance portfolio positions.80 0.85 0.40 0. and similarly higher quality credit buckets historically delivered higher Sharpe ratios than lower rated credit ones. From an estimation point of view.30 1.09 (the weighted average according to the market cap is 1. • Illiquid asset classes historically delivered higher Sharpe ratios with respect to liquid assets (see Ilmanen. In the long run. The Sharpe ratio is 1. The arithmetic average of the Sharpe ratio cross-countries is 1. We measured the Sharpe ratio for the EMU equity markets from 2002 to 2007. • Diversified indices delivered higher Sharpe ratios. we made an approximation of keeping the market cap constant over time). 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.32 1. The plots show clearly the outperformance of low volatility stocks over high volatility ones.62 0.09 1.73 1.19 1. right panel on the 1.60 1.06 1.000 largest stocks according to their market capitalisation. 2011). Bond markets and Credit markets. We tested this argument in chart 9.94 0.09 1.00 EMU avg SPAIN NETHERTLAND BELGIUM FINLAND EMU GERMANY ITALY EMU avg2 FRANCE Amundi Investment Strategy Collected Research Papers 191 .

4. as an example. higher order statistics and tail risk. the level of this anomaly may be sample specific. However. A more detailed approach than the constant Sharpe ratio method may be required to set up the expected returns to calculate the discount rate. timing risk. 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. make inflation a key variable in order to decipher the future market scenario. the exceptionally favourable Sharpe ratio of bonds has been driven by a significant decline in interest rates. and the Sharpe ratio becomes unreliable. therefore except in very particular case this approximation may be kept to building strategic allocation in diversified portfolios. valuation risk. or maximum draw down. Hence. In case of strong asymmetric and fat-tailed asset classes. stress and liquidity indicators. Nevertheless we should remind the reader that volatility misses some important aspects of risk: liquidity risk. a slowdown of inflation to lower levels) is commonly associated 192 Amundi Investment Strategy Collected Research Papers . and fundamental risk. x Non-normality of asset classes The Sharpe ratio implicitly assumes volatility as the risk measure. New trends in asset allocation look at assets as vehicles of more fundamental macroeconomic variables. and possible extended recession and hence deflation on the other hand. default risk. While rising inflation is usually related to poor conditions in the real economy.e. the constant Sharpe ratio approach leads to a better diversified portfolio in terms of risk. and that is going to have implications on the estimation of the Sharpe ratio. It is evident that potential inflation caused by unconventional monetary policy of central banks on one hand. In addition. model risk. the approximation is too rough.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. disinflation (i. In this case many practitioners and researchers prefer to correct the formula by replacing the volatility with the downside volatility (Sortino ratio). 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.

Secondly we analysed the historical performance of asset classes conditional to contemporaneous inflation rate variation (on an annual basis). on longer historical perspective. and Hong Kong from 1997 to 2004. Deflation consists in a fall in consumer prices: it hurts the real economy because of a potential spiral in consumer prices. Japan from 90s to present times. It is evident that the last millennium was mainly characterised by positive inflation rates. inflation and deflation were both well represented: according to their study the 20th century itself represents an anomaly. Chart 10a reports the annual inflation rates for US (from 1871) and Sweden (from 1900). x low inflation: inflation rates between 0% and 2%. Deflation affected many countries in the past: the US’s Great Depression in the 30s. hence it is questionable whether deflationary periods were less likely than inflationary ones. As Reinhart and Rogoff (2011) demonstrated.with better economic conditions. and the consequent recession and depression. The first issue we want to address is the likelihood of deflationary and inflationary scenarios on a historical basis. 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 Amundi Investment Strategy Collected Research Papers 193 . We divided the inflation rates into buckets to separate different scenarios: x deflation: negative inflation rates.

showing how to segment asset classes according to their attitude to polarise with respect to variations of macroeconomic variables (rising or falling scenarios). The sample from 1970 only has three entries for very high inflation scenarios and zero 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. US corporate investment grade intermediate maturity. 7 In this case results on extreme scenarios (deflation and very high inflation) should be taken with care. x commodity GSCI is a good hedge against inflationary scenarios. x the high dispersion of returns for equity markets highlights that inflation is not the main driver in this case. The sample from 1927 only presents four entries for very high inflation scenarios and seven for deflationary scenarios. Time series are shorter in this case: our sample is from 1970 for investment grade corporate and commodity GSCI. commodity GSCI and gold. the very high inflation scenario is not significant in this case).x high inflation: inflation rates between 5% and 10%. 194 Amundi Investment Strategy Collected Research Papers . Three main messages emerge: x nominal bonds are a good hedge for deflationary scenarios. and for Swedish bonds and equity from 1900 to September 2012. For each bucket we plot the average. The main take-home messages are: x corporate investment grade mimics the behaviour of bonds. Chart 10b reports the outcome of the analysis for US 10-year bonds and equity from 1871 to September 2012. hence inflation hurts corporate investment grade. and from 1927 for gold 7. In Pola (2013) we will more carefully illustrate the dependency of asset returns on macroeconomic variables. the minimum and maximum performance as a proxy of the dispersion. Then we computed the annualised real performance of major asset classes conditional to the above inflation scenarios. x gold does not reward significantly over inflation on high inflation scenarios (as noted in footnote 7. In chart 10c we performed the same exercise for US corporate investment grade all maturities. x very high inflation: inflation rates higher than 10%. We investigated the dependency with respect to the US inflation rates.

5%] >5% >10% min avg max min avg max Amundi Investment Strategy Collected Research Papers 195 . 2%] (2%. 10%] >10% <=0% (0%. 10%] >10% <=0% (0%.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) real performance (annualized) 30% 20% 20% 0% 10% 0% -20% -10% -40% -20% -30% -60% <=0% (0%. 5%] (5%. 2%] (2%. 2%] (2%. 2%] (2%. 5%] (5%. 10%] >10% min avg max min avg max 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% real performance (annualized) real performance (annualized) 20% 20% 10% 10% 0% 0% -10% -10% -20% -20% -30% <=0% (0%. 5%] (5%. 2%] (2%. 10%] >10% <=0% (0%. 5%] (5%. 2%] (2%. 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% real performance (annualized) real performance (annualized) 60% 40% 40% 20% 20% 0% 0% -20% -20% -40% -40% -60% -60% <=0% (0%. 2%] (2%. 5%] (5%. 10%] >10% <=0% (0%. 5%] (5%. 10%] >10% min avg max min avg max Commodity GSCI 1970-2012 conditional to US CPI rate Gold 1927-2012 conditional to US CPI rate buckets buckets 80% 60% 60% 40% real performance (annualized) real performance (annualized) 40% 20% 20% 0% 0% -20% -20% -40% -40% -60% -60% <=0% (0%.

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. according to the Capital Asset Pricing Model (CAPM). and then to elaborate a consistent Bayesian framework to integrate qualitative views.g. allows fund managers to build their portfolios more coherently.5. In this case the most popular approach consists in determining the risk aversion parameter according to a strong prior information or conviction (e. and then the expected views in a portfolio (optimisation). the demand for these assets will exactly equal the outstanding supply” (He & Litterman 1999). It is the rate at which more return is required for more risk. “prices will adjust until the expected returns of all assets in equilibrium are such that if all investors hold the same belief. This approach is more general than its application in the Black & Litterman (1990) model: according to the reverse optimisation technique. what can it be used for? The Markowitz model strongly depends on expected returns assumptions. 196 Amundi Investment Strategy Collected Research Papers . and sometimes produces results that are extreme and not particularly intuitive. 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. and it can be expressed as the ratio of risk premium and variance of the market portfolio. implicit returns can be extracted from any strategic asset allocation. The iterative process that enables the allocation to be transformed in implied views (reverse optimisation). commodities and alternative assets. this issue remains a critical one. This approach becomes decreasingly intuitive when dealing with a diversified investment universe including bonds. The main assumption beyond the equilibrium expected return is that. this procedure is known in literature as reverse optimisation (Cantaluppi 1999). The remedy of Black & Litterman (1990) was to firstly identify a reference point for expected return assumptions (equilibrium expected return). Given the debt crisis in the Eurozone. Reverse optimisation is a valuable tool for achieving consistency between model portfolios and the fund manager’s portfolio. Reverse optimisation. equities. the return of a short-term bond).

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). rates tended to also be more stable and more in line with nominal growth: this represented the first assumption. such as the home bias. In the past this approach was able to produce coherent figures. 6. This allowed monetary policy rates to be generally lower and more stable than in the '70s and the '80s. for example. In particular. The developed world economy had been characterised by decreasing inflation and decreasing volatility in growth and inflation in the '90s (see chart 11). Therefore. according to which investors tend to overweight assets of their own country. 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. Equilibrium-based models usually assume that the valuation of markets against fundamentals converge to some historical average or fundamental level. we need to forecast: the long-term inflation rate. hence the approach suffers from increasing the returns of assets which are in a bubble.x favours overshot assets (no mean reversion mechanism) when applied to the market cap According to the reverse optimisation technique. Amundi Investment Strategy Collected Research Papers 197 . Indeed it does not include some bias. in this more stable environment. inflation expectations stabilised at fairly low levels over the '90s. 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. x assumes no segmentation in markets and no constraints The model does assume that investors can buy any markets without any regulatory constraints. According to this model. returns are proportional to the market cap. No mean reversion mechanisms are present to alleviate the problem.

-63 déc.-93 déc. Chart 12 reports the time-series of US bonds.-81 déc.-01 déc.-67 déc.-85 déc.-91 déc. 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.-87 déc.-83 déc. The best way to overcome difficulties in equilibrium-based models and valuation approaches rely on diversifying the estimation process with complementary models. such as valuation ratios.Chart 11 US GDP & CPI rolling volatility (5 yrs rolling) 3. we should remind the reader that returns sometimes may never revert over a practical time frame. and they make it difficult to apply this approach because econometric relations to estimate risk premia might not be robust in non-stationary markets.50% 0.-61 déc. The plots show clearly that the return-to-the-mean sometimes is very slow. Current market conditions are now far more uncertain.-65 déc.-03 déc.50% 2.-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.-77 déc.-97 déc.-57 déc.00% 0.-69 déc.50% 3.-09 déc.-59 déc.-75 déc. have a better track record in forecasting asset class returns than rearview-mirror measures” (Ilmanen 2011). and that risk premia can exhibit robust trends which can last for many decades.-89 déc.-79 déc.-71 déc. 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. 198 Amundi Investment Strategy Collected Research Papers .-07 déc.-05 déc.00% 1. The figures are contrasted to a hypothetical long-term average and dispersion (see annex for details).-73 déc.-95 déc.00% 2. Furthermore. US equity and Japanese equity in specific time periods.00% déc. This environment favoured equilibrium-based model approaches.-99 déc.50% 1.

12 1996.05 1978.05 1999.03 1937.12 1929.06 1932.5 -1 -0.8 1 1.10 1975.12 1998.11 median median 1936.12 Chart 12 1990.12 1934.12 1978.11 2003.11 1940.11 1931.02 1939. 7.12 1992.06 1974.07 1977.07 1970.06 1994.1 1991.08 1941.04 -2sigma -2sigma -2sigma US Bond 1965-1981 (Cumulative Return) 2003.06 1942.12 1981.06 1968.12 1997.5 -1 -0.12 2010.4 0.5 -0.10 1942.12 2002.06 1935.10 rid of expected returns? 1933.08 1937.08 1973.03 1992.01 1932.04 1966.12 2009. Cumulative Performance (logarithm) Cumulative Performance (logarithm) Cumulative Performance (logarithm) -1.02 1934.12 median 1999.06 1991.06 1996.02 1977.06 2001.5 -2 -1.06 2005.12 The last decade in the equity market has been particularly dramatic for diversified asset allocation such as pension funds and balanced portfolios: poor performances.09 Japan Equity 1990-2012 (Cumulative Return) 2004.01 1974.6 0.06 1973.05 1934.01 1967.10 1972.09 1971.03 1933.5 1 1.06 1993.12 2007.06 1938.01 1936.08 1980.12 2006.06 1995.06 1940.07 1971.01 1980.09 1939.06 mkt mkt mkt 1998.03 1979. large draw- How to handle uncertainty in expected returns? Can strategic asset allocation get 199 .2 0.06 1931.06 2010.12 1938.06 1937.12 1995.04 1969.05 1981.5 1.06 +2sigma +2sigma +2sigma 2002.09 1969.02 1965.06 2008.5 0 0.2 1989.06 2006.12 US Equity 1929-1942 (Cumulative Return) 1976.02 1970.2 0 0.07 1930.04 1939.06 2009.03 2000.12 1964.07 1976.12 2001.09 1966.05 1965.05 1975.10 1979.12 1935.12 Amundi Investment Strategy Collected Research Papers 2005.12 1930.04 1972.5 3 3.11 1968.12 1994.08 1932.01 2008.03 1941.5 2 0 0.5 1 2 2.08 1967.

The virtue of the risk-parity approach is robustness in portfolio construction. The portfolio construction implication of these approaches is the overweight of low risk assets. elaborating investment processes that can completely neglect assumptions on expected returns. thus complementing traditional asset classes (bonds and equities) with commodities. and the effect of incorrect assumptions in the portfolio optimisation process. 2012). inflation. and factors (growth. Meucci (2011) consider robust Bayesian allocations that also account for the estimation errors in co-variances. The most famous Bayesian method is the Black-Litterman (1990) model. Chopra and Ziemba (1993) demonstrated that the sub-optimality due to estimation risk can be dramatic. investment styles (value. volatilities). Moreover. In the eighties and nineties. the maximum diversification portfolios. trend. Meucci (2009) proposed a diversification measure which allows investors to diversify portfolios on independent factors (principal components). These portfolio construction schemes allow the investor to make allocation according (only) to a risk model for the asset-classes. the choice of this range is quite arbitrary. the seconds are good candidates for diversified allocation. Robust allocations deal with uncertainty of input parameters by choosing the best allocation in the worst market condition within a given uncertainty range. illiquidity. the financial industry moved even further. Best and Grauer (1991). While the former has been used for equity portfolios. and thus the need for leveraging allocations in order to match medium-high risk profile without losing equilibrium among portfolio’s bets. Risk-parity approach is proven to be superior to the traditional mean-variance solution in case of high uncertainty in the input parameters. (1980). the approach allows investors to modify the region according to their specific views. and the risk parity approach (see Clarke et al. and tail risks). the limit is that it is difficult to match it with a specific 200 Amundi Investment Strategy Collected Research Papers . The main approaches are the Bayesian methods and the robust models. 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 most popular were the minimum variance. Jobson and Korkie. More recently the need for diversification led researchers and practitioners to explore new directions. carry. Robust Bayesian allocations enable the uncertainty region for the input parameters to be defined in a more coherent way. real estate.downs and very slow recovery again call into question the predictability of risk premia. Indeed. More recently.

In Pola and Facchinato (2013).macroeconomic view. Main results are: x in inflationary scenarios nominal bonds suffer. 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. and equilibrium-based approaches: high non-stationary patterns for the Sharpe ratio on one hand. x in deflationary scenarios. the increase in macroeconomic volatility on the other hand. In Pola and Facchinato (2013) we introduce a new reference portfolio where different macroeconomic scenarios are in equilibrium. We investigated the effectiveness of statistical methods. gold does not provide excess return over inflation.g. 8. in the observed risk premia and the risk-return relationship among asset classes. it might be wiser to define the reference expected returns from a different perspective. 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. 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). Many financial variables are in an uncharted region. never reached going back many decades. Increasing the length of history and identifying the observed risk-return under different market regimes may help to clarify the picture. In some cases. high dispersion of Amundi Investment Strategy Collected Research Papers 201 . Conclusion The recent crisis exhibited major changes in the risk level. Euro peripherals’ bond market). 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. risk-parity models end up being pure statistical tools to diversify unobservable information. We briefly investigated the dependency of main asset classes on inflation. whereas more diversified commodity indices offer better real performances. we illustrate a new approach for strategic asset allocation (DAMS) which can be helpful to build more robust estimates for expected returns in uncertainty. thus expressing the current uncertainty in financial markets. nominal bonds are the best hedge. Given the limits of the CAPM hypothesis. make it difficult to use recent historical time-series to calibrate econometric models.

econometrics and quantitative finance. 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. Investigating the Sharpe ratio as a function of volatility. higher Sharpe ratios. less liquid assets. This pluri-disciplinary approach may overcome some of the difficulties experienced by both statistical and equilibrium-based approaches. x relying more on risk to construct portfolios (e. On a more practical basis.25 is a reasonable estimate for the Sharpe ratio of equity markets for single countries and a ten-year horizon. 0. we should point out that a normative hypothesis of equal Sharpe ratio for each asset class allows us to build more diversified portfolios. relying on pluri-disciplinary skills (macroeconomics. 202 Amundi Investment Strategy Collected Research Papers . 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. Indeed whenever the portfolio expected return is relevant per se (e. Despite these empirical evidences. maximum diversification and risk parity portfolio construction).g.conditional equity returns on inflation buckets. on historical basis. global equity indices are likely to deliver higher Sharpe ratios due to index diversification. determining the discount rate for pension funds and insurance companies). higher quality credit bonds. minimum variance. suggests to us that inflation is not the primary risk driver.g. behavioural models are important as well to explain many documented anomalies in the markets). we prefer a more detailed description of asset classes’ Sharpe ratio. we show a rich phenomenology according to which low volatility investments. and more diversified indices delivered.

and Marc- Ali Ben Abdallah for very stimulating discussions and very constructive suggestions. Sergio Bertoncini. Amundi Investment Strategy Collected Research Papers 203 .Acknowledgements We would like to thank Eric Tazé-Bernard. Jean-Renaud Viala.

all volatilities increased. and Japanese bonds. The 2008 crisis was a dramatic year for most of the spread markets. Canadian.ANNEX A1 In chart 3b we measured the volatilities. Bond markets. With the exception of Australian. French bonds did not exhibit any draw-downs not priced before. The increase in volatility of CRB was more marked with respect to the GSCI and gold. 204 Amundi Investment Strategy Collected Research Papers . Inflation linked markets. and Canada did not register worst draw downs. Credit markets. France. The issue is the decoupling of the Eurozone: Italy’s MDD doubled in the crisis period. Germany. gold performed better in the crisis (the average compound return increased). the difference between CRB and GSCI indices is mainly due to the larger exposure of CRB to energy commodities. The volatilities increased but only marginally with respect to the pre-crisis period. euro investment grade suffered less. The crisis increased the MDDs in US. CRB and GSCI indices registered new MDDs in the crisis sample. Japan. 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. 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). and performance of some asset classes in two historical periods: pre-crisis sample (1990-2007) and the full sample (1990-2012). Commodity markets. Equity markets. UK. Nevertheless Germany suffered between the end of 2011 and first quarter of 2012. In the Eurozone. MDD for US investment grade more than doubled. maximum draw downs (MDD). Eurozone debt crisis drove the main changes: the euro index presents the higher increase in volatility. Euro peripherals. bold indicates an increase in volatility. The grey shaded areas in chart 3b signal markets that exhibit a worse MDD than before. World high yield doubled its MDD. Australia and Emerging markets. All volatilities increased except for the emerging market debt in hard currencies.

Chart 3b Amundi Investment Strategy Collected Research Papers 205 .

24 0. Denmark.23 0.28 0. It reports the performance of various US markets from 1960 to 2009. All time-series are monthly and in local currency.19 Japan 0.09 0. 2008) for financial markets and the global economy. volatility. Netherlands. and Maximum Draw Down of various indices in two samples: pre-crisis sample (1990-2007) and the full-sample (1990-September 2012). Data provider is Bloomberg. Chart 3a Chart 3a is redrawn from Ilmanen (2011). Switzerland. Austria.21 0. The scatter plot relates the compound average real return (Y-axis) to the average real losses in the three worst years (1974. Italy. Japan.20 0.10 Germ any 0. Spain.29 0. Chart 2 The plot reports the two-year yields for international countries: Germany. France.19 0.18 0.28 UK 0. Sweden. Canada. US.21 0.27 0.27 0. Data provider is Schiller database.35 USA 0. Chart 3b Chart 3b contrasts the performance.14 0. 5 years.33 0.10 France 0.21 0.24 0.29 0.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.32 UK 0.40 Japan 0. Belgium.29 Germ any 0.26 0. Greece.24 France 0. 3 years) Table1 MEDIAN 10 yrs 5 yrs 3 yrs AVERAGE 10 yrs 5 yrs 3 yrs USA 0.19 0. A few indices are shorter due to their 206 Amundi Investment Strategy Collected Research Papers .38 ANNEX A3 Chart 1 The plot reports the US Treasury yield 10-year. UK.30 0. 1981. Portugal.

long-term corporate investment grade. commodity CRB since January 1994. Chart 4 The plot reports the historical 1-year rolling volatility (top panel) for German. all-maturity Amundi Investment Strategy Collected Research Papers 207 . 3 years). 1993. Table 1 in annex A2 reports the summary of the results: median and average for countries and horizons (10. Australian equity since May 1992. 5. Data provider is Schiller database. post-war and new millennium. 1998. and their historical 1-year correlation to the EMU equity market. 1995. US equity. Time-series are monthly. data provider is Bloomberg. Chart 6 The plot reports the 10-year CDS for Italy and Germany since 2005. Time-series are daily. Chart 7 The plot reports computations of Sharpe ratio over rolling windows (10 and 3 years) of various equity markets (US. long-term treasury. Credit Euro HY since Dec. Germany. 1993. and compares them to the inflation rates in the same period. Spanish equity since Dec. Data provider is Bloomberg. 1997. Emerging Inflation Linked since Dec. Chart 5a & 5b The left chart reports the performance of US bonds. 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. Spanish and Italian 7-10 year bonds. all- maturity corporate investment grade. Japan. all-maturity treasury. France). 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. Performances are presented in decades (except for the last “decade” which includes 12 years). Data providers are Datastream and Bloomberg. Emerging Debt in Local currency since Dec. short-term corporate investment grade. Emerging Debt in Hard currency since Dec. Credit Euro IG since Dec. 1996. The risky assets are: short-term treasury. Data provider is Bloomberg.inception: World Inflation Linked since Dec. The vertical bar divides the sample in three periods: pre- war. 1999. Euro Inflation Linked since Dec. 2003. UK.

x constant drift. Data providers are Bloomberg and Schiller database. Historical volatilities in the previous decade (monthly observation).20) and a risk-free rate given by the 10-year bond yield at the beginning of the sample. We assumed a constant Sharpe ratio (0. 208 Amundi Investment Strategy Collected Research Papers . x constant volatility.

vol. 43. F.”. 315-342. 1991. H. 6-11. no. He. F. Benartzi. 35. Baker. Fama. and Bernstein.. D. Frazzini.. Chopra. E. 58. S. A. variances. J. Loss Aversion.”. M. “Benchmarks as Limits to Arbitrage: Understanding the Low-Volatile Anomaly.”. P.. and Coignard. “Risk Parity. S. 1988a. B. Cantaluppi. S.. 1999. and Litterman. Best. Quarterly Journal of Economics. G... Financial Analysts Journal.. New York University and NBER Working Paper 16601.”. 1989. no. “The Reverse Optimization. Campbell. 2000. F. R. and Ming H. 3 (July): 661–676. 2010. Asness. J.. Journal of Portfolio Management pp. AQR Capital Management..”. and covariances on optimal portfolio choice. 25. and Individual Stock Returns. R. and Wurgler. vol.. and Shiller. Goldman. vol.”.. and Minimum Variance: An Analytic Perspective. 2011. R.. and Kenneth. 4 (August): 1247–1292. Financial Analysts Journal. vol.. P.”.”. 1995. W. 56. Y. 2008..”.. 1993. Journal of Finance. “The effects of errors in means.”. “Myopic Loss Aversion and the Equity Premium Puzzle. M. Review of Financial Studies 4. 1999. Investment Management Research.. Barberis.. vol.. 2001. Journal of Finance.. “Betting Against Beta. and Litterman. “Stock Prices. 2 (March/April): 64–85. N. R. R. no.. 67. R. and Ziemba. no. Bonds: Explaining the Equity Risk Premium. Clarke. Amundi Investment Strategy Collected Research Papers 209 . no. and Thorley. C. L. L.. 2 (March/April): 96–113. Goldman Sachs.. 1990. 1 (Fall). “On the sensitivity of mean-variance-efficient portfolios to changes in asset means: Some analytical and computational results. Bradley. Earnings and Expected Dividends. “Business Conditions and Expected Returns on Stocks and Bonds. 1 (November): 23–49. 2012. Vol. Journal of Financial Economics.”.. V. J. “The Intuition Behind Black-Litterman Model Portfolios. vol. 110. “Mental Accounting..”. 56. Working Paper. Y. and Grauer.References Arnott. R.. no. “Stocks vs. Black. “What Risk Premium Is ‘Normal’?”.... R. 1. no. “Towards Maximum Diversification. “Asset Allocation: Combining Investors Views with Market Equilibrium. vol. Financial Analysts Journal. and Thaler. Y.”. De Silva. R. No. Journal of Portfolio Management.”. Financial Markets and Portfolio Management. H. 2002.. Maximum Diversification. Sachs & Company. working paper. Choueifaty. 1 (February): 73–92. and Lasse. T. Fixed Income Research.

G. R. 210 Amundi Investment Strategy Collected Research Papers . J. in preparation..Ibbotson. “Portfolio Selection”. Balancing macroeconomic scenarios”. “Risk and Asset Allocation”. Reinhart...”. Assets’ segmentation in response to macroeconomic changes”. “Estimation for Markowitz efficient portfolios”. R. 1980. 2011. H. 1976a. working paper. The Journal of Business. The Journal of Finance. and Facchinato. no. Bills. M. vol. vol. S. Bonds. “Plight of the Fortune Tellers: Why We Need to Manage Financial Risk Differently”. Ibbotson. “Managing uncertainty with dams. Meucci. 2013. D. NJ: Princeton University Press. 49. A. 49. Wiley Finance Jobson. Ilmanen. Markowitz.”.... S. B. 2013. Mandelbrot. 1952. G. Bonds. M.. K. R. 2011.. symmys. G. Rogoff. “Managing uncertainty with dams. Meucci. and Inflation: Simulations of the Future (1976–2000). 2008.. A. and Korkie. Journal of the American Statistical Association 75. “The variation of certain speculative prices”. A. and Sinquefield. “Expected returns. no... An Investor’s Guide to Harvesting Market Rewards”.. G. 1976b. Princeton. 2009. and Sinquefield. R. Journal of Business.com Pola. B.. “Stocks. R. 2007. 1 (January): 11–47. 1963. “Stocks. Journal of Business.. Rebonato. 544-554. A. C. 3 (July): 313–338. American Economic Review. in preparation. Pola. “From Financial Crash to Debt Crisis”.. A.. “Managing Diversification”. 101: 1676-1706. Springer Finance. Bills and Inflation: Year-by- Year Historical Returns (1926–1974).

we provide a formal description of two investment processes that successfully exploit this low risk anomaly: the minimum variance and the risk parity portfolio. We find that most of the relevant empirical studies focus on systematic risk. Some other explanations rather refer to behavioral bias. only a few exceptions refer to idiosyncratic volatility. and the Amundi Experience Alessandro Russo. 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. and the utility function of fund managers. Theories. According to theories referring to leverage constraints. In the last section of our research. Amundi Investment Strategy Collected Research Papers 211 . 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. WP-033 Low Risk Equity Investments: Empirical Evidence. benchmarking. Researchers have been documenting such anomalies since the early nineties. inflating their prices and thus lowering subsequent returns. while some of them state that the anomaly holds regardless of whether total or systematic risk is considered. Some theories focus on the distribution of equity returns: skewness and convexity. investors willing to leverage but facing leverage constraints buy high beta stocks. delegated portfolio management. At the same time.

1. as well as many subsequent studies proving the outperformance of low volatility stocks versus high volatility stocks. when corrected for size effect.by higher returns. and its persistence during recent decades. Jensen and Scholes (1972). while taking investment decisions. 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. investors maximize expected returns for a given amount of portfolio risk: this leads to a positively sloped and concave efficient frontier. 212 Amundi Investment Strategy Collected Research Papers . our goal is not to take a position in the everlasting dispute in favor or against the CAPM. more recent evidence shows that well-constructed low risk portfolios contradict this axiom. In his Portfolio Selection (1952). In fact. the latter exhibiting a higher expected return than the former. the slope of the linear relationship is flatter than anticipated by CAPM. Low Risk Equity Investing. as they deliver higher returns than riskier portfolios. According to the theory. quite surprisingly. In 1992. In Sharpe’s CAPM (1964) risk is defined as portfolio beta relative to the Market Portfolio. Yet. in Fama and MacBeth. and Fama and MacBeth (1973) form portfolios on the basis of in sample estimations of beta. The idea behind the CAPM is that any asset’s average return may be expressed as the sum of the risk-free rate and a risk premium. empirical evidence of success The first axiom of investing is that higher risk is compensated -on average. and then look at realized Beta and stock returns. Fama and French showed that value and size factors generate large returns in the US markets and. that the relationship between beta and returns is rather negative. the latter being proportional to the asset’s covariance with the Market Portfolio (its beta). Markowitz defines portfolio risk as the volatility (or variance) of its returns and states that. both the Minimum Variance portfolio and the Market Portfolio belong to the efficient frontier. For instance. finding that higher beta is positively correlated to higher returns. However it is worth mentioning some preliminary tests that provided some comforting results during the early seventies. Despite some statistical weakness. As mentioned. these contributions held for a long time as a general confirmation of the traditional positive relationship between risk and return. Black.

Amundi Investment Strategy Collected Research Papers 213 . The author’s controls for factors such as value. rather than systematic and total risk. companies with greater sensitivity to the VIX factor should exhibit higher idiosyncratic risk in a Fama French framework. after controlling for aggregate volatility. and via cross basket analysis: returns of the volatility factors are investigated among stocks with similar momentum. The significance of the low risk effect is little reduced. They explain such evidence by the premium that investors pay (thus reducing subsequent returns) for stocks that are positively correlated with volatility jumps. 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. As a consequence.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. 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. size or value characteristics. the negative relationship between returns and idiosyncratic risks is little changed. as the aggregate volatility (VIX factor) is significant indeed. Intuitively. while high variance stocks exhibit high beta. thus providing a hedge in case of market drops (as typically market drops take place together with volatility jumps). but missing in the Fama French model. momentum. replicating the VIX itself). Blitz and Vliet (2007) focus on systematic and total volatility. the authors associate the low-risk high-return anomaly to idiosyncratic risk. Surprisingly. They also find a strong negative relationship between idiosyncratic risk (residual component in a Fama French environment) and future returns. are both via linear regressions with those factors’ returns as explanatory variables (with little if any variance explained). and size. and only within baskets of stocks with similar market capitalization. and completely unexplained.

with baskets built according to beta showing somehow better regularity and monotony (Annex 1. Baker and Haugen first investigate Minimum Variance portfolios in the US equity market. They estimate a covariance matrix with Bayesian methods for shrinkage -in order to avoid error maximization problems. turnover (143% with monthly rebalancing). with no reduction in average returns. or any other factor that could help to explain the anomaly. They first detail some well-known portfolio characteristics like typical concentration (75 to 250 stocks with 3% cap on a single company). 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. they use an equally weighted basket and 24 months historical volatility as a measure of risk. size. tables 1 and 2. compared to high risk stocks. Clarke. compared to both a common US index and randomly selected portfolios. 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. pointing out a 30% reduction in portfolio volatility. 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. and zero mean but rather volatile exposure to momentum factor.Baker. positive exposure to size and value factors. Bradley and Wurgler (2011). in a rolling window of three years. They also point out that only rarely do the highest volatility deciles outperform lowest deciles. de Silva and Thorley (2006) build Minimum Variance portfolios on the largest 1000 US stocks over the period 1968 – 2005. apparently supporting argument by Blitz and Vliet (2007) for no need for a sophisticated risk model. In 1991. Actually no control is done for valuation. In their intentionally simple approach. They confirm Baker and Haugen’s evidence (1991) of 30% reduction in 214 Amundi Investment Strategy Collected Research Papers . 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). in any observed country.as well as principal component analysis.

Amundi Investment Strategy Collected Research Papers 215 . even though optimization remains our optimum for Minimum Variance portfolio construction. changes in the correlation matrix generate higher than justified turnover. In their more recent (2011) “Minimum Variance Portfolio Composition”. However their work is worth mentioning mostly because. 2. Furthermore in their view. and Moulin (2011) discuss typical low beta and small cap exposure of Minimum Variance portfolios. Why low risk stocks outperform Most of the recent literature addressing the low risk anomaly has also offered some theoretical framework. like the historical dividend yield. while controlling many other constraints. Thomas and Shapiro (State Street Global Advisors.volatility with substantially no reduction in average returns. while proposing an interesting analytical solution of the Minimum Variance portfolio composition. rather than optimizing. a weighting scheme may be a reasonable solution. de Silva and Thorley update previous (2006) statistics with basically no change. in order to avoid the typical drawbacks of a Minimum Variance investment (excessive concentration in a few low risk sectors and stocks. In our opinion. Lu. stocks with beta higher than a threshold are strictly excluded from the portfolio. 2007) show their encouraging results on a Minimum Variance portfolio built on the Russell 3000. Clarke. while high idiosyncratic risk “only” contributes to lowering the stock’s weight in the portfolio. Results are little changed after constraining fundamental factors’ exposures and reducing turnover to 56% through quarterly rebalancing. thus they suggest constructing portfolios by applying stable weighting schemes based on stocks’ beta. while enhancing their performance by tilting the portfolio toward some long term successful alpha strategy. they point out that systematic risk dominates in the construction of such a portfolio: in their simplified single factor model. lack of control for involuntary factor exposure). 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. as an optimization package enables turnover to be reduced. Carvalho. using a standard optimization package (BARRA).

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 delivering poorer returns than the CAPM would predict. The authors build a “betting against beta” factor by going long on low beta stocks and shorting a smaller amount of high beta stocks. or taking into account higher moments (and their relative premium) than the mean and the variance of returns. investing massively in high beta stocks thus lowering subsequent returns. Leverage Constraints Blake identifies borrowing restriction as one of the possible sources of low risk stocks outperformance in his “Beta and Returns” (1995). Lu. treasury. This slope is flatter as the tightness of the funding constraint increases. They provide empirical evidence for alpha decreasing with security beta. causing those assets to offer lower subsequent returns. Carvalho.1. However. equilibrium models addressing the utility function of fund managers (rather than of investors) in delegated portfolio management. and for the “betting against beta” factor exhibiting positive returns in equity. thus obtaining a beta neutral factor. the former buy low beta stocks and leverage their holdings by short selling the high beta stocks. credit. and create a demand imbalance. and investors for which leverage is forbidden. investors prefer riskier stocks. 216 Amundi Investment Strategy Collected Research Papers .Explanations account for borrowing constraints and other market frictions. thus reducing the slope of asset returns relative to beta itself. 2. and Moulin (2011) agree that leverage constraint has a major impact on the preference towards high beta stocks. theories referring to leverage constraints are probably the most represented. 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. thus it is not immediately clear why only high beta stocks should be overpriced. Intuitively both the low beta and the high beta stocks have their demand in the market. as they simply seek high returns. and currency markets. Among the others. behavioral hypothesis. they add that. They argue that while the latter overweight high beta assets. Investors are supposed to pass through such a limitation.

Delegated portfolio management. and fund managers’ utility function Cornell and Roll (2005) recognize that the considerable market share of investment being nowadays delegated to professional fund managers. make a step forward focusing explicitly on the low risk anomaly. Without a direct implication on performance of low risk versus high risk stocks. 2. (2) consider few cases of Amundi Investment Strategy Collected Research Papers 217 . The objective function of the fund manager is not the investor’s wealth maximization. The authors finally mention another well represented family of theories that will be further discussed in the next section. which are perceived to be similar to lottery tickets. among the possible explanations of the anomaly. the anomaly is not fully arbitraged away. they show how a pricing model based on delegated investments (where fund managers’ objective function dominates the investors’ utility function). for which they are much less risk averse. these portfolio managers may seek to maximize outperformance in the upward markets. These relationships violate CAPM. who (1) favor lottery-type investments.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). Baker. they take into account a behavioral theory by Shefrin and Statman (2000). The anomaly itself is indeed a consequence of irrational behavior of non-professional and professional investors. impose asset pricing models to incorporate the objective function of the agents. in equilibrium implies some cross sectional relationship between stocks’ alpha and their beta relative to the benchmark. investors will overpay for (often few and badly diversified) risky stocks. thus systematically preferring high beta stocks. In this case. but the maximization of active return versus the benchmark of the delegated mandate. together with the traditional utility function of the final investors. and a high aspiration layer.2. where investors allocate their wealth according to two layers: a low risk layer designed to avoid poverty. 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. Bradley and Wurgler (2009 and 2011). benchmarking. However.

or an absolute discretionary threshold). However. 218 Amundi Investment Strategy Collected Research Papers . for a higher probability of beating their benchmarks (or a given target). and are likely to disagree on future stocks returns. Supporting their intuition. as the risk return relationship is rather inverted. that is usually paid when performances are positive in absolute terms. Finally fund managers may find it easier to justify holdings or turnover of newsworthy stocks. fund managers would have a concrete incentive in buying low risk stocks. (3) are overconfident. provided by high volatility stocks. from 2000 to 2011– companies with higher institutional ownership exhibit higher volatility than stocks with low institutional ownership. analysts are often willing to recommend stocks “in the news”. and higher than a threshold (benchmark return plus the management fees. like Cornell and Roll (2005). Portfolio managers exchange higher expected returns of low beta stocks. Their explanation for fund managers not actually exploiting the anomaly is that they seek to maximize the probability of receiving a bonus. 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). but the latter still generate a higher tracking error relative to the mandate benchmark.success of high volatile stocks as representative of all the other high volatility stocks. the authors recognize the prominent role of delegated and benchmarked investment. they find that analysts’ coverage (number of recommendations) is positively correlated with volatility. as a limit to arbitrage: behavioral bias result in high absolute risk stocks delivering similar returns than low absolute risk stocks. the authors show that –for 1000 US stocks grouped in 10 homogeneous classes of market capitalization. or stocks whose news flow is quite intense and that tend to exhibit higher than average volatility. this time related to the delegated portfolio construction process: in order to impress colleagues. state that. The authors also mention some additional incentives to hold volatile stocks. trade more actively when optimistic than when pessimistic. Finally. 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). Backer and Haugen (2012). especially on high volatility stocks.

for any pair of baskets. while having aversion to variance.3. The signs and their significance prove some interesting monotony: higher skewness for higher beta stocks. Kraus and Litzenberger (1976) find that. it is transparent and easily replicable. 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. 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 hypothesis of skewness increasing with beta and volatility is confirmed. this latter is associated with a positive price (instead of a discount. we then formed equally populated baskets according their betas. (2003). however. Similar findings are provided by Prakasha et al. as it is the case for variance). if the capital asset pricing model is extended to include systematic skewness. Results are even more significant. as skewness has a positive price. We then computed differences and T- statistics of the average skewness. We regressed stocks’ weekly returns (excluding only those stocks with less than two years of available data) on the index returns. it should be higher among high beta or high risk stocks. We recognize our analysis is completely in sample and misses some predictive power. and this suggests that the premium that investors pay for a Amundi Investment Strategy Collected Research Papers 219 . As a consequence.2. investors exhibit a preference for positive skewness. especially for baskets whose differences in beta are high enough. so that these latter are priced at a premium compared to a CAPM equilibrium. Results are summarized in tables 3 and 4 in Annex 2: skewness generally increases with beta. We repeated the exercise using total ex-post volatility as a measure for sorting and grouping stocks (tables 5 and 6). as skewness increases almost monotonically with volatility. and the zero intercept for the security market line is not rejected. 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. The intuition behind these three contributions is that. The evidence also suggests that investors exchange expected return for positive skewness. sometimes significantly.

-33% in the case of a stock beta higher than 3. we apply the methodology described for skew in the previous paragraph. They explain that high beta stocks provide a call option payoff. and a second one with the series of MSCI World squared returns as an additional explanatory variable: Ri = β i RMsci + ui (1) and 2 Ri = γ i R Msci + λi R Msci + vi (2) 220 Amundi Investment Strategy Collected Research Papers . that is activated in case of market returns of.positive skewness may explain lower subsequent risk-adjusted returns for high beta and high volatility stocks. multiplied by a factor roughly equal to their beta (whatever the market return is. to test if we find increasing convexity. let’s say. at worst. or -50% in case of beta higher than 2). in case of negative market returns. while low beta stocks have negative convexity. However. losses are limited to 100% of invested capital. Cowan and Wilderman (2011) find that high beta stocks exhibit a positive convexity relative to broad market index returns. exactly like a leveraged position). Conversely. In the case of positive market returns. The difference with leveraged investments is straightforward as the latter generate payoffs exactly equal to the returns of the unleveraged positions times the leverage. 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. that is quite rarely (the convex profile takes place in the form of a stop loss. high beta stocks deliver market returns. In our view this explanation may be convincing only for very extreme market returns. with theoretically no limit to downside. 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. For each MSCI World constituent (from January 2003 to June 2012. for increasing beta.

3. They all belong to the absolute risk category: away from the notion of tracking error or information ratio. Table 9 and 10 report results for the same experiment. or on portfolio construction techniques as maximum Amundi Investment Strategy Collected Research Papers 221 . they focus on Sharpe ratio or risk-adjusted return. while table 8 reports differences in average convexity for any pair of baskets. Amundi has strongly invested in order to meet investors’ needs in such a challenging market context. We group stocks according to their estimate of ȕ from OLS regression 1. during the last decade. Results are even more significant as convexity increases almost monotonically with volatility. Table 7 reports average convexity and standard deviation across any basket. Their risk-return profile differs as well as their behavior in up and down markets. Results are summarized in Annex 3 and seem to support Cowan and Wilderman’s intuition. when we rank and group stocks according to total ex-post volatility. Differences are often significant and this may explain some premium paid by investors for high beta stocks. In the last few years. and we compute the average and the standard deviation of the estimate Ȝ for any basket. Over the last decade. and volatility metrics. 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.where a positive estimate of Ȝ implies positive convexity of stocks’ returns. relative to market returns. developing a range of innovative solutions aiming at Sharpe ratio improvement. They are based either on the use of instruments providing favorable asymmetry (options and other derivatives). We then test the significance of differences in average basket convexities. the larger and more significant the difference in convexity: high beta stocks exhibit a higher and thus more profitable convexity than low beta stocks. Equity investors have been faced with a major and unfavorable change in traditional risk return payoffs. 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). The higher the distance of betas.

at any time t of our sample.from now on). and risk parity (“Smart Beta” -or simply “Smart”. we provide some empirical evidence on two back-tested Minimum Variance and Smart portfolios. For this reason it is worth investigating whether Minimum Variance and Smart processes limit portfolio volatility mainly by selecting low risk stocks. during the period December 2003 – December 2011. with a 10 to 20% reduction for Smart. However. For each of them at every quarter of our back-test.85 (again significantly lower than 1 at 1% confidence level) and the ex-ante beta ranges from 0.87.7 to 0. and wi(t) is its weight at time t.diversification. In table 12 we further investigate the systematic risk characteristics of the two portfolios and the benchmark in a multifactor framework.2).48 to 0.systematic risk and beta are by far the most significant. 222 Amundi Investment Strategy Collected Research Papers . is computed as follows: CF(t) = Ȉi wi(t) CFi(t) where CFi(t) is the common factor risk of the ith stock at time t. and up to a 35% reduction for Minimum Variance (Annex 5. On an ex-ante basis with quarterly observations. portfolio beta ranges from 0. minimum variance. In Annex 4. we compute the weighted average common factor risks extracted from the BIM model by BARRA1. 1 Weighted average common factor risk. most of the relevant literature strictly identifies low risk strategies with the selection of low risk stocks and -among several measures of risk. On an ex-post basis. the volatility of the Global Minimum Variance and Global Smart portfolios is significantly lower than the market index. both of them are lower than the MSCI World. Weighted average common factor risk of the Minimum Variance portfolio is lower than that of Global Smart. table 11 shows that the ex-post beta relative to MSCI World is 0.55 (significantly lower than 1 at a 1% confidence level).65. The latter two are particularly relevant to this document as both exploit the low risk anomaly discussed previously.1 and 5. As for the Global Smart portfolio. Table 11 shows that the ex-post beta of the Minimum Variance portfolio on full data sample is only 0. or rather by enhanced diversifications.

The same result holds for average common factor risk. and the stock weighting scheme (inversely proportional to their total volatility). We use an optimization process to build a portfolio sitting on the very edge of the efficient frontier. median. with slightly higher frequency on stocks with average risk.Charts 2 and 3 show average portfolio weight over stocks grouped by beta and common factor risk2. and low risk). Amundi Global Minimum Variance: an enhanced process As shown. The two of them are described in detail in section 3. the most straightforward way to exploit the low risk anomaly is certainly building a portfolio of stocks with the lowest possible risk. Amundi Investment Strategy Collected Research Papers 223 . for any portfolio. compared to the MSCI World.1. Pacific ex Japan. while weights are distributed much more uniformly for the MSCI World. 2 Every quarter we build three equally populated baskets of stocks. while being fully invested. and other customized universes. 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). In fact. we focus on the Global Minimum Variance as it is the most complete case for descriptive purposes. Minimum Variance and Smart portfolios overweight stocks with lower than average beta. 3.2. and Moulin (2011). according to their beta and common factor risk (high. and our paper portfolios on developed World. Aggregate weights decrease with increasing risk for Minimum Variance and Smart. We then compute the aggregate weight of stocks in each group. 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. with two Europe portfolios (since 2007 and 2009 respectively). a very recent Global portfolio. Emerging markets. Japan. Lu. we further extend these findings to Smart Beta as well. We claim several years of experience in Minimum Variance management. Finally we compute the historical average of aggregate weights. This evidence supports the intuition by Carvalho. In building such a portfolio. 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. In this section we describe our Minimum Variance approach on a Global Developed Equity universe. expected returns are not needed as the only requirement is to minimize volatility. However in this document.

Thomas and Shapiro (2007) highlight the risk of the Minimum Variance portfolio being excessively concentrated on few low risk sectors. Although we recognize the advantage of such a process being transparent and intuitive. and Moulin). They also express their preference for tilting portfolios toward some successful stock ranking criteria. 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. Similarly. at least in the long run. At the same time we don’t want to renounce an optimization process which is completely independent from expected returns.Our simple objective function is thus: Min (wTVw) Such that eTw = 1 where w is the vector of the optimal portfolio weights. 224 Amundi Investment Strategy Collected Research Papers . Expected returns are very noisy in forecast and thus responsible for well-known “error maximization” problems. we are conscious of some typical drawbacks that may arise from Minimum Variance portfolios: as shown in Clarke. positive exposure to value and small capitalization stocks (with some relevant implications on liquidity). That’s why we implement our enhanced portfolio construction process in Barra One. may exhibit rather high turnover. and eT is a vector of ones. as suggested by Carvalho. Minimum Variance portfolios may be quite concentrated on a few low volatility stocks. as described below. V is the variance-covariance matrix. Lu. de Silva and Thorley (2006). and the lack of control for involuntary factor exposure. 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. Quality Stocks We believe that fundamental equity selection can provide some valuable enhancement in the risk return profile of equity portfolios. and some volatile exposure to momentum factor.

D is the number of days that we accept to liquidate the fund.the top quality 33% of MSCI constituents (equally weighted) have outperformed the market index with lower volatility. Turnover and liquidity High turnover is a critical issue in many systematic investment strategies like Minimum Variance. as suggested by Baker and Haugen (1991). In our case. the median basket has performed in line with the market. turnover in the investment universe is limited as the Piotroski score is based on balance sheet data that varies very little during one quarter. excluding the lowest quality stocks from the optimization. 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. more than turnover itself. we apply a qualitative filter to our investment universe.For this reason. Table 11 in Annex 4 reports the balance sheet. ADVi is the average daily volume over the last quarter. 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. At the same time. without using explicit expected returns. Furthermore we also rebalance our portfolio quarterly. the optimizer is left with a high degree of freedom and it tilts the optimal portfolio toward good quality stocks. 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. Nevertheless. and the bottom basket has underperformed with even higher volatility. Amundi Investment Strategy Collected Research Papers 225 . and table 4 shows that -in the last decade. To address this requirement. 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. without incurring significant market impact costs. Basically. income statement and corporate actions employed in the Piotroski score. Keeping 60% of constituents available for investments.

table 15). Minimum Variance portfolios outperformed the standard index by a minimum of 3. Morel. As for performance (Annex 5. while volatilities and draw-downs were reduced by 226 Amundi Investment Strategy Collected Research Papers . imposing a neutral exposure to the suspected factor. country. The goal of this monitoring is to detect bubbles or suspicious asymmetries like excessive positive skewness in recent performance (Sornette. Malongo. growth. and stock concentration As mentioned above. we have decided not to manage them systematically as –again– we don’t want to excessively constrain the optimization process. we regularly monitor the behavior of all the risk factors of the BARRA model (size. On countries and sectors we accept deviations from the market index of 500 to 1000 bp. On the other hand.6% in Pacific (All Countries) ex Japan. without preventing the optimizer from choosing solutions that are far enough from a market index. momentum. Minimum Variance portfolios may tend to be poorly diversified across sectors. countries or single stocks. while for single stocks we apply a general upper bound (GUB). 2013): in the case of significant alerts. to a maximum of 5% in the World developed Markets. thus modifying the actual upper bound as follows: ª ADVi º UB i = min «GUB. 2003. We observe that much of our size exposure is corrected away by the liquidity constraints. we are conscious that Minimum Variance portfolios may be exposed to fundamental factors as size. value. we punctually hedge the risk of an exploding bubble. results are very satisfactory indeed: from the beginning of 2003 to the third quarter of 2012. We have thus applied some constraints at these levels. As for other factor exposures. leverage…).Sector.1. value or momentum. and Lambinet.25% D» «¬ NOT »¼ Management of asymmetries in factor returns Furthermore.

stocks must be held on the Amundi emerging market flagship fund. Pacific ex Japan. or must be top-ranked (33%) according to a Piotroski score. Emerging Markets. Emerging Markets with Sharia filter). 3. and Teiletche (2009) have pointed out. Furthermore. if the number of assets involved is somehow relevant. Minimum Variance portfolios belong to the absolute return investment category. For this reason we have decided to split our portfolio construction process into two steps: the region-sector allocation. As a consequence. and we are investigating the behavior of such strategies in many others areas (World ex Japan. Our approach is consistent across the regions with the Developed World being probably the best example for descriptive purposes. in order to be eligible to minimum variance optimization. As Maillard.000 constituents of the MSCI World. we typically deal with 1. We recall that we currently manage several risk parity portfolios (Euro Area. Europe.2. as it is confirmed by ex-post tracking errors ranging from 9 to 10% roughly. As for the number of assets involved. Optimization (for instance. Roncalli. Risk parity means that each asset (asset class. Amundi Investment Strategy Collected Research Papers 227 . single stock) has an equal contribution to the total risk of the portfolio. risk adjusted returns are twice as high as for the standard index3. Amundi Global “Smart Beta”: A systematic risk parity approach In this section we discuss our risk parity approach on global developed markets. 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. the minimization of the cross-section standard deviation of assets’ contribution to risk) does not guarantee a full risk parity solution. Japan. That’s the main source of performance difference between them. equity sector. and correlations are very heterogeneous. 4 The former of the two European portfolios has been designed in order not to exceed the ex-ante volatility of 10%. 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. and may not be achieved either through optimization.one fourth to one third. World). and the stock weighting in each regional sector basket. it can be 3 As for the Emerging Markets portfolio (from December 2005).500 to 2.

**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 ı is the volatility of portfolio P, MCi is the
**

marginal contribution to risk of each asset, and is equal to:

∂σ

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.

228 Amundi Investment Strategy Collected Research Papers

**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 =

σ

Amundi Investment Strategy Collected Research Papers 229

**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

stock’s 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.

230 Amundi Investment Strategy Collected Research Papers

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.

Amundi Investment Strategy Collected Research Papers 231

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)

232 Amundi Investment Strategy Collected Research Papers

**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

Amundi Investment Strategy Collected Research Papers 233

ANNEX 1

(Ex-post risk and returns in World Developed markets)6

6

Source for all figures: Amundi, Factset

234 Amundi Investment Strategy Collected Research Papers

ANNEX 2

(Skewness in World Developed markets) 7

7

Source for all figures: Amundi, Factset

Amundi Investment Strategy Collected Research Papers 235

Factset 236 Amundi Investment Strategy Collected Research Papers .ANNEX 3 (Convexity in World Developed markets)8 8 Source for all figures: Amundi.

Factset Amundi Investment Strategy Collected Research Papers 237 .ANNEX 4 (Amundi NextGen Equities and low risk exposure)9 9 Source for all figures: Amundi.

ANNEX 5 (Amundi NextGen Equities)10 Annex 5. Factset 238 Amundi Investment Strategy Collected Research Papers .1 – Minimum Variance 10 Source for all figures: Amundi.

Factset Amundi Investment Strategy Collected Research Papers 239 .2 – Smart Beta (risk parity)11 11 Source for all figures: Amundi.Annex 5.

240 Amundi Investment Strategy Collected Research Papers .

Amundi Florian Ielpo. we find that a total of five factors are at work. Panthéon-Sorbonne April 2013 The maximum number of uncorrelated strategies to be included in a portfolio is a pertinent issue when building cross-asset strategies. Factors are identified. Commodity prices are affected by two common factors. Amundi Investment Strategy Collected Research Papers 241 . the concentration of correlation is weaker. interpreted and their stability over time is investigated via testing the significance of correlation between factors over a rolling window. This question can be formulated as the search for the optimal number of factors in a factor model. while credit spreads display one factor. Research Analyst – Investor Research Center. many of them in the framework of Principal Component Analysis. Using a refined information criterion. we find the usual three factors. a task for which various criteria have been proposed. even during economic downturns. Doctoral Student. During recessions. we estimate the number of factors to be associated to five datasets. the number of uncorrelated strategies drops for US Treasury Bond rates. In the case of US Treasury Bond rates. Yet for commodities. University of Paris 1. whereas currencies demonstrate only one factor. Portfolio Manager. WP-035 Determining the Maximum Number of Uncorrelated Strategies in a Global Portfolio Ling-Ni Boon. When bringing all these assets together. providing evidence to support their diversification potential.

While there are idiosyncratic factors influencing return variation. Chamberlain and Rothschild (1983) extend Arbitrage Pricing Theory using the factor model. Forni et al. There are (i) macroeconomic factors (observable economic or financial time series). or in policy analyses. (ii) fundamental factors (observable asset characteristics). In macroeconomics. (2001) analysed the factor structure of equity returns. mainly in macroeconomics and asset pricing. (2003) study the structure of the macroeconomy. An example of a well-known single factor model is the Capital Asset Pricing Model 242 Amundi Investment Strategy Collected Research Papers . among others. For example. Factor models have been widely studied. More recently. or numerous asset classes. the drivers of return variation may appear elusive. 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. prediction. they are used to determine the factors that influence measures of the economy. and to construct portfolios with desired characteristics. (2005) introduced the FAVAR model to analyse monetary policy.a topic of interest to investment managers. Uncovering and decomposing the importance of these common drivers help in cross-asset strategy building in portfolio management . or (iii) statistical factors (unobservable asset characteristics). there a r e a ls o common factors that account for the portfolio’s collective variation. (2005) compare static and dynamic principal components in estimating macroeconomic variables. Bernanke et al. but also to describe the return’s covariance structure. Factor models are categorized by type of factors.1. which has since been used not only to decompose risk and return into explicable and inexplicable components. In consumer demand theory. Introduction In a portfolio comprised of different assets from the same class. Merville et al. while Favero et al. Lewbel (1991) applied factor models to budget share data to reveal information about the demand system. In finance.

so tracking them yields only a tenth of the insights o b t a i n e d b y tracking investor sentiment. Alessi et al. BN’s criterion often does not converge. For instance.(CAPM). In this paper. (2007). Conversely. hence i t deserves heightened attention. we find that the risk appetite factor in the Global Macro Hedge Fund accounts for up to 45% of the data’s variance. and established conditions ensuring the consistency of their methods. (2009) (ABC) refine BN’s criterion and demonstrate that their criterion has superior performance. Asian market movements explain only about 4% of variances. which describes the relationship between risk and return. investors would be better able to evaluate t h e risk-adjusted performance of portfolios allocated to various kinds of assets. IC is based on the idea that an (r + 1)-factor model has to fit at least as well as an r-factor model. Numerous methods to determine the number of factors in the case of unobserved factors have been proposed. hence they fail to consistently estimate the number of factors when the factors are unobserved. The well. as demonstrated in Forni et al. Alternate approaches include analysing the factor loadings (Connor and Amundi Investment Strategy Collected Research Papers 243 . Arguably the most popular is via information criteria (IC).known Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) cannot be directly adopted as they are functions of N or T alone. The operational value of exposing common factors that drive variances in these markets is immense. but is less efficient. Bai and Ng (2002) (BN) propose a set of six penalty functions to replace the ones in AIC and BIC. and the factors are computed using Principal Component Analysis. In addition to theoretical interest regarding the suitability of methodologies on empirical data. and to build cross-asset strategies. the focus is on statistical factor models. Despite its wide empirical adoption. 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.

94% for USTB. two for Commodity Prices. USCS’s sole factor corresponds to mid-range risky assets while that for currencies is labelled as the carry factor. in theory. The stability of the number of factors over time is investigated by testing the significance of correlation between factors.Korajczyk. The pair of factors for commodity prices is linked to energy and metal. Kapetanios (2005)). numerous tests on the rank of the covariance matrix (Lewbel (1991). Even though by definition of principal components. Onatski (2009b). commodities. ABC’s is deemed to be the overall best in terms of accuracy and precision. and one each for US Credit Spreads (USCS) and currencies. tests on the eigenvalues of the covariance matrix of returns (Onatski (2009). the estimated factors may not be so. as it is 74% for GMHF. T h e a pplication of ABC’s criterion to five datasets yields the following number of factors: five for Global Macro Hedge Fund (GMHF). Forni and Reichlin (1998)). 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 factors are. By comparison of the criteria in a Monte Carlo study. 1993). t he US dollar. 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. slope and curvature. fewer factors are required to explain variances in the data due to increased cross-market 244 Amundi Investment Strategy Collected Research Papers . three for US Treasury Bond Rates (USTB). 49% for USCS. 1997). orthogonal to each other. T h e t otal variation explained by the factors varies. The five factors for GMHF are associated with risk appetite. 27% for commodity prices. Those for USTB fit the description of factors found in previous research and are labelled as level. and a graphical method that is rarely used due to its lack of theoretical basis (Donald. the Japanese market and Asian stock markets. and 59% for currencies.

When all asset classes are considered together in the GMHF dataset. The results are interpreted a n d t h e i r stability over time is analysed. After a Monte Carlo Study for selected methods.correlation. Yet for commodity prices. We first present the factor model and briefly explain the numerous criteria proposed to determine the number of factors. For traditional asset classes.. . it is less clear. This paper is organized as follows.e. and suggest that cross-asset correlation may be a leading indicator of economic cycles. but the largest eigenvalue of the idiosyncratic Amundi Investment Strategy Collected Research Papers 245 . 2. the different factor interactions that are observed within each asset class manifest in a more complex manner. it is evident that correlation between factors increased.. F t is the r × 1 vector of common factors. Ȝ i is the r × 1 vector of factor loadings. Methodology 2. we apply the methods to datasets relevant to the investment management industry.T (1) whereby Xit is the observed data for the ith cross section at time t. Ȝt Ft is the common component of Xit .1. as a spike in the correlation is observed prior to the 2007 to 2010 financial crisis. and t denotes the complex conjugate transpose of the matrix. Presentation of the Factor Model An r-factor approximate factor model is specified as: Xit = Ȝt Fit + eit i = 1 . eit is the idiosyncratic component. N and t = 1. . Ft and et are assumed to be uncorrelated and the matrix comprised of cov(ei. allows non- pervasiveness in asset-class-uncertainty). ej) is not necessarily diagonal (i. such as USTB.

or log(V (r. Hence. F r ) + rcg(N. ȜN). while the generation of the random subsamples is described in Alessi et al. (2009) (ABC) Alessi et al. or log(V (r. F r ))+rı̄ 2 g(N. with whereby F is the matrix of r factors. 2.component’s covariance matrix is bounded to limit the degree of correlation. and is a consistent estimate of . Two examples of g(N. Alessi et al. 2008). Furthermore. Detailed explanations of the role of c are provided in Hallin and Liska (2007). can be replaced by . T ) is the penalty for over-fitting. F r )+rg(N. (2009) propose a refinement of BN that multiplies a constant. V (r. c. Xt is explained by both common components and specific factors. . T ). ȁ = (Ȝ1 . The number of estimated factors remains as the one yielding the lowest value for these modified loss functions. and neighboring values of c. T ). and which has been shown to possess good properties when errors are cross-correlated (Bai and Ng. T ) are ) which is frequently used in empirical works. The tests considered in this paper are: 1. . the authors suggest evaluating the loss functions over random subsamples of the data to find an estimate that is insensitive to the sample size. In practice. to the penalty function as follows. F r )) + rcı̄ 2 g(N. and it is not any more 246 Amundi Investment Strategy Collected Research Papers . T ). g(N. T ). (2009). This criterion has been shown to provide a solution when BN’s criterion fails. r is a constant. with as the maximum number of factors considered. 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.

defined as . are computed. ¨N īí . . with as the OLS estimated residuals. 3.pointed out in Bai and Ng (2002) to be undesirable . eiT ). Connor and Korajczyk (1993) (CK) An alternate approach developed by Connor and Korajczyk (1993) (CK) is based on the idea that an r factor model’s (r + 1)st factor can have nontrivial factor loadings for some assets.and that ei = (ei1 . a t-test is carried out on the estimates . in essence. Then the even month’s for the regression with r + 1 factors is subtracted from the odd month’s μ for the r factor model. Onatski (2009b) developed a statistical test for Ho : r = r0 versus Amundi Investment Strategy Collected Research Papers 247 . 4. . and bounded (r + 1)st eigenvalue. 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. A cross- sectional mean for both ı̂it ’s. In order to establish the distribution of the idiosyncratic components. giving a value ¨ˆ N . . complex in implementation because it requires. . with ī as the 1 covariance matrix of ¨ is asymptotically standard normal as n ڀŃۻ. i = 1. but only a small proportion of them. . It proceeds by running two regressions by Ordinary Least Squares (OLS). for both regression models. The adjusted squared residuals. one with factors while the other with r + 1 factors. the authors made the assumption of homoscedasticity across time periods . ei2 . hence in practice. ۻŃis a mixing process. is calculated next. . . A statistical test for this is developed to test whether the (r + 1)st factor is pervasive. Under the null hypothesis that the model has r factors. multiple repetitions of BN.

which are determined by prior knowledge. Critical values for the test are provided in Onatski (2009b) for up to r = 18 factors. imposing a valid distribution requires fairly strong assumptions such as having idiosyncratic components that follow a Gaussian distribution. Beginning from r0 . I n El Karoui (2006). the Discrete Fourier Transforms (DFTs). H1 : r0 < r 助Ń r1 . with r as the number of factors. 2. Similar to the case of CK. we perform a Monte Carlo test to evaluate their relative performance on simulated data with various qualities. 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. The test statistic is whereby Ȗi is the largest eigenvalue of the covariance matrix of X̂t. X̂t is shown to be asymptotically distributed as Tracy- Widom. Monte Carlo Study Before choosing one method over another.2. r1 and r0 are the upper and lower bounds for the number of factors. for each successive r. (2009). This is similar to the DGPs used in Alessi et al. 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 Ȝij are normally distributed with zero mean and unit variance. 248 Amundi Investment Strategy Collected Research Papers . but fewer r1 factors. are computed at pre-specified frequencies Ȧj .

t h e com mon component has a smaller variance than the idiosyncratic component: Amundi Investment Strategy Collected Research Papers 249 . same variance for the common and idiosyn cratic compone nt: 6. 5. 1) and r = 2ș 4. Seri al corr elation across idiosy ncrat ic parts. 1) and . same variance for the common and idiosyncratic component: 3. same variance for the common an diosyncratic component: eit ƦŃN (0. Homoskedastic idiosyncratic component. Small cross-section correlation across idiosyncratic parts. the common component has a larger variance than the idiosyncratic component: eit ƦŃN (0. Homoscedastic idiosyncratic component. Heteroskedastic idiosyncratic component. Homoskedastic idiosyncratic component. 1) and r = ș 2. the common component has a smaller variance than the idiosyncratic component: eit ƦŃN (0.1.

CK and Onastki’s test always began with the lower bound of 1. The corresponding rmax for BN and ABC. the parameters to determine the random subsamples a r e : nJ = (see Alessi et al. The correlation between the common and idiosyncratic components is ȡ = 0. correctly identifying the number of factors. the common component has a larger variance than the idiosyncratic component. 500 Monte Carlo replications are performed for each instance. 120). it is unlikely to have prior knowledge beyond the belief that there should be at least one factor. such as in DGP 5 to 7. and the upper bound on CK and Onatski test is rmax = 8 when r = 1.e. and cmax = 13 with step size 0. the criterion slightly overestimates the number of factors. Serial and small cross-section correlation across idiosyncratic parts. (2009)). (100. N and T ) on the results. 10. ȕ = 0.5. (150. 3. rmax = 20 when r = 15. r is chosen to be 1. 500)Ơ. T Ơ. 3. rmax = 15 when r = 8.2. 8. given that the data does indeed have a factor structure. Additionally. T ) = ƞ(70. we test for the pairs of time and cross-section dimension (N. 5. The true number of factors. which is consistent with BN’s claim that their criterion yields precise 250 Amundi Investment Strategy Collected Research Papers . 70). all of which are consistent with the requirement r < minƞN. However. and 15.7. while H = max .01. There is no obvious effect of dimensions (i. BN’s criterion has perfect performance for DGPs 1-4. to test ABC’s criterion. to suggest that in many financial datasets. when the DGP demonstrates cross-section or serial correlation across the idiosyncratic components. 10. H ș For all seven DGPs. 5.

In general. when the time dimension is small. the number of estimated factors is 1 more than 80% of the time. CK’s test has a similar tendency of underestimating the true number of factors as 1 close to or exceeding 50% of the time when r 努Ń5. While ABC does not display perfect performance. b e c a u s e it is generally plagued by mild overestimation. In cases Amundi Investment Strategy Collected Research Papers 251 . regardless of the actual number of factors. as it estimates that the true number of factors is 1 close to 60% of the time. i. the test performs reasonably well for all DGPs except for DGP 2 and 7. t = 500.estimates for minƞN. but also in terms of its ease of implementation2 . can be set since it must be fewer than minƞN. and methods such as ABC are developed to be less sensitive to the upper bound. 2 about 30%. T ƠŃ> 40. and overestimating the factor by 1 for DGP 5. the adoption of ABC is justified since most financial portfolio time series demonstrate cross-section and serial correlation. its performance is more accurate for DGP 5. the Monte Carlo study substantiates BN’s criterion as superior not only in terms of accurate estimates. 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. and 3 about 10% of the time. time and cross-section dimensions. like BN d o e s for DGP 1-4.e. no prior knowledge is available to determine the lower bound. T Ơ. However. for the sample with N = 150. correctly identifying the number of factors at least 50% of the time for DGP 1-4 and 6. however. Even though BN has stellar performance in most cases. This result is similar to ABC’s own Monte Carlo study. An upper bound. hence a larger upper bound can always be selected. being insensitive to the true number of factors. Onatski’s criterion performance pales in comparison with the other criteria in almost all cases. In the case of DGP 2. large cross-section and time dimensions.

3-5 years). from January 1997 to March 2012. AA. financial rating (e. either due to cross-sectional or serial dependence. dated between January 1999 and March 2012. CK’s test may perform well when the time and cross-section dimensions are large. BBB). AAA. T h e d aily closing prices of commodities such as gold. the stability of factors over time is analysed by testing the significance of correlation between them. credit spreads. currency exchange rate and oil futures. currencies and commodities. currency exchange rates. The US Credit Spreads (USCS) data is comprised of daily closing rates categorized by industry (e. and duration (e. corn. so the difficulty i n r e c o n c i l i n g market closing times worldwide is mitigated. 1-3 years. Global Results The Global Macro Hedge Fund (GMHF) dataset is comprised of major indices. The three best criteria: BN. ABC’s criterion is applied to five datasets covering major asset classes such as equities. and energy). then ABC’s criterion should be executed. 3.1. 252 Amundi Investment Strategy Collected Research Papers .where BN’s criterion does not perform well. ABC and CK are implemented on five sets of data comprising equities. interest rates. and re-estimating the number of factors after splitting each dataset along the time dimension according to the economic cycle. 3.g. natural gas. aluminium. The estimated number of factors are first identified and labelled. financial corporation. Empirical Results In this section. or difficulty in estimating rmax .g. from January 1997 to March 2012. insurance. and currencies between 1997 and 2012.g. Then. government bonds. The data is of weekly frequency due to its cross-boundary nature. 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. commodities. credit spreads. US Treasury Bonds.

Using the AIC criterion to determine the number of lags to include in the VAR model. BN.rmax is always estimated. and GMHF and currencies with VAR(1). commodity prices with VAR(2). CK’s criterion yields slightly different estimates. ABC and CK criteria to the datasets. USCS. CK’s estimates do not always coincide with ABC’s estimates. and it is invariant to whether linear dependencies exist in the data. Commodity Prices and Currencies datasets. Amundi Investment Strategy Collected Research Papers 253 . New Zealand Dollars. The fifth dataset is called Currencies. 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. Norwegian Krone. Japanese Yen. the number of factors estimated and the proportion of variance explained by each factor. ABC and CK criteria are applied to log prices or log spread variations for GMHF. Despite its commendable performance in the Monte Carlo study. BN and ABC criteria provide the same outcome when the analysis is done on the residuals as on the returns data. but the latter is taken to be more accurate due to better performance in the Monte Carlo study. Australian Dollars. the S&P GSCI (Goldman Sachs Commodity Index) etc. Canadian Dollars. rate variation is used since with two-year rates close to zero or negative.wheat. their percentage variation are extreme. whereas for USTB. In Exhibit 6. from October 1998 to March 2012 are compiled in the dataset of Commodities. Swiss Franc. BN’s criterion fails to converge on all datasets . and the Swedish Krona in terms of US Dollars from January 1999 to December 2012. USTB and USCS datasets are fitted with VAR(3). Great Britain pound. The composition of each dataset is presented in Exhibit 1. and includes the daily prices of the Euro. then applying the same criteria on the residuals. since the Monte Carlo study indicates that the selected criteria generally have poorer performance when cross- section correlation exists. Aside from applying BN.

i. factor loadings. hypothetical factors according to their relationship with the assets. Next.e. adding those with the largest absolute correlation first. that is. as it is merely the sum of the corresponding asset in the portfolio. Moreover.the lowest among the datasets considered. accounting for around 49% and 59% of the variances respectively. and progressively added to the selection. Commodities’ low concentration of correlation is consistent with the findings of Gorton and Rouwenhorst (2004) on the asset class’s diversification potential. is investigated and presented in bar charts. Using ABC’s criterion. USTB has three factors that explain 94% of the variances. An asset class or portfolio for which correlations across asset variations are high should have a high percentage explained.e. until the selection achieves at least 95% R2 when used as explanatory variables in a simple regression model for the asset returns. which collectively explain about 74% of variances in the dataset. loadings treated as weights) is zero. are presented. The commodity prices dataset is estimated to have two factors. After obtaining the number of factors. portfolios that are insensitive to a particular factor can be constructed by selecting assets such that their weighted sum (i. This notion of sensitivity can be extended to that of a portfolio. Factor loadings also help in identifying the factor. i. These 254 Amundi Investment Strategy Collected Research Papers .e. the sensitivity of each asset’s return to the factor.e. The proportion of variances explained by the estimated number of factors suggests the concentration of correlation. 2011). the GMHF is estimated to have five factors. and becomes the rationale for investors to increase portfolio allocation to commodity assets (Daskalaki and Skiadopoulos. o n e factor each is estimated for US credit spreads and currencies. by labelling the latent. i. Absolute correlation of the factors with the assets’ returns is first placed in descending order. ratio of the ith eigenvalue and the sum of all eigenvalues of the covariance matrix of the returns. Together they explain 27% of the variances .

a n d hence impact. The slope factor implies larger shocks for bonds with s h o r t maturities. Conversely. investors are more willing to take risks. During bullish periods. Curvature affects medium-term interest rate. as compared to bonds with longer maturities. A level shock shifts the curve in a parallel manner. investments are diverted into bonds. Factor 2 relates to oil and GSCI. Amundi Investment Strategy Collected Research Papers 255 . hence the y prefer to invest in equities. and has a larger correlation. on bonds with shorter maturities. Factor 4 represents the Japanese market. hence it is labelled a commodities factor. whereas factor 5 is linked to Asian markets. slope. just as a level factor would. as should a slope factor. A s ummary of the results is provided in the final table in Exhibit 1. which have lower risk. These names describe the shift of the yield curve in response to a shock. and curvature. This suggests that factor 1 corresponds to a risk appetite factor. Exhibit 3 shows that factor 1 has relatively uniform correlation across bonds of all maturity. T h e c orrelation for factor 2 changes sign once. Thus. Factor 3 has a hump in its correlation figure. the findings are consistent with existing results. during bearish periods. hence i t s h o w s as a “hump” on the yield curve. Global Macro Hedge Fund Exhibit 2 shows that factor 1 of GMHF is highly correlated to major equity indices worldwide. Its name is derived from the effect of the yield curve becoming less steep as a result of a slope shock. Indeed. fitting the description of a curvature factor.correlations are presented as bar plots in Exhibits 2 to 5. 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. resulting in an almost equal effect on bonds of all maturity. Factor 3 is a dollar factor due to its association with the US dollar.

e. Similar to the case of USTB. such as AAA-rated assets and treasury bonds. Daskalaki et al. as shown in Exhibit 3. assets in USCS have a relatively more uniform correlation (i. such as consumer cyclicals (e. Similarly. along with those that are highly dependent on the state of the economy. industries responsible for providing goods with low substitutability. Factor 2. are not among those that possess the highest correlation with the factor.). being correlated with numerous metals. since among all datasets. automotive etc. Our results are consistent with those of Daskalaki et al. healthcare and energy. suggesting that they play a relatively smaller role in the movement of returns.g. commodity price factors collectively explain the least amount of variance in the data. such as utilities. as is evident in Exhibit 4. macro and equity-motivated factor models. Investments with low credit risks and conventionally small credit spreads. and principal component factor models to find that none of them satisfactorily prices commodity assets. implying that it is associated with assets with mid to low credit risk. an observation that supports the commodity diversification effect. (2012) investigate the common components of a cross-section of commodity futures data using numerous asset pricing models intended for equities. possess lower correlation with the factor. financial corporates and industrial credit spreads that dominate those with utilities and consumer cyclicals. entertainment. Commodity Prices Interpretation for the two factors of commodity prices is straightforward: factor 1 is the energy factor. and segmentation of equity and commodity markets.US Credit Spreads The sole factor for USCS is correlated to A-rated investments. made 256 Amundi Investment Strategy Collected Research Papers . The authors attribute this poor pricing model performance to heterogeneity in commodity markets. all above 50%) with the factor. In comparison with other datasets. is the metals factor. the number of factors in commodity datasets is commonly studied.

Stability Analysis Since all datasets span ten years. as shown in Exhibit 5. such as the Australian Dollar. These factors are estimated and identified by analysing their correlation with returns. 3. Currencies For currency datasets. It is labelled as the carry factor as it is correlated to currencies with high average rates. Attempts to devise formal tests include those by Audrino et al. including the most recent financial crisis in 2008.popular by Gorton and Rouwenhorst (2004). ABC estimates a single factor. It could also be understood as the dollar factor as all currency prices are positively correlated with the US dollar. Thus. Amundi Investment Strategy Collected Research Papers 257 . His hypothesis is that if the factor loadings appear to be similar a c r o s s all sub periods. Bliss (1997) divided his sample into three sub-periods and investigated the factor loadings on each. but factor volatility varies over time. the number of salient factors is determined. the identification of factors by commodity group and the low correlation across these groups suggest a sectorial framework when studying the commodity market.2. then the factor structure is stable. Furthermore. Most authors have either assumed stability or relied on some graphical method. it is of interest to know whether the number of estimated factors is stable over time. the authors conclude that the factor structure is stable. there is no empirical evidence in the literature regarding the number of factors in the currency datasets. Norwegian Krone and Swedish Krona. 2011). Literature on t h e stability of factor models is sparse. Since the PCA results are similar in both samples. To the best of our knowledge. by applying ABC’s criterion to the datasets. Chantziara and Skiadopoulos (2008) analysed the term structure of petroleum futures. Pérignon and Villa (2002) found that factor loadings are stable. also by splitting the sample into two. and which motivated investors to increase their portfolio allocation in this asset class (Daskalaki and Skiadopoulos.

To investigate the stability of the factors on our dataset. the t-test for significance of correlation is used. from -1 down to -12. over time. whereas the latter involves constructing a bootstrap distribution for the test statistic. but the correlation between the factors may not be.2. In the latter. 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.1. a n d specifically t h e i r correlation. the dataset is split into expansion and contraction economic periods. (2007). To reveal the evolution of the relationship between factors. In other words. 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. The former relies on testing t h e equivalence of the factor loadings on sub-periods. Correlation between the factors is tested over six-month rolling windows for GMHF. and the focus is on their dynamics. who focused on inte re st rate term structure. 3.(2005). The key difference between the two approaches is that by assuming the factor structure is stable in the former. Correlation between factors when the factor structure is stable is closely related to the concentration of eigenvalues. Significance of Correlation Between Factors In this section. the estimate of the number of factors is done only once. To take into consideration t h e f a c t that financial market performance is a leading indicator of macroeconomic situation. The first assumes that the factor structure is stable. and Philip et al. and the corresponding number of estimated factors in each sub-period is obtained. and one-year rolling windows for 258 Amundi Investment Strategy Collected Research Papers . Next. discarding the assumption of a stable factor structure. the periods of contraction and expansion are lagged by a negative number of months. the number of factors is estimated for each variation in the economic cycle using ABC’s criterion. two approaches are attempted. the factor structure estimated in the previous section is assumed to be stable but the dynamics between factors evolve over time.

Moreover.US Treasury Bond rates as well as commodity prices. factor 1 affects bonds of all maturity evenly. energy and metals are essential to Amundi Investment Strategy Collected Research Papers 259 . 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. factor 2 has a strong impact on short maturity bonds while factor 3 has the highest influence on bonds of mid-term maturity. Indeed. i.e. the correlation between the factors is insignificant. This is in line with the result in Kat and Oomen (2006) that there is weak correlation across commodity groups. are distinct commodity groups. i. Hence. energy and metals.e. Since the factors were identified by their impact on bonds of different maturity. the level. using the t-test for significance of correlation. t ƦŃStudent with degree of freedom equalling N ǙŃ2. the two factors identified. 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. Exhibit 7 shows a marked drop in the number of uncorrelated factors to only one factor during the most recent financial crisis. 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. Plots of the number of uncorrelated factors suggest that the factors tend to be correlated during recession periods. This is attributable to historically low interest rates as investors sought safe investments. slope and curvature factors merged into a single factor. Increased correlation between factors is particularly obvious for USTB.

Since it is also fewer than five in many other instances. which could have given rise to an interim factor influencing the returns. followed by fluctuations in correlation. an observation that motivated the analysis in the next sub-section. it is less clear whether higher correlation between factors is a unique feature of recessions. With as many as five factors.many industries and are not substitutable. For the 5% test on GMHF. Hence. 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. To further investigate this. As for currencies.during recession. stability analysis is not applicable. 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. when occasionally two factors are estimated. The stability of the factors is dependent on the asset class.the number estimated using ABC’s criterion over the entire horizon . Exhibit 8 demonstrates that the number of factors r e m a ins constant at one for most of the time period. the dynamics between factors do evolve over time. Since USCS and currencies have only one factor. prior to the most recent financial crisis. and apply the same t-tests to investigate the evolution of the number of uncorrelated factors over time. Thus. the number of factors that 260 Amundi Investment Strategy Collected Research Papers . This period coincides with the introduction of the Euro. It could also be argued that the number of factors reduced before recession. Out of curiosity. they represent immutable drivers of return among commodity assets. For USTB. except in early 1999. the mean absolute correlation between factors is plotted in Exhibit 9. on which there is an indisputable spike in correlation in 2007. the correlation between factors rises substantially. the number of uncorrelated factors is never five . the sectorial view of the commodity market remains valid throughout ups and downs in the economy. we over-estimate the number of factors at five each. which is considered to be the lowest risk among those considered.

Commodity price factors seem invariant to the economic climate. coherent with the t-test for significance of correlation results. To investigate this. the number of factors is stable. On the contrary. After reviewing the literature on existing approaches to determine the number of factors. 4. it is possible that the change in the number of factors occurs prior to contractions in the economy. these interactions become more complex and are realized as fluctuating correlations between factors. GMHF and US Credit Spreads have a constant number of factors throughout expansion periods.affects its return is lower during recessions.2. The Monte Carlo analysis suggests that the criterion by Alessi et al. T h e Amundi Investment Strategy Collected Research Papers 261 . For USTB. (2009) is most reliable. For example. The results are presented in Exhibit 10. are lagged by a negative number of months. Estimated Factors by Economic Cycles As financial markets are often thought of as leading indicators of economic cycles. As for commodities. Conclusion This paper investigates the number of cross-asset uncorrelated strategies that are available to portfolio managers. When combined in the GMHF portfolio. and to a lesser extent by currencies. the view of financial market performance as a leading indicator of economic cycles is supported by USTB rates only. Therefore. “Lag = -1 month” of the most recent financial crisis between December 2007 and June 2009 refers to the interval November 2007 to May 2009.2. especially for that which occurred between late 2007 and 2010. four methods are selected and tested. 3. the number of factors fluctuates between one and five during expansion periods. The number of factors in the currencies dataset falls to zero prior to contraction periods. the contraction periods. as determined by NBER. supporting the view that the correlation between factors changes prior to recessions.

The evidence is strongest for US Treasury Bond rates.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). aligned with the observation of low correlation between commodity groups claimed in previous studies. which is most likely due to US post-financial crisis macroeconomic policies. investors would have a better understanding of the common sources of risk. With more information on the factors t h a t drive the returns o f different asset classes. US Treasury Bond Rates (3). 262 Amundi Investment Strategy Collected Research Papers . Thus. However. ba r ring credit spreads. a strategy that is built upon exploiting these common sources of risk may have to take the economic climate into account. the results regarding stability have to be considered with caution. and Currencies (1). Plots of the number of uncorrelated factors do not all support the hypothesis of increased cross-market correlation during economic recession. yielding fluctuating correlation between the factors during economic downturns. demonstrate a combination of the observed outcomes on the rest of the datasets. Depending on the asset class in mind. Commodity Prices (2). as not every financial crisis is followed by a recession period. but i s w e a k e s t for commodity prices. US Credit Spreads (1). GMHF. there could be a change in correlation between assets occurring outside the NBER-determined recession periods. comprised of a mix of these assets.

AUD futures. Eurostoxx. SMI. EUR- JPY. 3-5 years. CHF-JPY. Range: January 1997 to March 2012 List of Assets in the US Credit Spreads Dataset Master index. 1-3 years. consumer cyclicals. USD-SGD. FTSE. AUD. JPY. soybean. aluminum. OZ 10Y. GE 2Y. A. Others (1 asset) Mini SP 500. (ABC) Global Macro 12 (5) 5 US Treasury 4 (5) 3 US Credit Spreads rmax is always estimated. Range: October 1998 to March 2012 List of Assets in the Commodity Prices Dataset Gold. industrials. GE 5Y. Commodities (2 assets) Oil. USD-BRL. 7-10 years. MSCI Singapore Free Index (Singapore). GE 10Y. S&P GSCIs: agriculture. rice. WTI. US 5Y. precious metals. consumer non-cyclicals. capital goods. AAA. JPY futures. 30Y. copper. corn. IBEX. Nasdaq 100. USD. USD-KRW. AEX. healthcare. banks. 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. Range: January 1997 to March 2012 List of Assets in t h e US Treasury Bond Rates Dataset (Maturity) 3M. BBB. EUR- Currencies (29 assets) SEK. GBP futures. coffee. USD-CNY. insurance. EUR-NOK. AA. Bonds (12 assets) US 30Y. 6M. heating oil. CAN 10Y. 8Y. cotton. GBP-JPY. Summary Table of Results Number of Factors Estimated by Method Criterion Connor and Bai and Ng (BN) Korajczyk (CK)* essi et al. EUR-CAD. EUR-AUD. CAC 40. 1Y. Brent. Amundi Investment Strategy Collected Research Papers 263 . 7Y. Nikkei. CAD futures. S&P GSCI. AUD- NZD. financial corporates. wheat. energy. USD-ZAR. 9Y. JP 10Y. USD-JPY. EUR-GBP. 20Y. NOK-SEK. platinum. utilities. USD-TWD. AUD-JPY. EUR-CHF. gas-oil. TOPIX. 15Y. cocoa. ED 4 (Euro-Dollar bond). DAX. natural gas. AUD-CAD. zinc. Hang Seng. CAD. energy. EUR-USD. 5Y. USD-CAD. 10Y. NZD. Futures (5 assets) CHF futures. nickel. industrial metals. 3Y. Range: January 1999 to December 2012 List of Assets in t h e Currencies Dataset Daily price in USD of: EUR.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. 5-7 years. 2Y. USD-CHF. NOK SEK. sugar. US 2Y. CHF. 25Y. silver. US 10Y. AUD-USD. CAD-JPY. NZD-USD. GBP. 4Y. GBP-JPY. lead. EUR-PLN. UK 10Y.

10Y EURGBP Singapore ED.4 í 0.10Y USDKRW USDBRL Hang.fut EURUSD DAX CAC.SP500 GBP.fut USDCHF FTSE CHF.Seng OZ.0 0.6 0.fut Dow.100 GE.4 0.5 Correlation Correlation Factor 5 (Global Macro) Eurostoxx CAD.30Y USDTWD US.6 í0.4 CADJPY GSCI CAD.2 0.0 0.Global Macro Hedge Fund Data Factor 1 (Global Macro) Factor 2 (Global Macro) Eurostoxx SMI EURJPY Nikkei CHFJPY SMI USDCAD Hang.Exhibit 2: Factor Identification .8 í0.1 AUDUSD AUD.Jones AUD.4 Correlation Correlation Factor 3 (Global Macro) Factor 4 (Global Macro) CAC.fut Mini.0 0.8 í í í 0.fut USD CHF.2 Correlation 264 Amundi Investment Strategy Collected Research Papers .10Y US.SP500 USDCAD IBEX EURGBP CHFJPY EURJPY CHF.0 0.Seng Singapore Nikkei Topix 16 í0.0 0.2 0.Seng CAD.fut GBPJPY.5 0.fut GBPJPY GBP.2 0.40 USD GSCI Eurostoxx Oil 0.4 í 0.fut EURJPY AUDJPY USD GBPJPY.4 0.1 CADJPY USDJPY NZDUSD JPY.5Y EURAUD GE.fut EURJPY USDCHF EURUSD CHFJPY í0.5Y US.fut USDCAD USDTWD USDKRW Oil USDSGD EURCAD AUDCAD CHFJPY EURCAD EURUSD USDSGD USDCHF AUDJPY CHF.fut IBEX NZDUSD USDSGD AUDUSD AEX GBPJPY Mini.40 Nasdaq.fut USDCHF USD EURUSD Hang.

8 0. one and two.4 0. respectively.2 í0.0 0. This is satisfied by the above figures.6 0.2 0.6 0.2 0.2 0.2 0.1 0.8 Correlation Correlation A partial characterization of USTB factors is the number of sign changes.2 0.4 0. zero. Amundi Investment Strategy Collected Research Papers 265 .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 í0.4 í0.0 0.3 í0.6 Correlation Correlation Factor 1 (Credit Spreads) Factor 3 (Treasury Bond) 30Y Master Index 25 20Y A 15Y 10Y Financial Corporates 9Y 8Y Industrials 7Y 5Y í Years 4Y 3Y BBB 2Y 1Y í Years 6m 3m AA í0.0 0.1 0.4 0.0 0.

Commodity Prices Factor 1 (Commodities) Factor 2 (Commodities) GSCI.Ind.Exhibit 4: Factor Identification ..Oil Natural..Currencies Factor 1 (Currencies) SEK NOK NZD AUD CAD JPY CHF GBP EUR 0.0 0.7 í0.00.4 í í í0.Agri.Prec.1 0.Metals GSCI.6 0.6 í0.Agri.2 0. Rice Soybean Cotton Cocoa Sugar Coffee Wheat Corn Heating.8 Correlation Correlation Exhibit 5: Factor Identification .4 0.Oil GSCI.Gas Gasoil Brent WTI Lead Zinc Nickel Copper Aluminium Platinium Silver Gold WTI GSCI.Energy GSCI.4 0.Energy Heating.0 0.5 í0.2 0.6 0. Natural.Gas í0.Metals GSCI.8 Correlation 266 Amundi Investment Strategy Collected Research Papers .

469 0.022 0.014 Proportion (%) 45 13 7 4.8 74 US Treasury Bond Rates Factor 1 2 3 Label Level Slope Curvature Eigenvalue 12.676 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 0.181 Proportion (%) 15 13 Cumulative (%) 15 27 Currencies Factor 1 Label Carry factor Eigenvalue 0.8 4.2 Cumulative (%) 45 58 65 69.305 1.11 Proportion (%) 59 Amundi Investment Strategy Collected Research Papers 267 .135 0.015 0.206 0.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 0.041 0.

USTB and Commodity Prices Top to Bottom: 1% level t-test for significance of correlations for GMHF. USTB.5 3. fact. fact.Exhibit 7: Number of Uncorrelated Factors .e. March to November 2001.0 2. 4 3 2 1 2000 2002 2004 2006 2008 2010 2012 Time 1. 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.0 1. fact. and December 2007 to June 2009 (National Bureau of Economic Research (2012)). of uncorr.5 2. i. and Commodity 5 Prices No.0 No. of uncorr.GMHF. 268 Amundi Investment Strategy Collected Research Papers . 2000 2005 2010 Time 5 No. of uncorr.

of uncorr.Exhibit 8: Number of Uncorrelated Factors . of uncorr. and December 2007 to June 2009 (National Bureau of Economic Research (2012)).0 1998 2000 2002 2004 2006 2008 2010 2012 Time The shaded areas mark the intervals of contraction (peak to trough) of business cycles.0 2000 2005 2010 Time 2.4 1.0 1.5 1. fact. 2.0 1. 1.8 No.e. fact. Amundi Investment Strategy Collected Research Papers 269 .US Credit Spreads (10% level t-test. top) and Currencies No. March to November 2001. i.2 1.

35 0. by Business Cycle Expansions and Contractions. Exhibit 10: Number of Estimated Factors by Dataset.e. and December 2007 to June 2009 (National Bureau of Economic Research (2012)).15 0.Exhibit 9: Mean Absolute Correlation of Factors .20 0.Global Macro Hedge Fund 0. i. March to November 2001. 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.25 0. The horizontal dotted line indicates the median. 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 0 270 Amundi Investment Strategy Collected Research Papers . lagged by negative number of months Dataset Global Macro Hedge Fund US Treasury Bonds Rate US Credit Spreads Lag (No.10 2000 2002 2004 2006 2008 2010 2012 Time The shaded areas mark the intervals of contraction (peak to trough) of business cycles.30 Mean Absolute Correlation of Factors 0.

Should investors include commodities in their portfolios after all? N ew evidence. and Ng. and Eliasz. and Rothschild. (1997). Marcellino. G. Economic Review. Arbitrage. Technical report. Journal of Applied Econometrics. (2000). (2008). Barone-Adesi. S. F. Bai. G.augmented vector autoregressive (favar) approach. A test for the number of factors in an approximate factor model. Chamberlain. Chantziara.. S. (2008). C. R. Are there common factors in commodity futures returns. and Skiadopoulos. Econometrica. B. (Q 4):16–33. J. factor structure. and Skiadopoulos. Journal of Financial Econometrics. Daskalaki. El Karoui. Q. M... T. (2006). S. Determining the number of factors in approximate factor models. The Journal of Finance. 120(1):387– 422. A. J. 3(2):89–163. Energy Economics. Journal of Banking and Finance. (2005). Donald. Bliss. The stability of factor models of interest rates. Barigozzi. Econometrica.. 35(10):2606– 2626. Measuring the effects of monetary policy: A factor. P. LV(5):1943–1979. Inference concerning the number of factors in a multivariate nonparametric relationship. and Singleton. 70(1):191–221.. and Neglia. Can the dynamics of the term structure of petroleum futures be forecasted? E vidence from major markets. Foundations and Trends in Econometrics. (2002). 65(1):103–132. (1993).. Connor. Movements in the term structure of interest rates.. G. Boivin. 3(3):422–441. A.. and Mira. G. 48(4):1263–91. Audrino. J. and Skiadopoulos. 20(5):603–620. 35(2).. G. (2011). K. 51(5):1281–304. and mean- variance analysis on large asset markets. Specification analysis of affine term structure models. (2005). R. (2009). 30(3):962–985. F. (1997).. and Capasso. M.References Alessi. R. Journal of Finance. G. A. A robust criterion for determining the number of factors in approximate factor models. (2012). Bernanke. Daskalaki. Favero. and Ng. Bai. Kostakis. M. N. S. Amundi Investment Strategy Collected Research Papers 271 . J. The Quarterly Journal of Economics. and Korajczyk. S. Econometrica. 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. M. A. C. Large dimensional factor analysis. Annals of Probability. (1983). G. Dai. L. C. Principal components at work: the empirical analysis of monetary policy with large data sets. (2005).

pages 54–62. National Bureau of Economic Research (2012). J. and Reichlin.rica. C.. CEPR Discussion Papers 4133.. C. To appear in the Review of Economics and Statistics. M. G. National Bureau of Economic Research. (2002). K. Kat. M. Journal of Portfolio Management.Forni. Lippi. and Xu. G. Lewbel. and Rouwenhorst. and Oomen.. International Center for Financial Asset Management and Engineering. Testing hypotheses about the number of factors in large factor models. R. R. Opening the black box: Structural factor models versus structural vars. L. and Sheinkman. G. Lippi.. Let’s get real: a factor analytical approach to disaggregated business cycle dynamics. Inc. (2006).R.. and Villa. Cass Business School. Onatski. Determining the number of factors from empirical distribution of eigenvalues.. US business cycle expansions and contractions. Merville. Economic Letter ..P. and Liska. Common factors affecting bond returns. Y. Onatski. A testing procedure for determining the number of factors in approximate factor models with large datasets. 102:603–617. Technical report. What every investor should know about commodities. Technical report. Philip. A. and Reichlin. Federal Reserve Bank of San Francisco. (2004). Econometrica. The rank of demand systems: Theory and nonparametric estimation. G. NBER Working Papers 10595. ULB – Universite Libre de Bruxelles. number 2003-15. M.E. C. M. (2007). (2003). Pérignon. (1991).What Makes the Yield Curve Move?. C. Litterman. H. and Urga. 27(4):51–61. Hayes-Yelken. Kapetanios. L. (2007). C. (1991). Forni. Forni. A. D. A. ULB Institutional Repository 2013/10147. part ii: Multivariate return analysis. D. Discussion Papers. Opening the black box - structural factor models with large gross-sections. Testing for instability in factor structure of yield curves. A. J. European Central Bank. Technical Report 33. The Journal of Fixed Income. (2009b). Facts and fantasies about commodity futures. Identifying the factor structure of equity returns. S. M. (2009a). Working Paper Series 712. Hallin. 272 Amundi Investment Strategy Collected Research Papers . Journal of the American Statistical Association. Cass Business School. M. Determining the number of factors in the general dynamic factor model.. R. 77(5):1447–1479. Gorton. Kao. Giannone.. (2005). Goff. and Reichlin. L. L. editor (2003). (2007). 59(3):711–30. M. Economet. (2001). Permanent and transitory factors affecting the dynamics of the term structure of interest rates.. J. (1998).

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.

Amundi Investment Strategy Collected Research Papers 273

1. Introduction

In Markowitz’s (1952) setting, portfolio selection is driven solely by financial parameters and

the investor’s 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)

274 Amundi Investment Strategy Collected Research Papers

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 Markowitz’s (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 investor’s 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 SRI’s 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.

Amundi Investment Strategy Collected Research Papers 275

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 investor’s 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.

276 Amundi Investment Strategy Collected Research Papers

In the traditional mean-variance portfolio selection (Markowitz, 1952), the investor

maximizes her portfolio’s 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 investor’s 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 & Poor’s 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.

Amundi Investment Strategy Collected Research Papers 277

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

278 Amundi Investment Strategy Collected Research Papers

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 investor’s

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 Grauer’s (1990)

methodology. Proposition 2 summarizes the results.

Amundi Investment Strategy Collected Research Papers 279

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 . the best SR-rated portfolios are at the top. For O ! O0 . For O O0 . The SR-sensitive frontier is ! I0 L ' ¦ 1 L identical to the SR-insensitive For O ! O0 . resp. L ' ¦ 1 I The SR frontier differs totally I0 from the SR-insensitive frontier. if G 1 ! 0 . the SR constraint impacts 280 Amundi Investment Strategy Collected Research Papers . Proposition 2 makes explicit the situations in which there is an SRI cost. the SR-sensitive frontier is another hyperbola lying L ' ¦ 1 I below the SR-insensitive frontier. (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. the SR-sensitive frontier frontier is identical to the SR- insensitive frontier. resp. As showed in Proposition 1. G 1 0 . the SR-sensitive frontier is identical to the SR- insensitive frontier. of the SR-insensitive frontier: by consequence.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 . at the bottom. The impact of the constraint on the SR ratings depends on the parameter G 1 and on the strength of the constraint.

that is to say the higher the threshold I 0 . L ' ¦ 1 I In the case where G 1 ! 0 and ! I 0 (see Figure 1). There is only an SRI cost for investors whose risk aversion parameter is above the threshold O0 . 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 . 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). In addition. the efficient frontier is even not modified at all. For portfolios with lower expected returns and variances. In the case where the threshold I 0 is below the minimum rating of the SR-insensitive frontier. Amundi Investment Strategy Collected Research Papers 281 . resp. We illustrate the four possible cases through Figures 1 to 4. This is the most favourable case. the SRI constraint induces less efficient portfolios. the more the investor wants a well-rated portfolio. at the top. the SR-sensitive and the SR- L ' ¦ 1 L insensitive frontiers are the same and there is no SRI cost at all.first the efficient frontier at the bottom. the bigger the portion of the efficient frontier being displaced.

For portfolios with higher expected returns and variances. the SR-sensitive and the SR- L ' ¦ 1 L insensitive frontiers are the same below the corner portfolio ( E 0 . L ' ¦ 1 I Figure 3 SR-sensitive frontier versus SR-insensitive frontier with G 1 0 and ! I0 L ' ¦ 1 L E0 Mean ı0 Standard Deviation SR-insensitive frontier SR-sensitive frontier 282 Amundi Investment Strategy Collected Research Papers . There is only an SRI cost for investors whose risk aversion parameter is below the threshold O0 . the SRI constraint induces less efficient portfolios. V0 ) . L ' ¦ 1 I Figure 2 SR-sensitive frontier versus SR-insensitive frontier with G 1 ! 0 and I0 L ' ¦ 1 L E0 Mean ı0 Standard Deviation SR-insensitive frontier SR-sensitive frontier L ' ¦ 1 I In the case where G 1 0 and ! I 0 (see Figure 3).

The SRI constraint induces less efficient portfolios for every investor.e.e. This is the worst case. The significance of the mean- variance efficiency loss may be assessed using the test of Basak et al. L ' ¦ 1 I Figure 4 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 To sum up. the SR-sensitive and the SR- L ' ¦ 1 L insensitive frontiers are totally different. Indeed. a cost for high-risk-aversion investors only). c) only the right portion of the efficient frontier is penalized (i. 9 This section highlights the impact of a constraint on the portfolio rating in the mean-variance optimization. b) only the left portion of the efficient frontier is penalized (i. we show in this Section that this parameter matters in the cost of responsible investing9. L ' ¦ 1 I In the case where G 1 0 and I 0 (see Figure 4). a cost for low-risk aversion investors only). Amundi Investment Strategy Collected Research Papers 283 .e. a cost for all the investors). we show that this SR cost depends on the link between SR ratings and financial returns and on the investor’s risk aversion. the cost of investing responsibly. if non-zero. no cost). and d) the full frontier is penalized (i. may be non-significant. (2002) or any spanning test (see de Roon and Nijman (2001) for a literature review). four cases being possible: a) the SR- sensitive frontier is the same as the SR-insensitive frontier (i. However. There is an SRI cost for everybody. while the investor’s risk aversion is generally left out of the story in the SRI practice.e.

3. we seek to determine the portfolio ratings I p of the optimal portfolios on the “SR-insensitive capital market line”. we refer it here as “SR-insensitive capital market line”. 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. So. we add responsible ratings to the story. It corresponds to Problem 3. Indeed. we assume the existence of a risk-free asset and we explore. Denote by r the return of the risk-free asset and by Zr the fraction of wealth invested in this risk-free asset. 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 social responsibility of this risk-free asset should also be taken into account. Henceforth. 284 Amundi Investment Strategy Collected Research Papers . we assess first the social responsibility of the optimal portfolios obtained by an SR-insensitive investor. the impact of considering responsible ratings in the mean-variance portfolio selection. And then we study whether an SR-sensitive investor is penalized by requiring portfolios with high SR standards. As in Section 2. The standard mean-variance portfolio selection in the presence of a risk-free asset has been extensively studied by Lintner (1965) and Sharpe (1965). in this case. 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 * . Portfolio selection with a risk-free asset In this section. In the following.

(6) has the same form as eq.Proposition 3 (i) Along the SR-insensitive capital market line. 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 * . the riskier the optimal Amundi Investment Strategy Collected Research Papers 285 . 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 . along the capital market line. In the same way as in Section 2. If G 1* ! 0 . (6). Proof: see Appendix 3 Proposition 3 attributes an SR rating of any portfolio of the SR-insensitive capital market line. G 1* 0 . Here. (3) in the case without a risk-free asset. the sign of the parameter G 1* is crucial because it represents where the trade- off appears between risk aversion and SR rating. 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. resp. From the optimality conditions comes the fact that. 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. It is striking that this relationship expressed by eq. the direction of this link is determined by the sign of the parameter G 1* .

the worse. resp. the better.portfolio. the constraints in the mean-variance optimization are also linear. the best SR-rated portfolios are at the top. Similarly to Section 2. 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. G 1* 0 . 286 Amundi Investment Strategy Collected Research Papers . Thus. as we did for Problem 2. resp. of the SR-insensitive capital market line. Proposition 4 summarizes the results. we employ Best and Grauer’s (1990) methodology. resp. we now consider the case of SR investors wishing high SR standards and so. In other words. This corresponds to Problem 4. its SR rating. if G 1* ! 0 . requiring the portfolio rating I p Z 'I Z r I * to be above a threshold I0 . at the bottom.

the SR-sensitive capital market line is a hyperbola lying below the SR-insensitive capital market line. The SR-sensitive capital market line is the same as the SR- For O ! O*0 . For O O*0 . line differs totally from the SR- insensitive capital market line and becomes a hyperbola. As in Proposition 2. ( 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 Proof: see Appendix 4 Proposition 4 makes explicit the situations in which there is an SRI cost in the presence of a risk-free asset.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 . For O ! O*0 . the SR-sensitive capital market line is a hyperbola lying below the SR-insensitive I* ! I0 capital market line. the SR-sensitive capital market line is identical to the SR-insensitive capital market The SR-sensitive capital market I* I0 line. the impact of the constraint on the SR ratings depends on Amundi Investment Strategy Collected Research Papers 287 . the SR-sensitive insensitive capital market line capital market line is identical to the SR-insensitive capital market line.

the capital market line becomes a hyperbola in the mean-standard deviation plan. 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). Below this portfolio. the SR constraint impacts first the capital market line at the bottom. the modified part of the capital market line has a different mathematical form: for this segment. This is the best case. resp. the best SR-rated portfolios are at the top. contrary to the case without a risk-free asset. resp. There is an SRI cost only for investors with cold feet. at the top. at the bottom. 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* . that is to say with a risk aversion parameter above the threshold O*0 . Figures 5 to 8 illustrate the four cases. 288 Amundi Investment Strategy Collected Research Papers . As showed in Proposition 3. if G 1* ! 0 . the SR- sensitive capital market line becomes a hyperbola. G 1* 0 . the SR-sensitive and the SR-insensitive capital market lines are the same and there is no SRI cost at all. In the case where G 1* ! 0 and I * ! I0 (see Figure 5). However.the parameter G 1* and on the strength of the constraint. resp. of the SR- insensitive capital market line: by consequence.

There is an SRI cost only for investors with a risk aversion parameter below the threshold O*0 . Amundi Investment Strategy Collected Research Papers 289 . Figure 6 SR-sensitive capital market line versus SR-insensitive capital market line with G 1* ! 0 and I * I0 Mean E0* ı0* 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 * ! I 0 (see Figure 7). The SR-sensitive capital market line is no longer a line but a hyperbola. the SR-sensitive capital market line becomes a hyperbola. Figure 7 SR-sensitive capital market line versus SR-insensitive capital market line with G 1* 0 and I * ! I 0 Mean E0* ı0* 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 8). This is the most disadvantageous case: there is an SRI cost for all the investors. V0* ) . the SR-sensitive and the SR-insensitive capital market lines are the same for portfolios above the corner portfolio ( E 0* . the SR-sensitive and the SR-insensitive capital market lines differ entirely. Above this portfolio.

e. Figure 8 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 mean-variance portfolio selection in the presence of a risk-free asset also. 4. We consider the EMBI+ indices from JP Morgan as proxy for emerging bond returns. Brazil. Philippines. 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. These indices track total returns for actively traded external debt instruments in emerging markets. the investor’s risk aversion matters in the SRI cost. a cost for all the investors). Bulgaria. Russia. no cost).10 The indices are expressed in US dollars and taken at a monthly frequency from January 1994 10 Argentina. Venezuela. a cost for low- risk aversion investors only). Panama. Peru. a cost for high-risk-aversion investors only). and d) the full capital market line is penalized (i. 290 Amundi Investment Strategy Collected Research Papers . 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. In order to be ready-to-use for practitioners.e.e.e. c) only the right portion is penalized (i. b) only the left portion is penalized (i. Ecuador. Mexico.

(2).94 RUSSIA 83. we use the Environmental Performance Index (EPI) as responsible ratings. the portfolio EPI is defined in the same way as in eq. These objectives are gauged using 25 performance indicators tracked in six well-established policy categories.00 Gˆ1 0.03 Sources: Universities of Yale and Columbia.47 ECUADOR 84.78 BRAZIL 82.65 BULGARIA 78.30 Amundi Investment Strategy Collected Research Papers 291 . In the same way as Scholtens (2009).00 Standard Deviation 2. We obtain the following estimates for the parameters G 0 and G1 : Gˆ0 76.05 Mean 81. EPI focuses on two overarching environmental objectives: reducing environmental stress to human health and promoting ecosystem vitality and sound management of natural resources.06 PERU 78. Table 3 Environmental Performance Index 2008 ARGENTINA 81.08 PHILIPPINES 77. They are extracted from Datastream.85 VENEZUELA 80.44 UNITED STATES 81.36 MEXICO 79.80 PANAMA 83. Here. We start by estimating the portfolio EPI along the SR-insensitive frontier. which are then combined to create a final score. which corresponds to estimating the relationship (3) of Proposition 1.to October 2009. 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 scores attributed in 2008 are reported in Table 3. Descriptive statistics are available in Appendix 5.

Standard Deviation (%/year) SR-insensitive Frontier SR 80 SR 82 SR 84 SR 86 SR 88 292 Amundi Investment Strategy Collected Research Papers .00% 40. 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. January 1994 to October 2009 30. The corner portfolios for which there is a disconnect between the SR-sensitive and SR-insensitive frontiers are in Table 4.00% 50.00% 10.00% 30. Figure 9 exhibits the SR-sensitive frontiers for several thresholds I 0 . Mean (%/year) 20.30 in the EPI portfolio. The minimal EPI portfolio on the SR-insensitive frontier is obtained for the L ' ¦ˆ 1 I minimum-variance portfolio and is equal to 78.00% 0.00% 15.00% Ann.00% Ann.00% 20.00% 5. As Gˆ1 ! 0 . 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.00% 10. Figure 9 SR-sensitive frontiers versus SR-insensitive frontier for the EMBI+ indices.26 .00% 0. L ' ¦ˆ 1 L As an illustration of Problem 2.00% 25.

the EPI portfolio is 0.26 52.48 88 1. The minimal EPI portfolio on the SR-insensitive frontier is obtained 11 This variable is extracted from Datastream. the portfolio EPI increases with the expected return of the SR-insensitive capital market line.84%/year. for a 1%/year increase in expected returns. Then.13 22.22 84 2. we estimate the parameters G 0* and G 1* : Gˆ0* 80. the SR-sensitive frontier is the same L ' ¦ˆ 1 L L ' ¦ˆ 1 I as the SR-insensitive frontier.02 As Gˆ1* ! 0 .26 .26 .52 33.55 42.84%/year and differ for expected returns below 26.14 20.03 . we rely on the US 1-month interbank rate as a risk-free asset. According to the estimations. with a threshold equal to 84 on the portfolio EPI. the SR-sensitive frontier differs from L ' ¦ˆ 1 L the SR-insensitive frontier at the bottom and is the same at the top.21 40.02 higher.17 86 1. Amundi Investment Strategy Collected Research Papers 293 .11 Its responsible rating corresponds to the EPI of the United States I * 81. the SR-sensitive and the SR-insensitive frontiers are the same for expected returns above 26. For I0 ! 78.05 26.95 Gˆ1* 0. In order to illustrate Problems 3 and 4.93 L ' ¦ˆ 1 I As expected from Proposition 2.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. for I 0 78.84 32. For instance. improving the portfolio EPI costs more for investors with high risk aversion. In the case of emerging bonds.

Figure 11 exhibits the SR-sensitive capital market lines for several thresholds I 0 . we impose a set of constraints I p ! I 0 in the same way as in Problem 4. Mean (%/year) 20% 15% 10% 5% 0% 0% 10% 20% 30% 40% 50% Ann.03 . we consider SR investors wishing to adopt high environmental standards: thus. Standard Deviation (%/year) Non-SR Frontier Non-SR Capital Market Line From now on. January 1994 to October 2009 30% 25% Ann. and Table 5 displays the corner portfolios. The SR-insensitive frontier is shown in Figure 10. 294 Amundi Investment Strategy Collected Research Papers . Figure10 SR-insensitive capital market line for the EMBI+ indices.for the minimum-variance portfolio and is equal to I * 81.

We notice that the corner portfolios have particularly high expected returns and volatilities (see Table 5): this can be explained by the Amundi Investment Strategy Collected Research Papers 295 .01 63. the SR-sensitive capital market line is the same as the SR- insensitive capital market line.16 334. for I 0 I * 81.97 193. January 1994 to October 2009 30% 25% Ann. the SR-sensitive capital market line differs from the SR-insensitive one at the bottom and is the same at the top.38 144.06 84 0.13 As expected. Standard Deviation (%/year) 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 1.11 88 0. the SRI cost appears for investors with high risk aversion.89 453. Mean (%/year) 20% 15% 10% 5% 0% 0% 10% 20% 30% 40% 50% Ann. For I0 ! I * 81.93 323.Figure 11 SR-sensitive capital market lines versus SR-insensitive capital market lines for the EMBI+ indices.16 50.03 .23 239.08 86 0.03 . Here also.

00% Ann. (1). However.particularly low sensitivity Gˆ1* . We seek 0.00% Ann. Figure 12 Displaced optimal portfolios for the “representative investor” 30. Mean (%/year) 20. Let us now focus on a typical investor.00% 0.00% 10. This numerical application highlights that the cost implied by high environmental requirements in an emerging bond portfolio differs according to the investor’s risk aversion.00% 20.00% 15. it costs more to be green for investors with cold feet. meaning that for expected returns below 50.00% 30. if we consider a threshold I 0 82 . In this particular case. Standard Deviation (%/year) SR-insensitive Frontier SR 82 SR 84 SR 86 SR 88 SR 90 296 Amundi Investment Strategy Collected Research Papers .01%/year.00% 5. Sharpe (2007) suggests that the “representative investor” has a risk 2 aversion parameter O in the traditional mean-variance optimization of eq.00% 50.01%/year.00% 40. For example.7 to determine the consequences of SR thresholds for this “representative investor” by computing the optimal portfolios for different thresholds on the portfolio EPI.00% 0. the SR-sensitive and SR-insensitive capital market lines are disconnected. Figure 12 displays these portfolios (the means and variances of the optimal portfolios are given in Appendix 6).01%/year and an expected volatility of 63.00% 10.00% 25.06%/year. we observe in Figure 11 that the SR- insensitive and SR-sensitive capital market lines are very close for expected returns slightly below 50. the corner portfolio has an expected return of 50.

a cost for every investor). Our results are detailed so that they are ready to use by practitioners. depending on the link between the returns and the responsible ratings and on the strength of the constraint. Conclusion The rapid growth of the SRI fund market has given birth to a burgeoning academic literature. at the top) of the frontier. the “representative investor” is directly concerned by the disconnect between SR-sensitive and SR-insensitive frontiers for reasonable SR thresholds.37. a cost for low risk-aversion investors only). Indeed. d) totally different from the SR- insensitive efficient frontier (i. 1952). a cost for high risk-aversion investors only). b) penalized at the bottom only (i.e. if portfolio ratings increase (resp. an SRI cost appears and the optimal portfolio is no longer on the SR-insensitive frontier. (2006) highlight that limiting the investment universe to SRI funds can seriously harm diversification. Most academic studies show that there is little difference between financial returns of SRI funds and conventional funds. we show that a threshold on the responsible rating may impact the efficient frontier in four different ways. Geczy et al. SRI is introduced in the mean-variance optimization as a constraint on the average responsible rating of the underlying entities. This SRI cost rises with the strength of the constraint. the SRI cost arises first at the bottom (resp. c) penalized at the top only (i. no cost at all). In our study. 5. the constraint on the portfolio EPI has no cost while the threshold is below 82. our paper aimed at modelling SRI in the traditional mean-variance portfolio selection framework (Markowitz.37. However. The results are the same in Amundi Investment Strategy Collected Research Papers 297 . In other words.e. The SR-sensitive efficient frontier can be: a) identical to the SR-insensitive efficient frontier (i. In this case.e. To shed light on this debate. When the threshold is above 82. which is slightly above the average EPI rating of the sample’s countries.or the “representative investor”.e. decreases) with the expected return along the traditional efficient frontier.

Our work highlights the fact that the investor’s risk aversion clearly matters in the potential cost of investing responsibly. we believe it could find other applications in the asset management industry. We strongly believe that this finding could help portfolio managers of SRI funds. 298 Amundi Investment Strategy Collected Research Papers . this cost being zero in some cases. However. As the calculations in our paper are very general.the presence of a risk-free asset. notably for portfolio selection with asset liquidity constraints. 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.

Princeton University Press. R. Journal of Finance 7.. 1195-1215. 2009. Testing for mean-variance spanning: a survey.. Theory of social returns in portfolio choice with application to microfinance. C. D. M. 1965. 2006. Working Paper No 09-34. 105-120. Journal of Economic Dynamics and Control 26. Wharton School of the University of Pennsylvania. The valuation of risky assets and the selection of risky investment in stock portfolios and capital budgets. Nair V. G. R. Journal of Economics and Business 42.. Basak. Working Paper. Available at: http://www. N.. Nijman T. A cautionary note on the pricing of ethics. European Sustainable Investment Forum. European SRI Study. Levin. Ariane Szafarz and Kokou Topeglo for their helpful comments. Why do we invest ethically? Journal of Investing 14. Van Der Wijst.. A direct test of mean-variance efficiency of a portfolio. 2007... Dorfleitner. Princeton. 2002. A. Lintner. Markowitz. Jagannathan. J. Portfolio selection. Journal of Empirical Finance 8.. P. R. Beal. Paris. 13-37. 2005. Kim Oosterlinck.. 111-156.. Working Paper. G. Review of Economics and Statistics 47.. Investing For Change. A. G. Hedge Funds: An Analytic Perspective. Oxford. An ethical investments evaluation for portfolio selection.eurosif. 1239-2685. Landier. Amundi Investment Strategy Collected Research Papers 299 .. Journal of Business Ethics and Organization Studies 9.. 66-77. Nice guys with cold feet: the cost of responsible investing in the bond markets”. 2008. 2001. The efficient set mathematics when mean-variance problems are subject to general linear constraints. 53-56.. 77-91.. 2005. Benjamin Lorent. Leidl. Reeder J. Journal of Investing 14. B. Grauer.. de Roon. Université Libre de Bruxelles. Valérie Mignon.. Geczy. Farmen.. Oxford University Press. F. Centre Emile Bernheim. M. 2008..org/publications/sri_studies . European Sustainable Investment Forum. T.Acknowledgements The author thanks Marie Brière. H.. M.. Best. 1952. D. Ombretta Signori.. Sun.. University of Regensburg... 2009.. 1990.. Lo. Goyen. 2009. Investing in socially responsible mutual funds. Hugues Pirotte. Stambaugh. C. Philips. References Barracchini. Drut.

pdf .. Principles for Responsible Investment.. B. Measuring sustainability performance of investments. 300 Amundi Investment Strategy Collected Research Papers .pdf . the case of Dutch bond funds. Steinbach.. 2008.org/files/Scholtens_PRI2009.socialinvest.pdf . Journal of Finance 19. Renneboog. Society for Industrial and Applied Mathematics Review 43. Sharpe. 2007. Report on socially responsible investing trends in the United States. L.org/files/PRI%20Annual%20Report%2009. W. Paper prepared for the PRI Academic Conference 2009. Available at http://www. W. J. Scholtens. Journal of Banking and Finance 32.unpri.. Sharpe.org/resources/pubs/documents/FINALExecSummary_2007_SIF_Tre nds_wlinks. New York. Ter Horst.. Available at: http://www.unpri. Available at: http://www. 18-30. Socially responsible investments: institutional aspects. 1723-1742. 425-442. 31-85. M. Markowitz revisited: mean-variance models in financial portfolio analysis. PRI Annual Report.Principles for Responsible Investment. 1965. Social Investment Forum. C. 2007. 2009. and investor behaviour. Expected utility asset allocation. performance. Capital asset prices: a theory of market equilibrium under conditions of risk. 2001... 2009. Zhang. Financial Analysts Journal 63.

the weights vector that solves the problem is: ¦ 1 L 1 ª 1 L ' ¦ 1 P º Z «¦ ( P L ) L' ¦ L O ¬ 1 L ' ¦ 1 L »¼ The corresponding expected return P P stands as: 1 ª 1 º PP P 'Z L ' ¦ 1 P (( P ' ¦ 1 P )(L ' ¦ 1 L ) (L ' ¦ 1 P ) 2 )» L ' ¦ 1 L «¬ O ¼ And the expected variance V P2 stands as: 1 ª 1 º V P2 Z ' 6Z 1 2 (( P ' ¦ 1 P )(L ' ¦ 1 L ) (L ' ¦ 1 P ) 2 )» L ' ¦ 1 L «¬ O ¼ 1 The corresponding responsible rating I P is also a linear function of : O 1 ª 1 º Ip I 'Z L ' ¦ 1 I (( P ' ¦ 1 I )(L ' ¦ 1 L ) (L ' ¦ 1 P )(L ' ¦ 1 I )» L ' ¦ 1 L «¬ O ¼ 1 As the expected return is also a linear function of . (L ' ¦ 1 L )( P ' ¦ 1 P ) (L ' ¦ 1 P ) 2 Amundi Investment Strategy Collected Research Papers 301 . 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 . in the standard mean-variance case.Appendix 1 Following Best and Grauer (1990). 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.

this corner portfolio is the one for which the portfolio rating is 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. 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.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. 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. Indeed. Two cases have to be distinguished: where G 1 ! 0 and where G 1 0 . If I 0 . If ! I 0 . Case G 1 ! 0 In this case. This is verified if the portfolio rating of the efficient portfolio from the traditional mean-variance optimization is above the threshold I 0 . The minimal portfolio rating is . 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 ) 302 Amundi Investment Strategy Collected Research Papers . To compute the portion of the efficient frontier for which the linear constraint on the portfolio rating is binding. the portfolio rating increases linearly with the expected return in the traditional L ' ¦ 1 I L ' ¦ 1 I mean-variance optimization.

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 . the portfolio solutions stand as: 1 1 Z ¦ ( P rL ) > @ O 1 Zr 1 >L ' ¦ 1 ( P rL ) @ O In the mean-variance plan. Appendix 3 Following Best and Grauer (1990). the L' ¦ L1 L ' ¦ 1 L L ' ¦ 1 I constraint on the portfolio rating is binding. 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 . The maximal portfolio rating is . it possible to write I p G 0* G 1* P p with Amundi Investment Strategy Collected Research Papers 303 . the constraint on the portfolio L ' ¦ 1 L rating is binding at the top of the efficient frontier and inactive at the bottom. 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. Indeed. If ! I 0 . in the traditional mean-variance case.Case G 1 0 In this case. If I 0 . the portfolio rating decreases linearly with the expected return in the traditional L ' ¦ 1 I L ' ¦ 1 I mean-variance optimization.

the constraint on the portfolio rating is binding at the bottom of the efficient frontier and inactive at the top. If I * I 0 . This is verified if the portfolio rating of the efficient portfolio from the traditional mean-variance optimization is above the threshold I 0 . The minimal portfolio rating is I * . 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 ) 304 Amundi Investment Strategy Collected Research Papers . the portfolio rating increases linearly with the expected return in the traditional mean-variance optimization. If I * ! I 0 . Case G 1* ! 0 In this case. Two cases have to be distinguished: where G 1* ! 0 and where G 1* 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. (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.

Appendix 5 Descriptive statistics for the EMBI + indices in US dollars.40% 0.82% 11.18% VENEZUELA 13.07% 0.88% 22.47% -27.48% 0.50% -29. Std.63% PERU 14. If I * ! I 0 .75% -20.51% 0.94% 28.78 7.83% 11.44% RUSSIA 18.54 0. Mean Skewness Kurtosis Maximum Minimum Dev.05% -39.50 34.98 13.49 -0.84% -14. The modification of the Capital Market Line occurs for the same risk aversion parameter O*0 as for the case G 1* ! 0 .22 26. Case G 1* 0 In this case.47 10.27 13.90% BRAZIL 15.92 20.03% 21.79% 23.37 33.42 -0.84 12.19% 33.13% Amundi Investment Strategy Collected Research Papers 305 .73% 0.63% -72.10 28. the constraint on the portfolio rating is always active: the entire Capital Market Line becomes a hyperbola in the mean-standard deviation plan.77% -36. The maximum portfolio rating is I * .42 -1.26 8.96 25. January 1994 to October 2009 Ann.17% 0. Sharpe Ann.50 -0.26% 0.36 -1. the constraint on the portfolio rating is binding at the top of the Capital Market Line.05% 0.62 8.31% 20. Ratio ARGENTINA 4. (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.38% ECUADOR 16.59% PANAMA 15.84 34. the Capital Market Line is no longer a linear function in the mean-standard deviation plan but a hyperbola.41 7.91% 0.58 10.93% PHILIPPINES 7.55 -0.27% 34.03 -1. the portfolio rating I p decreases linearly with the expected return P p in the traditional mean-variance optimization.10% 20.88% -22.17% BULGARIA 15.07 9. when the constraint on the portfolio rating is binding.27 35.84% 0.18 28.29% -55. If I * I 0 .72 12.51 -0.78% MEXICO 9.32 -2.80% -43.

98 27.86 29.25 86 22.38 24.10 25.17 88 23.03 84 22.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 21.48 306 Amundi Investment Strategy Collected Research Papers .

WP-018 The Management of Retirement Savings revisited Didier Maillard. or yield negative returns. is fundamental. as risk free assets tend to vanish. and in particular which degree of risk to tolerate. provided the cost of risk may be mitigated. Amundi Investment Strategy Collected Research Papers 307 . With reasonable parameters. Finally. and thus expected purchasing power at old age. labour supply flexibility gives a reason for the optimal share of risky assets to decline with age. whatever the consequences in bad circumstances. Whether channelled through pension funds or individual accounts. Revised: September 2012 Retirement is not the only motivation of saving but it is a prominent one. However. such flexibility provides a significant incentive to increase investment in risky assets and provide significant welfare gains. or a way to optimise the likelihood of reaching a future consumption target. the question of how to allocate retirement savings. Investing in risky assets should not be viewed as a way to compensate for insufficient savings during a life time. Professor at Conservatoire National des Arts et Métiers Senior Advisor on Research – Amundi December 2011. 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. investing in risky assets is a way to improve expected returns on savings.

or objectives. is very low today and will undoubtedly remain so for a long period of time. a way to mitigate the consequences of an unfavourable configuration for return on investment in risky assets: supplementing inadequate purchasing power with labour income. if such assets truly exist. 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. the labour supply at the time of retirement and the portfolio allocation choice for retirement savings. which risk measures such as VaR or CVaR accurately capture. 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). In the specific case of saving for retirement. with conventional utility and expected utility functions. 1992) and generalizes it for different levels of risk tolerance (or aversion). We examine the importance of labour supply flexibility and the impact of such flexibility on asset allocation. such income can be secured by putting off retirement or by getting a job. if investment is made only in risk-free assets or moderately risky assets. In practice. however. 1 As Zvi Bodie has shown.1. the expectations of return are higher. there is a risk factor: in some adverse configurations. Introduction The return on risk-free assets. When risk is rewarded. however. even if part-time. expressions such as I cannot afford not to invest in risky assets should be banished. In light of these results. We find that that the portion allocated to risky assets can be substantially increased. Merton and Samuelson. Risky investing is not an appropriate way to make up for insufficient savings 1. If wealth is invested in risky assets. during the first few years of retirement. purchasing power could be perceived as catastrophically low. Therefore. In this research paper. 308 Amundi Investment Strategy Collected Research Papers . we will simultaneously model. the individuals concerned have. This result is consistent with the work done on labour supply flexibility and portfolio choice (Bodie. we can at last give thought to the identification of pension fund commitments.

it would be ten years or so. 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. and to satisfy the desire to transmit purchasing power to their heirs – but it is an important motivation. savings for retirement must be substantial. On average. which we will refer to generically as pension funds. to meet temporary interruptions in labour income. Amundi Investment Strategy Collected Research Papers 309 . Saving is therefore a key factor in having funds available and consuming during retirement. Under the institutional format. It is not the sole motivation – people also save as a precaution. 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. This length varies depending on the age of the working individual of interest: for a young person entering the workforce. with one major exception: defined benefit retirement plans guaranteed by a sponsor who is often the individual's employer. For an individual about to enter retirement. In this case. Under the individual format.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. One characteristic of saving for retirement is the length of the time horizon (delayed purchasing power). Conversely. the fund itself is responsible for asset allocation. this period is almost fifty years – a half-century. 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. a period of approximately thirty years elapses between the time savings are amassed and when they are used for consumption. All methods combined. Usually retirement lasts 20 years and follows forty years of work. retirement is a motivation to amass savings during a person's working life. the saver (assisted by his or her advisers) has primary responsibility for asset allocation. There are two principal schemes for saving for retirement: an individual format and an institutional format. 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.

assuming that savings is amassed mid- career and its fruits spent midway through the retirement period. It is worth bearing in mind that the orders of magnitude obtained are nonetheless significant. 2011b). The effort can be substantially reduced for higher real returns but they can only be achieved through risk- taking. With other saving channels. Arguments are made that the needs of consumption are lower after retirement due to.155 4% 0. in particular through pay-as-you-go regimes. tax generally eats into real returns. which is optimistic today for risk-free savings. among other reasons. the savings effort is very dependent on expected real return. approximately one-third of all labour income must be set aside for savings.379 0% 0.093 6% 0. In practice.184 0.In fact. the more so inflation is high (Maillard. tax does not impact real return. Savings here should be understood in a sense widened to include contributions to mandatory plans.611 0. Naturally.250 2% 0.333 0.103 0.058 0. If income tax rates do not differ between the two periods. the model should be fine-tuned to take account of the characteristics of the labour income time profile and mortality tables.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. replacement rate = 2/3 Required Saving Rate Real return (a) (b) -2% 0. so that a certain lifestyle can be maintained after retirement. But there are also arguments in favour of aiming at higher level of resources to cover care and medical bills. children leaving the household. 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. This is obviously very important for risk-free investing. 310 Amundi Investment Strategy Collected Research Papers . it is achievable consumption that is the aim. where expected real return is very low and sometimes even negative. at least partially. with a desirable replacement rate of two-thirds. Table 1 The savings effort required based on real return T = 30 years. With an actual rate of return (after tax 2) on investment near zero.

we assume that savings have been accumulated and that the individual has. most often. the capital secured is known with certainty. That reduces somewhat the savings ratio target but it remains important for low returns. 3.1.1. at moment in time 0. sufficient to comply with commitments or reach targets. into annuities 3. which undoubtedly satisfies the desires of most retirees. the initial wealth will be invested.In case of low real returns. or working age income net of retirement savings (column (b) in Table 1). W0 . wealth or capital. 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. If savings is not managed by an institution. r f . as least in part. Risk-free investment (nominal or corrected for inflation) can produce a return. If there is a risk-free investment between dates 0 and T. the individual may also think in terms of a target. WT . The time horizon to retirement is T. WT *. and if the initial wealth is invested in this risk-free asset. Limitations on management in terms of target. The issue In this section. The question of managing savings for retirement 3. secured on this horizon. that he or she must invest in one way or another (or that must be invested on their behalf).e. Amundi Investment Strategy Collected Research Papers 311 . to express the target in terms of old-age potential consumption as a fraction of working age potential consumption. 3 The hypothesis is therefore formulated that pensions are not exposed to risk once liquidated. 3. or the targets reached if: WT t WT * There are two possible cases. and that this final capital reflects the degree of achievement of this person's goals. We assume that are efficient mechanisms to transform wealth. Commitments will be kept. i. the saving effort is so huge that it is certainly fair to compute the ratio on a net-of-saving basis. in risky assets and the capital secured will be exposed to risk. However.1.

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. or. the functions of allocating the lifetime risk and handling the financial administration of the savings). the optimum solution would be found in investing in a binary option backed by a portfolio of risky assets. 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). in risk-free assets. Going beyond that. and exactly W T * with the highest possible probability. especially in the early 2010s. e WT * . We come back to the problem of optimising for the benefit of the ultimate investor as the pension fund is transparent (but ensures. 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. 312 Amundi Investment Strategy Collected Research Papers . by the same token. It can consist of investing a portion of wealth. of course. 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. 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. t WT * rf T W0 e The question of asset allocation is straightforward in that case. WT * rf T W0 e Complying with commitments or reaching targets is no longer certain. paying 0 with a low probability. Often it is impossible to meet commitments or to reach targets by means of a risk-free asset. But if we go deeper. and the remainder in high-risk assets. The first example is perhaps not the most typical. How then should be question of asset allocation be formulated? Should the probability of complying with commitments be maximised.

Amundi Investment Strategy Collected Research Papers 313 . 2011).2. In the interests of simplicity. with sufficient risk aversion (Maillard. is: ª 1 2 2º « r f D ( P r f ) 2 D V » T DV T H WT W0 e ¬ ¼ where İ is a random variable with zero mean and unitary standard deviation.We therefore assume that the charge of the shortfall does not land on the shoulders of the plan sponsor. Room is left for active portfolio management to improve the performance of the risk portfolio.1. 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. in particular because of the distant horizon. ı is annualised volatility and t is the Sharpe ratio. 5 We are speaking of "lite" dynamic allocation in the sense that the parameters are deemed constant over the period (no predictability). We also do not take into account fund's possible regulatory constraints on asset allocation. The choice of dynamic allocation allows an analytic treatment of the optimization problem. or it can be invested in an optimally-managed risk portfolio 4 (instantly optimising the Sharpe ratio) if adopting a dynamic management style 5. 4 The portfolio of risky assets is not necessarily the market portfolio. The target will be reached or exceeded if: ª r * r f 1 º H t H* « t DV » T ¬ DV 2 ¼ Pr(WT ! W *T ) 1 ) (H *) ĭLVWKHODZRIFXPXODWive frequency distribution of the risk. P rf t V The eventual value of the wealth. if a portion. Here ȝ is average annual expectation of return over the risk portfolio period. 3. Į. is (continually) invested in the risk portfolio. if one assumes they exist. Modelling Wealth can be invested in risk-free assets. Its results will differ very little from that of static allocation. we will use a Gaussian risk distribution.

İ = . investing all wealth in a risk-free asset is sufficient to achieve a probability of 1 in reaching the target (trivial.. Į = 0.If this threshold performance is less than the actual risk free rate.

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.2 0 0% 50% 100% 150% 200% 250% 300% 350% 400% -0. there is a funding gap at the effective risk free rate and the difference represents a shortfall in annualized returns. The theoretical optimum is found.4 0.4 -0. it is logical to place the asset allocation question within a framework of optimisation. which can be assessed using conventional measurements. a risk free rate of 2%. The threshold therefore decreases with the portion of risky assets.6 Share of risky assets Pr(WT>WT*) VaR 99% CVaR 99% Given the limitations of reasoning in terms of targets. If it is greater. for an infinite investment in the risk portfolio. Chart 1 Probability of meeting the target and risk 1 0. a credit spread of 4%.2 -0. aka Expected Shortfall).6 0. a annualised volatility of the risk portfolio of 20% and a funding gap of 2% per year. the standard deviation of final wealth or Value-at-Risk (VaR) or Conditional Value-at-Risk (CvaR. with leverage over the risk-free asset. As an illustration. The quid pro quo is obviously risk. with an investment term of 20 years.8 0. 314 Amundi Investment Strategy Collected Research Papers .

Amundi Investment Strategy Collected Research Papers 315 . the optimal proportion of risky assets decreases with the initial wealth of the savers. 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.2.2. Management in response to optimisation 3.1. In the catalogue of standard utility functions. which is counter-intuitive (See Annex 4). 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).3. 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 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. and the constant absolute risk aversion (CARA) function. we have rejected the quadratic function (with which utility decreases with consumption after a certain threshold). With this last class of functions. 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. 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. MaxE (U (WT )) As to the form of the utility function. However. we have opted for a function with the feature of constant relative risk aversion (CRRA).

2. Flexibility in the labour supply on retirement To take into account labour supply flexibility. with a weighting dependent on coefficient b. L) b 1 J 1 J C is the consumption for the period. the utility function is augmented to: C 1J ( L L)1J U (C . L is the labour supplied. optimization is written: MaxU (W wL. 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. W. The first term is the conventional utility function (CRRA) with constant relative aversion equal to Ȗ. the elasticity of substitution is confined in a range 0 to 1. L is the maximum amount of work that is possible to provide and L . Using the same exponent for consumption and for leisure firstly cross-references a conventional utility function of the constant elasticity of substitution (CES) type. If w represents the labour compensation rate. and labour income. the individual can move forward (if rules permit) the age at which he or she can receive his or her pension.L is therefore leisure. under a budget constraint. 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. using the same exponent provides an analytical solution to the problem of optimising utility. 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 reasonable. which is tolerable: if performance of the risky assets is very strong. The funds available for spending are in fact made up of accumulated wealth.2.3. 316 Amundi Investment Strategy Collected Research Papers . at the moment of interest. Furthermore. The second term represents the contribution of leisure to total utility.

representing minimum consumption. apart from the fact that the term is constant. the optimal portion of the risky assets therefore decreases with age. optimal allocation is composed of: . assumed to be constant. To demonstrate this point. . The impact of flexibility on the risk asset portion is not immaterial. at a risk free rate. or w Le rf T . is normally negative in this type of function. a dynamic portfolio optimally combining risky assets and risk-free assets. while the earnings from potential future work are independent. We can use the results of Bajeux. If this person can envisage working the equivalent of two years Amundi Investment Strategy Collected Research Papers 317 . a short position in the risk free asset corresponding to the present value. L) ¨1 b J w J ¸ V (W ) 1 J ¨ © ¸ ¹ The function V(W) is a HARA-type function. a risk-free investment ultimately yielding exactly the minimum consumption. the optimal allocation is the following combination: . 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). 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. At the outset. a long position in the portfolio combining the risk portfolio and the risk-free asset using dynamic management. 1 1 J (W w L)1J § 1 · U (C . The second phase entails maximising V(W). Jordan and Portait (2003) when maximising HARA utility. The proportion of the risky assets will change over time with the value of the portfolio. For the question at issue. the same as that resulting from optimising a CRRA. it is: rf T rf T W0 w Le § w Le · D D ¨¨1 ¸ ¸ W0 © W0 ¹ 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. of the maximum labour income that the retiree can receive. In this case. As accumulated savings usually grows with the age of the saver.

the opportunity and willingness to get an old-age job or to defer retirement. against the inability to work. 4. The ideal would be to offer à la carte allocation. Part of those gains stem from the ability to invest more in risky assets that labour supply flexibility provides. the optimal proportion of risky assets is increased by half. with little risk (but not without pain…). by making it possible to invest in riskier assets and to capture the rewards of risk. 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). that labour supply flexibility at the time of retirement can improve economic welfare. we would recommend promoting flexibility on the liquidation age. The optimisation question ultimately translates into optimisation with the constraint of a fixed investment in a given asset. As to pension funds. in any event. it happens that working during retirement or deferring retirement procures additional wealth at no risk or. taking account of risk aversion on the one hand and. Discussion and conclusions We have shown. In fact. This assumes that there are good assurances against potential joblessness and. on the other. using a simple model without being unrealistic. and especially in the case where they manage most of the retirement savings of their principals. above all.subject to the same pay conditions and assuming the risk free rate is zero. both directly and indirectly. 318 Amundi Investment Strategy Collected Research Papers .

“Dynamic versus Static Asset Allocation: From Theory (halfway) to Practice”. Vol. Didier. "Dynamic Asset Allocation for Bonds. 2010. SSRN. Felix Goltz. Working Papers Series N° 1942901. Isabelle. Samuelson. 2011b. Z. “Labor Supply Flexibility and Portfolio Choice in a Life Cycle Model. Working Paper.” Journal of Economic Dynamics and Control. SSRN. R. Boston University School of Management Maillard. Working Papers Series N° 1968985... Journal of Business. Jordan and Roland Portait. “New Frontiers in Benchmarking and Asset Liability Management”. 2001. 2011a. C. 16. Apr 2003. Z. Lionel Martellini and Vincent Milhau. F. Edhec Risk Institute. 2003. “Tax and Investment Return”. 427-449 Bodie. Working Paper 2001-03. Issue 2 Bodie. 76. Merton and W. 1992. James V. March 2011 Maillard. Didier. Stocks and Cash". September 2011 Amundi Investment Strategy Collected Research Papers 319 .References Amenc. "Retirement Investing": A New Approach. September 2010 Bajeux-Besnainou. Noël.

and a portion.Annex 1 Target is achieved iff: ª 1 2 2º « r f D ( P r f ) 2 D V » T DV T H WT W0 e ¬ ¼ t WT * ª 1 2 2º WT * «r f D ( P r f ) 2 D V »T DV T H t ln W ¬ ¼ 0 WT * ª 1 2 2º ln «r f D ( P r f ) D V »T W0 ¬ 2 ¼ ª r * r f 1 º Ht « t DV » T H* DV T ¬ DV 2 ¼ Pr(WT ! W *T ) 1 ) (H *) Annex 2 If a portion of wealth Į is invested in a risk portfolio with an expected return of ȝ and volatility on this return of ı. Using . 1-Į is invested in a risk-free asset whose return is r. and İ designing a zero mean unitary variance random variable ª 1 2 2º ª 1 2 2º « 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 ¬ ¼ WT W0 e ¬ ¼ If we assume that the utility function is CRRA with parameter Ȗ. 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.Wǀ VOHPPDand integrating leads to: ª 1 2 2º d ln W «¬r D ( P r f ) 2 D V »¼ dt DVdz Assuming the random process is Gaussian. 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 320 Amundi Investment Strategy Collected Research Papers .

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) ª 1 1 1 º 1 1 L «b J w J » b J L w JW «¬ »¼ 1 1 J J b w L 1 1 L 1 1 W 1 1 J J J J b w b w Amundi Investment Strategy Collected Research Papers 321 .

L) b¨ b J w J ¸¨ 1 b¨ b J w J ¸ 1 J ¨ © ¸¨ 1 ¹© b J w1 J ¸ ¸ 1 J ¨ © ¸ ¹ ¹ 1 1 J (W w L)1J § 1 · U (C . final wealth is: ª 1 2 2º « r f D ( P r f ) 2 D V » T DV T H WT W 0 w Le rf T . L) b bw J » (W w L) 1J 1 J « ¨ 1 ¨ 1 1 ¸ ¸ ¨ 1 ¨ J 1 1 ¸ ¸ « ©b J w J ¹ ©b w J ¹ » ¬ ¼ 1J 1 § · J (W w L)1J § 1 1 ·¨ 1 ¸ (W w L)1J § 1 1 1 · U (C .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 ¹ 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 ¹ b w b w 3) Utility function 1 1J 1 1J ª§ · § · º 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 ¹ ©b w ¹ ¼» 1J 1J ª 1J § · 1J § · º 1 « J ¨ 1 ¸ ¨ 1 ¸ » U (C . L) ¨1 b J w J ¸ V (W ) 1 J ¨ © ¸ ¹ Optimally.

e ¬ ¼ wL As to the funds available for consumption. they are 1 1 ª 1 2 2º « r f D ( P r f ) 2 D V » T DV T H J J b b WT wL 1 1 1 (WT w L) 1 1 1 W 0 w Le rf T .

e ¬ ¼ J J J J b w b w The utility is: 322 Amundi Investment Strategy Collected Research Papers .

1 1 J (WT w L)1J § 1 · U (C . L) ¨1 b J w J ¸ 1 J ¨ © ¸ ¹ 1 1 J §ª 1 · (1J ) ¨¨ « r f D ( P r f ) D 2V 2 » T DV T H ¸¸ º § 1 · 1J (1 J ) ¨1 b J w J ¸ W0 w Le f ¨ 1 ¸ r T .

e ©¬ 2 ¼ ¹ © ¹ 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 © ¹ 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 1 J t2 § 1 · u rf A( w) ¨1 b J w J ¸ 2 ¨ © ¸ ¹ 1J 1J W 0 e uT .

bL 1J A( w) (W0 w Le rf T )e uT .

1 1J W0 e uT «¬ 0 ª A( w) (W w Le r f T )e uT .

bL 1J º 1J »¼ Amundi Investment Strategy Collected Research Papers 323 .

Annex 4 Thoughts on the choice of the utility function The utility functions commonly used are the quadratic function. if Į 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 . In fact. the final portfolio is worth. 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. the constant relative risk aversion function (CRRA) and the hyperbolic absolute risk aversion (HARA) function. the constant absolute risk aversion (CARA) function. but should not differ significantly). you obtain a result that is relatively counter-intuitive. 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.

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 @.

then D '* D * maximises W '0 E (U (W 'T )) exp> aW ' 0 A@E exp> aW ' 0 BD @. W0 If Į maximises this expression.

324 Amundi Investment Strategy Collected Research Papers . 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.

To maximise the expectation of a HARA function. With labour flexibility. on the other. below which the question of utility is irrelevant.CARA function It has the form: C 1J U (C ) 1 J HARA function It has the form: (C Cˆ )1J U (C ) 1 J Ĉ is irreducible consumption. 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 Ĉ and the present value of the maximum of labour income in the risk-free asset. Amundi Investment Strategy Collected Research Papers 325 . placing in this asset exactly what is needed to attain Ĉ . 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. on the one hand the risk-free asset must exist and. It should not be confused with a target such as that discussed in the body of the article.

326 Amundi Investment Strategy Collected Research Papers .

We study this question taking the International Capital Asset Pricing Model approach. Revised: June 2012 Since many governments over the world have started issuing inflation-protected securities.and a local specific component. the inflation-linked bond yields and their yield spreads called the inflation breakeven rates. It is not evident how to incorporate the new securities into the investment processes. alongside the conventional nominal bonds. The challenge is to fit a linear factor model onto the triangular relationship that exists between the nominal bond yields. the investment universe for fixed-income investors has de facto been extended. or linkers. Head of Fixed Income Quantitative Research. in particular. The model we develop provides a consistent view on risk for internationally invested portfolios containing nominal bonds. WP-022 Incorporating Linkers in a Global Government Bond Risk Model Marielle de Jong. Amundi February 2012. linkers and/or inflation swaps. where in essence risk is decomposed in a global systematic. how to build them into a bond risk model such that the overall global price covariance is captured. Amundi Investment Strategy Collected Research Papers 327 .

and as such place a directional bet on the inflation level in a country. Naturally. We choose to do this in an international setting focusing on the cross-border price dependency between national bond markets as a whole. Since the pricing of inflation has been made explicit in a way via the bond yield differentials called the breakeven inflation rates. It creates the possibility to take a relative position on real debt versus nominal debt. this market innovation calls for a revision of the existing risk models serving government bond investment strategies. The currency risk inherent to purchasing foreign assets is not considered either. This was not conceivable before. but the price influence of that cannot be made explicit in an easy way without a real bond comparative. 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 328 Amundi Investment Strategy Collected Research Papers . We adopt the International Capital Asset Pricing Model introduced by Solnik [1974]. towards an international scale including other asset classes as well. The dichotomous decomposition of risk facilitates the assessment of the tactical positions split into global directional bets and individual country bets. rather than between individual financial instruments. The essence of the model is that price returns are being decomposed into a systematic global market risk component and country-specific residual components. inherently exposed to inflation risk since their payoff is typically not compensated for the inflation incurred over the time to maturity. in this article. This model is a generalization of the Sharpe-Lintner CAPM that was initially developed for equity instruments within one country. The model design typically suits an international bond investor who runs a country allocation strategy. We explore. it would in principle be possible to model inflation risk in an explicit way as well. unlike real bonds.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 world’s markets. The term structure in the bond yields is not considered in this article. how to build the new inflation dimension into a bond risk modeling framework. This market innovation does not merely increase the choice in sovereign debt securities. We have not found studies in the literature making such attempt. Nominal bonds are. it essentially adds a new dimension to the scope of the bond investor.

and the yield spreads. Within the modeling setting that we have chosen we uncover an intriguing price correlation relationship between the bonds. Correlations are measured between the time-variation of the nominal bond yields (NBY). the Euro Area. which would not be easy to detect in a national context. over an eight-year period from June 2002 to July 2010. Six countries have been retained in this study: Australia. and section 5 concludes. Canada. the inflation-linked (real) bond yields (RBY). In section 4 we present which modeling approach would suit best.5 years as calculated by Barclays Capital. The global price covariance between nominal. the breakeven inflation rates (BEIR). are measured on weekly bond returns in the seven-to-ten-year maturity range with an average duration of 7. which correspond to the members in the Barclays Developed World Inflation- Linked Bond Index that have started issuing linkers before 2002. 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. see for example Fabozzi [1998]. Great-Britain. In Exhibit 1 the price correlation matrix is displayed between the two types of bond traded in the major developed world markets. Sweden and the United States. as well as the annual volatilities which are displayed on the diagonal.1 The correlations.and inflation-linked bonds We start by exploring the price covariance of the inflation-linked bonds with the nominal bonds over the recent past. The particularly volatile months between October 2008 and May 2009 have been discarded for reasons discussed separately below. In next section we discuss this pricing phenomenon.the usually more general fixed-income models used by practitioners and discussed in the literature. Amundi Investment Strategy Collected Research Papers 329 . In section 3 we formulate the challenge it represents for modeling the risk.

16) 0.7% Canada 0.14(-0.2% United States 0.44 0.65 0. It reveals a global common price movement.37 (0.31 0.5% Euro Area 0.15) 0.38 0.36 0.01 0.36 4.56 0.49 0.27 0. Why would those variables be systematically less correlated within countries than between? Cette and de Jong [2008] give an explanation.82 0.81 5.31 0.56 0.11 0. Data source: Barclays Capital.48 0.67 5.12 0.4% Canada 0.52 0.29 0.54 0.27 3.45 0.56 0.43 0.1% United States 0.1% Euro Area 0.30 0.69 0.65 0.59 0.19 0.43 0.37 0.35 0.55 4.66 0.24 0.22(-0.30 0.05) 0.72 0. The numbers on the diagonal are annual return volatilities (bond duration 7.01(-0.4% NBY Australia 0.52 0.5% United States 0.3% Euro Area 0.6% BEIR Australia (0.38 0.43 0.67 0.59 0.42 0.29 3. see Christensen.31 0.61 0. The same holds for the breakeven rates with respect to nominal yields.24 0.25 0.46 0.19 0.56 0.48 0. The observation concerning the RBY and BEIR seems odd though.11) 0.74 0.1% Sweden 0.41 0.28 0.51 5.21 0.46 6.28) 0.60 0.80 0.25 0.21 0.10.74 0.65 0.46 0.2% Sweden 0.0% Period: June 2002 to July 2010. Note that the correlations are significantly positive.49 0.49 0.23 0.4% Canada 0.27 0.76 0.28 0.52 0.24) 0. Data frequency: weekly (Friday to Friday).33 0. NBY and BEIR variation RBY NBY BEIR es es es n 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 d at at at ed ed ed ite ite ite na na na ro ro ro st st st