Академический Документы
Профессиональный Документы
Культура Документы
Editors forewords p. 2
- Amundi Research in a nutshell p. 4
PASCAL BLANQU
Global Chief Investment Officer
of Amundi Group
Dear client,
The new normal, evolution, revolution these are all concepts that have
driven the financial markets since the mid-2000s. Financial crises, recession,
banking crises, debt crises, deflation, and so forth all these have been at the sides
of asset allocators and risk managers, more closely intertwined and requiring
more substantive and independent research. Active management, SMART beta,
risk analysis, diversification and entropy are all themes that one should revisit.
Proprietary research matters more than ever, and it may even be time to give it
a bigger role in assessing risks better.
Amundi has therefore chosen to expand its research teams over the past few
years and to get them implicated in its investment decisions. Our approach has
consisted in expanding both our top-down and bottom-down resources and
in involving them with the management teams and asset allocation decisions
on a regular basis. Research teams also take part in the various investment
committees and help create new investment processes. All this makes portfolio
investment returns a shared objective.
This has also helped enhance Amundis visibility worldwide. It is important to
provide guidelines that explain and publicise Amundis views both externally
and internally. The new line of research products (the cross-asset line) has made
a big contribution to improving this duty to explain. In this way, research work
promotes the groups investment strategies and themes at all times.
We have organised our research to achieve the very important objective of
remaining close to the academic world. Publishing of working papers and
financing research chairs, resulting in conferences and calls for papers, help
achieve this objective.
We have provided our research team with the resources that reflect the role that we
have given it. In addition to proximity with the management teams and clients, the
team offers a broad diversity of profiles and publications. This book shows clearly,
if there was a need, how important research is in Amundis set-up, as well as
the depth and diversity of our research capabilities. It highlights both the role
of research in the investment process and its ability to make lasting analyses that
form the basis of future action.
I wish you all a good reading.
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.
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
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
E
Eri
The
A RT
Pa
IC
LE
S
an
ac
ad
em
ic
24 ulso an
2nd 150
2014 Ranking of Amundi equity research Number of conferences each year
in Extel Thomson survey
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
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
May 2014
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.
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
Min (w Vw) T
Such that e w = 1 T
w
=
, ww
Since correlations among any pairs of assets are lower than one, the denominator
is lower than the numerator and the ratio is always higher than one. Maximizing this
Better diversification and lower correlations explain why the risk of the portfolio
maximizing the diversification ratio is always lower than the risk of a standard
market index. In addition to that, the optimisation contains a pseudo-minimization
of the denominator that is satisfied via the selection of low systematic risk stocks.
On the other hand, at the numerator, the optimisation results in the selection of high
specific risk stocks since they increase average volatility, while having little impact
on the denominator: specific risk doesnt matter at the denominator as it is easily
diversified away.
As a result, the portfolio maximizing the diversification ratio may show an average
total volatility that is not statistically different from that of a standard market index,
but will necessarily result in below average systematic risk stocks (the denominator
effect), and above average specific risk stocks (the numerator effect).
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.
Risk parity means that each asset (asset class, equity sector, single stock) has an
equal contribution to the total risk of the portfolio.
= = =
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
~
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.
TELECOM
INDUSTRIALS ENERGY
20%
10%
UTILITIES
FINANCIALS
0%
INF. TECH
CONS. DISCRET
CONS. STAPLES
MSCI EMU Eq. Weighted Risk Weighted (1 Step) Risk Weighted (2 Steps)
Source: Amundi Research
TELECOM
INDUSTRIALS ENERGY
UTILITIES FINANCIALS
CONS. STAPLES
Risk Weighted (2 Steps) Risk Weighted (2 Steps with Correlations) 10% Target
Source: Amundi Research
In this section we show with a practical example that all of the three smart beta
strategies discussed so far are exposed to the low risk anomaly.
We build three portfolios (in Barra One, at the model date of 12/31/2012),
restricting the investment universe to the constituents of the MSCI World Index.
We impose that no stock can exceed a 5% weight. We then group stocks into
three equally populated families, according to their risk: low risk, average risk
and high risk stocks. Finally we observe the percentage allocated to each family
of stocks, for each of the three portfolios as well as for the MSCI World.
90%
80%
70%
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, the risk parity portfolio has only a slight tilt
toward low risk stocks, compared to the standard index. The portfolio maximizing the
diversification ratio is apparently well balanced in absolute terms toward low or high risk
stocks, while it clearly underweights average risk stocks. As a conclusion, using total
risk as a grouping criterion, we see a clear and intuitive exposure to low risk anomaly
for the minimum variance, a slight but intuitive exposure for the risk parity portfolio,
and no exposure at all but rather a barbell allocation for the portfolio maximizing the
diversification ratio.
However we traditionally distinguish two components of risk: the systematic
component (or common factor component according to Barra One terminology) and
the specific component. This distinction is needed because, as we have documented,
the systematic risk is the most relevant measure when addressing the low risk anomaly
and, if we restrict our analysis to this component only, the picture changes.
90%
80%
70%
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. The low risk feature of the risk
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, the portfolio maximizing the diversification ratio exhibits almost 50% of
the weight invested in high specific risk stocks, and only marginal weight in low
specific risk stocks.
We have explained in section 1 that the maximization of the diversification ratio
contains a pseudo-minimization of the denominator that is satisfied via the
selection of low systematic risk stocks. On the other hand, at the numerator, the
optimisation results in the selection of high specific risk stocks since the latter
increase the numerator, while having little impact on the denominator: specific risk
doesnt matter at the denominator as it is easily diversified away.
As a result, the portfolio maximizing the diversification ratio may show an average
total volatility that is not statistically different from that of a standard market index,
but will necessarily result in below average systematic risk stocks (the denominator
effect), and above average specific risk stocks (the numerator effect).
2.2 Diversification
2,8
2,6
2,4
2,2 Diversification
2
Minvar
1,8
1
Diversification Ratio Total Risk Diversification Ratio Common
Factor Risk Source: Amundi Research
Unsurprisingly, we find that the minimum variance portfolio is well diversified indeed,
while the risk parity portfolio also provides some diversification improvement,
relative to the standard market index.
We compute the same measure excluding the specific risk component both at
the numerator and at the denominator and, while finding the same hierarchy,
we confirm that the specific risk inflates diversification measures, and better
explains why a process maximizing diversification is tilted toward high specific
risk stocks.
We than compute the average correlation of stocks, according to the CBOE
methodology:
N n
2-- wi w jm im j
i 1 j i
Average Correlation
70%
60%
50%
Diversification
40%
Minvar
30%
Risk Parity
20%
Msci
10%
0%
Average Correlation Total Risk Average Correlation Common Factors
Source: Amundi Research
Entropy
10000
1000
LOG Scale
100
1 276 740
10
47
39
1
Diversification Minvar Risk Parity Msci
Source: Amundi Research
In the Table below we show the correlation matrix of active returns relative to the
corresponding benchmark for each strategy.
FTSE TOBAM M.D. 29% 78% 66% 64% 82% 85% 84%
Amundi Risk Parity 35% 61% 64% 79% 48% 48% 48%
MSCI World MinVol 20% 83% 82% 65% 48% 91.4% 91.2%
Amundi MinVar - Piot 10% 86% 84% 54% 48% 91.2% 95.4%
We can easily recognize the three family blocs with the FTSE Edhec Risk Efficient
somehow being an outlier among its family as well as among the full sample of
strategies. This is due to the specific constraints that affect holdings on each stock:
any constituent cannot be weighted less than one-third of an equal weighting
schemes, neither more than 3 times such a quantity. Though these constraints are
sound, they make this index half way between a market weighted and an equally
weighted portfolio, and not that close to an unconstrained portfolio maximizing
diversification. Not surprisingly, this index is well correlated to the MSCI World Risk
Weighted index that applies similar constraints. Interestingly we notice that the
diversification bloc is highly correlated with the minimum variance block, while the
risk parity block stays somewhere in the middle.
In any case, the correlation matrix suggests that there is some common behavior
behind the active returns of each strategy and this intuition is confirmed by the principal
component analysis (always on active returns), summarized in the chart below.
75% 2.0%
70% 1.5%
65%
1.0%
60%
0.5%
55%
50% 0.0%
PC 1 PC 2 PC 3 PC 4 PC 5 PC 6 PC 7 PC 8
Cumulative Percentage Variance (LS) Variance of Each Component Source: Amundi Research
The common behavior is confirmed by the 85% variance explained by the first factor,
and by the 91% variance explained by the first two factors alone. One can argue
that we have such a high percentage explained as we use redundant information,
since many strategies in our analysis (almost all the strategies within each family
bloc) are very similar to each other, thus resulting in overlapping behaviors. For this
reason we run a simplified PCA on a restricted sample of one strategy per family
(FTSE Tobam Maximum Diversification, Amundi Risk Parity, Amundi Minimum
Variance).
Explained Variance
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
PC 1 PC 2 PC 3
With no common factor in place (that is, with perfectly uncorrelated strategies) any
principal component would coincide with a stand-alone strategy, while we can see
that the first component of our simplified sample explains as much variance as the
two most volatile strategies. There is definitely some common behavior underlying
the active returns of smart beta strategies and the true challenge is to identify such
a common performance drivers.
In order to detect those drivers, we regress the first two principal components of
the complete sample of eight strategies, over the explanatory variables listed below.
All variables are adjusted for market beta in order to avoid double counting for
the market beta effect and to limit multicollinearity. The dividend yield factor
has been simultaneously regressed over the market index and the value factor,
to delete positive correlation between value and dividend. As for the low risk
anomaly, we have built a long basket of stocks belonging to the lowest quintile
according to systematic risk (cf. common factor risk, estimated by Barra One),
and a short basket with stocks belonging to the highest quintile; baskets are then
weighted inversely proportional to their ex ante Beta in order make the long-short
beta-neutral, and residual (ex-post) market exposures as well as any involuntary
exposure to other factors are canceled out via a multiple regression over all
explanatory variables.
2 2,5
1,8 2,2
1,6 1,9
1,4 1,6
1,2 1,3
1 1
0,8 0,7
01/06/03 01/04/04 01/02/05 01/12/05 01/10/06 01/08/07 01/06/08 01/04/09 01/02/10 01/12/10 01/10/11 01/08/12 01/06/13
Low Syst. Risk Sector Reversal Moment. Small Cap Value Dividend Msci World
The sample period has been characterized by strong equity markets despite
the massive drawdown of 2008, a strong low risk anomaly effect (except during
the rebound of 2009), positive momentum, positive sector reversal (the latter is
interesting as it exhibits very low volatility), and small caps. Value and dividend yield
have been flat.
The first principal component has a very significant negative market beta, and
significant exposures to all the other explanatory variables with the exception of
momentum (positive but not significant). The variance explained is 60% for market
beta, 15% for the low risk anomaly, 5% for the dividend factor, and about 1% for
value, small caps and sector reversal.
The second component has small cap and sector reversal exposure, both of them
significant, but with small caps only explaining a non-negligible portion of variance (5%).
Overall we would argue that the active performance of smart beta strategies is
finally due to low market beta, low risk anomaly, small caps, and sector reversal.
However we recognize that each strategy may have different exposure to these
Unexpl. + Interact. MKT Low Syst. Risk Sector Reversal Momentum Small Cap value Dividend Total
The following chart instead shows the contribution to ex-post tracking error
(computed as the percentage explained variance times the realized tracking error)
for each of them.
10.00%
8.00%
6.00%
4.00%
2.00%
0.00%
FTSE EDHEC-R.E. Amundi Diversif. FTSE TOBAM MD MSCI World RW Amundi Risk Parity MSCI World MinVol Amundi MinVar Amundi MinVar -
Piot
Unexpl. + Interact. MKT Low Syst. Risk Sector Reversal Momentum Small Cap value Dividend
With the exception of the FTSE Edhec Risk Efficient, the regression model explains
80% to 90% of the variance of active returns and its F-test is significant for all the
strategies investigated. The model is thus overall well specified.
Intuitively the low market beta has a negative effect during upward markets, and
it explains a big percentage of the variance of active returns. Interestingly, those
strategies exhibiting the lowest market beta offset much of this negative effect with
a positive contribution by the low risk anomaly.
3.2 Smart Beta for active or absolute returns, a new equity core?
The choice whether smart beta should be used in an absolute or in an active risk-
return framework, depends on the utility function of the investor (or the mandate of
the fund manager in the case of delegated asset management), and the governance
of the investment process.
While a fund manager with the objective of maximizing information ratio -under
a limited tracking error constraint- may find it difficult to massively move toward
smart beta equities, an institutional investor aiming to maximize wealth under
some absolute risk constraint, could use smart beta equity to make up the bulk
(or the new equity core) of its equity investments. In an investment process that
is based on top-down strategic asset allocation by the investment board, and
equity allocation by the equity department thereafter, if the board allocates wealth
based on traditional benchmarks allowing limited tracking error deviations, the
equity department is likely to exploit the enhanced risk-return profile of smart beta
equities only in some satellites of the global equity allocation, since smart beta
equities bring high tracking error relative to a standard market index. On the other
100% 100%
80% 80%
60% 60%
40% 40%
20% 20%
0% 0%
d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20 d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20
10 Bln Usd Prog. Trade 25 Bln Usd Prog. Trade 50 Bln Usd Prog. Trade
Source: Amundi Research
100% 100%
80% 80%
60% 60%
40% 40%
20% 20%
0% 0%
d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20 d1 d2 d3 d4 d5 d6 d7 d8 d9 d10 d11 d12 d13 d14 d15 d16 d17 d18 d19 d20
10 Bln Usd Prog. Trade 25 Bln Usd Prog. Trade 50 Bln Usd Prog. Trade
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. The Minimum Volatility index is
the least liquid, not really because it is more exposed to small caps, but rather
because it is concentrated over a lower number of stocks (248), than the Risk
Weighted Index (1600).
As for smart beta in general, only a USD 10 billion program trade allows a relevant,
though not exhaustive, completion after 10 days. In detail, this is the percentage
completion after 10 days.
100%
95%
90%
85%
80%
75%
70%
65%
60%
55%
50%
Risk Weighted 43%R.W. Minimum Risk Weighted 43%R.W. Minimum Risk Weighted 43%R.W. Minimum
57%M.V. Volatility 57%M.V. Volatility 57%M.V. Volatility
10 BLN USD 25 BLN USD 50 BLN USD
In order to effectively complete the program trades in 10 days, the investor cannot
hold 100% of equity in smart beta and should dilute his holding with traditional
and more liquid equity investments. In the table below, we show the maximum
allocation in smart beta that the investor can afford, in order to complete each
program trade in 10 days.
100%
95%
90%
85%
80%
75%
70%
65%
60%
55%
50%
Risk Weighted 43%R.W. Minimum Risk Weighted 43%R.W. Minimum Risk Weighted 43%R.W. Minimum
57%M.V. Volatility 57%M.V. Volatility 57%M.V. Volatility
10 BLN USD 25 BLN USD 50 BLN USD
If the investor is not likely to incur program trades bigger than USD 10 billion, risk
weighted, minimum volatility (to a lesser extent), and a mix of the two indices may
all become a new equity core, as the investor can hold up to 100% of total equity
in smart beta. For higher sizes of program trades, smart beta allocation should be
kept residual with respect to market weighted equity, thus smart beta would be
more suited to being a satellite bucket of the portfolio.
However, if comfortable with the USD 10 billion hypothesis, the investor that
goes for smart beta as a new equity core, should seriously consider changing its
strategic benchmark.
In the next chart, we show the tracking error relative to the MSCI World Index
of all equity allocations from the example above, and the tracking error of an
allocation with 40% in the smart beta above and 60% in global bonds, relative
to a classic balanced benchmark (40% MSCI World Index, and 60% JPM Global
Bond Index).
Risk Weighted 43%R.W. Minimum Risk Weighted 43%R.W. Minimum Risk Weighted 43%R.W. Minimum
57%M.V. Volatility 57%M.V. Volatility 57%M.V. Volatility
10 BLN USD 25 BLN USD 50 BLN USD
The lower the impact of liquidity issues, the easier the move toward smart beta
equities as a new equity core. But massive investments in smart beta equities
As historical returns may suggest, as far as the risk is lower while returns are
higher, the risk-return profile of some traditional bond-equity allocation is improved
by simply switching from market weighted equities to smart beta. In the chart
below, we trace two simplified efficient frontiers using the JP Morgan Global Bond
index for fixed income, and the MSCI World Index or the MSCI World Minimum
Volatility for equities. The chart is based on historical data only (returns, variance
and covariance).
Despite a slightly higher correlation with bonds, and thanks to the far better risk
return profile of the MSCI Minimum Volatility Index, the improvement in the efficient
frontier is straightforward:
12%
11%
10%
9%
8%
7%
6%
5%
4%
4% 6% 8% 10% 12% 14% 16% 18%
------- MSCI Min Vol: historical data ------- MSCI Min Vol: historical data
Source: Amundi Research
We can state that, for the same level of risk of a traditional bond-equity allocation,
we can increase the relative weight of smart beta equities in the allocation (as
smart beta equities are more conservative than traditional equities), thus improving
performance, via both the higher percentage of equity and the higher return of
This is equivalent to the hypothesis that the MSCI Minimum Volatility Index is simply
a low beta index, as it was a combination of cash and a traditional index. Thus,
investing in low beta equity with the same risk adjusted return of traditional equity
would be equivalent to imposing the constraint of a minimum holding in cash, thus
making the frontier less attractive than using unconstrained equity.
12%
11%
10%
9%
8%
7%
6%
5%
4%
4% 6% 8% 10% 12% 14% 16% 18%
------- MSCI Min Vol: historical data ------- MSCI Min Vol: historical data
- - - - MSCI Min Vol: same risk adjusted return and correlation of MSCI World Source: Amundi Research
12%
11%
10%
9%
8%
7%
6%
5%
4%
4% 6% 8% 10% 12% 14% 16% 18%
------- MSCI Min Vol: historical data ------- MSCI Min Vol: historical data
- - - - MSCI Min Vol: same risk adjusted return and correlation of MSCI World
- - - - MSCI Min Vol: same risk adjusted return and correlation of MSCI World historical correlation Source: Amundi Research
We may wonder if this evidence is limited to the MSCI World Minimum Volatility
Index, or correlation with bonds is higher for any smart beta. Actually this is rather
generalized evidence, with the exception of the FTSE EDHC Risk Efficient Index
that is close to traditional equity, even from a correlation standpoint.
There are several ways to diversify across smart beta strategies, and diversification
is possible even if the number of strategies involved is very limited. Diversifying
across smart beta may also be useful in addressing the issue of building a
reasonable multi-strategy smart beta benchmark.
We provide several cases of multi smart beta allocation and for illustrative purposes
we stay within the MSCI family (MSCI World Minimum Volatility, and MSCI Risk
Weighted). We first consider the case of an investor within an absolute risk-return
framework. If the investor wants to achieve diversification by equalizing the two
indices contribution to absolute risk, according to a long term (10 years) covariance
matrix, he would allocate 57% of his assets to the minimum variance and 43% to
the risk weighted indices respectively.
100%
80%
57% 50,0%
60% 77%
40%
43% 50,0%
20%
23%
0%
Weights Total Risk Contribution Active Risk Contribution
100% 100%
Unexplained 10,06%
80% Momentum 80%
12,96%
Sector Reversal
60% 60% 15,43%
Value
92,08%
40% Dividend 40%
Low Syst. Risk
48,67%
20% 20%
Small Cap
0% MKT Beta 0%
Factor Contributions Factor Contributions
Source: Amundi Research
80%
33% 26.7% 50% 5.61% Momentum
80% 12.37%
Sector Reversal
60% 73.3% 16.69%
67% 60% Value
40%
50% 17.19% Dividend
40%
20%
Low Syst. Risk
By doing so, active risk would be balanced across the two strategies and, as a
side effect, diversification across the performance drivers would improve as well,
while the percentage of active risk with an unknown source would be reduced.
However, the investor might be much more sensitive to diversification across
the factors than across the two indices themselves. In this third example, we
assume that the investor is willing to maximize the diversification of the sources
of active risk. We thus maximize the measure of entropy as defined in section 2.2,
computed over the active risk contributions by factors, or performance drivers.
In this case, in order to reduce the still dominant contribution of low market beta,
the allocation in the risk-weighted index would increase. Contrarily, small cap,
dividend yield, and low risk anomaly contributions are increased. The change in
the latter, however, is mainly due to a base effect: portfolio exposure to the low
risk factor decreases, but as the active risk decreases as well (risk weighted index
has a much lower tracking error than minimum volatility, relative to the standard
index), its risk contribution as a percentage marginally increases. As a side effect,
unexplained active variance is further reduced.
As a fourth and last case, we now consider an investor that is comfortable with an
objective of diversification across factors in an active management framework, but
that is willing to introduce into his allocation some exposure to the sector reversal
factor, because of its regular and low-volatility historical contribution to performance.
The exposure to this factor is negligible in the three previous allocations.
The smart beta portfolio that is most exposed to sector reversal is the Amundi Risk
Parity, because of the two-step stock-sector construction process. We thus repeat
the last case study, adding Amundi Risk Parity to the set of available strategies.
As several factors drive smart beta performance, the most straightforward way to
implement some timing over the different indices or strategies, should be to time
the underlying factors and consistently allocate strategies.
Investors may develop a reliable style rotation model, and may apply some allocation
where risk contributions match return expectations, rather than maximizing some
diversification measure as we have done in previous case studies.
Another way to time smart beta strategies might be to investigate their behavior
according to different market conditions. The following chart exhibits the 12-month
cumulative outperformance of each of the three Amundi strategies relative to the
standard index, with a quarterly frequency, from June 2003 to December 2013.
Results are interesting and often intuitive as well.
0.25 3.5
0.2 2
0.15 3 3.0
1
0.1
0.05 2.5
0
-0.05 4 5 2.0
-0.1
-0.15 1.5
-0.2
-0.25 1.0
2011 03
2011 06
2011 09
2011 12
2003 06
2003 09
2003 12
2004 03
2004 06
2004 09
2004 12
2005 03
2005 06
2005 09
2005 12
2006 03
2006 06
2006 09
2006 12
2007 03
2007 06
2007 09
2007 12
2008 03
2008 06
2008 09
2008 12
2009 03
2009 06
2009 09
2009 12
2010 03
2010 06
2010 09
2010 12
2012 03
2012 06
2012 09
2012 12
2013 03
2013 06
2013 09
2013 12
In long and steady bull markets as was the case from 2003 to mid-2007, the
risk parity portfolio often exhibits the best returns, while during market crashes
minimum variance is by far the winning strategy.
When the impressive rebound of March 2009 starts, minimum variance starts
lagging the two other strategies, while diversification and especially risk parity react
well since they keep on delivering some positive outperformance.
When the market is suffering some higher volatility without exhibiting a clear trend
as in the period between mid-2011 and mid-2012, minimum variance is the winning
strategy with some nice resistance by the diversification strategy as well.
0.0%
25.0%
-5.0% 20.0%
15.0%
-10.0%
10.0%
-15.0%
5.0%
-20.0%
0.0%
Risk Parity Diversif. Minvar MSCI Risk Parity Diversif. Minvar MSCI
15.0% 10.0%
5.0%
10.0%
0.0%
5.0%
-5.0%
0.0% -10.0%
Risk Parity Diversif. Minvar MSCI Risk Parity Diversif. Minvar MSCI
The intuition behind this is that when correlation is very high (left chart), there is
less benefit in searching for diversification over risk factors, while searching for
diversification across assets directly is probably more effective. When average
correlation is lower (right chart), strategies based on diversification across risk
factors benefit more than those diversified on stocks directly, since the former
favor those stocks exposed to uncorrelated factors. Overall, returns on the right-
hand chart are lower, as the low correlation across sectors and countries includes
a typical pre-crisis situation (October 2008 and July 2011).
Average Correlation and Returns in the Eurozone
Estimation Period: 01/2003 - 06/2012
0,2
0,15
0,1
0,05
0
-0,05
-0,1
-0,15
-0,2
02/01/03
05/01/03
08/01/03
11/01/03
02/01/04
05/01/04
08/01/04
11/01/04
02/01/05
05/01/05
08/01/05
11/01/05
02/01/06
05/01/06
08/01/06
11/01/06
02/01/07
05/01/07
08/01/07
11/01/07
02/01/08
05/01/08
08/01/08
11/01/08
02/01/09
05/01/09
08/01/09
11/01/09
02/01/10
05/01/10
08/01/10
11/01/10
02/01/11
05/01/11
08/01/11
11/01/11
02/01/12
05/01/12
08/01/12
High Corr Low Corr Neutral Delta 2W - 2Y Av Corr Source: Amundi Research
Our third indicator is a turbulence index. We define market turbulence as the cross-
section dispersion of returns, computed over the GICS industry group indices. As this
indicator is closely correlated with the VIX index, we normalize by the VIX itself, as we want
to capture the turbulence that is not already explained by the market implied volatility.
15.0% 8.0%
6.0%
10.0%
4.0%
5.0%
2.0%
0.0% 0.0%
Risk Parity Diversif. Minvar MSCI Risk Parity Diversif. Minvar MSCI
Results are less intuitive than in the two previous cases, but the turbulence
indicator is uncorrelated with the implied volatility and average correlation signals.
As mentioned, our final dynamic strategy consists in the portfolio that averages
the three model portfolios based on the three market signals above. Each model
portfolio overweights the best performing strategy according to the observed
signal at the end of the previous month.
The chart below shows the gross total return performance of the market-weighted
index, an equally weighted basket of the three smart beta strategies, and the
dynamic strategy described so far. Starting mid-September 2012 data are out-of
sample, while starting from June 2013 the strategy feeds a real money portfolio.
Active Returns
2.50
Returns
1.03
2.00
1.02
1.50 1.01
1.00 1.00
0.50 0.99
2014
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Strategy with Timing (LS) Const. Mix (LS)
Mkt W Index (LS) Cum. Active (RS) Source: Amundi Research
The case for smart beta benchmark is different. Lets have an example focusing on
some minimum variance benchmarks (such as the MSCI Minimum Volatility index),
or on some diversification benchmarks (such as the FTSE EDHEC Risk Efficient
Index, or the FTSE TOBAM Maximum Diversification).
These indices do not rely on some objective and easily measurable metrics, as in
the case for market capitalization: rather, all of them depend on:
1. Some risk measure estimation (the variance covariance matrix of stocks)
2. An optimizer, and its numerical algorithms
3. The objective function that is maximized: portfolio variance is minimized
(or some diversification measure maximized), subject to some constraints
(minimum stocks threshold, stocks upper bounds, sector concentration
constraints, etc.)
All the points mentioned above contain provider-specific features (risk model,
optimizer), while some of them (the set of constraints) are also very discretionary
and better shaped to design an investment strategy than to build a traditional
investment benchmark.
Although reasonable, realistic, and prudent, these rules are discretionary and lead
to a benchmark that is provider-specific, rather than universally recognized, as it
would be required instead.
In conclusion, while the disclosure of parameters and models may satisfy the
transparency condition, the dependency on different risk models and optimizers,
and the common practice of applying various and sometimes heterogeneous
constraints prevent the benchmark to be easily recognized and universally
representative.
Conclusion
Smart Beta equities are the asset management industrys answer to some well-
known drawbacks of market capitalization-based equity indices such as price
noise, overrepresentation of large caps, absence of an auto-corrective mean-
reversion mechanism. Some of these features may result in high volatility and
massive drawdowns, thus potentially compromising the risk-return payoff of
traditional equities, at least when the investment horizon is shorter than 8-10 years.
In this study we provide a formal description of three popular risk-based smart beta
strategies minimum variance, diversification, and risk parity.
We show that low market beta and the low risk anomaly explain a relevant
portion of the variability of the active returns of the minimum variance strategies,
with some variance explained by sector reversal and dividend yield. Yet the
unexplained variability corresponds to some non-negligible positive contribution to
performance (thus further investigation is needed), while filtering the universe for
some quality criteria proves to provide additional value.
Performance drivers behind the risk parity strategies are basically the same, but
we notice that the low beta and low risk anomaly are less explanatory than
small cap and sector reversal. Sector reversal as a source of outperformance
is more relevant for risk parity than for any other smart beta, especially where (as
is the case in Amundis process) risk parity is achieved through a two-step stock-
sector construction process.
Smart beta may become a new equity core if the investors relevant risk measure
is absolute risk. In this case, however, the liquidity of those strategies must be
consistent with the amount of assets the investor holds. If the investors relevant
risk measure is relative risk, smart beta might still become a new equity core,
Quality Stocks
We believe that fundamental equity selection can provide some valuable
enhancement in the risk return profile of equity portfolios, at least in the long run.
At the same time we do not want to renounce an optimisation process which is
completely independent from expected returns. Expected returns are very noisy in
forecast and thus responsible for well known error maximization problems. For this
reason, we apply a qualitative filter to our investment universe, excluding the lowest
quality stocks from the optimisation. Basically, each quarter we rank the constituents
of the MSCI World Developed Markets according to a Piotroski (2000) score and
we exclude the two bottom quintiles. Keeping 60% of constituents available for
investments, the optimizer is left with a high degree of freedom and it tilts the optimal
portfolio toward good quality stocks, without using explicit expected returns.
Turnover and liquidity
High turnover is a critical issue in many systematic investment strategies like
Minimum Variance and other optimisation-based strategies. In our case, turnover
in the investment universe is limited as the Piotroski score is based on balance
sheet data that varies very little during one quarter. Furthermore we also rebalance
our portfolio quarterly, as suggested by Baker and Haugen (1991).
Nevertheless, more than turnover itself, our concern is indeed liquidity: we aim to
avoid small illiquid companies as we want to be able to liquidate our portfolio in a
reasonable time lag, without incurring significant market impact costs.
To address this requirement, we limited the amount held in any stock to the
following percentage:
where UB i is the upper bound on the ith stock, D is the number of days that we
accept to liquidate the fund, ADV i is the average daily volume over the last quarter,
and NOT is a notional amount of assets under management of USD 1 billion: quite
conservative as it is still far above the current size of our fund.
Sector, country, and stock concentration
As mentioned above, Minimum Variance and Diversification portfolios provide
excellent diversification across risk factors, but may tend to be poorly diversified
across sectors, countries or single stocks. We have thus applied some constraints
at these levels, without preventing the optimizer from choosing solutions that are
far enough from a market index.
Bibliography
Ang A., Hodrick R.J., Xing, Y., Zhang, X. 2006. The Cross Section of Volatility and Expected
Returns. The Journal of Finance. Vol. LXI, No. 1 (February)
Baker M., Bradley B., Wurgler J. 2009. A Behavioral Finance Explanation for the Success
of Low Volatility Portfolio. New York University, Working Paper
Baker M., Bradley B., Wurgler J. 2011. Benchmarks as Limits to Arbitrage: Understanding
the Low Volatility Anomaly. Financial Analysts Journal, Vol. 67, No. 1, CFA Institute
Baker N. L., Haugen R. A. 1991. The Efficient Market Inefficiency of Capitalization-Weighted
Stock Portfolios. The Journal of Portfolio Management, Vol. 17, No. 3 (March)
Clarke R., De Silva H., CFA, Thorely S., CFA. 2011. Minimum Variance Portfolio Composition.
The Journal of Portfolio Management, Vol. 37, No. 2 (Winter)
Clarke R., De Silva H., CFA, Thorely S., CFA. 1991. Minimum Variance Portfolios in the US
Equity Market. The Journal of Portfolio Management, Vol. 33, no. 1 (Fall)
Clarke, R., De Silva, H., Thorley, S. 2012. Risk Parity, Maximum Diversification, and
Minimum Variance: An Analytic Perspective. Journal of Portfolio Management, Vol. 39, No.
3,(Spring 2013)
Carvalho R.L., Lu X., Moulin P. 2011. Demystifying Equity Risk-Based Strategies: A Simple
Alpha Plus Beta Description. The Journal of Portfolio Management, Vol. 38, no. 3 (Spring)
Choueifaty, Y., Coignard, Y. 2008. Towards Maximum Diversification. The Journal of
Portfolio Management, Vol.35, No. 1 (Fall)
Markowitz H. 1952. Portfolio Selection. The Journal of Finance, Vol. VII, No.1 (March)
Maillard S., Roncalli T., Teiletche J. 2009. On the Properties of Equally-Weighted Risk
Contributions Portfolios. Social Science Research Network, Working Papers Series
(September)
Piotroski J. D. 2000. Value Investing: The Use of Historical Financial Statement Information
to Separate Winners from Losers. Journal of Accounting Research, Vol. 38, Supplement
2000
Russo A. 2013. Low Risk Equity Investments: Empirical Evidence, Theories, and the Amundi
Experience. Amundi Working Papers, WP-033-2013 (March)
Sharpe W.F. 1964. Capital Asset Prices: A Theory of Market Equilibrium under Conditions of
Risk. Journal of Finance, Vol. XIX, No. 3 (September)
Shefrin H., Statman M. 2000. Behavioral Portfolio Theory. Journal of Financial and
Quantitative Finance, Vol. 35, No. 2 (June)
Thomas R., CFA, Shapiro R., CFA. 2007. Managed Volatility: A New Approach to Equity
Investing. State Street Global Advisors
March 2014
Introduction
and Previous Literature
The relationship between firms environmental, governance and social practices
and their financial performance has attracted much debate in recent years.
This controversy has been fuelled by arguments from economics, management
and finance. As reminded by Kacperczyk (2009), the two main theses in play
could be described as the shareholder theory and the stakeholder theory. A
stakeholder as defined by Freeman (1984) is any group or individual who can
affect or is affected by the achievement of an organizations purpose. Both
theories defend different views on the role CSR (Corporate Social Responsibility)
should play in the definition of a firms objectives.
According to the shareholder theory corporate managers should focus solely
on increasing their shareholders wealth. The responsibility towards shareholders
should always be considered as more important than the responsibility towards
non-shareholding stakeholders such as employees, customers, natural
environment or communities. This thesis is notably upheld by Friedman (1970),
Jensen and Meckling (1976) and Fama and Jensen (1983).
The stakeholder theory (Freeman (1984) and Freeman et al. (2007)) states that
corporations should consider the interests of each stakeholder in their decision
making. In a modern pluralistic society, a firm cannot simply maximize one
objective function in order to deal with all potential contingencies. Furthermore,
no stakeholder should have a prima facie obligation over another (Kacperczyk
2008). According to Freeman (2004), the stakeholder theory asks for the purpose
of a firm and shared values with all stakeholders. Efficient corporate governance
for instance may foster financial performance and facilitate debt financing. Having
close relationships with suppliers and being attentive to customers needs might
establish a form of loyalty that helps to reduce uncertainty and strengthen a
I - Amundis extra-financial
analysis process
1.1 Philosophy
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
3 Regulation 3 9
36
By performing this exercise for all criteria, we get a score per criterion. For
instance, on the criteria of utilities E dimension, we get the following scores:
Criteria Score
Energy consumption & GHG emissions 18
Water 36
Biodiversity, pollution & waste management 10
t The weight of the dimensions E, S and G: this is the ratio of the sum of the
scores of criteria for a given dimension to the sum of scores for all criteria. In
the utilities sector, we get:
E S G
Weights 38% 30% 32%
Sector E S G
Automobile 37% 32% 31%
Bank 26% 33% 41%
Pharmacy 28% 42% 30%
Thus, the more potential impact a criterion has on the value of the business, the
more it will be weighted in the analysis model, in compliance with our desire to have
a pragmatic approach, i.e. based on the most tangible risks and opportunities.
A. A proprietary ESG data analysis and processing tool was developed with
the intent to:
t Collect and process data from extra-financial rating agencies so as to make
them comparable,
t Calculate the ratings,
t Generate alert signals for extra-financial analysts in case of insufficient,
obsolete or contradictory data,
t Include the ESG evaluation of businesses done by the extra-financial analysis
team and return it to the management teams,
t Spread the ESG rating of an issuer to all underlying issues.
B. A complementary qualitative approach
In addition to the automatic rating calculation, an active, in-depth analysis is done
on more than 250 securities. To enrich their analyses, extra-financial analysts rely
on several sources of extra-financial data:
t Meetings with businesses and their sustainable development ratio,
t Brokers, who are producing a greater quantity of increasingly refined studies
on the topic of SRI and sustainable development,
t NGOs,
t Analyses by Crdit Agricole Group,
t Media and public documents,
t Scientific reports...
To perform the analysis, each security is compared to securities belonging to a
sector whose ESG issues are homogeneous.
For example, the utilities sector is subdivided into three homogeneous sectors:
t Water utilities,
t Electric utilities,
t Networks.
> 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
Thus for each criteria in the benchmark, the businesses get a risk exposure/risk
management score of between 0 and 10.
On a given criterion, businesses that properly manage their extra-financial risks are
those who succeed in managing their financial risk as well. Indeed, in our example
of the water criterion, good risk management means lower operating losses due to
leaks (as the lost water has been previously treated by the business and therefore
incurred a cost) but also environmental risks (lack of water) in view of the rarity of
water, particularly in certain geographic areas.
Therefore there is a close link between financial and extra-financial risk.
Finally, each business is placed on a graph tracing its risk exposure to its risk
management (for compliance reasons, the businesses are not explicitly cited):
On the graph below, water utilities are shown in blue.
X-axis:
The business operating in the United Kingdom is not exposed to much of a water
risk - thus it is on the left in the graph (4/10 risk), while the American business
operates in water stress areas and is therefore given a risk score of 8.8/10.
Y-axis:
The business operating in the United Kingdom has the lowest leakage rate and
thus gets a risk management score of 9/10, while the other businesses, which are
down on this criterion, get a lower score.
10.00
English company
9.00
A
8.00
7.00
American
B
Company
6.00
Global Company 1
5.00 Global Company 2
4.00
C
3.00
2.00
1.00
D E F G
0.00
0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 9.00 10.00
Low exposure High exposure Risk exposure
Ultimately, the businesses placed at the upper left of the graph will be given the
best ratings on the criteria analyzed (the English business rating on the water
criterion is B) while the businesses at the lower right of the graph will be given the
lower ratings (the US business rating on the water criterion is D).
On the basis of these ratings, SRI managers must abide by these constraints:
t the overall portfolio ESG rating has to be over 0.5 (C rating)
t the overall portfolio ESG rating has to be over the benchmark rating
t the portfolio cannot be invested in stocks with an ESG rating below -0.5 (E, F
and G rated stocks)
Because the choice of criteria included in ESG ratings is based on their possible
impact on the businesses economic performance, it is reasonable to expect
that portfolios that pick their stocks according to ESG criteria may outperform.
However, the demonstration of this potential outperformance comes up against
difficulties with the rating method.
One of the most traditional - and simple - ways to test the added value of a
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
The two curves - total and systematic - are closely correlated, which alerts us
to the importance of these geographic biases in this long shorts performance.
It is hard to justify attributing the relative out- or underperformance of the
geographic areas to differences in ESG ratings. There are too many economic,
cyclical and political factors influencing such out- or underperformance.
Thus we have also carried out an ex ante risk analysis of the long short portfolio.
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
The graph above shows that the geographic allocation effect explains between
50% and 90% of the ex-ante risk. As such, this effect is quite dominant, and there-
fore totally masks the effect of the SRI in the long shorts gross performance.
For bond portfolios, additional challenges emerge in the construction of a long
short. Depending on the maturity of the bonds, sensitivity to rate fluctuations is
highly variable. A long short portfolio, created without precautions, is very likely to
be exposed to interest rate risk, moreover, with an unstable magnitude over time.
Furthermore, many issuers have several bonds with different characteristics; then
the question arises of choosing and weighting the bonds. An alternative is to use
fixed-maturity CDs instead of bonds, but the credit risk is very different depending
on the issuers rating, and varies in a much broader spectrum than on equities.
Thus a long short portfolio, even if built with CDs, has every chance of showing
strong systemic biases that are unstable over time.
We think the potential added value of ESG criteria is more likely to be detected in
the performance after stripping out the systematic effects, at the level of security
picking. Several methods can be used to measure this added value:
Measuring the minimum tracking error (TE) induced by an SRI process is critical
information regarding the development of SRI. Such information represents the
cost of SRI implementation from a risk perspective. If this minimum TE is low, there
is little risk turning a classic index into an SRI-compliant index while integrating
ESG convictions. On the contrary a high TE would be the sign of a greater risk
to undergo in order to be compliant and would then be less favorable for SRI
advocates. Additionally we have previously seen that geographic biases are of
paramount importance and as such further tests need to be carried over different
geographic zones.
To determine the minimum level of TE induced by transforming some well-known
2.00 2.00
1.50 1.50
1.00 1.00
0.50 0.50
0.00 0.00
World North America Pacific Europe
Median Source: Amundi Research
For Europe and World we can state that SRI compliance is consistent with low
tracking error processes, with respectively 0.30% and 0.55% on average.
However, this is no longer the case for the North American and Pacific zones where
minimum TE levels are higher, above 1%, and even reach 2% TE levels in some
specific conditions.
Ex-post levels of TE are globally in line with the ex-ante observations.
TE ex-post ESG
World 0.60%
North America 1.07%
Pacific 1.21%
Europe 0.35%
2.0
1.5
1.0
0.5
0.0
01/01/05
01/04/05
01/07/05
01/10/05
01/01/06
01/04/06
01/07/06
01/10/06
01/01/07
01/04/07
01/07/07
01/10/07
01/01/08
01/04/08
01/07/08
01/10/08
01/01/09
01/04/09
01/07/09
01/10/09
01/01/10
01/04/10
01/07/10
01/10/10
01/01/11
01/04/11
01/07/11
01/10/11
01/01/12
01/04/12
01/07/12
01/10/12
01/01/13
01/04/13
-------- World North America Pacific Europe Source: Amundi Research
Date TE
08/31/2008 1.12%
01/31/2009 2.38%
%CR %CR
Ptf weight Ptf weight ESG ESG Total Total
to Active to Active
Company as of as of score score risk risk
Total Risk Total Risk
08/2008 01/2009 08/2008 01/2009 08/2008 01/2009
08/2008 01/2009
Company 1 1.83% 0.00% -0.15 -0.60 23.1 37.1 0.09% 5.12%
Company 2 1.22% 0.00% 0.05 -0.57 37.1 65.4 0.06% 5.36%
Company 3 1.21% 0.00% 2.32 -0.70 43.6 101.8 0.05% 2.61%
Company 4 0.93% 0.00% 0.03 -1.18 43.4 59.1 0.02% 5.61%
Company 5 0.92% 0.00% 1.38 -0.53 36.9 67.2 0.06% 4.21%
(2) What happened in 2010, with a sudden reduction of the TE level, is almost the
opposite of the previous situation:
Date TE
01/31/2009 2.38%
01/31/2010 0.57%
Computing similar statistic tests on the full sample period IRs does not prove more
significant.
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
We have seen so far that making an index SRI compliant proved to be neutral from
a performance perspective. We now investigate whether an investment process
aiming to maximize the ESG profile of a world developed markets equity portfolio
under some constraint of tracking error may improve performance relative to
a standard benchmark. While in the previous exercise we tried to quantify the
impact of ESG constraints on performance, we are now focusing on a more direct
exploitation of ESG rating.
From January 2005 to June 2013, with a monthly frequency, we maximize a
classical risk-return utility function as:
Utility=ESG - 2
Where ESG is the weighted average ESG score of the portfolio, is the risk aversion
coefficient, and is the tracking error of the portfolio relative to its benchmark.
The risk aversion parameter describes the tradeoff between the average ESG
rating of the portfolio and the relative risk of the portfolio: the lower the investor
tolerance to relative risk, the lower its ambition should be in terms of ESG rating;
the higher the portfolios required ESG rating, the higher the relative risk that the
investor should accept. In order to assess the impact on performance (as well as
on portfolio characteristics) of a different target tracking error level, we run several
simulations allowing the risk aversion parameter to vary.
Every month of our sample period from January 2005 to June 2013, we maximize
our utility function applying a risk aversion of 9 for the low tracking error portfolio,
and a risk aversion of 3 and 1 for the average and high tracking error portfolios,
respectively. The risk aversion parameters have been calibrated in order to obtain
average TE levels of roughly 2%, 3%, and 4.5% respectively.
Obviously, as in the previous simulation on tracking error minimization, we continue
to impose Amundis SRI rules:
t the overall portfolio ESG rating has to be higher than or equal to 0.5;
t the overall portfolio ESG rating has to be higher than benchmark rating;
t the portfolio cannot be invested in stocks with an ESG rating below -0.5.
We are aware that the ESG signal may be biased toward some styles and some
geographical areas. As we want to capture the specific component of the ESG
signal rather than some implicit (and involuntary) allocation effect, we use a risk
budgeting constraint: the common factors component of risk is limited to 10%,
leaving the remaining 90% of the risk budget available for specific risk.
Although some observations may differ substantially from zero on an annual basis
(table below), none of the results turns out to be significant at an acceptable level
of confidence. The t-stats in the previous table have been computed over 100
monthly returns, while those in the table below have been computed over eight
annual information periods.
min TE 0.73 -0.55 0.51 -2.00 0.70 1.42 -0.04 -0.41 0.12
We can infer from this analysis that whatever risk aversion we use, the added value
of maximizing the ESG rating is neutral and does not depend on the level of tracking
error. Thus, we should expect an ESG portfolio to deliver performance in line with
that of a classic portfolio and its benchmark, at least with a tracking error ranging
from 2.0% - 4.5%.
However, stating that the added value of maximizing the ESG rating on a global equity
portfolio is neutral does not necessarily imply that none of the environmental /social/
governance (E/S/G) stand-alone ratings have some relationship with performance,
or that none of the stand-alone ratings or the aggregate ESG rating is neutral in any
specific geographical area.
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
As a conclusion, we can state that whatever the extra-financial criteria and the
geographic areas we investigate, the added value of the SRI signal is neutral.
Combining our finding on ESG value added with those on typical tracking error
of an ESG equity portfolio construction, we can infer that investors may achieve
full compliance with SRI rules at the costs of reasonable tracking error (especially
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
For more details, in the table and in the chart below we rank stand-alone styles
according to their contribution to performance.
10.0%
8.0%
6.0%
4.0%
2.0%
0.0%
e
th
ue
ity
ge
ld
l
ta
ilt
er
rit
iz
rs
tio
ur
ie
w
tu
id
To
al
ra
at
S
ea
ve
th
os
ro
Y
ia
en
qu
V
ve
ol
O
ar
G
ni
in
xp
om
V
Li
Le
U
-L
V
on
M
n
gn
ng
io
N
at
ei
ni
e
im
r
ar
Fo
iz
st
By far, the best contributor to performance is the size factor (the portfolio is exposed
to small caps, with small cap risk factor performing extremely well over the last 8 years).
Some additional positive contribution comes from low risk stocks (negative
exposure to volatility), and growth (negative exposure, significant contribution).
Contribution from value is also significant (especially in the US and Australia).
Among the worst performers, yield and momentum have limited impact and are
not statistically significant.
If we impose a risk budget constraint on common factors, the picture changes
dramatically. Limiting the common factors to 10% of the total ex ante risk, the
cumulative active return of the Governance portfolio is much lower, and interestingly
it fits the unconstrained portfolios specific component very well.
125
120
115
110
105
100
95
90
6 7 8 9 0 11 11 11 11 12 12 12 12 13 13
05 05 05 05 06 06 06 0 07 07 07 0 08 08 08 0 09 09 09 0 10 10 10 1
v- 20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 20 -20 -20 -20 -20 -20 20 -20 20 -20 20 -20
jan pr- Jul- ct an- pr Jul- ct an- pr Jul- ct an- pr Jul- ct an- pr Jul- ct an- pr Jul- ct Jan Apr Jul Oct an- pr Jul- ct an- pr
A O J A O J A O J A O J A O J A O J A O J A
Specific no risk budgeting Total with risk budgeting Source: Amundi Research
We believe that the governance filter has gained popularity in academic studies as
a well performing signal, thanks to both a favorable timeframe up to 2009-2010,
and thanks to some bias that should be taken into account properly. All in all, an
investment process that is compliant with governance criteria may be achieved
at a cost of low tracking error, and without negative impact on performance, but
we cannot state that the process is likely to benefit from superior performance.
We have seen that ESG investing is not expensive in terms of risk and is neutral as
regards to performance, especially in Europe and in the global developed markets.
Consequently there is no relationship between the added value of an ESG process
and the target tracking error level of the portfolio that maximizes the average ESG
score. However, as far as we allow higher tracking error, we can come out with a
higher average portfolio ESG score.
2.00 2.00
1.50 1.50
1.00 1.00
0.50 0.50
0.00 0.00
minTE AV9 AV3 AV1
Median Source: Amundi Research
0.90 0.90
0.80 0.80
0.70 0.70
0.60 0.60
0.50 0.50
0.40 0.40
0.30 0.30
0.20 0.20
0.10 0.10
0.00 0.00
minTE AV=9 AV=3 AV=1
Median Source: Amundi Research
10%
Marginal Contribution to Active Total Risk
5%
R = 0.00
Good
0%
R = 0.00 Average
R = 0.00
Bad
-5%
-10%
-4.00 -2.00 0.00 2.00 4.00
ESG score
Source: Amundi Research
As for the minimum tracking error portfolio, the average transfer coefficient is
very low among the three groups. Low transfer coefficients among any groups of
stocks, however, may result in a (artificially) positive transfer coefficient overall.
This depends on the highly asymmetrical distribution of the scatter plot of the
marginal contribution to risk and the ESG score. The negative contribution to risk
of stocks with an ESG rating lower than -0.5 (thus excluded, thus with negative
active weights, thus with negative contribution to risk), gives the charts an overall
positive slope (positive information coefficient).
The picture changes completely if we investigate one of the ESG maximizing
portfolios. Below we show the case for risk aversion equal to 3.
10%
Marginal Contribution to Active Total Risk
5%
R = 0.32
R = 0.00
R = 0.00 Good
0%
Average
Bad
-5%
-10%
-4 -2 0 2 4
ESG score Source: Amundi Research
Here the transfer coefficient is flat among bad and average stocks, while
it is significantly positive among good stocks. The positive overall transfer
coefficient is thus explained by an excellent transfer coefficient among good
stocks: the stocks that contribute the most to the portfolio tracking error are
those with the best ESG score (ideally those with the more pronounced positive
active weights).
Generally speaking, transfer coefficient is a correlation measure, and it is very
explicative when dealing with two normal distributions. When leading with
asymmetrical distribution (as the marginal contribution to risk of stocks in a long
only portfolio), we should better investigate the transfer coefficient among several
partitions of the scatter plot.
In order to allow for some generalization we compute transfer coefficients at any
date of our sample period, for any portfolio, and among any group of stocks. We
plot their median in the following chart.
0.50
0.00
-0.50
Good Average Bad
The ESG compliant portfolio with the lowest possible TE has a different profile with
respect to the three optimizations with an explicit ESG maximization objective.
These latter are quite similar to each other, since all of them better exploit the
signals relative to the minimum tracking error, especially among well-rated stocks.
Among them, the lower the tracking error, the better the transfer coefficient is.
In order to gain coherence between portfolio composition and stocks ESG rating,
an ESG process should not be limited to pure compliance, but should rather allow
for some further ESG rating improvement, even though it should ideally remain in
a low tracking error range.
However, the validity of this assumption may depend on how ambitious the ESG
compliance constraints are. If our investment universe has an average ESG rating
above our minimum threshold (0.5), and if in our investment universe there are only
a few small companies that must be excluded because of their negative rating,
ESG compliance would be respected by simply investing in the benchmark without
any further ESG improvement needed. On the opposite hand, if the average ESG
rating of our benchmark is much lower than our minimum threshold, and if many
stocks are excluded because of their negative rating, targeting ESG compliance
implies a significant ESG improvement, as it forces the optimizer to choose well
rated stocks to reach the minimum threshold.
In order to distinguish between periods where ESG compliance is straightforward
and periods where ESG compliance is harder to accomplish, we have monitored
the average ESG score of the minimum tracking error portfolio. We consider periods
where the average ESG rating is higher than our threshold, as easy periods, that
is periods where the average ESG score is naturally above the threshold. In the
same way, we consider periods where the average ESG rating is equal to our
thresholds as difficult periods as the optimizer matches the threshold constraint
while minimizing the tracking error.
0,80
0,60
0,40
0,20
0,00
Good Average Bad
When the ESG profile of the investment universe is poor, and compliance is
an ambitious goal to accomplish, a minimum tracking error process forces the
optimizer to exploit the good and average stocks even better than in each of the
three processes with an explicit ESG maximization objective.
10%
Marginal Contribution to Active Total Risk
5%
R = 0.27 R = 0.95
R = 0.00 Good
0%
Average
Bad
-5%
-10%
-6 -4 -2 0 2 4 6
ESG score
Source: Amundi Research
102
101.5
101
100.5
100
99.5
99
98.5
01/01/10
01/03/10
01/05/10
01/07/10
01/09/10
01/11/10
01/01/11
01/03/11
01/05/11
01/07/11
01/09/11
01/11/11
01/01/12
01/03/12
01/05/12
01/07/12
01/09/12
01/11/12
01/01/13
01/03/13
01/05/13
01/07/13
------ ESG Europe ESG US ESG World ESG Asia/Pacific Source: Amundi Research
As one can see in the table below, the outperformance of the European portfolio is
mainly driven by the factors E and G. The portfolio optimized with E as the expected
return shows a non-significant outperformance of 0.4%. The portfolio optimized with G
as the expected return exhibits a significant outperformance at the 10% level of 0.6%.
Zone ESG E S G
Ann Act Ret 0.30% -0.20% -0.30% 0.20%
World IR 0.47 -0.31 -0.41 0.29
T-stat 0.9 (-0.58) (-0.77) 0.55
Ann Act Ret 0.40% 0.40% -0.20% 0.60%
Europe IR 0.6 0.65 -0.31 0.88
T-stat 1.13 1.23 (-0.59) 1.67
Ann Act Ret 0.00% -0.10% -0.30% -0.10%
US IR -0.05 -0.3 -0.36 -0.07
T-stat (-0.10) (-0.57) (-0.68) (-0.13)
Ann Act Ret 0.00% 0.10% 0.10% 0.00%
Asia/Pacific IR 0.13 0.23 0.3 0.06
T-stat 0.25 0.44 0.56 0.11
By looking at the graph, of all portfolios optimized with the factor G as the expected
return, we see that the governance factor has an impact on the performance only
in Europe.
103
102
101
100
99
98
97
01/01/10
01/03/10
01/05/10
01/07/10
01/09/10
01/11/10
01/01/11
01/03/11
01/05/11
01/07/11
01/09/11
01/11/11
01/01/12
01/03/12
01/05/12
01/07/12
01/09/12
01/11/12
01/01/13
01/03/13
01/05/13
01/07/13
-------- Europe US World Asia/Pacific Source: Amundi Research
The optimized portfolios incorporating the ESG rating do not exhibit any significant
outperformance. Only the portfolio containing European bonds incorporating the
governance factor has a significant positive outperformance.
We use control variables specific to the issue and to the issuer. All issuer specific
data stems from FactSet. The control variables specific to the issuer are the
following: The Size is measured by the natural logarithm of the total assets of the
firm. Large firms are widely perceived as less risky and thus benefit from a lower
cost of debt. The Leverage is defined as total liabilities over total assets. It indicates
how much debt a firm has. The higher the leverage ratio, the riskier the firm. A
high leverage ratio should increase the cost of debt. The ROA is the accounting
return on assets. It represents the profitability of the firm and thus the ability to
pay back its debt. The Capital Intensity is the ratio of fixed assets to total assets.
Since the fixed assets could be claimed by a creditor in case of a default, a high
capital intensity should decrease the level of the spread. Loss is a dummy variable
that equals 1 if the firms net income before extraordinary items is negative in the
current and prior fiscal year. The control variables specific to the bond issue are
World
E 0.006 0.013
S 0 0.017
G -0.015 0.019
Europe
E 0.029 0.021
S 0.054*** 0.019
G 0.058*** 0.018
North America
E 0.018 0.015
S -0.009 0.022
G -0.039 0.026
Asia/Pacific
E -0.032 0.024
S -0.005 0.037
G 0.068 0.059
Only in Europe, we find a link between the cost of debt of a firm and the
environmental, social and governance ratings. Interestingly, the cost of debt
increases with a better ESG performance. It seems that the spreads of well rated
firms in Europe decrease during the 3 years of our backtests. This is mainly driven
by the factor governance. Portfolios using the factor governance as expected
return outperform their benchmark in Europe. Event though the overperformance
portfolios optimized using the ESG ratings is positive in Europe, it is not significant
according to our statistics.
Risk-Free Assets:
What Long-Term Normalized
Return?
Sylvie de LAGUICHE,
Head of Quantitative Research
March 2014
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, over long and forward-looking timeframes, cash is not without its
risks. If, for each horizon, we determine risk in the form of an equivalent normal
volatility which offers the same dispersion (i.e. by multiplying it by the root of the
horizon), we see that equivalent volatility increases.
This means that, over a 30-year horizon, the risk of reinvestment linked to cash
is high. In fact, dispersion is similar to that of an asset with a volatility of 12%
(namely around 60-80% of equities). While the natural response, to avoid this risk
of reinvestment, is therefore to define zero-coupon investments with a nominal rate
over the horizon in question as risk-free, this definition poses several problems.
First, although it is true that nominal returns are known, this is no longer the case
if we reason in real terms. A zero-coupon investment over the long term is
seriously exposed to the risk of erosion by inflation. As such, a regularly
renewed short-term investment (i.e. for which there is a certain adjustment in short
rates in relation to inflation) is therefore less risky in real terms.
3%
2.5%
2%
1.5%
1%
0.5%
0
1929-2013 1959-2013 1989-2013
Second, depending on the issuer, such an investment may carry a credit risk.
Before the crisis, and at least for strong currencies, the tendency was to consider
government bond rates as risk-free. Within the eurozone, the sovereign debt crisis led
to a serious discrepancy in government borrowing rates depending on the country,
much in line with the drop in ratings for peripheral countries. Unless we accept the
highly debatable view that returns on assets in euros would be different according
to country, a common reference curve is a fundamental requirement. Swap rates are
of course an option as their collateralisation means they incur very little credit risk.
Having said that, the crisis in confidence in the banking system and the increase in
Libor rates led to a distortion in swap rates, linked not to the risk of these instruments,
but rather to the very strong distortion of the Libor 3-month rates used as a reference
for variable rate swaps. One remedy was to replace the Libor 3-month rate by a daily
rate that was therefore less distorted. OIS rates (overnight interest swaps) do indeed
appear to bear less credit risk even if there are no long-term time series on OIS rates.
5 0.5
4 0.4
3 0.3
2 0.2
1 0.1
0 0
01-07 01-08 01-09 01-10 01-11 01-12 01-13
OIS 10yrs euro Libor 10yrs swap Difference (right hand scale)
Source: Amundi Research
III - Approaches
3.1 Macroeconomic equilibrium model
In a balanced economy, the interest rate at which the economy finances itself is
equal to the sum of the equilibrium values for (real) GDP growth and inflation. This
approach is particularly useful in establishing a reasonable target value for a short
rate associated with a short-term investment in the far distant future. This perfect
equilibrium is obviously not achieved today which is why this approach cannot be
used to directly forecast future cumulative returns for cash starting immediately.
Furthermore, the rate at which the economy finances itself is a mix of short and long
rates. Ascertaining the cumulative return on cash over a long period starting today
using this method means establishing a path between the current and target rates.
Finally, this method also requires an estimation for growth and inflation at equilibrium.
The great merit of methods that use the yield curve is that they are based on observable
data and not on forecasts. A first solution is a pure and simple extrapolation of the
current short rate. The virtue of this solution is its simplicity, but it is only justified if
investors anticipate that rates will remain stable which will mean a flat yield curve. The
slope between long and short rates is, however, much too variable over time to make
it an acceptable hypothesis. Another approach is to use the zero-coupon rate over
the investment horizon as a normative forecast for the return on cash. This approach
is based on two arguments. The first is to say that, in nominal terms at least, the
return on a buy and hold bond investment is known in advance and can therefore
be considered as risk-free. We have seen that this point of view is debatable and, in
any case, in no way guarantees the link between the initial long rate and the return
11
16
21
26
31
36
41
46
0.00%
DIFFERENCE MODEL - SPOT
-0.10%
-0.20%
-0.30%
-0.40%
-0.50%
-0.60%
-0.70%
-0.80%
-0.90%
-1.00%
Horizon (yrs)
HJM Source: Amundi Research
0.00%
Difference model - spot ZC
-0.20%
-0.40%
-0.60%
-0.80%
-1.00%
-1.20%
%
1%
2%
25
50
50
0.
0.
1.
Volatility
Source: Amundi Research
While these approaches are interesting, they require complex calculations and the
calibration of a substantial number of parameters. It is nonetheless useful to keep
in mind the implications of this theory.
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
We examined whether it is more useful to know the initial long rate, and the
long rate for which we have very long series is the 10-year bond yield. However,
because a coupon bond with a 10-year maturity has a shorter duration (around
7 years), we examined the cumulative returns on a short-rate investment over
7-year periods, and analysed the forecast errors made using the initial long rate
for each sub-period.
The dispersion is a little lower than with the initial short rate but is still strong.
3%
2,5 %
2%
1,5 %
1%
0,5 %
0
Observed Knowing initial Knowing initial
short terme rate long term rate
Horizon 7 yrs Source : Amundi Research
Here, we employ a 7-year horizon as an approximate duration for the long rate
used. Our aim is to identify the relationships between the forecast errors using
the short rate or the initial long rate and other variables.
Forecasting error on cash return and initial yield curve slope: 1930-2013
7.00%
FORECASTING ERROR
5.25%
3.50%
1.75%
0%
-1.75%
-3.50%
-5.25%
SLOPE
-7.00%
-2.00% -1.00% 0% 1.00% 2.00% 3.00% 4.00%
VForecasting error using initial short term rate Linaire (forecasting error using initial short term rate)
Source : Amund Research
While this relationship is significant and robust, it is also variable over time with a
sensitivity of around 50% on average.
More specifically, we limited our analysis to periods where initial conditions are
not fundamentally different (initial inflation between 0% and 5%). The graph below
shows the relationship between the forecast errors and the slope depending on
whether we use the long rate or the initial short rate.
Forecasting error on cash return and initial slope: 1929-2013 with filter
8,00
FORECASTING ERROR
6,00
4,00
2,00
0
-8,00 -6,00 -4,00 -2,00 0 2,00 4,00 6,00
-2,00
-4,00
-6,00
-8,00
-10,00
SLOPE -12,00
Foracasting error using initial long term rate Forecasting error using initial short term rate
Linaire Linaire
(forecasting error using initial short term rate) (foracasting error using initial long term rate)
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. Using the initial long rate on its own
creates an error with a negative and often downward bias in relation to the slope.
This means that the initial long rate globally overestimates the future return on
cash, all the more so when the slope is steep.
The graph below shows that there is also a link with the volatility of rates.
4.50%
3.00%
1.50%
0%
-1.50%
-3.00%
-4.50%
INITIAL VOLATILITY OF LONG TERM RATE
-6.00%
0 0.75 1.50 2.25 3.00
VForacasting error using initial long term rate Linaire (foracasting error using initial long term rate)
Source: Amundi Research
Using the initial long rate means we overestimate future returns, all the more so
when volatility is high. The scope of the phenomenon is also significant since,
for 1% volatility, using the initial long rate overestimates the return on cash by an
average 0.8%. While this is consistent with the arbitrage pricing theory, it is even
higher than what is predicted by the theory with a Sharpe ratio of 0.3. There are
two possible explanations here. The first is that the decline in rates during the
period in question meant that the Sharpe ratio was higher. The second is that the
increase in risk aversion during periods of strong volatility appears to accentuate
the phenomenon.
Given todays extremely positive curve and the low volatility in rates, using short
and long rates together therefore seems more effective than using either one or
the other on their own. The correction linked to volatility, which involves complex
calibration, is less useful.
We can also see that the forecast error is linked to the real short rate.
Forecasting error on cash return and initial real rate: 1929-2013 with filter
7.00%
FORECASTING ERROR
5.25%
3.50%
1.75%
0%
-1.75%
-3.50%
INITIAL REAL RATE
-5.25%
-6.0% -4.0% -2.0% 0% 2.0% 4.0% 6.0% 8.0%
VForecasting error using initial short term rate Linear (forecasting error using initial short term rate)
Source: Amundi Reseach
Mitigating the short rate with inflation means we can improve the forecast as we
avoid having to extrapolate a short rate level, distorted due to a very restrictive or
4.50%
3.00%
1.50%
0%
-1.50%
-3.00%
OBSERVED - INITIAL INFLATION
-4.50%
-5.3% -3.5% -1.8% 0% 1.8% 3.5% 5.3% 7.0%
V - Operational implications
In order to marry operational simplicity and coherence with theory and observation,
we recommend using a combination of short rates (25%), inflation (25%) and
zero-coupon long rates (50%) over the horizon as a normative forecast for
returns on short-term investment.
The initial long rate segment (50%) factors in the data on the future trend in rates
extracted from the yield curve.
The initial short rate segment (25%) reflects the fact that, on average, short
rates are lower than their forecast using long rates and present a strong level of
autocorrelation.
The inflation segment (25%) is used to correct the distortion between short rates
and the macroeconomic fundamentals linked to a temporary episode of very
accommodating or very lax monetary policy.
Today, we recommend using the OIS curve rates or, failing that, government
bond rates where governments have an acceptable rating or are in control of their
currency.
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,
t a significant increase in the volatility of long rates,
t an inversion in the yield curve.
I would like to thank my colleagues Jean Gabriel Morineau, Gianni Pola and Eric
Taz-Bernard at Amundi for their valuable input in improving this paper.
REFERENCES
Black F, Scholes M [1972] The Valuation of Option Contracts and a Test of Market Efficiency,
Journal of Finance, Vol. 27.
Heath DC, Jarrow BA, Morton A, [1992] Bond Pricing and the Term Structure of Interest
Rates: A New Methodology for Contingent Claims Valuation, Econometrica, Vol. 60.
Hull JC, White A [1997] Options, Futures and Other Derivatives, 3rd Edition, Prentice-Hall.
Musiela M, Ruthowski M [2004] Martingale Methods in Financial Modelling, 2nd Edition,
Springer.
Ltz FA [1940] The Structure of Interest Rates, Journal of Economics, Vol 55.
Modigliani F, Shiller R [1973] Inflation, Rational Expectations and the Term Structure of
Interest Rates, Economica, Vol 40.
Vasicek O [1977] An Equilibrium Characterization of the Term Structure, Journal of Financial
Economics, Vol 5.
May 2013
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 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.
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.
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
We investigated asset segmentation through empirical analyses. The point of view is that one
of a US investor who is managing a portfolio of domestic and international investments
including traditional and alternative asset classes; in total we investigated 139
assets/strategies. The aim of the paper is to segment them in order to identify the best hedge
for each macroeconomic scenario.
We conclude the section with a remark. In this work we investigate the relationship between
long positions of assets and fundamental factors. Short positions on assets are certainly to
reverse the relationship with factors, hence in principle providing interesting opportunity to
diversify. We work only with long positions because we want to be long on asset risk
premia 6. Indeed our horizon is long term (we deal with strategic asset allocation), and net
short positions may lead to structural negative risk premia over time.
Investigating the behaviour of asset classes to factors dynamics is particularly crucial today as
e.g. we expect inflation will become extremely relevant in the near future. The optimal
portfolio to hedge inflation risk has been studied by Atti and Roache (2008), Amenc et al.
(2009) in developed countries and Brire and Signori (2011) in Brazil.
The paper is organized as follows. In section 1 we briefly review traditional approaches for
asset segmentations. In section 2 we introduce the fundamental factors, and investigate firstly
6
The investment universe is made up mainly by long positions on assets/strategies that deliver a risk premium in
the long term. However we included also some assets/strategies that are not supposed to bear any long-term
return as fx-rates, equity spread indices (long equity sectors and styles and short the S&P500).
Traditional approaches for asset segmentations consist in clustering together assets according
to:
x similarity of asset type (bonds, equities, commodities, etc.);
x risk argument: assets are mainly segmented according to their volatility, VaR, or CVaR;
x correlation measure: assets are similar if they are correlated (metric mainly based on
standard Pearsons coefficient).
The recent crisis highlighted some limitations of the above mentioned methods. Let us
consider for example the Italian BTP:
x The debt crisis in the Eurozone was characterized by the decoupling of the Euro core
bonds from the peripherals: despite the similarity of asset type between the Italian BTP
and Euro core bonds (e.g. German Bund), it is evident that their dynamics during the
debt crisis have been profoundly different.
x As shown in Pola and de Laguiche 2012, the volatility of BTP increased dramatically
during the debt crisis. Any classification of asset classes based on risk argument suffers
from the instability of asset volatility.
x Another effect of the decoupling of the Eurozone is the change in the correlation
between bond and equity markets: in Pola and de Laguiche 2012, we showed clearly
that the effect of the debt crisis on BTP was to make it positively correlated to the
equity market.
We worked in a different direction: our goal is to cluster together assets that exhibit similarity
in their response to changes in macroeconomic variables and stress indicator. Assets
8
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.
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.
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.
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
12
Inflation swaps and ZEW indicators permit one to evaluate expectations on inflation and growth.
Unfortunately time-series are very short, and do not allow one to perform a robust statistical analysis.
13
It is worth mentioning that our point here is simply to name a specific time interval as rising or falling; the
evaluation is ex-post and we do not need to forecast market scenarios.
2.2 Are fundamental factors orthogonal? Is the relationship stable over time?
The first issue to address is whether the identified fundamental factors are independent or
redundant. For the sake of simplicity let us name the four factors (CPI, real GDP, PMI, and
market stress) as f1, f2, f3, f4. We remind the reader that fundamental factors are expressed
by Boolean time-series of +1 and -1, hence the standard Pearsons correlation coefficient is
not applicable 14.
We quantify the dependency between fundamental factors as the distance between the joint
probability P(f1, f2, f3, f4) and the product of marginal probabilities P(f1)P(f2)P(f3)P(f4):
x a non-zero distance signals a statistical relationship among some factors. The Kullback-
Leibler distance (KL distance in the following) permits one to quantify the distance
between two probability measures (Cover and Thomas 2006; Annex B for a brief
review), and hence the dependency among factors.
This approach does require large amount of data samples (see Annex B for a discussion on
data sampling); moreover it is not very appropriate for our sample because time-series
inceptions are not homogeneous across factors. For this reason we preferred to investigate
pair-wise orthogonality 15.
14
The standard Pearsons correlation coefficient measures the relationship between two continuous variables; it
cannot be applied to discrete random variables.
15
Indeed pair-wise orthogonality does not guarantee full independence among factors. But usually, pair-wise
relations are more relevant than higher order ones.
12
The scenarios are well distributed; this means that rising and falling scenarios are about
equally likely (see Annex A for series inceptions).
In Chart 5 we compute the six pair-wise joint probabilities. Probabilities are estimated from
the frequency table. In order to have perfect orthogonality between factors, we expect to find
a joint probability close to 25%, this corresponding to rising and falling scenarios of the two
factors being completely unrelated each other. Among the computations, those involving the
stress scenario are less significant because of the short data sample.
x CPI vs. real GDP. The equilibrium is slightly moved towards off-diagonal elements:
this means that rising (resp. falling) inflation scenarios are more likely linked to falling
(resp. rising) growth ones;
x CPI vs. STRESS. The off-diagonal figures are more likely than the diagonal ones;
x Real GDP vs. STRESS. The most likely joint scenario is falling real GDP and falling
stress;
x Real GDP vs. PMI. The equilibrium is clearly moved towards the diagonal elements,
indicating a strong (positive) relationship between them;
In the second last row in Chart 5 we reported the KL distance: we remind the reader that the
KL distance is zero when the variables are independent (i.e. the joint probability is identical to
the product of marginal probabilities), and greater than zero otherwise (see Annex B for more
details). Test of significance has been achieved through a non-parametric approach 16.
x orthogonality among real GDP, CPI and market stress holds only as a first
approximation.
The main take-home message of this section is that fundamental factors are not strictly
orthogonal: the analysis shows a deviation from independence in some pairs taken from CPI,
real GDP, PMI, and market stress. We believe that, at least, part of the anomaly can be
explained by the sample specificity of the analysed period: in order to assess the significance
of the result, a deeper historical analysis is advisable.
16
In order to clarify the approach, let us consider the first pair: CPI vs. real GDP. In order to compute the KL
distance we only need the two Boolean sequences indicating rising or falling scenarios. In order to test
significance, we firstly randomly reshuffled the Boolean sequence of GDP, and computed the KL distance. We
performed this procedure 10,000 times, in order to get 10,000 computations of the KL distance. It is worth
stressing that computations from reshuffled pairs should not signal any relationship between CPI and real GDP;
any possible relationships are certainly spurious and originate from the data sample. This way allows one to
evaluate if the KL distance is significant or spurious.
14
Chart 6. Historical joint and marginal probabilities (focus on CPI and GDP)
Joint Probability of rising-falling scenarios on CPI and real GDP - 10 yr rolling analysis in US
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
1963
1965
1966
1968
1969
1971
1972
1974
1975
1977
1978
1980
1981
1983
1984
1986
1987
1989
1990
1992
1993
1995
1996
1998
1999
2001
2002
2004
2005
2007
2008
2010
2011
(- ; -) (- ; +) (+ ; -) (+ ; +)
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1963
1965
1966
1968
1969
1971
1972
1974
1975
1977
1978
1980
1981
1983
1984
1986
1987
1989
1990
1992
1993
1995
1996
1998
1999
2001
2002
2004
2005
2007
2008
2010
2011
CPI - CPI +
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1963
1965
1966
1968
1969
1971
1972
1974
1975
1977
1978
1980
1981
1983
1984
1986
1987
1989
1990
1992
1993
1995
1996
1998
1999
2001
2002
2004
2005
2007
2008
2010
2011
In this section we illustrate the methodology to assess the relationship between asset returns
and factors scenarios. Ideally it is desirable to investigate the relationship between assets and
each factor separately (e.g. studying the dependency of nominal bonds versus movements of
inflation, keeping the dependency on growth and market stress fixed). This is not possible
because in financial markets all factors act synchronously to determine asset prices; moreover
factors are orthogonal only as a first approximation. In small data sample, it might be the case
that some factors are more important than other (e.g. inflation in 70s, growth in the Great
Depression, market stress in 2008), leading potentially to misleading results. A longer data
sample enables this problem to be alleviated.
We start with a nave approach: for each asset and factor we divide asset return in
correspondence of rising and falling scenarios. In Chart 7 we perform this exercise for the US
treasury. Chart 7 (left panel) reports the (conditional) average of excess-returns over 3-month
T-bills in different scenarios 17; the chart includes also the unconditional average 18. Two main
findings emerge:
x nominal bonds present positive risk premia;
x nominal bonds exhibit clear characteristics in terms of factors scenarios: they are good
hedge against falling inflation, falling growth and rising stress scenarios.
1.80% 15%
1.60%
1.40% 10%
1.20%
1.00% 5%
0.80%
0.60% 0%
0.40%
0.20% -5%
0.00%
-0.20% -10%
CPI GROWTH STRESS CPI rising CPI falling GROWTH GROWTH STRESS STRESS
rising falling rising falling
rising avg falling
17
Cash level is mostly determined by central bank, whereas asset excess return over cash responds mainly to
surprises in expectations of macroeconomic variables and market stress.
18
The average figure for the stress indicator is significantly different from those of CPI and growth because of a
shorter data sample. Growth in this chart refers to real GDP. Observations are quarterly.
16
The parametric and non-parametric tests allow one to define the polarization coefficient P 20.
(We name with P param and P non-param respectively the parametric and the non-parametric
indicators; see Annex D for all the technical details regarding the derivation of the indicators.)
The main properties of P are:
x sign of P indicates a preference of the asset class for scenario +1 (rising) or -1 (falling);
19
We report the maximum and minimum observations, as well as figures within +1 standard deviation and -1
standard deviation.
20
We label the polarization measure with capital rho P: this is to indicate the analogy with the standard
Pearsons correlation coefficient . P shares with to be in between -1 and 1, main difference is that P measures
the relationship between a continuous variable (asset returns) and a Boolean variable (+1 and -1 standing
respectively for rising and falling scenarios), whether compares two continuous variables.
In the following we report P param and P non-param for the US Treasury in Chart 7:
x Falling Inflation: P param = -99.88% ; P non-param = -99.85%;
x Falling Growth: P param = -99.65% ; P non-param = -99.85%;
x Rising Stress: P param = 99.67%; P non-param = 99.53%.
The figures above demonstrate that nominal bonds are sensitive to falling inflation, falling
growth and rising stress with a high confidence level. Indeed the parametric and non-
parametric measures are coherent.
In this section we compute the polarization coefficient P for a large investment universe from
a US investors perspective. We firstly introduce a two-dimensional representation (the
polarization plan), then we move towards a more comprehensive three-dimensional chart that
expresses polarization of assets to the three fundamental factors (the polarization cube).
4.1 Asset class universe
We consider the point of view of a US investor investing in domestic bonds (nominal bonds,
inflation-linked bonds, corporate bonds), international emerging market bonds (local currency
and hard currency), domestic equity, domestic equity styles (value, growth, large, small) and
sectors, commodities, volatility, hedge fund strategies (including CTAs, long-short trend
follower strategies on long term nominal bonds and fx-rates, long-only trend follower
strategies on a diversified investment universe), and a global risk parity balanced strategy.
Moreover we considered zero-duration credit indices (in order to disentangle credit dynamics
from interest rate ones), alpha equity indices (long equity sectors and styles, short S&P 500),
and beta-neutral equity indices (long equity sectors and styles, short S&P 500 to neutralize
beta to S&P500). The point of view is that one of a US investor who wants to understand the
relationship of asset returns (domestic and international) versus his own countrys
18
21
Indices obtained taking the excess returns over government bond of same maturity (Merrill Lynch
computation).
22
Synthetic indices obtained going long on the sector or style and going short 100% on the S&P 500; calculation
on a monthly basis.
23
Synthetic indices obtained going long on the sector or style and going short on the S&P 500 for a quantity
equal to the index beta to the S&P 500; calculation on a monthly basis. Beta estimation is on the whole data
sample.
24
Long-short strategy is designed on a monthly basis (the last Thursday of the month), and invests in 7-10-yr
bonds from US, Germany, UK, Japan, Australia, Canada, Italy. Portfolio construction is given by asset historical
volatility for the previous year (risk budget is the same for all the markets). Long-short signals are given
comparing the index with the moving average for the last 100 days: if the price is above (resp. below) the signal
is long (resp. short). Portfolio is not leveraged. Long-short trend follower on fx follows the same investment
scheme. Investment universe is EUR vs. USD, CAD, AUD, GBP, JPY, and USD vs. CAD, AUD, JPY, NZD,
NOK, BRL, MXN, KRW, TWD, ZAR (carry returns indices). Our hypothesis is that strategies are implemented
through derivatives instruments. We considered transaction costs equal to 15 bps for 100% portfolio turn-over.
25
They are labelled in the following as riskparity8 trend and cstra95.
26
Risk-parity allocation is derived within a monthly rebalancing. Risk budget is equally divided between
nominal 7-10-yr bonds (US, Germany, UK, Japan, Australia, Canada), Equity (Euro, US, UK, China, Japan),
Commodity (Gold, Copper, Agriculture, Oil WTI). Each month the portfolio is determined according to asset
historical volatility for the previous year; the portfolio then is leveraged in order to match a target volatility of
8% (estimated on the previous year historical observations). Risk allocation is rebalanced once a month (the last
Thursday of the month), volatility targeting is on a daily basis.
An effective way to represent polarization of asset classes to two factors is to scatter plot their
polarization coefficients P. Chart 8 shows the behaviour of a selection of asset classes versus
CPI and real GDP as a proxy of growth (here, we report P param for quarterly observations).
This section is devoted to presenting the main ideas; in the next section we present a more
comprehensive analysis. We remind the reader that, according to the definition of P, the
closer the spots are to the border, the more significant the asset response to macroeconomic
changes.
Let us summarize the main findings:
x Nominal bonds are excellent hedges in case of falling growth and falling inflation.
x Equity is an excellent hedge in case of rising growth, and to a lesser extent to falling
inflation scenarios. The equity energy sector tracks commodity investments, performing
better in rising growth and rising inflation, whereas real estate stocks are very sensitive
to growth, but weakly related to inflation changes.
x Corporate bonds do not present a homogeneous response. Investment grade bonds are
an excellent hedge in case of falling inflation, to a lesser extent in case of falling
growth. High-yield bonds offer a good preference for rising growth scenario; they are
rather insensitive to inflation changes. Emerging Market Debt in Local currency
(EMDL) behaves like commodity CRB, preferring rising growth and rising inflation.
On the other hand, Emerging Market Debt in Hard currency (EMDH) results almost
blend versus changes in inflation and growth.
x The 1-5 segment of Inflation-linked bonds is an excellent hedge for rising inflation
scenarios (and rather insensitive to growth). The Inflation-linked all maturity index
responds well to falling growth scenarios, and, surprisingly, do not offer a very
significant hedge against rising inflation (in this case the short data sample may have
had a strong impact in the estimation).
27
The strategy invests in a large, diversified, and global investment universe including more than 150 indices
(fx-rates, nominal bonds, equities, commodities). The strategy is designed to invest on the best ten indices in
terms of realized historical Sharpe ratio on a given time interval. Explicit constraints allow us to control portfolio
turnover. The strategy is long-only on all assets, with the exception of fx-rates where we take long-short
positions.
20
AGRICULTURE
60%
US INFLATION 1-5
EMDH
Polarization to Growth ( P )
40%
20%
0%
-20%
-40%
-60%
CORPORATE IG
-80%
US INFLATION ALL MATURITY
GOVY VIX
-100%
-100% -50% 0% 50% 100%
Polarization to Inflation (P )
The analysis above shows clearly that traditional asset classes present mixed characteristics
(e.g. commodities are commonly reported as proxy of inflation, indeed they are sensitive to
growth as well). In order to individuate diagonal proxies 28 for the rising/falling scenarios, it
might be wiser to consider combinations of assets, as follows:
x proxy rising growth: portfolio of equities and commodities;
x proxy falling growth: portfolio of nominal bonds and inflation-linked bonds;
x proxy rising inflation: portfolio of commodities and inflation-linked bonds;
x proxy falling inflation: portfolio of nominal bonds and equities.
28
With this term we mean proxies that exhibit a clear relationship versus a factor, but are rather insensitive to
other factors: e.g. commodity is not a proper proxy of inflation, because it is strongly polarized to growth as
well.
In order to express the polarization of asset returns to the three factors, we scatter plot the
polarization coefficients P in a three-dimensional space. By construction all assets will be
represented by spots in the interior of a cube (see Chart 9; top panel). Different colours refer
to different asset classes, as follows: nominal bond (blue), inflation-linked bonds (magenta),
credit markets (light blue), zero-duration credit markets (cyan), volatility (red), equity
(orange), equity alpha indices (light orange), equity beta-neutral indices (yellow),
commodity (green), fx-rates (brown), hedge funds, trend-follower, CTAs and dynamic
strategies (black). Chart 9 lower panel reports the two-dimensional projection along the stress
axis. Coherently with Chart 8, the plot refers to the parametric evaluation of the polarization
coefficient P for quarterly data; moreover we consider the real GDP as proxy of growth.
22
x asset classes are placed about in the corners of the inflation-growth two-dimensional
projection;
x despite the behaviour of credit markets is not homogeneous, the zero-duration indices
are definitely placed in the rising-growth rising-inflation corner.
x the rising stress slice is defined clustering together assets with P stress between 0.70 and
1;
x the blend stress slice includes assets with P stress between -0.70 and 0.70;
x the falling stress slice is defined for assets exhibiting P stress between -1 and -0.70.
For the sake of simplicity we plot the three slices integrating away the dependency on the
market stress factor (see Charts 10, 11, and 12). At a first glance it is evident that the rising
stress scenario is less populated than the falling stress one: most of the assets rarely hedge
against market stress. In the following we describe each slice in detail, sorting asset classes
according to their joint polarization to inflation and growth.
x Rising-inflation AND rising-growth corner. The best hedge is given by the Information
Technology equity-sector alpha index (the beta-neutral index is placed in the same
corner, but it is less significant in terms of polarization to growth) and, to a lesser
extent, Russell 1000 Growth and S&P 500 Growth (alpha and beta-neutral indices), and
the equity spread Growth vs Value.
Chart 10
Chart 11 includes assets that are rather independent from market stress factor:
x Rising-inflation AND rising-growth corner. The best hedge is given by the Information
Technology equity-sector. Less significant are Inflation-linked 1-5 years (high
confidence level on inflation axis, but poor polarization to growth direction), and
Agriculture.
x Falling-inflation AND falling-growth corner. The best hedge is given by the S&P 500
High Dividend alpha index. Less significant are the Corporate Investment Grade (all
Chart 11
x Rising-inflation AND rising-growth corner. The best hedge is given by the TWDUSD
fx-rate, Industrial Metals, Commodity CRB, Emerging Market Debt in Local currency,
Corporate Investment grade zero-duration index, KRWUSD fx-rate, ZARUSD fx-rate,
Gold Miners stocks, High Yield zero-duration index, Asian dollar index, Crude Oil
WTI, HFRI Equity Hedge, S&P500 Energy equity-sector (pure and beta-neutral
indices), EURUSD fx-rate, HFRI Fixed Income Convertible arbitrage, Gold miners
stocks beta-neutral index, and Small Cap 600 growth (alpha and beta-neutral index). See
Annex E for more details.
x Falling-inflation AND rising-growth corner. The best hedge is given by the S&P 500,
and to a lesser extent, Industrials equity-sector (alpha and beta-neutral indices), S&P
5000 High Dividend, Russell 2000 Value (alpha and beta-neutral indices), Consumer
Staples and Financials equity-sectors. See Annex E for more details.
26
Chart 12
From the previous section, it emerges that nominal bond, inflation-linked bond, equity, and
commodity CRB are placed about in the corners of the polarization cube. This plot suggests
us two relationships among them:
Nominal bond + Commodity CRB 0 (1)
Inflation-OLQNHGERQG(TXLW\ (2)
The equations above indicate that an ad-hoc combination of nominal bonds and commodities
might lead to a blend polarization versus inflation, growth and market stress changes;
similarly inflation-linked bonds and equity might be combined together to neutralize their
environmental biases. On this basis, we will say that nominal bonds and commodity are dual
with respect to inflation, growth and market stress; the same for equity and inflation-linked
Portfolio weights in Eq. (3) have been determined according to the following receipt:
x portfolio weights are inverse proportional to the long term volatility of assets,
x and the portfolio leverage has been set in order to target the long term volatility of CRB.
We then computed P param for the synthetic series (quarterly data) computed according to Eq.
(3):
P param (inflation)=94.21%,
P param (growth)=89.80%,
P param (stress)=-99.41%.
P param (inflation)=99.61%,
28
P param (stress)=-99.99%.
We conclude that the basket accurately represents the response to macroeconomic changes of
CRB, with the differences coming from market specificities that are not captured by the
factors dynamic.
6. Conclusion
The recent crisis poses serious doubts on the effectiveness of diversification to reduce draw-
downs in balanced portfolios: diversification failed when it was mostly needed was the
leitmotif of many institutional investors after 2008. Nevertheless recent portfolio construction
schemes, like risk parity and maximum diversification, make of diversification the kernel of
asset allocation. While most approaches diversify on asset class level, the new challenge in
asset allocation suggests diversifying on fundamental factors that are believed to be the main
drivers of asset price dynamics.
The factors approach provides a new, challenging, and powerful way to interpret financial
markets. Even non-stationary patterns of correlation between asset classes might be brought
back to the dominance of one factor over the others (e.g. inflation in the seventies, growth in
the Great Depression, and stress in 2008).
This new way forces us to rethink asset segmentation. The main assumption of our approach
is:
asset price dynamics can be largely explained in terms of changes
in expectations of macroeconomic variables and market stress.
This statement suggests a new way to segment asset classes according to their similarity in
responding to changes in expectations. In this work, we found that each asset class presents
mixed relationships with fundamental factors. Asset polarization to factors is represented
through their position in a three-dimensional cube:
(i) assets placed in the corners present strong polarization towards the three factors
(e.g. nominal bond, equity, commodity);
(ii) assets at the edges present high polarization to two factors, and rather blend the
response to the third factor (e.g. high yield and real estate stocks);
(i) building asset allocations that are better equipped to navigate different
macroeconomic environments (in Pola & Facchinato 2013 we will explore the
implications in terms of portfolio construction, and we will illustrate the DAMS
investment process);
I would like to thank S. Facchinato and C. Casadei for very illuminating discussions and
suggestions regarding this research topic. The ideas investigated in this paper constitute the
foundation of the DAMS investment process, which has been developed by Amundi Milan
Investment Management and G. Pola (Amundi Paris Quant Research); DAMS is in place at
Amundi Milan within a range of flexible funds (diversified DAMS), absolute return (income
DAMS) and equity funds (equity DAMS) since December 2011. Moreover I would like to
thank Ph. Ithurbide and S. de Laguiche for their contribution which improved the research
study and quality of the manuscript.
32
x Inflation. US CPI Seasonally Adjusted (Bloomberg code CPI INDX Index). Time-series
from December 1953;
x Real Growth. US Real GDP Seasonally Adjusted (Bloomberg code GDP CHWG
Index). Time-series from September 1948;
We investigated 139 assets/strategies as detailed in tables A1, A2 and A3. Time-series are up
to December 2012.
30
The Bloomberg U.S. Financial Conditions Index combines yield spreads and indices from U.S. Money
Markets, Equity Markets, and Bond Markets into a normalized index. The values of this index are Z-scores,
which represent the number of standard deviations that current financial conditions lie above or below the
average of the January 1994-June 2008 period.
The Kullback-Leibler distance (KL distance in the following) measures the distance between
two probability distributions (Cover and Thomas, 2006). Let us consider two random
variables x and y taking values from a discrete set 31; we do not need the set to be of finite
dimension. Let P(x) and P(y) be (respectively) their probability distributions. Let P(x,y) be the
joint probability of outcomes (x,y). x and y will be independent if and only if
P(x,y)=P(x)P(y). This notion of independence is stronger than that one given by the standard
Pearsons correlation coefficient 32. The KL distance permits one to measure the distance
between P(x,y) and P(x)P(y), hence to quantify the degree of dependence. The KL distance is
defined as follows:
P ( x, y )
DP ( x, y ) || P( x) P( y ) P( x, y) log P( x) P( y) , (B1)
x, y
DP ( x, y ) || P ( x) P ( y ) H ( x) H ( y ) H ( x, y ), (B2)
where H(x), H(y) and H(x,y) stand respectively for the Shannon Entropy 34 of x, y, and the
joint variable (x,y). This formulation expresses the KL distance as the difference between the
sum of the variability of variables x and y minus the variability of (x,y). This representation
allows one to write down the following inequality:
31
We present here the KL distance for discrete random variables, but the formalism is more general and can be
applied to continuous random variables (Cover and Thomas, 2006).
32
In fact the standard Pearsons correlation coefficient can even wrongly interpret two deterministic related
random variables: let x be uniformly drawn in the interval [-1, 1], and let y=x2. It can be proven easily that the
Pearsons correlation coefficient is zero in this case. The other important aspect is that the statement according to
which positive correlated assets imply that once one asset increases (decreases) the other increases (resp.
decreases) as well is wrong, as shown clearly in Lhabitant 2011. Finally the Pearsons correlation coefficient is
a linear measure; it completely misses non-linear terms like x2y, xy2, x2y2, etc. The KL distance enables us to
alleviate many of these drawbacks. The only con, as we will see in this section, is that computing a reliable
estimate of KL distance requires large data samples.
33
However it should be stressed that this measure is not symmetric. There are several ways to get a symmetric
KL measure. We prefer here to work with the simplest version of the KL distance.
34
The Shannon entropy for random variable x is defined as H(x)=-x P(x) log P(x).
N N
DP ( x, y ) || P( x) P( y ) log x y ,
N
xy
where N x , N y and N xy are respectively the average number of relevant states 35 of P(x), P(y),
and P(x,y).
N
P ( x1 ,..., x N )
D P ( x1 ,..., x N ) || P ( xi ) P( x ,..., x
1 N ) log N
. (B3)
i 1 x, y
P( x )
i 1
i
Analogously it is easy to derive the upper bound and the formulation of the KL distance in
terms of relevant states.
We conclude this section briefly discussing the estimation problem in Eq. (B3). For the sake
of simplicity let us assume that random variables x i can take only two values (+1 and -1). In
this case the dimension of the response space is 2N. Let us come back to the practical example
with four factors (CPI, real GDP, PMI, market stress). If we require at least 30 observations
(on average) for each state we come out with 3024=480, which is equivalent to 120 years of
quarterly observations. This simple evaluation clarifies why measuring the KL distance on all
the factors might be difficult in practice, and why we prefer to evaluate pair-wise dependency
(in this case we can lower 120 years of quarterly observations to 30 years; i.e. 3022=120).
Nevertheless it is worth pointing out that there is a vast literature which faced the sampling
problem of the KL distance: many contributions might be found within the field of
Computational Neuroscience, where the KL distance was extensively used to study neural
coding of sensory information. A brief review can be found in Pola et al 2002. In Pola et al
2005 we proposed a tight lower bound of the mutual information which can be useful to
delineate robust bounds for the KL distance.
35
Given an entropy measure H=-i Pi log Pi , index i running from 1 to N, the average number of relevant states
DUH GHILQHG DV H[S+ ,W FDQ EH SURYHQ HDVLO\ WKDW LI 3i= 1/M for each i= 1, , M1 0 7KLV DQDO\WLFDO
result justifies the definition. Meucci (2009) introduced an analogous definition.
38
The aim of this section is to provide the methodology according to which we define rising and
falling scenarios starting from the time-series of the fundamental factors (CPI, real GDP,
PMI, market stress). We will follow a heuristic approach 36. In case of quarter data, the
approach consists in:
x CPI. Take the quarterly variations. At each point compare the current quarter variation
to the average of the last four variations (including the current figure). If the last
observation is above the average, we will name the quarter as a rising-inflation quarter
(symbol +1), otherwise as a falling-inflation (symbol -1) quarter.
x GDP. Take the quarterly variations. At each point compare the current quarter variation
to the average of the last four variations (including the current figure). If the last
observation is above the average, we will name the quarter as a rising-growth quarter
(symbol +1), otherwise as a falling-growth (symbol -1) quarter.
x PMI. At each point compare the current quarters figure (average of the three monthly
observations) to the average of the last twelve observations (including the current
quarters figure). If the last observation is above the average, we will name the quarter
as a rising-PMI quarter (symbol +1), otherwise as a falling-PMI (symbol -1) quarter.
x STRESS 37. At each point compare the current quarters figure (average of the three
monthly observations) to the average of the last twelve observations (including the
current quarters figure). If the last observation is above the average, we will name the
quarter as a rising-stress quarter (symbol +1), otherwise as a falling-stress (symbol -1)
quarter.
x CPI. Take the monthly variations. At each point compare the current months variation
to the average of the last twelve variations (including the current figure). If the last
observation is above the average, we will name the month as a rising-inflation month
(symbol +1), otherwise as a falling-inflation (symbol -1) month.
36
Recently Kritzman et al 2012 investigated this issue in the context of Markov switching models.
37
In order to simplify notation, we preferred to compute rising/falling scenarios to the negative of the Bloomberg
Stress indicator: in this way rising stress would be linked to symbol +1 and falling stress to symbol -1.
x PMI. At each point compare the current months figure to the average of the last twelve
observations (including the current figure). If the last observation is above the average,
we will name the month as a rising-PMI month (symbol +1), otherwise as a falling-PMI
(symbol -1) month.
x STRESS35. At each point compare the current months figure to the average of the last
twelve observations (including the current figure). If the last observation is above the
average, we will name the month as a rising-stress month (symbol +1), otherwise as a
falling-stress (symbol -1) month.
40
The aim here is to define a polarization indicator, quantifying how much an asset class
responds to rising or falling scenarios. In order to answer to this question we perform a
conditional historical analysis.
Let us consider that we want to verify if there is a relationship between asset A and
fundamental variable V. Firstly we define symbol +1 and -1 in correspondence of a rising or
falling scenario for indicator V. Secondly we divided the performance sample of A in
correspondence of symbol +1 and -1, we will name them respectively as A + and A - . We
question whether there is an ordering between sets A + and A - . If we prove that A + is on
average significantly greater (resp. lower) than A - , then we conclude that asset A is a good
candidate to hedge against rising (resp. falling) scenario of fundamental factor V.
x Parametric. t-test of the null hypothesis that data in the samples A + and A - are
independent random samples from normal distributions with equal means and different
and unknown variances (Behrens-Fisher problem), against the alternative that the means
are not equal. The t-statistics, under the null hypothesis, is distributed according to an
approximate Students t distribution with a number of degrees of freedom given by
Satterthwaite's approximation (see Timm, 2002). The p-value 38 is reported as p param.
It is worth stressing that p param >0.50 (resp. p param <0.50) indicates a preference of asset A for
scenario +1 (resp. -1); the same holds for p non-param .
38
We considered the general case of different variances for sample A+ and A-, and one-tailed distribution.
39
Asset performance is measured at time k+1, whereas macroeconomic scenarios are evaluated at time k.
Amenc, N., Martellini, L., Ziemann, V., 2009. Alternative Investments for Institutional
Investors, Risk Budgeting Techniques in Asset Management and Asset-Liability
Management. The Journal of Portfolio Management, 35(4), p. 94-110.
Atti, A.P., Roache, S.K., 2009. Inflation Hedging for Long-Term Investors. IMF Working
Paper 09-90, April.
Brire, M, Signori, O., 2011. Hedging Inflation Risk in a Developing Economy: the case of
Brazil, Amundi Working Paper WP-012-2011.
Cover, T. M., Thomas, J. A., 2006. Elements of Information Theory, Wiley series in
Telecommunications and Signal Processing.
Kritzman, M., Page, S., and Turkington, D., 2012. Regime Shifts: Implications for Dynamic
Strategies, Financial Analysts Journal, 68 (3), 22-39.
Pola, G., and de Laguiche, S., 2012. Unexpected Returns. Methodological considerations on
Expected Returns in Uncertainty, Amundi Working Paper WP-032-2012.
Pola, G., and Facchinato, S., 2013. Managing uncertainty with dams. Balancing
macroeconomic scenarios, in preparation 2013.
Pola, G., 2013. Rethinking strategic asset allocation in terms of diversification across
macroeconomic scenarios, Amundi Special Focus.
Pola, G., Petersen, R. S., Thiele, A., Young, M. P., Panzeri S., 2005. Data-robust tight lower
bounds to the information carried by spikes times of a neuronal population, Neural
Computation, 17, 1-44.
Pola, G., Schultz, S. R., Petersen, R. S., Panzeri, S., 2002. A practical guide to information
analysis of spike-trains. Book chapter in A practical guide to neuroscience databases and
associated tools, Kluwer, London.
October 2013
Interest in sovereign wealth funds (SWFs) as key players in financial markets has grown
rapidly over the last years. A large number of sovereign wealth funds (SWFs) have been set
up to collect and manage the tax revenues that states receive from natural resources or
exports. SWFs serve various economic objectives, such as budget stabilization, diversification
from commodities, saving for future generations. They may also pursue political strategies,
(Avendano and Santiso, 2009; Ang, 2012). SWFs can be managed by different institutional
A large body of empirical research has analysed the public investment strategies of sovereign
wealth funds and their performance. Although this takes into account only a fraction of SWF
investments, mainly equity stakes in listed firms, it shows that SWFs tend to invest in large
foreign firms, often in the finance and energy sectors, with low diversification and poor
medium-term performance (Bernstein et al., 2013; Chhaochharia and Laeven, 2009; Dyck and
Morse, 2011; Bortolotti et al., 2013). SWFs also served as investors of last resort during the
last crises, intervening to support their domestic financial markets (Clark and Monk, 2010;
(2009a and b), Brown et al. (2010), Martellini and Milhau (2010) have addressed the optimal
benchmark. But the example of the recent crisis clearly shows that other sovereign liabilities
have to be taken into account: debt, contingent liabilities, etc. Moreover, when a government
is short of liquidity to meet its debt payments, the SWFs assets are often available to
substitute for the funds initially earmarked for this purpose. In 2010, for example, in the wake
of the subprime crisis, Russia, Ireland, Kazakhstan and Qatar used SWFs or public pension
2014), we proposed estimating the whole sovereign economic balance sheet using the theory
of contingent claims and considering the joint management of all sovereign assets and
liabilities in an ALM framework. The sovereign is considered in the broad sense, including
all the related institutions (budgetary government, central bank, SWFs, pension funds and
Managing the wealth of a sovereign is not very different from managing the wealth of an
individual (Merton, 1969; Bodie et al., 1992; Bodie et al., 2008), a pension fund (Bodie et al.,
2009) or a foundation (Merton, 1993). The central government receives tax revenues each
year. Part of this income can be spent, and the residual saved in the SWF, central bank
reserves, or the public pension fund. How much should be saved and how it should be
invested is a classic ALM problem. The optimal allocation and expenditures of the sovereign
will crucially depend on the nature and size of its assets and liabilities, and the sources of their
uncertainty. Merton (1993) solved a similar problem for a university endowment fund. In our
sovereign case, the optimal sovereign allocation differs slightly. It can be broken down into a
performance-seeking portfolio and three additional portfolios hedging for the variability of
the fiscal surplus and external and domestic debt. Financial management of government
resources and expenditures raises difficult issues in practice. Standard macroeconomic tools
are ill-suited to estimating sovereign economic balance sheets. Most of the macroeconomic
variables monitored at present describe flows, not stocks, and are unsuitable for valuing
intangible assets such as human and natural capital (Aglietta, 2010). Moreover, traditional
macroeconomic data lack a significant dimension, namely risk (Gray et al., 2007). This lack
of aggregate data makes it difficult to coordinate sovereign wealth management with fiscal
empirical point of view, and we show how real-life SWFs asset allocations differ from
theoretical ones (Section 2). We present our conceptual framework for optimal sovereign
wealth management (Section 3). We then discuss its practical implementation, giving country
examples and suggesting possible institutional arrangements that would enable efficient
While there is an abundant literature on the allocation of foreign-exchange reserves, there are
only a few papers devoted to SWF optimal asset allocation. The two topics are nevertheless
interlinked, since the funds invested in SWFs often come from foreign exchange reserves. We
start this section with a state of the art review for these two topics.
Caballero and Panageas (2005a and b), Beck and Rhababi (2008), Beck and Weber (2011)
examine the optimal allocation of foreign exchange reserves in the event of a sudden
slowdown in private capital inflows (sudden stop). The central bank uses its reserves to
repay the short-term foreign debt and minimize the variance of its portfolio in real terms. In
this framework, optimal portfolio weights depend, in addition to the standard minimum
variance demand term, on the extent to which the assets can be used to hedge against sudden
stops. In their empirical investigation, Caballero and Panageas (2005b) suggest the use of
assets based on the S&P 500 implied volatility index, providing efficient protection against
sudden stops in emerging markets, often linked to global liquidity crises. Beck and Rhababi
(2008) show that dollar-denominated assets are a better hedge for global stops and for
regional stops in Asia and Latin America, whereas the euro is a better hedge in Emerging
Europe.
that governments assign to SWFs in practice. Aizenman and Glick (2010) compare the
optimal allocations of foreign-exchange reserves by the central bank and by an SWF, which
have different objectives: (1) reducing the probability of sudden stops for the central bank,
and (2) maximizing the expected utility of a domestic representative agent for the SWF. In
this framework, the authors show that the SWF must hold a riskier foreign-asset allocation
than the central bank. Brown et al. (2010) propose an allocation model for different types of
SWFs, with either a pure return objective or a fiscal smoothing objective. Scherer (2009a and
tradable wealth, and shows that in this case the optimal asset allocation of the SWF should
include a hedging demand against commodity price variations. Martellini and Milhau (2010)
propose a dynamic asset allocation framework for SWFs having liabilities exhibiting inflation
indexation. In a recent study (Bodie and Briere, 2013), we proposed a framework for optimal
asset allocation of sovereign wealth, taking explicit account of all sources of risk affecting the
sovereigns balance sheet. We used Mertons approach (1974) to estimate the process of the
country's assets, and then we optimized the balance sheet using the ALM approach.1 This
framework expanded previous results on SWFs optimal asset allocations by introducing three
additional sources of risk affecting the sovereign balance sheet. We showed that the optimal
hedging demand terms for the variability of the fiscal surplus and external and domestic debt.
Comparing theory on optimal SWF asset management with real-life data could provide
interesting insights. Unfortunately, a large portion of SWF investments remains private, and
most authors concentrate on SWFs equity interests in listed companies. Dyck and Morse
(2011) and Bernstein et al. (2013) show that SWF portfolios tend to be insufficiently
1
Das et al. (2012) offer a literature review on the use of ALM techniques applied to sovereign fund
management.
telecommunications (Bertoni and Lugo, 2012, Bortolotti et al, 2013; Chhaochharia and
Laeven, 2009), contradicting the principles of sound diversification. They also tend to take
large stakes in companies facing financial difficulties, both abroad and domestically
(Raymond, 2010). During the subprime crisis, some SWFs3 played the role of investor of last
resort, rescuing major Western banks or recapitalizing their home equity markets. The
performance of those investments is generally poor in the long run, even if the announcement
of SWF investments yields positive abnormal stock-price returns in the very short run
3. Conceptual Framework
We consider the concept of sovereign in the broad sense, including not just the states
budgetary institutions and monetary authorities (central bank), but also the other institutions
related to it, such as pension funds, SWFs and state-owned enterprises.4 The sovereign has a
multitude of objectives. Some are purely financial, such as debt repayment and setting aside
foreign exchange reserves to cope with liquidity crises. Others are social, including pensions
and financing of social services (infrastructure such as hospitals, roads, education, defence,
etc.). Still others are economic, such as investment in key sectors or industries for future
growth. To achieve its objectives, the sovereign has a variety of resources, particularly future
2
Even when the country is producing commodities
3
For example in China, Hong Kong, Kuwait, Qatar, Russia, Saudi Arabia and Singapore.
4
Distinctions among various state entities are less and less meaningful, as recent crises have shown. In 2010
several countries turned to public institutions for assistance in coping with the crisis-related credit crunch. Some
countries used the assets of SWFs or national pension funds to invest in bank deposits (Russia and Kazakhstan)
or to support equity-market liquidity (Kuwait). Others used the resources to directly recapitalise ailing banks
(Ireland, Kazakhstan and Qatar). For this purpose, states modified their funds investment rules on a
discretionary basis, exposing them to new risks. Finally, in some countries with greater borrowing capacities, the
state tweaked the funds regulations to allow them to buy a larger share of the sovereign debt. These recent
examples clearly show that a state facing a crisis can elicit contributions from the off-budget entities that it
owns or controls in order to meet its short-term obligations without unduly worsening the fiscal deficit.
seigniorage, and possibly a stock of financial assets (foreign exchange reserves, SWF assets,
The sovereigns global economic balance sheet is key to a full understanding of its situation
and risks (Gray et al., 2007). The idea is to estimate all the states assets and liabilities at
market price, and to measure the risks (volatility and sensitivity to economic shocks)
associated with each balance sheet item. Just as a companys balance sheet is regularly used
to assess the risk of bankruptcy (Merton, 1974 and 1977; KMV, 2002), the same analytical
framework may be applied to a state. This is useful not only with regard to the states debt
repayment capacity (Gray et al., 2007; Gray and Malone, 2008), which is obviously a minimal
objective, but more generally, as we shall see, with regard to its ability to meet its long-term
social and economic objectives. Table 1 gives a simplified example of a sovereign balance
sheet.
ASSETS LIABILITIES
Foreign reserves, gold, Special Drawing Base money
Rights
Local currency debt
Pension fund assets
Foreign currency debt
SWF
Pension fund liabilities
Other public-sector assets (state-owned
enterprises, real estate) Contingent claims: implicit guarantees (to
banks, etc.)
Present value of future taxes, fees,
seigniorage Present value of expenditures on economic
and social development, security,
government administration, benefits to other
sectors
market price and volatility of all its component assets and liabilities separately. However, to
do this, the present value of future income and expense flows has to be estimated. An
alternative method is to estimate the markets valuation of the balance sheet, as described by
Merton (1974, 1977) and Gray et al. (2007). An implied value for the sovereign's assets can
be estimated from the observed prices of liabilities. To do this, it is necessary to rearrange the
balance sheet entries and adopt an integrated presentation, subtracting the present value of
expenses from the present value of income, and subtracting the value of contingent liabilities
from assets. The two liabilities can then be valued as contingent claims on sovereign assets.
The foreign currency debt is considered as a senior claim, and the local currency debt plus
base money as a junior claim, which can be modelled as a call option on the total value of
the sovereign's assets. The value of the sovereigns assets and their volatility can then be
estimated as a function of the default barrier (promised payments in foreign currencies), (Gray
From a theoretical standpoint, managing the wealth of a sovereign is similar to managing the
wealth of an individual (Merton, 1969; Bodie et al., 1992; Bodie et al., 2008), a pension fund
(Bodie et al., 2009) or a foundation (Merton, 1993). The sovereign receives tax revenues each
year. Part of these revenues are spent, and the residual is saved in SWFs, central bank
reserves, or public pension funds. Determining how much should be saved and how it should
We assume that the sovereigns objective is to maximise its expected utility, which is a
function of its Global Sovereign Surplus (GSS),5 depending on the allocation of the
5
Measured as sovereign assets minus sovereign liabilities.
crucially depend on the nature and size of the fiscal asset and unconditional liabilities, and the
sources of their uncertainty. Bodie and Brire (2014) solve this problem analytically and show
that the optimal portfolio w* can be broken down into a performance-seeking portfolio and
three hedging demand terms for the variability of the fiscal surplus and external and domestic
debt:
1 (1 ) 1 (1 ) 1
w* = FA1 FA ,t FA FA, FS + FA1 FA, FL + FA FA, DL (5)
( 1)
with FA the vector of annualized expected returns of the n financial assets in the portfolio
over the investment horizon, FA their covariance matrix, the fraction of total sovereign
assets dedicated to financial wealth (the remainder is the fiscal surplus), the fraction of total
sovereign liabilities dedicated to foreign debt (the remainder is domestic debt), FA, FS FA, FL ,
FA, DL the covariance of the financial asset returns with the fiscal surplus, foreign liabilities
These results shed new light on the optimal allocation of the sovereigns wealth. We
generalize previous results on SWFs asset allocations by introducing three additional sources
of risk affecting the sovereign balance sheet. Martellini and Milhau (2010) express the SWFs
preference in real terms and observe a hedging demand against realized inflation. Scherer
(2009a and b) identifies the optimal asset allocation of an SWF with non-tradable wealth and
observe a hedging demand against oil price variations. In a more general framework, taking
explicit account of all sources of risk affecting the sovereign balance sheet, three hedging
demand terms are added to the speculative portfolio. We recommend taking into account not
only the risks from inflation and fluctuations in natural resource prices, which both influence
6
We disregard other potential macroeconomic decision variables (tax rate, etc.), considered as constant, in order
to concentrate on the asset allocation choice.
from foreign and domestic liabilities. Moreover, the fiscal surplus variability is influenced not
only by commodity prices and inflation volatility, but also by the sovereigns policies on
4. Practical Implementation
The practical implementation of sovereign ALM raises several difficulties. Traditional public
finance data are often incomplete and ill-suited to accurately estimation of the sovereign
economic balance sheet. This lack of data compromises the coordination of sovereign wealth
management with fiscal policy, monetary policy and public debt management. We discuss
To implement sovereign ALM, what really needs to be measured is the actual nature of
macroeconomic and financial risks, with their non-linear features (contingent liabilities
modelled as options, etc.), and the accumulation phenomena that lead to systemic risks. Flow
of funds statistics available in many countries provide balance sheet estimates of the
government sector but do not fully correspond to what is actually needed. The definition of
the government entity differs between countries8 and may not correspond exactly to our
broad definition of the sovereign. The IMF's GFS database, created in 2001, remedies these
differences with a unified base of 153 countries data on government balance sheets, with a
7
This leads to another important difference from the previous literature. In our framework, the variability of the
flow of revenues from the sale of natural resources needs to be hedged, not the fluctuations in commodity prices
themselves (Scherer (2009a, 2009b)). This has important implications, as the fiscal surplus may not have a
sensitivity of one to natural resource prices, as we will see in our estimation for Chile in Section 3.
8
In the US, the Flows of Funds statistics consider state and local governments (excluding employee retirement
funds), the federal government (including government-owned corporations and agencies that issue securities
individually) and the monetary authority. In Europe, the European Central Bank and Eurostat Euro Area
Accounts have a more restrictive definition. The general government sector comprises only central, state
(regional) and local government and the social security or pension funds belonging to it. It does not include
public enterprises, which are included in the corporate or financial sector and cannot be disentangled from it.
limitations. There is no evaluation of the present value of future tax revenues, or expenditures.
the financial sector and implicit guarantees to provide social benefits when various needs
arise. Finally, these data, which are purely accounting-based and generally available on an
annual basis, are not sufficient to measure the risks associated with each item. In the case of
sovereign balance sheets, risks are related on the one hand to market price fluctuations (for
commodities, exports, wage costs, etc.) that cause the governments income and expenditures
to fluctuate, and on the other hand to inventory changes (natural resource depletion,
In 2000 the World Bank took the unprecedented step of measuring the wealth of nations
(World Bank, 2006 and 2011). The total wealth of each nation is estimated as the present
value of future flows of consumption. Consumption levels are based on past historical data
but are adjusted to be sustainable.10 Total wealth is broken down into: (1) produced capital
(machinery, structures and urban land), (2) natural capital (energy resources, mineral
resources, timber resources, non-timber forest resources, cropland, pastureland and protected
areas) and (3) intangible capital (human, etc.), calculated as a residual, the difference between
total wealth and the sum of produced and natural capital. These data are a very useful
supplement to the existing figures because they provide an estimate of stocks11 of natural
resources and intangible assets. World Bank estimates of natural and human capital can be
used to estimate the present value of the fiscal surplus, given a certain level of desired
9
It comprises not just the central government budgetary authority but also the central bank, SWFs, pension
funds, deposit insurance funds, state-owned enterprises, subnational governments and other government
agencies.
10
For years when adjusted net savings are negative, the actual consumption rate is added to adjusted net savings.
11
Flow variables are also available: depletion of natural resources, investment in education, domestic net
investment.
2006 and 2011 reports and are not available as a historical series.
between institutions that control sovereign assets and sovereign liabilities (at least the central
bank, the debt management office, the treasury and the ministry of finance). What the most
efficient institutional arrangement would be is still an open question, and the few country
examples show that very different organizations are possible. New Zealand, Canada,
Denmark, Britain, South Africa and Turkey are the handful of countries that have made
significant steps in the direction of developing an ALM framework. In New Zealand and
South Africa, there is a specialized asset-liability management unit that analyses the
sovereigns balance sheet. In New Zealand, the mandate of the debt management office is to
keep the net foreign currency position close to zero, explicitly matching foreign currency
assets and liabilities and hedging exchange-rate movements. In Canada, ALM was introduced
for the tactical management of foreign reserves in 1997, with the goal of minimizing currency
and interest-rate risks by matching the assets to the liabilities funding them. In Turkey, debt
management is also defined in an ALM framework, in close cooperation with the reserve
management office.
In most of the example countries cited (Canada being an exception), the ALM exercise has
been performed by the debt management office, already responsible for cash management and
treasury services. This is not without drawbacks since the issuance of government debt might
also respond to other, possibly conflicting, objectives. Government debt has public good
characteristics, including setting the risk-free yield curve and providing highly liquid
securities. In Australia and Norway, for example, the government decided to continue debt
issuance even though there was no need for government borrowing, because of the
the asset management offices would also make sense. The example of Canada, which gave the
central bank tactical reserves management office an ALM mandate, is a good example of this.
But responsibility for the wider government balance sheet would sit uneasily with central
bank independence, and there could be potential conflicts of interest with monetary policy.
The sovereign wealth fund would actually be an excellent candidate for the job of
implementing the sovereign ALM. In many countries, this may be facilitated by the fact that
the finance ministry is responsible both for debt issuance and fiscal policy and for
In any case, a coordinated approach to the management of the national balance sheet would
necessitate central responsibility. Probably the most realistic scenario would be to encourage
more links and consultation between the different agencies, with detailed instructions from
the ministry of finance. South Africa has organized such a framework with a common
committee bringing together the South African reserve bank and the treasury. When South
Africa had a net negative forward currency position in the late 1990s, a strategy was
developed jointly by the reserve bank and the treasury to bring down this exposure. The
finance ministry might be the best candidate to lead this coordination, but the optimal
institutional arrangement may in the end depend on the political organization of each country.
5. Conclusion
This paper presents an analytical framework for sovereign wealth and risk management,
extending the theory of contingent claims analysis, and discusses its practical implementation.
A complete approach to the sovereign balance sheet is necessary to fully understand the
country's risks and determine how it can best manage its wealth. This supposes the broadest
possible definition of the sovereign, including, in particular, entities subordinated to the state,
enterprises. The reason is that the funds, even if located in different entities, become fungible
if a crisis arises. This approach also requires all balance sheet items, both assets and liabilities,
as well as their risks, to be measured precisely. To do this, it is necessary to measure not only
the sovereigns financial wealth, but also its human and natural capital. Similarly, a relatively
contingent liabilities are also needed. A sovereign ALM strategy can thus be developed for
managing asset risks in a way that is consistent with the sovereign entitys liabilities. One
under direct state control. The optimal allocation of sovereign wealth should involve a
performance-seeking portfolio and three hedging portfolios for the variability of the fiscal
Our ambitious approach has limitations. First, to concentrate on asset allocation, we consider
many more policy instruments, including taxation level. It can also inflate or repudiate its debt
(Landon-Lane and Oosterlinck, 2006). A general equilibrium model endogenizing all of the
states decision variables would be more realistic, but also much more complex. Second, the
regular basis. Moreover, strong coordination is needed between the sovereign entities. This
coordination involves the institutions that manage both sides of the balance sheet: the central
bank and sovereign wealth fund on the asset side, and the debt management office on the
liability side. The ministry of finance is particularly well positioned as a central institution to
facilitate this coordination. However, even if the implementation of the ALM framework for
SWF asset allocation is an unfeasible first-best solution for many countries, far removed from
current practice, it can nevertheless be thought of as providing useful guidelines for efficient
of sovereign assets and the hedging of important risk factors affecting the sovereign balance
sheet.
The authors are grateful to Ariane Szafarz for her comments on a previous version of this
paper.
References
Alfaro, L., Kanczuk, F., 2009. Optimal reserve management and sovereign debt, Journal of
International Economics, 77(1), 23-36.
Aizenman J. and Glick R., Asset Class Diversification and Delegation of Responsibilities
between Central Banks and Sovereign Wealth Funds, NBER Working Paper, N16392.
Ang A. (2012), The Four Benchmarks of Sovereign Wealth Funds, in Bolton P., Samama
F., and Stiglitz J., eds., Sovereign Wealth Funds and Long-Term Investing, p 94-105,
Columbia University Press.
Avendano R. and Santiso J., 2009, Are Sovereign Wealth Funds Investments Politicall
Biased? A Comparison with Mutual Funds, OECD Development Center Working Paper, N
283.
Bagliano F. C., Fugazza C. and Nicodano G. (2009), Pension Funds, Life-Cycle Asset
Allocation and Performance Evaluation, Univesita di Torino Working Paper, March.
Beck R. and Rahbari E. (2008), Optimal Reserve Composition in the Presence of Sudden
Stops, the Euro and the Dollar as Safe Haven Currencies, European Central Bank Working
Paper, No 916, July.
Beck R. and Weber S. (2011), Should Larger Reserve Holdings Be More Diversified?,
International Finance, 14(3), p. 415-444.
Bernardell C., Cardon P., Coche J., Diebold F.X. and Manganelli S. (2004), Risk
Management for Central Bank Foreign Reserves, European Central Bank Publication, May.
Bernstein, S., Lerner J. and Schoar A. (2013), The Investment Strategies of Sovereign
Wealth Funds, Journal of Economic Perspectives, 27(2), April, p. 219-238.
Bodie Z. and Brire M. (2014), Sovereign Wealth and Risk Management: A Framework for
Optimal Asset Allocation of Sovereign Wealth, Journal of Investment Management,
Forthcoming.
Bodie Z., Merton R. and Samuelson W.F. (1992), Labor Supply Flexibility and Portfolio
Choice in a Life Cycle Model, Journal of Economic Dynamics and Control, 16, p. 427-449.
Bodie Z., Ruffino D., and Treussard J. (2008), "Contingent Claims Analysis and Life-Cycle
Finance." American Economic Review, 98(2), p. 29196.
Bortolotti B., Fotak V., Megginson W.L. (2013), The Sovereign Discount in Sovereign
Wealth Fund Investments, University of Oklahoma Working Paper.
Brown A., Papaioannou M. and Petrova I. (2010), Macrofinancial Linkages of the Strategic
Asset Allocation of Commodity-Based Sovereign Wealth Funds, IMF Working Paper, No
10/9, January.
Caballero R.J. and Panageas S, (2005b), A Quantitative Model of Sudden Stops and External
Liquidity Management, NBER Working Paper, N 11293.
Chhaochharia V. and Laeven L., (2009), The Investment Allocation of Sovereign Wealth
Funds, SSRN Working Paper, N1262383, available at:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1262383
Clark G.L. and Monk A.H.B. (2010), Government of Singapore Investment Corporation:
Insurer of Last Resort and Bulwark of Nation State Legitimacy, Pacific Review, 23(4), p.
429-451.
Das U.S., Lu Y., Papaioannou M. G., and Petrova I. (2012), Sovereign Risk and Asset and
Liability ManagementConceptual Issues, IMF Working Paper, No 12/241, October.
Dyck I.J. and Morse (2011), Sovereign Wealth Fund Portfolios, Chicago Booth Research
Paper, N11-15.
Gray D.F., Merton R.C. and Bodie Z. (2007), Contingent Claim Approach to Measuring and
Managing Sovereign Credit Risk, Journal of Investment Management, 5(4), p. 1-24.
Gray D.F. and Malone S. (2008), Macrofinancial Risk Analysis, Wiley Finance, UK.
Landon-Lane J., K. Oosterlinck (2006). Hope Springs Eternal: French Bondholders and the
Soviet Repudiation (1915-1919), Review of Finance, 10, p. 507-535.
Merton, R.C. (1969), Lifetime portfolio selection under uncertainty: The continuous-time
case, Review of Economics and Statistics, 51, p. 247-257.
Merton, R.C. (1974), On the Pricing of Corporate Debt: The Risk Structure of Interest
Rates, Journal of Finance, 29, p. 449-470.
Merton, R.C. (1993), Optimal Investment Strategies for University Endowment Funds,
Studies of Supply and Demand in Higher Education, Clotfelter C.T. and Rothschild M. ed.,
University of Chicago Press, p. 211-242.
Raymond H. (2010), Sovereign Wealth Funds as domestic investors of last resort during
crises, International Economics, 123, p. 121-159.
Scherer B. (2009a). A Note on Portfolio Choice For Sovereign Wealth Funds, Financial
Markets and Portfolio Management, 23(3), 315-327.
Scherer B. (2009b). Portfolio Choice for Oil Based Sovereign Wealth Funds, Journal of
Alternative Investments, 13(3), 24-34.
World Bank (2006). Where is the Wealth of Nations? Measuring Capital for the 21s Century,
The International Bank for Reconstruction and Development/The World Bank.
World Bank (2011). The Changing Wealth of Nations, The International Bank for
Reconstruction and Development/The World Bank.
Unexpected Returns.
Methodological Considerations on
Expected Returns in Uncertainty
Sylvie de Laguiche,
Head of Quantitative Research, Amundi
Gianni Pola,
Balanced Quantitative Research, Amundi
November 2012
Expected returns are closely related to the portfolio allocation: according to the Markowitz
portfolio selection (1952), once agreed on the market risk model and the investors risk
aversion, expected returns unambiguously determine the portfolio, and, conversely, reverse
optimisation techniques (Cantaluppi; 1999) allow us to relate a given (optimal) portfolio to a
set of expected returns. While the finance industry and academia are aware of the inadequacy
of the Markowitz model, it is clear that estimation of expected returns is a key issue for
strategic asset allocation. A pension fund, for example, needs expected returns on asset
classes for:
x assessing returns on assets and setting an appropriate discount rate for liabilities.
While in the former, risk-adjusted hierarchy between assets is more relevant, in the latter the
level of returns itself is crucial to determine discount rates. In this document, we refer to long
horizons - at least ten years - somewhat in line with the average liabilities in pension funds.
Many financial variables are in a peculiar territory, never reached going back to many
decades, and more than a century in some cases:
x the 10-year US Treasury yield is at its lowest level since 1871 (see chart 1);
x the decoupling of the Eurozone: the two-year yield divergence (see chart 2, right
panel).
1 Historical averages are good estimators if and only if the (underlying) stochastic process is stationary. Rapidly changing regimes and non-stationary dynamics prevent us
from estimating returns from historical averages, making them effectively unobservable.
16
12
0
1871
1874
1878
1882
1886
1890
1894
1898
1902
1906
1910
1914
1918
1921
1925
1929
1933
1937
1941
1945
1949
1953
1957
1961
1965
1968
1972
1976
1980
1984
1988
1992
1996
2000
2004
2008
2012
Chart 2
Low Yield (two-year bond yield) Eurozone Divergency (two-year bond yield)
10 10
8 8
6
6 EURO
4 PERIPHERALS
4
2
2
0
EURO
Sep-90
Sep-92
Sep-94
Sep-96
Sep-98
Sep-00
Sep-02
Sep-04
Sep-06
Sep-08
Sep-10
Sep-12
0
-2 CORE
Jun-10
Jun-11
Jun-12
Dec-09
Feb-10
Apr-10
Aug-10
Oct-10
Dec-10
Feb-11
Apr-11
Aug-11
Oct-11
1
ec-11
Dec-11
eb-12
Feb-12
Apr-12
Aug-12
ug-12
-2
u
GERMANY FRANCE NETHERLANDS
BELGIUM AUSTRIA JAPAN GERMANY FRANCE ITALY
UK CANADA SWITZERLAND NETHERLANDS BELGIUM AUSTRIA
DENMARK SWEDEN US SPAIN PORTUGAL GREECE
During the fifties and sixties, expected returns were considered to be time-dependent. They
were estimated from asset fundamentals (e.g. the dividend discount model for stocks, the
yield for a bond, etc.). In the next two decades, thinking on the subject changed according to
the work of Ibbotson and Sinquefield (1976a and 1976b). They modelled equity expected
returns as a time-varying baseline given by cash or bonds, plus a constant term, the long-term
equity risk premium. In the eighties, the financial community came back to the origin: risk
premia were believed to be time-varying quantities themselves. This counterrevolution started
with Campbell and Shiller (1988a, 1988b) and continued with the works of Asness (2000),
Arnott (2002), and Fama & French (1989). Recently, Ilmanen (2011) faced the problem of
estimating expected returns on major asset-classes broadening the investments to non-
traditional assets (commodities, real estates), investment styles (value, trend, carry,
volatilities), and underlying factors (growth, inflation, illiquidity, and tail risks).
Time-varying risk premia can be profitable if and only if investors are able to predict them.
Nevertheless the recent crises again call into question the predictability of asset-class risk
premia: neither a very simple normative approach like the Sharpe ratio can be applied easily
given the uncertainty on risk-free and high volatility of risk premia, nor can equilibrium-based
models be easily estimated given the rise in macroeconomic volatility. Today, researchers and
practitioners prefer to incorporate uncertainty and estimation errors in expected returns,
leading to Bayesian approaches (Black & Litterman model, 1990) and robust asset allocation
models (Meucci, 2011). Recently more extreme approaches emerged in the financial arena,
aiming to construct portfolios without any specific views on expected returns (e.g. minimum
variance, maximum diversification, risk parity approaches). Indeed, while the needs for
expected returns might be questionable in an optimisation process, it is evident that the long-
term estimates are absolutely crucial for pension funds and insurance companies to design
investment strategies to match their liabilities.
Rather than providing a specific recipe for estimating expected returns and computing return
figures, our aim here is to investigate this issue, stimulating the reader with relevant questions,
considerations, and few side empirical analyses to sustain our ideas. This manuscript is about
expected returns. Nevertheless, we will complement our considerations with historical
performance figures: even if the recent history is probably too peculiar to make extrapolations
from long-term time-series, there are always lessons that can be learned from history.
The paper is organised as follows. In section 1 we will start investigating how the crisis
changed the relationship between risk and return, and then questioning the possibility today of
Past average returns are rarely related to future performance. Nevertheless past bad
performance can be somehow indicative of future returns: risk premia correspond to the
premium that an investor requires in order to be compensated for poor performance in bad
times. Chart 3a (redrawn from Ilmanen 2011) plots the compound average real returns from
1960 to 2009 in some US markets as a function of the average returns in bad times (1974,
1981, 2008) for financial markets and global economy. Even if the relationship is not clearly
linear, it is evident that assets that performed poorly in crisis periods delivered higher returns
in the long-run. This intuition is the key idea of modern theories on asset-pricing which, rather
than deriving prices from future discounted cash flows, build expected returns from asset
returns co-variation with bad times.
Given this argument, we measured the volatilities, maximum draw downs (MDD) 2, and
performance of some asset classes in two historical periods: pre-crisis sample (1990-2007)
and the full sample (1990-2012); see chart 3b in annex A1 for more details. The aim is to
show how the crisis period (2008-2012) brought new stress that was not priced in the pre-
2
While the MDD is related to the tail-risk and is sometimes related to specific historical events (local measure),
the volatility is a more general measure of returns dispersion (global measure).
The Eurozone debt crisis determined a marked increase in MDDs and volatilities in Euro
peripherals nominal bonds (the MDD of Italian bond doubled) and Euro inflation-linked
indices. The 2008 crisis was dramatic for spread markets: all volatilities increased markedly,
MDD for US investment grade more than doubled, World high-yield doubled its MDD, Euro
credit markets suffered less with respect to the pre-crisis period. The worst equity markets in
the crisis period were US, Euro peripherals, Australia and Emerging markets; the volatilities
of equity markets increased but only marginally with respect to the pre-crisis period.
Commodity indices registered new MDDs in the crisis sample, gold performed better during
the crisis (the average compound return increased); volatilities increased.
The crisis period increased long-term volatility in most of the asset classes. Most volatility
estimation models for strategic asset allocation are mainly based on Exponential Smoothing
techniques which overweight the recent history with respect to more distant history.
According to these approaches, the ex-ante volatility of many asset allocations is likely to
increase especially in the low-medium risk investor profiles where bond markets are mostly
allocated.
Chart 3 a
8%
Compound Average Real Return, 1960-2009
7%
Small-Cap Stocks
6%
Stock Markets 5%
Real Estate
Commodity Futures 4%
0%
5% 0% -5% -10% -15% -20% -25% -30%
Average Real Return in Bad Times (1974, 1981, 2008)
While recent history (few months) can moderately reflect near future performance (as trend-
follower strategies and CTAs demonstrated in the long run), it is rather dangerous to
extrapolate over multi-year time windows when reversals often take place.
In historical estimates, the sample period is crucial: long time windows reduce sample
specificity and allow for more robust statistical inference, but may miss structural changes
happening in the market and may not be able to catch up non-stationary risk premia dynamics.
On the other hand, small samples may lead to nonsignificant results, and give an incomplete
picture of the financial dynamics. Given the level of many financial indicators today, and the
emergence of structural changes in the world economy, it is rather difficult to select an
appropriate sample to estimate econometric models: recent history may or may not be
relevant. Chart 4 reports the rolling historical volatilities of German, Italian, and Spanish 7-10
year bonds (upper panel), and their rolling historical correlation to the EMU equity markets
(lower panel): the charts clearly show that the Eurozone moved from divergence to
convergence in the new millennium, and recently from convergence back to divergence again
with the debt crisis. The most relevant sample to estimates markets returns in this case is not
obvious. Financial markets are moving from pre-crisis equilibrium to a new one: at the
moment we are in between, and uncertainty prevents us from tracing a clear picture.
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
1
0,8
0,6
0,4
0,2
0
-0,2
-0,4
-0,6
-0,8
-1
Oct-94
May-95
Mar-94
Jul-96
Oct-97
May-98
Jul-99
Feb-97
Feb-00
Apr-01
Jul-02
Feb-03
Apr-04
Apr-07
Mar-10
Oct-10
Dec-95
Dec-98
Nov-07
Jan-93
Aug-93
Sep-00
Nov-01
Sep-03
Nov-04
Jun-05
Jan-06
Sep-06
Jun-08
Jan-09
Aug-09
Jun-11
Jan-12
Aug-12
germany 7-10 italy 7-10 spain 7-10
Chart 5a
20%
15%
10%
5%
0%
-5%
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2012 avg
20%
1930
15%
1960 2000
1870
10%
Equity perf (ann)
-5% 1920
-10%
-10% -5% 0% 5% 10% 15% 20%
Bond perf (ann)
Equity in the new millennium did not offer any spread over inflation: this pattern on equities
was similar to the seventies when high inflation caused a bear equity market 3. On the bond
side, performance in the new millennium was higher than the long-term average: bonds
benefited from the marked fall in interest rates (see chart 1). A final remark on correlation
between performance of bond and equity: the scatter plot shows that, on a ten-year horizon,
they were on average positively correlated. This paradigm was not verified only from the
beginning of 1941 to the end of 1980, when bonds suffered from the increase in 10-year
interest rates from 1.95% to 12.84%, delivering a negative real performance.
This analysis highlights the importance of extending the historical sample. In addition, in
order to get more robust estimates it might be wiser to complement model-free historical
figures with models from financial theories, models from behavioural theories, forward-
3
In the 1970-1980 decade the annual inflation rate was +8.05%.
2. Should expected returns reflect an unbiased view about the market, or should
they incorporate economic scenario expectations and/or recent history
information such as valuation?
The approaches to estimate expected returns can be divided in two main categories: statistical
methods and equilibrium-based models. The separation between them is not very neat. Both
of them are certainly statistical as most of the models are derived according to econometric
relations, and both of them can be built on macroeconomic relationships among variables. The
main differences between them are that the former do not assume any specific
macroeconomic views, and are not equipped with valuation tools. We would say that while
statistical models estimate returns mainly looking at the past history, equilibrium-based
models forecast future returns incorporating forward looking macro variables.
x but are not very appropriate for computing discount rate for pension funds and
insurance companies.
In this document we do not face the class of Behavioural models. These approaches enable us
to explain many observed anomalies in risk premia 5, and might be useful to diversify
statistical and equilibrium models. This way may lead to more robust estimations.
Sharpe ratio is a risk-adjusted performance measure, and it is usually used to assess the
efficiency of an asset-class, investment strategy or fund. Sharpe ratio is usually computed as
the average excess return of a risky asset over the risk-free rate and normalised for historical
volatility. We prefer instead to compute the Sharpe ratio as the difference between the
annualised compound returns of the risky asset and the risk-free rate divided by the
annualised volatility. According to this formulation the expected return can be expressed as
the risk-free plus a term given by the ex-ante volatility times the Sharpe ratio.
4
Maximum-diversification portfolios are obtained maximising the (ex-ante) Sharpe ratio under the hypothesis of
constant Sharpe ratio across all asset-classes (Choueifaty and Coignard, 2008).
5
The equity premium puzzle indicates the difficulty of explaining the observed equity risk premium within
standard macroeconomic models. Behavioural models provide an explanation (see Bernatzi and Thaler 1995, and
Barberis and Huang 2001).
Chart 7 plots the Sharpe ratio for the US, Japanese, UK, German and French equity markets.
Computations have been performed on rolling windows of 10- and three-year horizons. The
longer the time window, the more stable the Sharpe ratio. Table 1 in annex A2 reports some
descriptive statistics: the median and the average figures are reported for each country and
three-, five- and 10-year time horizons. For the equity markets, 0.25 is a reasonable estimate
for the 10-year horizon.
Chart 6
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
2,5
1,5
0,5
-0,5
-1
-1,5
-2
juil.-65
juil.-67
juil.-69
juil.-71
juil.-73
juil.-75
juil.-77
juil.-79
juil.-81
juil.-83
juil.-85
juil.-87
juil.-89
juil.-91
juil.-93
juil.-95
juil.-97
juil.-99
juil.-01
juil.-03
juil.-05
juil.-07
juil.-09
juil.-11
USA JAPAN GERMANY FRANCE UK
2,5
1,5
0,5
-0,5
-1
-1,5
-2
juil.-65
juil.-67
juil.-69
juil.-71
juil.-73
juil.-75
juil.-77
juil.-79
juil.-81
juil.-83
juil.-85
juil.-87
juil.-89
juil.-91
juil.-93
juil.-95
juil.-97
juil.-99
juil.-01
juil.-03
juil.-05
juil.-07
juil.-09
juil.-11
x Differences among asset classes. Does it make sense to make a normative assumption
of equal Sharpe ratios for all asset classes?
Recently, practitioners and researchers (see Frazzini & Petersen 2010) documented some
anomalies regarding the Sharpe ratio. We report in the following the most relevant:
x On historical basis low volatility asset-classes delivered higher Sharpe ratio6. This
evidence inspired minimum-variance investments in the equity markets. We tested this
hypothesis in different baskets, as follows:
6
As reported by Asness et al. (2011), if investors are leverage averse, low-beta assets will offer higher risk-
adjusted returns, and high-beta assets lower risk-adjusted returns. Leverage aversion breaks the standard CAPM.
Equity markets. In chart 8 (top right panel) we investigated the bias of Sharpe ratio
towards low volatility segments measuring the Sharpe ratios for each sector of US
equity market, and plotting them against the volatilities. The plot confirms the
anomaly, even if the relation is noisy (linear regression; R2 is 0.2210). In chart 9
(lower panels) we report two pictures taken from Baker et al. 2011. They studied the
US equity market from January 1968 to December 2008.
Chart 8
0,60 0,50
0,50 0,40
0,30
0,40
sharpe (ann)
sharpe (ann)
0,20
0,30
0,10
0,20
0,00
0,10 -0,10
0,00 -0,20
0% 5% 10% 15% 20% 25% 0% 10% 20% 30% 40%
vol (ann) vol (ann)
Bond markets and Credit markets. As reported in Ilmanen (2011) a similar rule holds
for bond and credit markets: lower maturity bonds reported a higher Sharpe ratio than
longer maturity ones, and similarly higher quality credit buckets historically delivered
higher Sharpe ratios than lower rated credit ones.
Diversified indices delivered higher Sharpe ratios. We tested this argument in chart 9.
We measured the Sharpe ratio for the EMU equity markets from 2002 to 2007. The
Sharpe ratio is 1.30 which is greater than most of each countrys ones. The arithmetic
average of the Sharpe ratio cross-countries is 1.09 (the weighted average according to
the market cap is 1.19; we made an approximation of keeping the market cap constant
over time). The result confirms the benefit of diversification.
Illiquid asset classes historically delivered higher Sharpe ratios with respect to liquid
assets (see Ilmanen, 2011). In the long run, less liquid assets offer a compensation for
the higher trading costs and lower flexibility to rebalance portfolio positions. From an
estimation point of view, illiquid assets tend to exhibit smoothed prices thus leading to
underestimation of the true risk; this estimation error is certain to artificially increase
the Sharpe ratio.
Chart 9
2,00 1,73
1,60 1,32 1,30
1,09 1,06 1,09 1,19
1,20 0,85 0,94
0,80 0,62
0,40
0,00
EMU avg
SPAIN
NETHERTLAND
BELGIUM
FINLAND
EMU
GERMANY
ITALY
EMU avg2
FRANCE
A more detailed approach than the constant Sharpe ratio method may be required to set up the
expected returns to calculate the discount rate. However when it comes to portfolio
construction we should note that taking advantage of high Sharpe ratio for low volatility asset
classes is only possible when leveraging portfolios or for very low risk profiles. In addition,
the constant Sharpe ratio approach leads to a better diversified portfolio in terms of risk;
therefore except in very particular case this approximation may be kept to building strategic
allocation in diversified portfolios.
The Sharpe ratio implicitly assumes volatility as the risk measure. In case of strong
asymmetric and fat-tailed asset classes, the approximation is too rough, and the Sharpe ratio
becomes unreliable. In this case many practitioners and researchers prefer to correct the
formula by replacing the volatility with the downside volatility (Sortino ratio), or maximum
draw down. Nevertheless we should remind the reader that volatility misses some important
aspects of risk: liquidity risk, default risk, higher order statistics and tail risk, model risk,
timing risk, valuation risk, and fundamental risk, and that is going to have implications on the
estimation of the Sharpe ratio.
Inflation refers to a rise in the consumer price level and consequently to a reduction in the real
value of money. While rising inflation is usually related to poor conditions in the real
economy, disinflation (i.e. a slowdown of inflation to lower levels) is commonly associated
The first issue we want to address is the likelihood of deflationary and inflationary scenarios
on a historical basis. Chart 10a reports the annual inflation rates for US (from 1871) and
Sweden (from 1900).
It is evident that the last millennium was mainly characterised by positive inflation rates;
hence it is questionable whether deflationary periods were less likely than inflationary ones.
As Reinhart and Rogoff (2011) demonstrated, on longer historical perspective, inflation and
deflation were both well represented: according to their study the 20th century itself represents
an anomaly. Secondly we analysed the historical performance of asset classes conditional to
contemporaneous inflation rate variation (on an annual basis). We divided the inflation rates
into buckets to separate different scenarios:
Chart 10a
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
Then we computed the annualised real performance of major asset classes conditional to the
above inflation scenarios. Chart 10b reports the outcome of the analysis for US 10-year bonds
and equity from 1871 to September 2012, and for Swedish bonds and equity from 1900 to
September 2012. For each bucket we plot the average, the minimum and maximum
performance as a proxy of the dispersion. Three main messages emerge:
x high inflation hurts nominal bonds: inflation rates greater than 5% lead to negative
average performance in US and close to zero in Sweden;
x the high dispersion of returns for equity markets highlights that inflation is not the main
driver in this case.
In chart 10c we performed the same exercise for US corporate investment grade all maturities,
US corporate investment grade intermediate maturity, commodity GSCI and gold. We
investigated the dependency with respect to the US inflation rates. Time series are shorter in
this case: our sample is from 1970 for investment grade corporate and commodity GSCI, and
from 1927 for gold 7. The main take-home messages are:
x corporate investment grade mimics the behaviour of bonds, hence inflation hurts corporate
investment grade;
x gold does not reward significantly over inflation on high inflation scenarios (as noted in
footnote 7, the very high inflation scenario is not significant in this case).
In Pola (2013) we will more carefully illustrate the dependency of asset returns on
macroeconomic variables, showing how to segment asset classes according to their attitude to
polarise with respect to variations of macroeconomic variables (rising or falling scenarios).
7
In this case results on extreme scenarios (deflation and very high inflation) should be taken with care. The
sample from 1970 only has three entries for very high inflation scenarios and zero for deflationary scenarios. The
sample from 1927 only presents four entries for very high inflation scenarios and seven for deflationary
scenarios.
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)
20%
0%
10%
0% -20%
-10%
-40%
-20%
-30% -60%
<=0% (0%; 2%] (2%; 5%] (5%; 10%] >10% <=0% (0%; 2%] (2%; 5%] (5%; 10%] >10%
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%; 2%] (2%; 5%] (5%; 10%] >10%
Chart 10c
US Corporate IG 1970-2012 conditional to US CPI rate buckets US Interm Corporate IG 1970-2012 conditional to US CPI rate
buckets
40% 40%
30%
30%
real performance (annualized)
20%
20%
10%
10%
0%
0%
-10%
-10%
-20%
-20% -30%
<=0% (0%; 2%] (2%; 5%] (5%; 10%] >10% <=0% (0%; 2%] (2%; 5%] (5%; 10%] >10%
Commodity GSCI 1970-2012 conditional to US CPI rate Gold 1927-2012 conditional to US CPI rate buckets
buckets
80% 60%
60%
40%
real performance (annualized)
40%
20%
20%
0%
0%
-20%
-20%
-40%
-40%
-60% -60%
<=0% (0%; 2%] (2%; 5%] (5%; 10%] >10% <=0% (0%; 2%] (2%; 5%] >5% >10%
The Markowitz model strongly depends on expected returns assumptions, and sometimes
produces results that are extreme and not particularly intuitive. The remedy of Black &
Litterman (1990) was to firstly identify a reference point for expected return assumptions
(equilibrium expected return), and then to elaborate a consistent Bayesian framework to
integrate qualitative views.
The main assumption beyond the equilibrium expected return is that, according to the Capital
Asset Pricing Model (CAPM), prices will adjust until the expected returns of all assets in
equilibrium are such that if all investors hold the same belief, the demand for these assets will
exactly equal the outstanding supply (He & Litterman 1999). The procedure to determine the
implicit returns in the market portfolio consists in reverse engineering the mean-variance
method: the aim is to determine the expected returns according to which the market portfolio
is optimal; this procedure is known in literature as reverse optimisation (Cantaluppi 1999).
This approach is more general than its application in the Black & Litterman (1990) model:
according to the reverse optimisation technique, implicit returns can be extracted from any
strategic asset allocation.
Reverse optimisation is a valuable tool for achieving consistency between model portfolios
and the fund managers portfolio. The iterative process that enables the allocation to be
transformed in implied views (reverse optimisation), and then the expected views in a
portfolio (optimisation), allows fund managers to build their portfolios more coherently.
The risk aversion parameter corresponds to the expected risk-return trade-off. It is the rate at
which more return is required for more risk, and it can be expressed as the ratio of risk
premium and variance of the market portfolio. This approach becomes decreasingly intuitive
when dealing with a diversified investment universe including bonds, equities, commodities
and alternative assets. In this case the most popular approach consists in determining the risk
aversion parameter according to a strong prior information or conviction (e.g. the return of a
short-term bond). Given the debt crisis in the Eurozone, this issue remains a critical one.
According to the reverse optimisation technique, returns are proportional to the market cap,
hence the approach suffers from increasing the returns of assets which are in a bubble. No
mean reversion mechanisms are present to alleviate the problem.
The model does assume that investors can buy any markets without any regulatory
constraints. Indeed it does not include some bias, such as the home bias, for example,
according to which investors tend to overweight assets of their own country.
6. Equilibrium-based model
The financial crisis questions the possibility of setting up reasonable forward looking
equilibrium figures due to uncertainty on long-term macroeconomic prospect trends.
What is the impact of this uncertainty on asset class return forecasts? And what is the
confidence on expected returns?
In the past this approach was able to produce coherent figures. The developed world economy
had been characterised by decreasing inflation and decreasing volatility in growth and
inflation in the '90s (see chart 11). This allowed monetary policy rates to be generally lower
and more stable than in the '70s and the '80s. In particular, inflation expectations stabilised at
fairly low levels over the '90s. Therefore, in this more stable environment, rates tended to also
be more stable and more in line with nominal growth: this represented the first assumption.
3,50%
3,00%
2,50%
2,00%
1,50%
1,00%
0,50%
0,00%
dc.-57
dc.-59
dc.-61
dc.-63
dc.-65
dc.-67
dc.-69
dc.-71
dc.-73
dc.-75
dc.-77
dc.-79
dc.-81
dc.-83
dc.-85
dc.-87
dc.-89
dc.-91
dc.-93
dc.-95
dc.-97
dc.-99
dc.-01
dc.-03
dc.-05
dc.-07
dc.-09
dc.-11
GDP CPI
The second assumption was represented by the empirical evidence of a positive relationship
between performances and risks measured by volatility over a long period. This environment
favoured equilibrium-based model approaches. Current market conditions are now far more
uncertain, and they make it difficult to apply this approach because econometric relations to
estimate risk premia might not be robust in non-stationary markets.
Furthermore, the monetary policy regime changed as well: extraordinary measures like
quantitative easing were introduced over recent years by many central banks as a new policy
response to the crisis. At the same time the assumption on the link between return and risk
over the long term has also come under scrutiny.
Even if forward-looking indicators, such as valuation ratios, have a better track record in
forecasting asset class returns than rearview-mirror measures (Ilmanen 2011), we should
remind the reader that returns sometimes may never revert over a practical time frame. Chart
12 reports the time-series of US bonds, US equity and Japanese equity in specific time
periods. The figures are contrasted to a hypothetical long-term average and dispersion (see
annex for details). The plots show clearly that the return-to-the-mean sometimes is very slow,
and that risk premia can exhibit robust trends which can last for many decades.
The best way to overcome difficulties in equilibrium-based models and valuation approaches
rely on diversifying the estimation process with complementary models.
-1,5
-1
-0,5
1,5
-2
-1,5
-1
-0,5
-0,2
0
0,5
1
2
2,5
3
3,5
0
0,5
1
1,5
2
0
0,2
0,4
0,6
0,8
1
1,2
1989,12 1929,12 1964,12
Chart 12
mkt
mkt
1998,02 1934,12
1998,09 1971,12
1935,05
1999,04 1972,06
1935,10
1972,12
median
1999,11
median
median
1936,03
2000,06 1973,06
2001,01 1936,08 1973,12
2001,08 1937,01 1974,06
+2sigma
+2sigma
+2sigma
2002,03 1937,06 1974,12
2002,10 1975,06
1937,11
2003,05 1975,12
1938,04
-2sigma
-2sigma
-2sigma
US Bond 1965-1981 (Cumulative Return)
2003,12
US Equity 1929-1942 (Cumulative Return)
1976,06
1938,09
2010,05 1981,06
1942,06
2010,12 1981,12
The last decade in the equity market has been particularly dramatic for diversified asset
allocation such as pension funds and balanced portfolios: poor performances, large draw-
How to handle uncertainty in expected returns? Can strategic asset allocation get
199
downs and very slow recovery again call into question the predictability of risk premia, and
the effect of incorrect assumptions in the portfolio optimisation process.
In the eighties and nineties, Jobson and Korkie, (1980), Best and Grauer (1991), Chopra and
Ziemba (1993) demonstrated that the sub-optimality due to estimation risk can be dramatic.
This evidence led researchers and practitioners to investigate more robust portfolio
construction schemes which can alleviate the estimation risk in the portfolio construction
process. The main approaches are the Bayesian methods and the robust models. The most
famous Bayesian method is the Black-Litterman (1990) model. Robust allocations deal with
uncertainty of input parameters by choosing the best allocation in the worst market condition
within a given uncertainty range. Indeed, the choice of this range is quite arbitrary. Robust
Bayesian allocations enable the uncertainty region for the input parameters to be defined in a
more coherent way. Moreover, the approach allows investors to modify the region according
to their specific views. Meucci (2011) consider robust Bayesian allocations that also account
for the estimation errors in co-variances.
More recently, the financial industry moved even further, elaborating investment processes
that can completely neglect assumptions on expected returns. The most popular were the
minimum variance, the maximum diversification portfolios, and the risk parity approach (see
Clarke et al. 2012). While the former has been used for equity portfolios, the seconds are
good candidates for diversified allocation. These portfolio construction schemes allow the
investor to make allocation according (only) to a risk model for the asset-classes. The
portfolio construction implication of these approaches is the overweight of low risk assets,
and thus the need for leveraging allocations in order to match medium-high risk profile
without losing equilibrium among portfolios bets.
In Pola and Facchinato (2013), we illustrate a new approach for strategic asset allocation
(DAMS) which can be helpful to build more robust estimates for expected returns in
uncertainty. The key assumption of the Black-Litterman approach is to identify a reference
point for expected returns from reverse engineering the market portfolio (according to the
CAPM hypothesis). Given the limits of the CAPM hypothesis, it might be wiser to define the
reference expected returns from a different perspective. In Pola and Facchinato (2013) we
introduce a new reference portfolio where different macroeconomic scenarios are in
equilibrium, thus expressing the current uncertainty in financial markets.
8. Conclusion
The recent crisis exhibited major changes in the risk level, in the observed risk premia and the
risk-return relationship among asset classes. Many financial variables are in an uncharted
region, never reached going back many decades, and more than a century in some cases:
whether they would revert to the mean or they would stabilise to new levels is not clear (e.g.
Euro peripherals bond market). The uniqueness of the level reached by many financial
variables should at least convince us to lower our confidence on predictability of asset
classes risk premia.
x in inflationary scenarios nominal bonds suffer, gold does not provide excess return over
inflation, whereas more diversified commodity indices offer better real performances;
Even if inflation hurt US stocks in the seventies and eighties, the dependency of equity on
inflation is not clear: going back to 1871 in the US and 1900 in Sweden, high dispersion of
The directions which we would like to advice to tackle the expected return issue in this new
environment are:
This pluri-disciplinary approach may overcome some of the difficulties experienced by both
statistical and equilibrium-based approaches, leading to better adapted figures both to build
robust allocations and set up a realistic future level of expected returns and discount rate of
liabilities.
We would like to thank Eric Taz-Bernard, Sergio Bertoncini, Jean-Renaud Viala, and Marc-
Ali Ben Abdallah for very stimulating discussions and very constructive suggestions.
In chart 3b we measured the volatilities, maximum draw downs (MDD), and performance of
some asset classes in two historical periods: pre-crisis sample (1990-2007) and the full sample
(1990-2012). The grey shaded areas in chart 3b signal markets that exhibit a worse MDD than
before, bold indicates an increase in volatility.
Bond markets. The issue is the decoupling of the Eurozone: Italys MDD doubled in the
crisis period. Nevertheless Germany suffered between the end of 2011 and first quarter of
2012. In the Eurozone, French bonds did not exhibit any draw-downs not priced before. With
the exception of Australian, Canadian, and Japanese bonds, all volatilities increased.
Inflation linked markets. Eurozone debt crisis drove the main changes: the euro index
presents the higher increase in volatility. The emerging inflation-linked market is not very
significant due to small data sample: anyway data indicate that they deliver the best
performance per unit of volatility across all asset classes.
Credit markets. The 2008 crisis was a dramatic year for most of the spread markets. MDD
for US investment grade more than doubled, euro investment grade suffered less. World high
yield doubled its MDD, even though the euro high yield did not suffer its worst draw down in
the crisis period (the MDD in the full sample was from February 2000 to September 2002).
All volatilities increased except for the emerging market debt in hard currencies.
Equity markets. The crisis increased the MDDs in US, Euro peripherals, Australia and
Emerging markets. Germany, France, UK, Japan, and Canada did not register worst draw
downs. The volatilities increased but only marginally with respect to the pre-crisis period.
Commodity markets. CRB and GSCI indices registered new MDDs in the crisis sample,
gold performed better in the crisis (the average compound return increased). The increase in
volatility of CRB was more marked with respect to the GSCI and gold; the difference between
CRB and GSCI indices is mainly due to the larger exposure of CRB to energy commodities.
In the following we report the median and average Sharpe ratio computed in different equity
markets and according to different-sized rolling windows (10 years, 5 years, 3 years)
Table1
ANNEX A3
Chart 1
The plot reports the US Treasury yield 10-year. Data provider is Schiller database.
Chart 2
The plot reports the two-year yields for international countries: Germany, France,
Netherlands, Belgium, Austria, Italy, Spain, Portugal, Greece, Japan, UK, Canada,
Switzerland, Denmark, Sweden, US. Data provider is Bloomberg.
Chart 3a
Chart 3a is redrawn from Ilmanen (2011). It reports the performance of various US markets
from 1960 to 2009. The scatter plot relates the compound average real return (Y-axis) to the
average real losses in the three worst years (1974, 1981, 2008) for financial markets and the
global economy.
Chart 3b
Chart 3b contrasts the performance, volatility, and Maximum Draw Down of various indices
in two samples: pre-crisis sample (1990-2007) and the full-sample (1990-September 2012).
All time-series are monthly and in local currency. A few indices are shorter due to their
Chart 4
The plot reports the historical 1-year rolling volatility (top panel) for German, Spanish and
Italian 7-10 year bonds, and their historical 1-year correlation to the EMU equity market.
Time-series are daily, data provider is Bloomberg.
Chart 5a & 5b
The left chart reports the performance of US bonds, US equity, and compares them to the
inflation rates in the same period. Performances are presented in decades (except for the last
decade which includes 12 years). The vertical bar divides the sample in three periods: pre-
war, post-war and new millennium. The right panel reports a scatter plot between real
performance for US bonds and US equity. Each spot corresponds to the annualised
performance of bonds and equity for a specific decade. Data provider is Schiller database.
Chart 6
The plot reports the 10-year CDS for Italy and Germany since 2005. Data provider is
Bloomberg.
Chart 7
The plot reports computations of Sharpe ratio over rolling windows (10 and 3 years) of
various equity markets (US, Japan, UK, Germany, France). Time-series are monthly. Table 1
in annex A2 reports the summary of the results: median and average for countries and
horizons (10, 5, 3 years). Data providers are Datastream and Bloomberg.
Chart 8
In chart 8 (top left panel) we computed the historical Sharpe ratio and plot against historical
volatility in a diversified investment universe in the US. The risky assets are: short-term
treasury, all-maturity treasury, long-term treasury, short-term corporate investment grade, all-
maturity corporate investment grade, long-term corporate investment grade, all-maturity
Arnott, R. D., and Bernstein, P. L., 2002. What Risk Premium Is Normal?, Financial
Analysts Journal, vol. 58, no. 2 (March/April): 6485.
Asness, C. S., 2000. Stocks vs. Bonds: Explaining the Equity Risk Premium., Financial
Analysts Journal, vol. 56, no. 2 (March/April): 96113.
Baker, M., Bradley, B., and Wurgler, J., 2011. Benchmarks as Limits to Arbitrage:
Understanding the Low-Volatile Anomaly., Financial Analysts Journal, vol. 67, no. 1.
Barberis, N., and Ming H., 2001. Mental Accounting, Loss Aversion, and Individual Stock
Returns., Journal of Finance, vol. 56, no. 4 (August): 12471292.
Benartzi, S., and Thaler, R., 1995. Myopic Loss Aversion and the Equity Premium Puzzle.,
Quarterly Journal of Economics, vol. 110, no. 1 (February): 7392.
Best, M. J., and Grauer, R. R., 1991. On the sensitivity of mean-variance-efficient portfolios
to changes in asset means: Some analytical and computational results.. Review of Financial
Studies 4, 315-342.
Black, F. and Litterman, R.. 1990. Asset Allocation: Combining Investors Views with
Market Equilibrium., Fixed Income Research, Goldman Sachs.
Cantaluppi, L., 1999. The Reverse Optimization., Financial Markets and Portfolio
Management.
Campbell, J. Y., and Shiller, R., 1988a. Stock Prices, Earnings and Expected Dividends.,
Journal of Finance, vol. 43, no. 3 (July): 661676.
Chopra, V., and Ziemba, W. T., 1993. The effects of errors in means, variances, and
covariances on optimal portfolio choice., Journal of Portfolio Management pp. 6-11.
Choueifaty, Y., and Coignard, Y., 2008. Towards Maximum Diversification., Journal of
Portfolio Management, Vol. 35, No. 1 (Fall).
Clarke, R., De Silva, H., and Thorley, S., 2012. Risk Parity, Maximum Diversification, and
Minimum Variance: An Analytic Perspective., working paper.
Fama, E. F., and Kenneth, R. F., 1989. Business Conditions and Expected Returns on Stocks
and Bonds., Journal of Financial Economics, vol. 25, no. 1 (November): 2349.
Frazzini, A., and Lasse, H. P., 2010. Betting Against Beta., Working Paper, AQR Capital
Management, New York University and NBER Working Paper 16601.
He, G., and Litterman, R., 1999. The Intuition Behind Black-Litterman Model Portfolios.,
Investment Management Research, Goldman, Sachs & Company.
Ibbotson, R. G., and Sinquefield, R. A., 1976b. Stocks, Bonds, Bills, and Inflation:
Simulations of the Future (19762000)., Journal of Business, vol. 49, no. 3 (July): 313338.
Ilmanen, A., 2011. Expected returns. An Investors Guide to Harvesting Market Rewards,
Wiley Finance
Jobson, J. D., and Korkie, B., 1980. Estimation for Markowitz efficient portfolios, Journal
of the American Statistical Association 75, 544-554.
Mandelbrot, B., 1963. The variation of certain speculative prices. The Journal of Business.
Pola, G., 2013. Managing uncertainty with dams. Assets segmentation in response to
macroeconomic changes, in preparation.
Pola, G., and Facchinato, S., 2013. Managing uncertainty with dams. Balancing
macroeconomic scenarios, in preparation.
Rebonato, R, 2008. Plight of the Fortune Tellers: Why We Need to Manage Financial Risk
Differently. Princeton, NJ: Princeton University Press.
Reinhart, C. M., Rogoff, K. S., 2011. From Financial Crash to Debt Crisis, American
Economic Review, 101: 1676-1706.
March 2013
The first axiom of investing is that higher risk is compensated -on average- by higher returns.
In his Portfolio Selection (1952), Markowitz defines portfolio risk as the volatility (or
variance) of its returns and states that, while taking investment decisions, investors maximize
expected returns for a given amount of portfolio risk: this leads to a positively sloped and
concave efficient frontier.
In Sharpes CAPM (1964) risk is defined as portfolio beta relative to the Market Portfolio.
The idea behind the CAPM is that any assets average return may be expressed as the sum of
the risk-free rate and a risk premium, the latter being proportional to the assets covariance
with the Market Portfolio (its beta).
According to the theory, both the Minimum Variance portfolio and the Market Portfolio
belong to the efficient frontier; the latter exhibiting a higher expected return than the former.
As mentioned, our goal is not to take a position in the everlasting dispute in favor or against
the CAPM. However it is worth mentioning some preliminary tests that provided some
comforting results during the early seventies, as well as many subsequent studies proving the
outperformance of low volatility stocks versus high volatility stocks, and its persistence
during recent decades.
Black, Jensen and Scholes (1972), and Fama and MacBeth (1973) form portfolios on the basis
of in sample estimations of beta, and then look at realized Beta and stock returns, finding that
higher beta is positively correlated to higher returns. Yet, the slope of the linear relationship is
flatter than anticipated by CAPM. For instance, in Fama and MacBeth, the hypothesis that
beta is uncorrelated with returns is significantly rejected only in the full sample period (1935-
1968) while it is not rejected in any of the 10-year sub-periods.
Despite some statistical weakness, these contributions held for a long time as a general
confirmation of the traditional positive relationship between risk and return.
In fact, more recent evidence shows that well-constructed low risk portfolios contradict this
axiom, as they deliver higher returns than riskier portfolios.
In 1992, Fama and French showed that value and size factors generate large returns in the US
markets and, quite surprisingly, that the relationship between beta and returns is rather
negative, when corrected for size effect.
Intuitively, as the aggregate volatility (VIX factor) is significant indeed, but missing in the
Fama French model, companies with greater sensitivity to the VIX factor should exhibit
higher idiosyncratic risk in a Fama French framework.
Surprisingly, after controlling for aggregate volatility, the negative relationship between
returns and idiosyncratic risks is little changed, and completely unexplained.
Blitz and Vliet (2007) focus on systematic and total volatility. They state that the low-risk
anomaly holds regardless of which of the two measures is used for stock selection: low
variance stocks exhibit low beta, while high variance stocks exhibit high beta.
The authors controls for factors such as value, momentum, and size, are both via linear
regressions with those factors returns as explanatory variables (with little if any variance
explained), and via cross basket analysis: returns of the volatility factors are investigated
among stocks with similar momentum, size or value characteristics. The significance of the
low risk effect is little reduced, and only within baskets of stocks with similar market
capitalization.
They also argue (probably underestimating the cost in terms of loss of flexibility and risk of
incurring an undesired exposure) that there is no need to rely on sophisticated risk models and
that an equal weighted portfolio of stocks with low historical volatility is all investors need to
achieve superior returns.
We have replicated these transparent tests on the MSCI World constituents from January 2003
to June 2012 (excluding stocks with less than two years of presence in the index), and we
confirm that grouping stocks for (ex-post) beta rather than (ex-post) volatility does not have a
big impact: in both cases average risk-adjusted returns decrease with risk measure, with
baskets built according to beta showing somehow better regularity and monotony (Annex 1,
tables 1 and 2, chart 1).
Baker and Haugen (2012) implement deciles analysis for each of the 21 developed and 12
emerging countries of their sample: they group stocks from 1990 to 2011 according to their
historical volatility and they find that lowest risk stocks exhibit higher returns and better
Sharpe ratio, compared to high risk stocks, in any observed country. They also point out that
only rarely do the highest volatility deciles outperform lowest deciles, in a rolling window of
three years.
In their intentionally simple approach, they use an equally weighted basket and 24 months
historical volatility as a measure of risk, apparently supporting argument by Blitz and Vliet
(2007) for no need for a sophisticated risk model. Actually no control is done for valuation,
size, or any other factor that could help to explain the anomaly.
In 1991, Baker and Haugen first investigate Minimum Variance portfolios in the US equity
market, pointing out a 30% reduction in portfolio volatility, compared to both a common US
index and randomly selected portfolios, with no reduction in average returns.
Clarke, de Silva and Thorley (2006) build Minimum Variance portfolios on the largest 1000
US stocks over the period 1968 2005. They estimate a covariance matrix with Bayesian
methods for shrinkage -in order to avoid error maximization problems- as well as principal
component analysis.
They first detail some well-known portfolio characteristics like typical concentration (75 to
250 stocks with 3% cap on a single company), turnover (143% with monthly rebalancing),
positive exposure to size and value factors, and zero mean but rather volatile exposure to
momentum factor. They confirm Baker and Haugens evidence (1991) of 30% reduction in
In their more recent (2011) Minimum Variance Portfolio Composition, Clarke, de Silva and
Thorley update previous (2006) statistics with basically no change. However their work is
worth mentioning mostly because, while proposing an interesting analytical solution of the
Minimum Variance portfolio composition, they point out that systematic risk dominates in the
construction of such a portfolio: in their simplified single factor model, stocks with beta
higher than a threshold are strictly excluded from the portfolio, while high idiosyncratic risk
only contributes to lowering the stocks weight in the portfolio.
Carvalho, Lu, and Moulin (2011) discuss typical low beta and small cap exposure of
Minimum Variance portfolios. Furthermore in their view, changes in the correlation matrix
generate higher than justified turnover, thus they suggest constructing portfolios by applying
stable weighting schemes based on stocks beta, rather than optimizing.
In our opinion, a weighting scheme may be a reasonable solution, even though optimization
remains our optimum for Minimum Variance portfolio construction, as an optimization
package enables turnover to be reduced, while controlling many other constraints.
Thomas and Shapiro (State Street Global Advisors, 2007) show their encouraging results on a
Minimum Variance portfolio built on the Russell 3000, using a standard optimization package
(BARRA). Their contribution is relevant to us because -as we will discuss deeper in section 3-
they recognize the advantage of using such an optimization package, in order to avoid the
typical drawbacks of a Minimum Variance investment (excessive concentration in a few low
risk sectors and stocks, lack of control for involuntary factor exposure), while enhancing their
performance by tilting the portfolio toward some long term successful alpha strategy, like the
historical dividend yield.
Most of the recent literature addressing the low risk anomaly has also offered some theoretical
framework.
Among the others, theories referring to leverage constraints are probably the most
represented.
Blake identifies borrowing restriction as one of the possible sources of low risk stocks
outperformance in his Beta and Returns (1995). Investors are supposed to pass through such
a limitation, investing massively in high beta stocks thus lowering subsequent returns.
Frazzini and Pedersen (2011) develop a model of asset equilibrium populated by two
categories of investors: investors with no leverage constraints but with margin requirements,
and investors for which leverage is forbidden. They argue that while the latter overweight
high beta assets, causing those assets to offer lower subsequent returns, the former buy low
beta stocks and leverage their holdings by short selling the high beta stocks. Intuitively both
the low beta and the high beta stocks have their demand in the market, thus it is not
immediately clear why only high beta stocks should be overpriced, thus delivering poorer
returns than the CAPM would predict. However the analytical solution of the required rate of
return in equilibrium shows that the alpha of each security monotonically decreases with its
beta, thus reducing the slope of asset returns relative to beta itself. This slope is flatter as the
tightness of the funding constraint increases.
The authors build a betting against beta factor by going long on low beta stocks and
shorting a smaller amount of high beta stocks, thus obtaining a beta neutral factor. They
provide empirical evidence for alpha decreasing with security beta, and for the betting
against beta factor exhibiting positive returns in equity, treasury, credit, and currency
markets.
Carvalho, Lu, and Moulin (2011) agree that leverage constraint has a major impact on the
preference towards high beta stocks. However, they add that, as they simply seek high returns,
investors prefer riskier stocks, and create a demand imbalance.
However, among the possible explanations of the anomaly, they take into account a
behavioral theory by Shefrin and Statman (2000), where investors allocate their wealth
according to two layers: a low risk layer designed to avoid poverty, and a high aspiration
layer, for which they are much less risk averse. In this case, investors will overpay for (often
few and badly diversified) risky stocks, which are perceived to be similar to lottery tickets.
The authors finally mention another well represented family of theories that will be further
discussed in the next section, which relates the low risk anomaly to the utility function of the
fund manager: as the largest inflows go toward outperforming fund managers and to well
performing asset classes, these portfolio managers may seek to maximize outperformance in
the upward markets, thus systematically preferring high beta stocks.
Cornell and Roll (2005) recognize that the considerable market share of investment being
nowadays delegated to professional fund managers, impose asset pricing models to
incorporate the objective function of the agents, together with the traditional utility function
of the final investors. The objective function of the fund manager is not the investors wealth
maximization, but the maximization of active return versus the benchmark of the delegated
mandate.
Without a direct implication on performance of low risk versus high risk stocks, they show
how a pricing model based on delegated investments (where fund managers objective
function dominates the investors utility function), in equilibrium implies some cross sectional
relationship between stocks alpha and their beta relative to the benchmark. These
relationships violate CAPM.
Baker, Bradley and Wurgler (2009 and 2011), make a step forward focusing explicitly on the
low risk anomaly.
Backer and Haugen (2012), state that, as the risk return relationship is rather inverted, fund
managers would have a concrete incentive in buying low risk stocks.
Their explanation for fund managers not actually exploiting the anomaly is that they seek to
maximize the probability of receiving a bonus, that is usually paid when performances are
positive in absolute terms, and higher than a threshold (benchmark return plus the
management fees, or an absolute discretionary threshold).
Portfolio managers exchange higher expected returns of low beta stocks, for a higher
probability of beating their benchmarks (or a given target), provided by high volatility stocks.
In other words they exchange a higher mean of returns distribution (low volatility stocks) for
a higher expected value in the right-end tail of the distribution (high volatility stocks).
The authors also mention some additional incentives to hold volatile stocks, this time related
to the delegated portfolio construction process: in order to impress colleagues, analysts are
often willing to recommend stocks in the news, or stocks whose news flow is quite intense
and that tend to exhibit higher than average volatility. Finally fund managers may find it
easier to justify holdings or turnover of newsworthy stocks.
Supporting their intuition, the authors show that for 1000 US stocks grouped in 10
homogeneous classes of market capitalization, from 2000 to 2011 companies with higher
institutional ownership exhibit higher volatility than stocks with low institutional ownership.
Finally, they find that analysts coverage (number of recommendations) is positively
correlated with volatility.
Chunhachinda et al (1997) find that the returns of the world's 14 major stock markets are not
normally distributed. Optimal portfolio compositions are computed (as allocations of 14
international stock indexes) incorporating investors preferences for skewness. The empirical
findings suggest that the incorporation of skewness into the investor's investment process
causes a major change in the construction of the optimal portfolio. The evidence also suggests
that investors exchange expected return for positive skewness.
Kraus and Litzenberger (1976) find that, while having aversion to variance, investors exhibit a
preference for positive skewness. As a consequence, if the capital asset pricing model is
extended to include systematic skewness, this latter is associated with a positive price (instead
of a discount, as it is the case for variance), and the zero intercept for the security market line
is not rejected.
The intuition behind these three contributions is that, as skewness has a positive price, it
should be higher among high beta or high risk stocks, so that these latter are priced at a
premium compared to a CAPM equilibrium, and finally deliver lower average returns.
We tested this hypothesis in the last decade on the constituents of the MSCI World Index in
the period from January 2003 to June 2012. We regressed stocks weekly returns (excluding
only those stocks with less than two years of available data) on the index returns; we then
formed equally populated baskets according their betas. We then computed differences and T-
statistics of the average skewness, for any pair of baskets. The signs and their significance
prove some interesting monotony: higher skewness for higher beta stocks.
Results are summarized in tables 3 and 4 in Annex 2: skewness generally increases with beta,
sometimes significantly, especially for baskets whose differences in beta are high enough.
We repeated the exercise using total ex-post volatility as a measure for sorting and grouping
stocks (tables 5 and 6). Results are even more significant, as skewness increases almost
monotonically with volatility.
We recognize our analysis is completely in sample and misses some predictive power;
however, it is transparent and easily replicable. The hypothesis of skewness increasing with
beta and volatility is confirmed, and this suggests that the premium that investors pay for a
Cowan and Wilderman (2011) find that high beta stocks exhibit a positive convexity relative
to broad market index returns, while low beta stocks have negative convexity.
They explain that high beta stocks provide a call option payoff. In the case of positive market
returns, high beta stocks deliver market returns, multiplied by a factor roughly equal to their
beta (whatever the market return is, exactly like a leveraged position). Conversely, in case of
negative market returns, losses are limited to 100% of invested capital, at worst.
The difference with leveraged investments is straightforward as the latter generate payoffs
exactly equal to the returns of the unleveraged positions times the leverage, with theoretically
no limit to downside.
The authors argue that investors exchange future returns for having this call-type convex
payoff: they pay an additional premium for high beta stocks just like they paid a premium to
buy a call option.
In our view this explanation may be convincing only for very extreme market returns, that is
quite rarely (the convex profile takes place in the form of a stop loss, that is activated in case of
market returns of, lets say, -33% in the case of a stock beta higher than 3, or -50% in case of
beta higher than 2).
However, we apply the methodology described for skew in the previous paragraph, to test if
we find increasing convexity, for increasing beta. For each MSCI World constituent (from
January 2003 to June 2012, excluding companies with less than 100 weekly returns) we run
two OLS regressions: the first with the series of MSCI World returns (in USD) as the only
explanatory variable; and a second one with the series of MSCI World squared returns as an
additional explanatory variable:
Ri = i RMsci + ui (1)
and
2
Ri = i R Msci + i R Msci + vi (2)
We group stocks according to their estimate of from OLS regression 1, and we compute the
average and the standard deviation of the estimate for any basket. We then test the
significance of differences in average basket convexities.
Results are summarized in Annex 3 and seem to support Cowan and Wildermans intuition.
Table 7 reports average convexity and standard deviation across any basket; while table 8
reports differences in average convexity for any pair of baskets.
The higher the distance of betas, the larger and more significant the difference in convexity:
high beta stocks exhibit a higher and thus more profitable convexity than low beta stocks.
Differences are often significant and this may explain some premium paid by investors for
high beta stocks.
Table 9 and 10 report results for the same experiment, when we rank and group stocks
according to total ex-post volatility. Results are even more significant as convexity increases
almost monotonically with volatility.
We have lived for several years in very challenging markets: international equity indexes have
exhibited high realized volatility and quite disappointing returns, during the last decade.
Equity investors have been faced with a major and unfavorable change in traditional risk
return payoffs.
In the last few years, Amundi has strongly invested in order to meet investors needs in such a
challenging market context, developing a range of innovative solutions aiming at Sharpe ratio
improvement. Their risk-return profile differs as well as their behavior in up and down
markets. Over the last decade, these strategies have all succeeded in enhancing risk return
trade-off (Sharpe ratio has been systematically superior to that of relevant equity index as the
MSCI World). They all belong to the absolute risk category: away from the notion of tracking
error or information ratio, they focus on Sharpe ratio or risk-adjusted return, and volatility
metrics. They are based either on the use of instruments providing favorable asymmetry
(options and other derivatives), or on portfolio construction techniques as maximum
The latter two are particularly relevant to this document as both exploit the low risk anomaly
discussed previously.
On an ex-post basis, the volatility of the Global Minimum Variance and Global Smart
portfolios is significantly lower than the market index, with a 10 to 20% reduction for Smart,
and up to a 35% reduction for Minimum Variance (Annex 5.1 and 5.2). However, most of the
relevant literature strictly identifies low risk strategies with the selection of low risk stocks and
-among several measures of risk- systematic risk and beta are by far the most significant. For
this reason it is worth investigating whether Minimum Variance and Smart processes limit
portfolio volatility mainly by selecting low risk stocks, or rather by enhanced diversifications.
In Annex 4, we provide some empirical evidence on two back-tested Minimum Variance and
Smart portfolios, during the period December 2003 December 2011.
Table 11 shows that the ex-post beta of the Minimum Variance portfolio on full data sample is
only 0.55 (significantly lower than 1 at a 1% confidence level). On an ex-ante basis with
quarterly observations, portfolio beta ranges from 0.48 to 0.65. As for the Global Smart
portfolio, table 11 shows that the ex-post beta relative to MSCI World is 0.85 (again
significantly lower than 1 at 1% confidence level) and the ex-ante beta ranges from 0.7 to
0.87.
In table 12 we further investigate the systematic risk characteristics of the two portfolios and
the benchmark in a multifactor framework. For each of them at every quarter of our back-test,
we compute the weighted average common factor risks extracted from the BIM model by
BARRA1. Weighted average common factor risk of the Minimum Variance portfolio is lower
than that of Global Smart; both of them are lower than the MSCI World.
1
Weighted average common factor risk, at any time t of our sample, is computed as follows:
CF(t) = i wi(t) CFi(t)
where CFi(t) is the common factor risk of the ith stock at time t, and wi(t) is its weight at time t.
This evidence supports the intuition by Carvalho, Lu, and Moulin (2011), who infer that the
exposure to low beta and low systematic risk plays a major role in explaining the superior
performance of Minimum Variance portfolios; we further extend these findings to Smart Beta
as well.
In fact, such exposure of Smart portfolios to low risk stocks is somehow intuitive if we
consider the sector allocation process (sector weights are inversely proportional to their
marginal contribution to risk), and the stock weighting scheme (inversely proportional to their
total volatility). The two of them are described in detail in section 3.2.
As shown, the most straightforward way to exploit the low risk anomaly is certainly building a
portfolio of stocks with the lowest possible risk.
In this section we describe our Minimum Variance approach on a Global Developed Equity
universe. We claim several years of experience in Minimum Variance management, with two
Europe portfolios (since 2007 and 2009 respectively), a very recent Global portfolio, and our
paper portfolios on developed World, Japan, Emerging markets, Pacific ex Japan, and other
customized universes. However in this document, we focus on the Global Minimum Variance
as it is the most complete case for descriptive purposes.
The efficient frontier represents the set of portfolios that earn the maximum rate of return for
every given level of risk. We use an optimization process to build a portfolio sitting on the
very edge of the efficient frontier. In building such a portfolio, expected returns are not needed
as the only requirement is to minimize volatility, while being fully invested.
2
Every quarter we build three equally populated baskets of stocks, according to their beta and common factor
risk (high, median, and low risk). We then compute the aggregate weight of stocks in each group, for any
portfolio. Finally we compute the historical average of aggregate weights.
Min (wTVw)
where w is the vector of the optimal portfolio weights, V is the variance-covariance matrix,
and eT is a vector of ones.
Although we recognize the advantage of such a process being transparent and intuitive, we are
conscious of some typical drawbacks that may arise from Minimum Variance portfolios: as
shown in Clarke, de Silva and Thorley (2006), Minimum Variance portfolios may be quite
concentrated on a few low volatility stocks, may exhibit rather high turnover, positive
exposure to value and small capitalization stocks (with some relevant implications on
liquidity), and some volatile exposure to momentum factor.
Similarly, Thomas and Shapiro (2007) highlight the risk of the Minimum Variance portfolio
being excessively concentrated on few low risk sectors, and the lack of control for involuntary
factor exposure. They also express their preference for tilting portfolios toward some
successful stock ranking criteria.
We agree that most of those are relevant issues in portfolio construction and we do believe that
handling them through an optimization package is strictly needed in order to come out with
reasonable and investable portfolios, as only few of them might be addressed correctly with
some clever weighting schemes (like in the case of limiting turnover through an equally
weighted basket of low beta stocks, as suggested by Carvalho, Lu, and Moulin).
Thats why we implement our enhanced portfolio construction process in Barra One, as
described below.
Quality Stocks
We believe that fundamental equity selection can provide some valuable enhancement in the
risk return profile of equity portfolios, at least in the long run. At the same time we dont want
to renounce an optimization process which is completely independent from expected returns.
Expected returns are very noisy in forecast and thus responsible for well-known error
maximization problems.
Keeping 60% of constituents available for investments, the optimizer is left with a high degree
of freedom and it tilts the optimal portfolio toward good quality stocks, without using explicit
expected returns.
Table 11 in Annex 4 reports the balance sheet, income statement and corporate actions
employed in the Piotroski score, and table 4 shows that -in the last decade- the top quality 33%
of MSCI constituents (equally weighted) have outperformed the market index with lower
volatility. At the same time, the median basket has performed in line with the market, and the
bottom basket has underperformed with even higher volatility.
High turnover is a critical issue in many systematic investment strategies like Minimum
Variance.
In our case, turnover in the investment universe is limited as the Piotroski score is based on
balance sheet data that varies very little during one quarter. Furthermore we also rebalance our
portfolio quarterly, as suggested by Baker and Haugen (1991).
Nevertheless, more than turnover itself, our concern is liquidity indeed: we aim to avoid small
illiquid companies as we want to be able to liquidate our portfolio in a reasonable time lag,
without incurring significant market impact costs.
To address this requirement, we limited the amount held in any stock to the following
percentage:
ADVi
UBi = 25% D
NOT
where UBi is the upper bound on the ith stock, D is the number of days that we accept to
liquidate the fund, ADVi is the average daily volume over the last quarter, and NOT is a
notional amount of assets under management of USD 1 billion: quite conservative as it is still
far above the current size of our fund.
As mentioned above, Minimum Variance portfolios may tend to be poorly diversified across
sectors, countries or single stocks. We have thus applied some constraints at these levels,
without preventing the optimizer from choosing solutions that are far enough from a market
index.
On countries and sectors we accept deviations from the market index of 500 to 1000 bp, while
for single stocks we apply a general upper bound (GUB), thus modifying the actual upper
bound as follows:
ADVi
UB i = min GUB;25% D
NOT
We observe that much of our size exposure is corrected away by the liquidity constraints. As
for other factor exposures, we have decided not to manage them systematically as again we
dont want to excessively constrain the optimization process.
On the other hand, we regularly monitor the behavior of all the risk factors of the BARRA
model (size, value, growth, momentum, leverage). The goal of this monitoring is to detect
bubbles or suspicious asymmetries like excessive positive skewness in recent performance
(Sornette, 2003; Morel, Malongo, and Lambinet, 2013): in the case of significant alerts, we
punctually hedge the risk of an exploding bubble, imposing a neutral exposure to the
suspected factor.
As for performance (Annex 5.1, table 15), results are very satisfactory indeed: from the
beginning of 2003 to the third quarter of 2012, Minimum Variance portfolios outperformed
the standard index by a minimum of 3.6% in Pacific (All Countries) ex Japan, to a maximum
of 5% in the World developed Markets, while volatilities and draw-downs were reduced by
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.
In this section we discuss our risk parity approach on global developed markets. We recall that
we currently manage several risk parity portfolios (Euro Area, Europe, World), and we are
investigating the behavior of such strategies in many others areas (World ex Japan, Pacific ex
Japan, Japan, Emerging Markets, Emerging Markets with Sharia filter). Our approach is
consistent across the regions with the Developed World being probably the best example for
descriptive purposes.
Risk parity means that each asset (asset class, equity sector, single stock) has an equal
contribution to the total risk of the portfolio.
As Maillard, Roncalli, and Teiletche (2009) have pointed out, full risk parity cannot be
obtained in a closed formula unless some unrealistic hypotheses (such as equal correlation
among all the assets in the investment universe) are made, and may not be achieved either
through optimization, if the number of assets involved is somehow relevant, and correlations
are very heterogeneous.
As for the number of assets involved, we typically deal with 1,500 to 2,000 constituents of the
MSCI World. For this reason we have decided to split our portfolio construction process into
two steps: the region-sector allocation, and the stock weighting in each regional sector basket.
Optimization (for instance, the minimization of the cross-section standard deviation of assets
contribution to risk) does not guarantee a full risk parity solution. Furthermore, it can be
3
As for the Emerging Markets portfolio (from December 2005), in order to be eligible to minimum variance optimization,
stocks must be held on the Amundi emerging market flagship fund, or must be top-ranked (33%) according to a Piotroski
score.
4
The former of the two European portfolios has been designed in order not to exceed the ex-ante volatility of 10%. Thats the
main source of performance difference between them.
While defining the region and sector allocation of our risk parity portfolio, we distinguish
three regions: North America, Europe, and Pacific. Then, within each geographical region, we
operate on the 10 regional sectors according to GICS definition (Level 1) as homogeneous
groups of stocks.
In the allocation of each economic sector within one region, as well as of each region within
the global portfolio, we reject the very popular method of weighting baskets by the inverse of
their volatility: this procedure ignores correlations that should be taken into account explicitly
instead, as they may be highly heterogeneous across sectors and regions.
One way to account for them is to use the measure of marginal contribution to total risk.
If W is the vector of weights of portfolio P, and is the volatility of portfolio P, MCi is the
marginal contribution to risk of each asset, and is equal to:
MCi =
Wi
In order to come out with full risk parity, the following relation must hold:
In other words the risk contribution should be the same for any basket:
RC = MC W = Ke
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.
1
W TGT ~
MC INIT
MC TGT MC INIT
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 =
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.
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.
The global portfolio (available since January 2012 only) so far exhibits the same risk-adjusted
return as the MSCI index.
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)
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)
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
6
Source for all figures: Amundi, Factset
7
Source for all figures: Amundi, Factset
8
Source for all figures: Amundi, Factset
9
Source for all figures: Amundi, Factset
10
Source for all figures: Amundi, Factset
11
Source for all figures: Amundi, Factset
April 2013
In a portfolio comprised of different assets from the same class, or numerous asset
classes, the drivers of return variation may appear elusive. While there are idiosyncratic
factors influencing return variation, there a r e a ls o common factors that account for
the portfolios collective variation. Uncovering and decomposing the importance of these
Factor models have been widely studied, mainly in macroeconomics and asset pricing.
In macroeconomics, they are used to determine the factors that influence measures of
the economy, or in policy analyses. For example, Bernanke et al. (2005) introduced the
FAVAR model to analyse monetary policy, Forni et al. (2003) study the structure of the
macroeconomy, while Favero et al. (2005) compare static and dynamic principal
Lewbel (1991) applied factor models to budget share data to reveal information about
the demand system. In finance, Chamberlain and Rothschild (1983) extend Arbitrage
Pricing Theory using the factor model, which has since been used not only to
decompose risk and return into explicable and inexplicable components, but also to
describe the returns covariance structure, prediction, and to construct portfolios with
desired characteristics, among others. More recently, Merville et al. (2001) analysed
Factor models are categorized by type of factors. There are (i) macroeconomic factors
example of a well-known single factor model is the Capital Asset Pricing Model
that stocks with a small capitalization and a high book-to-market ratio tend to perform
better led Fama and French to refine CAPM as a three-factor model. In this paper, the
focus is on statistical factor models, and the factors are computed using Principal
Component Analysis.
The operational value of exposing common factors that drive variances in these markets
strategies. For instance, we find that the risk appetite factor in the Global Macro Hedge
Fund accounts for up to 45% of the datas variance, hence i t deserves heightened
sentiment.
factors have been proposed. Arguably the most popular is via information criteria
(IC). IC is based on the idea that an (r + 1)-factor model has to fit at least as well as
an r-factor model, but is less efficient. The well- known Akaike Information Criterion
(AIC) and the Bayesian Information Criterion (BIC) cannot be directly adopted as they
are functions of N or T alone, hence they fail to consistently estimate the number of
factors when the factors are unobserved. Bai and Ng (2002) (BN) propose a set of
six penalty functions to replace the ones in AIC and BIC, and established conditions
ensuring the consistency of their methods. Despite its wide empirical adoption, BNs
criterion often does not converge, as demonstrated in Forni et al. (2007). Alessi et al.
(2009) (ABC) refine BNs criterion and demonstrate that their criterion has superior
performance. Alternate approaches include analysing the factor loadings (Connor and
(2009), Onatski (2009b), Kapetanios (2005)), numerous tests on the rank of the
covariance matrix (Lewbel (1991), Forni and Reichlin (1998)), and a graphical method
that is rarely used due to its lack of theoretical basis (Donald, 1997).
to five datasets yields the following number of factors: five for Global Macro Hedge
Fund (GMHF), three for US Treasury Bond Rates (USTB), two for Commodity Prices,
and one each for US Credit Spreads (USCS) and currencies. T h e t otal variation
explained by the factors varies, as it is 74% for GMHF, 94% for USTB, 49% for USCS,
27% for commodity prices, and 59% for currencies. Economic interpretation is attached
to the factors according to correlation between the factors and the r e t u r n s o f assets
whose variances they describe. The five factors for GMHF are associated with risk
appetite, commodities, t he US dollar, the Japanese market and Asian stock markets.
Those for USTB fit the description of factors found in previous research and are
labelled as level, slope and curvature. USCSs sole factor corresponds to mid-range risky
assets while that for currencies is labelled as the carry factor. The pair of factors for
The stability of the number of factors over time is investigated by testing the
components, the factors are, in theory, orthogonal to each other, the estimated factors
may not be so. Instantaneous correlation between these estimated factors can be
between the factors over rolling windows does not yield an overarching conclusion for
all datasets regarding the hypothesis that during periods of economic downturn, fewer
factors are required to explain variances in the data due to increased cross-market
between factors increased. Yet for commodity prices, it is less clear. When all asset
classes are considered together in the GMHF dataset, the different factor interactions
that are observed within each asset class manifest in a more complex manner, and
spike in the correlation is observed prior to the 2007 to 2010 financial crisis.
This paper is organized as follows. We first present the factor model and briefly explain
the numerous criteria proposed to determine the number of factors. After a Monte Carlo
Study for selected methods, we apply the methods to datasets relevant to the investment
analysed.
2. Methodology
whereby Xit is the observed data for the ith cross section at time t, F t is the r 1 vector
conjugate transpose of the matrix. Ft and et are assumed to be uncorrelated and the
matrix comprised of cov(ei, ej) is not necessarily diagonal (i.e. allows non-
value of loss function V (r, F r )+rg(N, T ), or log(V (r, F r ))+r 2 g(N, T ), with
been shown to possess good properties when errors are cross-correlated (Bai and
Ng, 2008).
The number of estimated factors remains as the one yielding the lowest value for
these modified loss functions. Furthermore, the authors suggest evaluating the
loss functions over random subsamples of the data to find an estimate that is
of the role of c are provided in Hallin and Liska (2007), while the generation of the
random subsamples is described in Alessi et al. (2009). This criterion has been
shown to provide a solution when BNs criterion fails, and it is not any more
BN.
on the idea that an r factor models (r + 1)st factor can have nontrivial factor
loadings for some assets, but only a small proportion of them. A statistical test
for this is developed to test whether the (r + 1)st factor is pervasive. It proceeds by
running two regressions by Ordinary Least Squares (OLS), one with factors
sectional mean for both it s, defined as , is calculated next, for both regression
models. Then the even months for the regression with r + 1 factors is
subtracted from the odd months for the r factor model, giving a value N .
Under the null hypothesis that the model has r factors, N , with as the
1
4. Onatski (2009b)
Drawing upon the property that an r factor panel of data has unbounded first r
lower bounds for the number of factors, which are determined by prior
postulated by the alternative hypothesis that the model has more r factors, but
fewer r1 factors. Critical values for the test are provided in Onatski (2009b) for
Before choosing one method over another, we perform a Monte Carlo test to evaluate their
relative performance on simulated data with various qualities. The experimental design
Ftj and ij are normally distributed with zero mean and unit variance. This is similar to
idiosyncratic component:
5. Small cross-section correlation across idiosyncratic parts, same variance for the
6. Seri al corr elation across idiosy ncrat ic parts, t h e com mon component has a
For all seven DGPs, we test for the pairs of time and cross-section dimension (N, T ) =
(70, 70), (100, 120), (150, 500). The true number of factors, r is chosen to be 1, 3, 5,
8, 10, and 15, all of which are consistent with the requirement r < minN, T . The
corresponding rmax for BN and ABC, and the upper bound on CK and Onatski test is
Onastkis test always began with the lower bound of 1, to suggest that in many financial
datasets, it is unlikely to have prior knowledge beyond the belief that there should be at
least one factor, given that the data does indeed have a factor structure. The
while H = max . 500 Monte Carlo replications are performed for each instance.
subsamples a r e :
nJ = (see Alessi et al. (2009)), and cmax = 13 with step size 0.01.
BNs criterion has perfect performance for DGPs 1-4, correctly identifying the number
overestimates the number of factors. There is no obvious effect of dimensions (i.e. N and
T ) on the results, which is consistent with BNs claim that their criterion yields precise
performance is more accurate for DGP 5. This result is similar to ABCs own Monte
Carlo study. Even though BN has stellar performance in most cases, the adoption of
ABC is justified since most financial portfolio time series demonstrate cross-section and
serial correlation.
Onatskis criterion performance pales in comparison with the other criteria in almost all
cases, as it estimates that the true number of factors is 1 close to 60% of the time, 2
about 30%, and 3 about 10% of the time, being insensitive to the true number of factors.
This could be due to having the lower bound of the test always set at one to reflect the
case that when the test is implemented on actual data, no prior knowledge is available
to determine the lower bound. An upper bound, however, can be set since it must be
fewer than minN, T , and methods such as ABC are developed to be less sensitive to
the upper bound, hence a larger upper bound can always be selected.
CKs test has a similar tendency of underestimating the true number of factors as 1
close to or exceeding 50% of the time when r 5, when the time dimension is small.
However, for the sample with N = 150, t = 500, i.e. large cross-section and time
dimensions, the test performs reasonably well for all DGPs except for DGP 2 and 7,
correctly identifying the number of factors at least 50% of the time for DGP 1-4 and 6,
and overestimating the factor by 1 for DGP 5. In the case of DGP 2, the number of
estimated factors is 1 more than 80% of the time, regardless of the actual number of
In general, the Monte Carlo study substantiates BNs criterion as superior not only in
terms of accurate estimates, but also in terms of its ease of implementation2 . In cases
CKs test may perform well when the time and cross-section dimensions are large.
The three best criteria: BN, ABC and CK are implemented on five sets of data comprising
equities, commodities, credit spreads, interest rates, and currencies between 1997 and
2012.
3. Empirical Results
In this section, ABCs criterion is applied to five datasets covering major asset classes
such as equities, US Treasury Bonds, credit spreads, currencies and commodities. The
estimated number of factors are first identified and labelled. Then, the stability of
factors over time is analysed by testing the significance of correlation between them,
and re-estimating the number of factors after splitting each dataset along the time
The Global Macro Hedge Fund (GMHF) dataset is comprised of major indices,
government bonds, currency exchange rates, currency exchange rate and oil futures,
dated between January 1999 and March 2012. The data is of weekly frequency due to
The US Credit Spreads (USCS) data is comprised of daily closing rates categorized by
industry (e.g. financial corporation, insurance, and energy), financial rating (e.g. AAA,
AA, BBB), and duration (e.g. 1-3 years, 3-5 years), from January 1997 to March 2012.
T h e d aily closing prices of commodities such as gold, aluminium, natural gas, corn,
March 2012 are compiled in the dataset of Commodities. The fifth dataset is called
Currencies, and includes the daily prices of the Euro, Great Britain pound, Swiss Franc,
Japanese Yen, Canadian Dollars, Australian Dollars, New Zealand Dollars, Norwegian
Krone, and the Swedish Krona in terms of US Dollars from January 1999 to December
2012. BN, ABC and CK criteria are applied to log prices or log spread variations for
GMHF, USCS, Commodity Prices and Currencies datasets, whereas for USTB, rate
variation is used since with two-year rates close to zero or negative, their percentage
Aside from applying BN, ABC and CK criteria to the datasets, since the Monte Carlo
study indicates that the selected criteria generally have poorer performance when cross-
section correlation exists, the influence of such dependencies on the accuracy of results
dynamic linear dependence, then applying the same criteria on the residuals. Using the
AIC criterion to determine the number of lags to include in the VAR model, USTB
and USCS datasets are fitted with VAR(3), commodity prices with VAR(2), and GMHF
and currencies with VAR(1). BN and ABC criteria provide the same outcome when
the analysis is done on the residuals as on the returns data. CKs criterion yields
slightly different estimates. Despite its commendable performance in the Monte Carlo
study, BNs criterion fails to converge on all datasets - rmax is always estimated. CKs
estimates do not always coincide with ABCs estimates, but the latter is taken to be
more accurate due to better performance in the Monte Carlo study, and it is invariant
In Exhibit 6, the number of factors estimated and the proportion of variance explained by
each factor,
of the returns, are presented. The proportion of variances explained by the estimated
for which correlations across asset variations are high should have a high percentage
explained. Using ABCs criterion, the GMHF is estimated to have five factors, which
collectively explain about 74% of variances in the dataset. USTB has three factors
that explain 94% of the variances. The commodity prices dataset is estimated to have
two factors. Together they explain 27% of the variances - the lowest among the
the findings of Gorton and Rouwenhorst (2004) on the asset classs diversification
potential, and becomes the rationale for investors to increase portfolio allocation to
estimated for US credit spreads and currencies, accounting for around 49% and 59% of
After obtaining the number of factors, the sensitivity of each assets return to the factor,
i.e. factor loadings, is investigated and presented in bar charts. This notion of sensitivity
can be extended to that of a portfolio, as it is merely the sum of the corresponding asset
in the portfolio. Moreover, portfolios that are insensitive to a particular factor can be
constructed by selecting assets such that their weighted sum (i.e. loadings treated as
weights) is zero. Factor loadings also help in identifying the factor, that is, by
labelling the latent, hypothetical factors according to their relationship with the assets.
Absolute correlation of the factors with the assets returns is first placed in
descending order, and progressively added to the selection, i.e. adding those with the
largest absolute correlation first, until the selection achieves at least 95% R2 when used
as explanatory variables in a simple regression model for the asset returns. These
Exhibit 2 shows that factor 1 of GMHF is highly correlated to major equity indices
worldwide. This suggests that factor 1 corresponds to a risk appetite factor. During
bullish periods, investors are more willing to take risks, hence the y prefer to invest in
equities. Conversely, during bearish periods, investments are diverted into bonds, which
have lower risk. Factor 2 relates to oil and GSCI, hence it is labelled a commodities
factor. Factor 3 is a dollar factor due to its association with the US dollar. Factor 4
US Treasury Bonds
Prior research by Dai and Singleton (2000) and Litterman and Sheinkman (1991) has
shown that observed variation in bond prices can be explained by three factors: level,
slope, and curvature. These names describe the shift of the yield curve in response to a
shock. A level shock shifts the curve in a parallel manner, resulting in an almost equal
effect on bonds of all maturity. The slope factor implies larger shocks for bonds with
from the effect of the yield curve becoming less steep as a result of a slope shock.
yield curve. Indeed, Exhibit 3 shows that factor 1 has relatively uniform correlation
across bonds of all maturity, just as a level factor would. T h e c orrelation for factor 2
changes sign once, and has a larger correlation, a n d hence impact, on bonds with
shorter maturities, as should a slope factor. Factor 3 has a hump in its correlation
figure, fitting the description of a curvature factor. Thus, the findings are consistent
The sole factor for USCS is correlated to A-rated investments, financial corporates and
industrial credit spreads that dominate those with utilities and consumer cyclicals, as
shown in Exhibit 3, implying that it is associated with assets with mid to low credit
risk. Investments with low credit risks and conventionally small credit spreads, such as
AAA-rated assets and treasury bonds, are not among those that possess the highest
correlation with the factor, suggesting that they play a relatively smaller role in the
movement of returns. Similarly, industries responsible for providing goods with low
substitutability, such as utilities, healthcare and energy, along with those that are highly
dependent on the state of the economy, such as consumer cyclicals (e.g. entertainment,
automotive etc.), possess lower correlation with the factor. In comparison with other
datasets, assets in USCS have a relatively more uniform correlation (i.e. all above 50%)
Commodity Prices
the energy factor. Factor 2, being correlated with numerous metals, is the metals
factor, as is evident in Exhibit 4. Similar to the case of USTB, the number of factors
pricing models intended for equities, macro and equity-motivated factor models, and
principal component factor models to find that none of them satisfactorily prices
commodity assets. The authors attribute this poor pricing model performance to
markets. Our results are consistent with those of Daskalaki et al. since among all
datasets, commodity price factors collectively explain the least amount of variance in
the data, an observation that supports the commodity diversification effect, made
their portfolio allocation in this asset class (Daskalaki and Skiadopoulos, 2011).
Furthermore, the identification of factors by commodity group and the low correlation
across these groups suggest a sectorial framework when studying the commodity
market.
Currencies
For currency datasets, ABC estimates a single factor. It is labelled as the carry factor
as it is correlated to currencies with high average rates, such as the Australian Dollar,
understood as the dollar factor as all currency prices are positively correlated with the
literature regarding the number of factors in the currency datasets. Thus, by applying
ABCs criterion to the datasets, the number of salient factors is determined. These
factors are estimated and identified by analysing their correlation with returns.
Since all datasets span ten years, including the most recent financial crisis in 2008, it is
of interest to know whether the number of estimated factors is stable over time.
assumed stability or relied on some graphical method. Bliss (1997) divided his sample
into three sub-periods and investigated the factor loadings on each. His hypothesis is
that if the factor loadings appear to be similar a c r o s s all sub periods, then the
factor structure is stable. Prignon and Villa (2002) found that factor loadings are
stable, but factor volatility varies over time. Chantziara and Skiadopoulos (2008)
analysed the term structure of petroleum futures, also by splitting the sample into two.
Since the PCA results are similar in both samples, the authors conclude that the factor
structure is stable. Attempts to devise formal tests include those by Audrino et al.
the latter involves constructing a bootstrap distribution for the test statistic. An
To investigate the stability of the factors on our dataset, two approaches are attempted.
The first assumes that the factor structure is stable, but the correlation between the factors
may not be. Correlation between factors when the factor structure is stable is closely
between factors, the t-test for significance of correlation is used. Next, discarding the
assumption of a stable factor structure, the dataset is split into expansion and
contraction economic periods, and the corresponding number of estimated factors in each
and expansion are lagged by a negative number of months, from -1 down to -12. In other
words, the number of factors prior to the start of a contraction or expansion period is
computed to determine if the change in number of factors occurs before the economy
takes a turn in its cycle. The key difference between the two approaches is that by
assuming the factor structure is stable in the former, the estimate of the number of
factors is done only once, and the focus is on their dynamics, a n d specifically t h e i r
correlation, over time. In the latter, the number of factors is estimated for each variation
In this section, the factor structure estimated in the previous section is assumed to be
stable but the dynamics between factors evolve over time. Correlation between the factors
is tested over six-month rolling windows for GMHF, and one-year rolling windows for
correlation. Exhibits 7 and 8 plot the number of uncorrelated factors as a result of the
Under the null hypothesis that r = 0, i.e. the correlation between the factors is
Plots of the number of uncorrelated factors suggest that the factors tend to be
particularly obvious for USTB. Exhibit 7 shows a marked drop in the number of
uncorrelated factors to only one factor during the most recent financial crisis. This is
attributable to historically low interest rates as investors sought safe investments. Since
the factors were identified by their impact on bonds of different maturity, i.e. factor 1
affects bonds of all maturity evenly, factor 2 has a strong impact on short maturity
bonds while factor 3 has the highest influence on bonds of mid-term maturity, having all
factors collapse onto a single factor suggests that the prolonged near-zero short-term
interest rates and low long maturity rates have removed the disproportionate impact
that shocks have on the yield curve of bonds of varying maturity. Hence, the level,
the number of uncorrelated factors remains stable at two throughout the period
studied. This is in line with the result in Kat and Oomen (2006) that there is weak
correlation across commodity groups. Indeed, energy and metals, the two factors
identified, are distinct commodity groups. Moreover, energy and metals are essential to
among commodity assets. Thus, the sectorial view of the commodity market remains
Since USCS and currencies have only one factor, stability analysis is not applicable.
Out of curiosity, we over-estimate the number of factors at five each, and apply the
same t-tests to investigate the evolution of the number of uncorrelated factors over time.
Exhibit 8 for USCS shows that the number of uncorrelated factors fluctuates between
one and two thus does not provide conclusive evidence that the number of factors is
lower during economic crisis. As for currencies, Exhibit 8 demonstrates that the number
of factors r e m a ins constant at one for most of the time period, except in early 1999,
when occasionally two factors are estimated. This period coincides with the introduction
of the Euro, which could have given rise to an interim factor influencing the returns.
For the 5% test on GMHF, the number of uncorrelated factors is never five - the
number estimated using ABCs criterion over the entire horizon - during recession. Since
it is also fewer than five in many other instances, it is less clear whether higher
the mean absolute correlation between factors is plotted in Exhibit 9, on which there is
an indisputable spike in correlation in 2007. Hence, prior to the most recent financial
correlation. It could also be argued that the number of factors reduced before recession,
an observation that motivated the analysis in the next sub-section. With as many as five
factors, the dynamics between factors do evolve over time, as shown by the mean
The stability of the factors is dependent on the asset class. For USTB, which is
considered to be the lowest risk among those considered, the number of factors that
between late 2007 and 2010. Commodity price factors seem invariant to the economic
climate. When combined in the GMHF portfolio, these interactions become more
is possible that the change in the number of factors occurs prior to contractions in
the economy. To investigate this, the contraction periods, as determined by NBER, are
lagged by a negative number of months. For example, Lag = -1 month of the most
recent financial crisis between December 2007 and June 2009 refers to the interval
November 2007 to May 2009. The results are presented in Exhibit 10. For USTB,
the number of factors fluctuates between one and five during expansion periods,
supporting the view that the correlation between factors changes prior to recessions.
On the contrary, GMHF and US Credit Spreads have a constant number of factors
coherent with the t-test for significance of correlation results. The number of factors in
the currencies dataset falls to zero prior to contraction periods. Therefore, the view of
4. Conclusion
This paper investigates the number of cross-asset uncorrelated strategies that are
determine the number of factors, four methods are selected and tested. The Monte
Carlo analysis suggests that the criterion by Alessi et al. (2009) is most reliable. T h e
number of factors by ABC in parenthesis: Global Macro Hedge Fund (5), US Treasury
Bond Rates (3), US Credit Spreads (1), Commodity Prices (2), and Currencies (1).
Plots of the number of uncorrelated factors do not all support the hypothesis of
for US Treasury Bond rates, which is most likely due to US post-financial crisis
combination of the observed outcomes on the rest of the datasets, yielding fluctuating
correlation between the factors during economic downturns. However, the results
regarding stability have to be considered with caution, as not every financial crisis is
assets occurring outside the NBER-determined recession periods. With more information
on the factors t h a t drive the returns o f different asset classes, investors would have a
better understanding of the common sources of risk. Depending on the asset class in
mind, a strategy that is built upon exploiting these common sources of risk may have to
Futures (5 assets) CHF futures, JPY futures, AUD futures, CAD futures, GBP futures.
Commodities (2 assets) Oil, S&P GSCI.
Others (1 asset) Mini SP 500.
Eurostoxx
SMI
EURJPY
Nikkei CHFJPY
SMI USDCAD
Hang.Seng
CAD.fut
Dow.Jones
AUD.fut
IBEX NZDUSD
USDSGD AUDUSD
AEX
GBPJPY
Mini.SP500
GBP.fut USDCHF
FTSE
CHF.fut EURUSD
DAX
CAC.40 USD
GSCI
Eurostoxx
Oil
0.0 0.2 0.4 0.6 0.8 0.8 0.0 0.2 0.4
Correlation Correlation
CADJPY GSCI
CAD.fut
USDCAD
USDTWD
USDKRW
Oil
USDSGD
EURCAD
AUDUSD
AUD.fut
GBPJPY
GBP.fut USD
CHF.fut
EURJPY
USDCHF
EURUSD
CHFJPY
Correlation Correlation
Eurostoxx
CAD.fut
Mini.SP500
USDCAD
IBEX
EURGBP
CHFJPY
EURJPY
CHF.fut
USDCHF
USD
EURUSD
Hang.Seng
Singapore
Nikkei
Topix 16
0.4 0.0 0.2
Correlation
30Y 30Y
25Y 25Y
20Y 20Y
15Y 15Y
10Y 10Y
9Y 9Y
8Y 8Y
7Y 7Y
5Y 5Y
4Y 4Y
3Y 3Y
2Y 2Y
1Y 1Y
6m 6m
3m 3m
0.2
0.0
0.2
0.4
0.6
0.8
0.0
0.2
0.4
0.6
Correlation Correlation
25
20Y A
15Y
9Y
8Y Industrials
7Y
5Y Years
4Y
3Y BBB
2Y
1Y Years
6m
3m AA
0.4
0.3
0.2
0.1
0.0
0.1
0.2
A partial characterization of USTB factors is the number of sign changes; zero, one and two, respectively. This is
satisfied by the above figures.
WTI
GSCI.Energy
Heating.Oil
Natural.Gas
0.7 0.6 0.5 0.4 0.1 0.0 0.00.2 0.4 0.6 0.8
Correlation Correlation
Factor 1 (Currencies)
SEK
NOK
NZD
AUD
CAD
JPY
CHF
GBP
EUR
Correlation
Factor 1 2 3 4 5
Label Global Equities Commodities US dollar Japanese Market Asian Stock Markets
Eigenvalue 0.135 0.041 0.022 0.015 0.014
Proportion (%) 45 13 7 4.8 4.2
Cumulative (%) 45 58 65 69.8 74
US Credit Spreads
Factor 1
Label Mid-range risky assets
Eigenvalue 0.895
Proportion (%) 49
Commodity Prices
Factor 1 2
Label Energy Metals
Eigenvalue 0.206 0.181
Proportion (%) 15 13
Cumulative (%) 15 27
Currencies
Factor 1
Label Carry factor
Eigenvalue 0.11
Proportion (%) 59
Top to Bottom: 1% level t-test for significance of correlations for GMHF, USTB, and Commodity
5 Prices
No. of uncorr. fact.
4
3
2
1
Time
1.0 1.5 2.0 2.5 3.0
No. of uncorr. fact.
Time
5
No. of uncorr. fact.
4
3
2
1
Time
The shaded areas mark the intervals of contraction (peak to trough) of business cycles, i.e. March
to November 2001, and December 2007 to June 2009 (National Bureau of Economic Research
(2012)).
2.0
1.5
1.0
Time
2.0
1.8
No. of uncorr. fact.
1.4
1.2
1.0
Time
The shaded areas mark the intervals of contraction (peak to trough) of business cycles, i.e. March
to November 2001, and December 2007 to June 2009 (National Bureau of Economic Research
(2012)).
0.35
0.30
Mean Absolute Correlation of Factors
0.25
0.20
0.15
0.10
Time
The shaded areas mark the intervals of contraction (peak to trough) of business cycles, i.e. March
to November 2001, and December 2007 to June 2009 (National Bureau of Economic Research
(2012)). The horizontal dotted line indicates the median.
Dataset Global Macro Hedge Fund US Treasury Bonds Rate US Credit Spreads
Lag (No. of months) Expansion Contraction Expansion Contraction Expansion Contraction
0 5 6 3 4 1 1
-3 6 6 5 3 1 1
-6 5 5 2 3 1 1
-12 5 4 2 3 1 2
Alessi, L., Barigozzi, M., and Capasso, M. (2009). A robust criterion for determining
the number of factors in approximate factor models. Technical report.
Audrino, F., Barone-Adesi, G., and Mira, A. (2005). The stability of factor models of
interest rates.
Journal of Financial Econometrics, 3(3):422441.
Bai, J. and Ng, S. (2002). Determining the number of factors in approximate factor
models. Econometrica, 70(1):191221.
Bai, J. and Ng, S. (2008). Large dimensional factor analysis. Foundations and Trends in Econometrics,
3(2):89163.
Bernanke, B., Boivin, J., and Eliasz, P. S. (2005). Measuring the effects of monetary
policy: A factor- augmented vector autoregressive (favar) approach. The Quarterly
Journal of Economics, 120(1):387 422.
Bliss, R. R. (1997). Movements in the term structure of interest rates. Economic Review,
(Q 4):1633.
Merville, L. J., Hayes-Yelken, S., and Xu, Y. (2001). Identifying the factor structure
of equity returns.
Journal of Portfolio Management, 27(4):5161.
National Bureau of Economic Research (2012). US business cycle expansions and
contractions. Onatski, A. (2009a). Determining the number of factors from empirical
distribution of eigenvalues. To
appear in the Review of Economics and Statistics.
Onatski, A. (2009b). Testing hypotheses about the number of factors in large factor
models. Economet- rica, 77(5):14471479.
Prignon, C. and Villa, C. (2002). Permanent and transitory factors affecting the
dynamics of the term structure of interest rates. Technical report, International Center
for Financial Asset Management and Engineering.
Philip, D., Kao, C., and Urga, G. (2007). Testing for instability in factor structure
of yield curves. Technical report, Cass Business School.
Social Responsibility
and Mean-Variance
Portfolio Selection
Bastien Drut, PhD, Fixed Income
and Forex Strategist Amundi
In Markowitzs (1952) setting, portfolio selection is driven solely by financial parameters and
the investors risk aversion. This framework may however be viewed as too restrictive since,
in the scope of Socially Responsible Investment (SRI)1, investors also consider non-financial
criteria. This paper explores the impact of such SRI concerns on mean-variance portfolio
selection.
SRI has recently gained momentum. In 2007, its market share reached 11% of assets under
management in the United States and 17.6% in Europe.2 Moreover, by May 2009, 538 asset
owners and investment managers, representing $18 trillion of assets under management, had
signed the Principles for Responsible Investment (PRI)3. Within the SRI industry, initiatives
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-
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,
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)
restricting the investment universe to SRI funds can seriously harm diversification. Taken at
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.
prior exclusion, with each investor being allowed to choose her own SR commitment (Landier
and Nair, 2009). This translates into a trade-off between financial efficiency and portfolio
ethicalness (Beal et al., 2005). Likewise, Dorfleitner et al. (2009) propose a theory of mean-
variance optimization including stochastic social returns within the investors utility function.
second-generation SRI is still missing from the literature. Moreover, the impact of risk
aversion on the cost of SRI has not been investigated so far. Our paper aims at filling those
two gaps. By delineating the conditions under which SRI is costly, it will furthermore help
elucidate the apparent contradiction found in the literature regarding SRIs influence on
diversification.
This paper measures the trade-off between financial efficiency and SRI in the traditional
and her SR-sensitive counterpart in order to assess the cost associated with SRI. Our
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.
investors only), and d) the full frontier is penalized (i.e. a cost for all investors). Despite its
crucial importance, practitioners tend to leave the investors risk aversion out of the SRI story.
Our paper on the other hand offers a fully operational mean-variance framework for SR
portfolio management, a framework that can be used for all asset classes (stocks, bonds,
commodities, mutual funds, etc.). It makes explicit the consequences of any given SR
threshold on the determination of the optimal portfolio. To illustrate this, we complement our
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.
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
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
5
Notations of Lo (2008) are used here.
variance V p2 Z p ' 6Z p , Let O ! 0 be the parameter accounting for the investors level of risk
Problem 1
O
max Z ' P Z' Z
^Z ` 2 (1)
subject to Z 'L 1
Let us now add an SR rating independent from expected returns and volatilities. Typically,
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.
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
6
For sake of simplicity, short sales are allowed here.
7
See for instance the Carbon Efficient Index of Standard & Poors with the carbon footprint data from Trucost
PLC.
8
For negative screening, Ii is binary and I p denotes the proportion of portfolio p invested in the admissible
assets.
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)
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
SR rating. In other words, if G 1 ! 0 , resp. G 1 0 , the best SR-rated portfolios are at the top,
Consider now the case of an SR-sensitive investor. For instance, she requires a portfolio that
threshold I 0 on I p . For positive screening, the threshold value is left to the investors
discretion (Beal et al., 2005, Landier and Nair, 2009). The SR-sensitive optimization is
summarized by Problem 2.
Problem 2
O
max Z ' P Z' Z
^Z ` 2
subject to Z 'L 1 (4)
Ip Z 'I t I 0
We derive the analytical solutions to this problem by following Best and Grauers (1990)
The shape of the SR-sensitive efficient frontier depends on the sign of G 1 and on the threshold
G1 0 G1 ! 0
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
Proposition 2 makes explicit the situations in which there is an SRI cost. The impact of the
constraint on the SR ratings depends on the parameter G 1 and on the strength of the constraint.
As showed in Proposition 1, if G 1 ! 0 , resp. G 1 0 , the best SR-rated portfolios are at the top,
resp. at the bottom, of the SR-insensitive frontier: by consequence, the SR constraint impacts
wants a well-rated portfolio, that is to say the higher the threshold I 0 , the bigger the portion of
the efficient frontier being displaced. In the case where the threshold I 0 is below the minimum
rating of the SR-insensitive frontier, the efficient frontier is even not modified at all. We
L ' 1 I
In the case where G 1 ! 0 and ! I 0 (see Figure 1), the SR-sensitive and the SR-
L ' 1 L
insensitive frontiers are the same and there is no SRI cost at all. This is the most favourable
case.
L ' 1 I
Figure 1 SR-sensitive frontier versus SR-insensitive frontier with G 1 ! 0 and ! I0
L ' 1 L
Mean
Standard Deviation
L ' 1 I
In the case where G 1 ! 0 and I 0 (see Figure 2), the SR-sensitive and the SR-
L ' 1 L
insensitive frontiers are the same above the corner portfolio defined by its expected return
E 0 and its expected variance V0 V 02 . For portfolios with lower expected returns and
variances, the SRI constraint induces less efficient portfolios. There is only an SRI cost for
E0
Mean
0
Standard Deviation
L ' 1 I
In the case where G 1 0 and ! I 0 (see Figure 3), the SR-sensitive and the SR-
L ' 1 L
insensitive frontiers are the same below the corner portfolio ( E 0 , V0 ) . For portfolios with
higher expected returns and variances, the SRI constraint induces less efficient portfolios.
There is only an SRI cost for investors whose risk aversion parameter is below the
threshold O0 .
L ' 1 I
Figure 3 SR-sensitive frontier versus SR-insensitive frontier with G 1 0 and ! I0
L ' 1 L
E0
Mean
0
Standard Deviation
insensitive frontiers are totally different. The SRI constraint induces less efficient portfolios
for every investor. There is an SRI cost for everybody. This is the worst case.
L ' 1 I
Figure 4 SR-sensitive frontier versus SR-insensitive frontier with G 1 0 and I0
L ' 1 L
Mean
Standard Deviation
To sum up, while the investors risk aversion is generally left out of the story in the SRI
practice, we show in this Section that this parameter matters in the cost of responsible
investing9. Indeed, we show that this SR cost depends on the link between SR ratings and
financial returns and on the investors risk aversion; four cases being possible: a) the SR-
sensitive frontier is the same as the SR-insensitive frontier (i.e. no cost), b) only the left
portion of the efficient frontier is penalized (i.e. a cost for high-risk-aversion investors only),
c) only the right portion of the efficient frontier is penalized (i.e. a cost for low-risk aversion
investors only), and d) the full frontier is penalized (i.e. a cost for all the investors).
9
This section highlights the impact of a constraint on the portfolio rating in the mean-variance optimization.
However, the cost of investing responsibly, if non-zero, may be non-significant. The significance of the mean-
variance efficiency loss may be assessed using the test of Basak et al. (2002) or any spanning test (see de Roon
and Nijman (2001) for a literature review).
In this section, we assume the existence of a risk-free asset and we explore, in this case, the
impact of considering responsible ratings in the mean-variance portfolio selection. Indeed, the
social responsibility of this risk-free asset should also be taken into account. So, we assess
first the social responsibility of the optimal portfolios obtained by an SR-insensitive investor.
And then we study whether an SR-sensitive investor is penalized by requiring portfolios with
high SR standards.
Denote by r the return of the risk-free asset and by Zr the fraction of wealth invested in this
risk-free asset. The standard mean-variance portfolio selection in the presence of a risk-free
asset has been extensively studied by Lintner (1965) and Sharpe (1965). It corresponds to
Problem 3.
Problem 3
O
max Z ' P Z r r Z ' Z
^Z ` 2 (5)
subject to Z 'L Z r 1
In the mean-standard deviation plan, the set of optimal portfolios is referred as the well-
known Capital Market Line (CML). As the investor does not consider responsible ratings in
responsible rating of the risk-free asset and the portfolio rating is defined as I p Z 'I Z r I * .
In the following, we seek to determine the portfolio ratings I p of the optimal portfolios on the
(i) Along the SR-insensitive capital market line, the responsible rating I p is a linear
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 )
Proposition 3 attributes an SR rating of any portfolio of the SR-insensitive capital market line.
From the optimality conditions comes the fact that, along the capital market line, both the SR
1
rating I p and the expected return P p are linear functions of the quantity It is therefore
O
straightforward that the SR rating I p can be written as a linear function of the expected
return P p as in eq. (6). It is striking that this relationship expressed by eq. (6) has the same
form as eq. (3) in the case without a risk-free asset. Note that the portfolio of an infinitely risk
averse investor would be fully invested in the risk-free asset and would have its SR rating I * .
Here, the direction of this link is determined by the sign of the parameter G 1* . In the same way
as in Section 2, the sign of the parameter G 1* is crucial because it represents where the trade-
off appears between risk aversion and SR rating. If G 1* ! 0 , resp. G 1* 0 , the riskier the optimal
best SR-rated portfolios are at the top, resp. at the bottom, of the SR-insensitive capital market
line.
Similarly to Section 2, we now consider the case of SR investors wishing high SR standards
and so, requiring the portfolio rating I p Z 'I Z r I * to be above a threshold I0 . This
corresponds to Problem 4.
Problem 4
O
max Z ' P Z r r Z ' Z
^Z ` 2
subject to Z 'L Z r 1 (7)
Ip Z 'I Z rI * t I0
In Problem 4, the constraints in the mean-variance optimization are also linear. Thus, we
employ Best and Grauers (1990) methodology, as we did for Problem 2. Proposition 4
The shape of the SR-sensitive capital market line depends on the sign of G 1* and on the
G 1* 0 G 1* ! 0
( P rL )' 6 1 (I I *L )
With O*0
(I0 I *)
The associated expected return E0* and the expected variance V0* are:
1
E 0* r ( P rL )' 1 ( P rL )
O*0
1
V0* V 02* 2
( P rL )' 1 ( P rL )
O*0
Proposition 4 makes explicit the situations in which there is an SRI cost in the presence of a
risk-free asset. As in Proposition 2, the impact of the constraint on the SR ratings depends on
resp. G 1* 0 , the best SR-rated portfolios are at the top, resp. at the bottom, of the SR-
insensitive capital market line: by consequence, the SR constraint impacts first the capital
market line at the bottom, resp. at the top. However, contrary to the case without a risk-free
asset, the modified part of the capital market line has a different mathematical form: for this
segment, the capital market line becomes a hyperbola in the mean-standard deviation plan.
In the case where G 1* ! 0 and I * ! I0 (see Figure 5), the SR-sensitive and the SR-insensitive
capital market lines are the same and there is no SRI cost at all. This is the best case.
Figure 5 SR-sensitive capital market line versus SR-insensitive capital market line with
G 1* ! 0 and I * ! I0
Mean
Standard Deviation
In the case where G 1* ! 0 and I * I0 (see Figure 6), the SR-sensitive and the SR-insensitive
capital market lines are the same for portfolios below the corner portfolio defined by its
expected return E0* and the expected variance V0* V 02* . Below this portfolio, the SR-
sensitive capital market line becomes a hyperbola. There is an SRI cost only for investors
with cold feet, that is to say with a risk aversion parameter above the threshold O*0 .
G 1* ! 0 and I * I0
Mean
E0*
0*
Standard Deviation
In the case where G 1* 0 and I * ! I 0 (see Figure 7), the SR-sensitive and the SR-insensitive
capital market lines are the same for portfolios above the corner portfolio ( E 0* , V0* ) . Above
this portfolio, the SR-sensitive capital market line becomes a hyperbola. There is an SRI cost
only for investors with a risk aversion parameter below the threshold O*0 .
Figure 7 SR-sensitive capital market line versus SR-insensitive capital market line with
G 1* 0 and I * ! I 0
Mean
E0*
0*
Standard Deviation
In the case where G 1* 0 and I * I0 (see Figure 8), the SR-sensitive and the SR-insensitive
capital market lines differ entirely. The SR-sensitive capital market line is no longer a line but
a hyperbola. This is the most disadvantageous case: there is an SRI cost for all the investors.
G 1* 0 and I * I0
Mean
Standard Deviation
In the mean-variance portfolio selection in the presence of a risk-free asset also, the investors
risk aversion matters in the SRI cost. In order to be ready-to-use for practitioners, we make
explicit the four cases are possible: a) the SR-capital market line is the same as the SR-
insensitive capital market line (i.e. no cost), b) only the left portion is penalized (i.e. a cost for
high-risk-aversion investors only), c) only the right portion is penalized (i.e. a cost for low-
risk aversion investors only), and d) the full capital market line is penalized (i.e. a cost for all
the investors).
This section illustrates the results of Sections 2 and 3 by considering the case of a responsible
We consider the EMBI+ indices from JP Morgan as proxy for emerging bond returns. These
indices track total returns for actively traded external debt instruments in emerging markets.10
The indices are expressed in US dollars and taken at a monthly frequency from January 1994
10
Argentina, Brazil, Bulgaria, Ecuador, Mexico, Panama, Peru, Philippines, Russia, Venezuela.
Appendix 5.
In the same way as Scholtens (2009), we use the Environmental Performance Index (EPI) as
responsible ratings. The EPI is provided jointly by the universities of Yale and Columbia in
collaboration with the World Economic Forum and the Joint Research Centre of the European
environmental stress to human health and promoting ecosystem vitality and sound
indicators tracked in six well-established policy categories, which are then combined to create
ARGENTINA 81.78
BRAZIL 82.65
BULGARIA 78.47
ECUADOR 84.36
MEXICO 79.80
PANAMA 83.06
PERU 78.08
PHILIPPINES 77.94
RUSSIA 83.85
VENEZUELA 80.05
Mean 81.00
Standard Deviation 2.44
UNITED STATES 81.03
Here, the portfolio EPI is defined in the same way as in eq. (2). We start by estimating the
portfolio EPI along the SR-insensitive frontier, which corresponds to estimating the
relationship (3) of Proposition 1. We obtain the following estimates for the parameters G 0
and G1 :
G0 76.00 G1 0.30
frontier: a 1%/year increase in expected returns corresponds to an increase of 0.30 in the EPI
portfolio. The minimal EPI portfolio on the SR-insensitive frontier is obtained for the
L ' 1 I
minimum-variance portfolio and is equal to 78.26 .
L ' 1 L
efficient frontier and we impose a set of constraints I p ! I 0 on the portfolio EPI. Figure 9
exhibits the SR-sensitive frontiers for several thresholds I 0 . The corner portfolios for which
there is a disconnect between the SR-sensitive and SR-insensitive frontiers are in Table 4.
Figure 9 SR-sensitive frontiers versus SR-insensitive frontier for the EMBI+ indices,
30.00%
25.00%
Ann. Mean (%/year)
20.00%
15.00%
10.00%
5.00%
0.00%
0.00% 10.00% 20.00% 30.00% 40.00% 50.00%
Ann. Standard Deviation (%/year)
SR-insensitive Frontier SR 80
SR 82 SR 84
SR 86 SR 88
(%/year) (%/year)
L ' 1 I
As expected from Proposition 2, for I 0 78.26 , the SR-sensitive frontier is the same
L ' 1 L
L ' 1 I
as the SR-insensitive frontier. For I0 ! 78.26 , the SR-sensitive frontier differs from
L ' 1 L
the SR-insensitive frontier at the bottom and is the same at the top. For instance, with a
threshold equal to 84 on the portfolio EPI, the SR-sensitive and the SR-insensitive frontiers
are the same for expected returns above 26.84%/year and differ for expected returns below
26.84%/year. In the case of emerging bonds, improving the portfolio EPI costs more for
In order to illustrate Problems 3 and 4, we rely on the US 1-month interbank rate as a risk-free
asset.11 Its responsible rating corresponds to the EPI of the United States I * 81.03 . Then, we
As G1* ! 0 , the portfolio EPI increases with the expected return of the SR-insensitive capital
market line. According to the estimations, for a 1%/year increase in expected returns, the EPI
portfolio is 0.02 higher. The minimal EPI portfolio on the SR-insensitive frontier is obtained
11
This variable is extracted from Datastream.
30%
25%
Ann. Mean (%/year)
20%
15%
10%
5%
0%
0% 10% 20% 30% 40% 50%
Ann. Standard Deviation (%/year)
From now on, we consider SR investors wishing to adopt high environmental standards: thus,
we impose a set of constraints I p ! I 0 in the same way as in Problem 4. Figure 11 exhibits the
SR-sensitive capital market lines for several thresholds I 0 , and Table 5 displays the corner
portfolios.
30%
25%
15%
10%
5%
0%
0% 10% 20% 30% 40% 50%
Ann. Standard Deviation (%/year)
SR-insensitive CML SR 82 SR 84 SR 86 SR 88
(%/year) (%/year)
As expected, for I 0 I * 81.03 , the SR-sensitive capital market line is the same as the SR-
insensitive capital market line. For I0 ! I * 81.03 , the SR-sensitive capital market line
differs from the SR-insensitive one at the bottom and is the same at the top. Here also, the SRI
cost appears for investors with high risk aversion. We notice that the corner portfolios have
particularly high expected returns and volatilities (see Table 5): this can be explained by the
meaning that for expected returns below 50.01%/year, the SR-sensitive and SR-insensitive
capital market lines are disconnected. However, we observe in Figure 11 that the SR-
insensitive and SR-sensitive capital market lines are very close for expected returns slightly
below 50.01%/year.
This numerical application highlights that the cost implied by high environmental
requirements in an emerging bond portfolio differs according to the investors risk aversion.
In this particular case, it costs more to be green for investors with cold feet. Let us now focus
on a typical investor. Sharpe (2007) suggests that the representative investor has a risk
2
aversion parameter O in the traditional mean-variance optimization of eq. (1). We seek
0.7
computing the optimal portfolios for different thresholds on the portfolio EPI. Figure 12
displays these portfolios (the means and variances of the optimal portfolios are given in
Appendix 6).
30.00%
25.00%
Ann. Mean (%/year)
20.00%
15.00%
10.00%
5.00%
0.00%
0.00% 10.00% 20.00% 30.00% 40.00% 50.00%
Ann. Standard Deviation (%/year)
SR-insensitive Frontier SR 82
SR 84 SR 86
SR 88 SR 90
threshold is below 82.37, which is slightly above the average EPI rating of the samples
countries. When the threshold is above 82.37, an SRI cost appears and the optimal portfolio is
no longer on the SR-insensitive frontier. This SRI cost rises with the strength of the
constraint. In this case, the representative investor is directly concerned by the disconnect
5. Conclusion
The rapid growth of the SRI fund market has given birth to a burgeoning academic literature.
Most academic studies show that there is little difference between financial returns of SRI
funds and conventional funds. However, Geczy et al. (2006) highlight that limiting the
investment universe to SRI funds can seriously harm diversification. To shed light on this
debate, our paper aimed at modelling SRI in the traditional mean-variance portfolio selection
average responsible rating of the underlying entities. Our results are detailed so that they are
ready to use by practitioners. Indeed, we show that a threshold on the responsible rating may
impact the efficient frontier in four different ways, depending on the link between the returns
and the responsible ratings and on the strength of the constraint. The SR-sensitive efficient
frontier can be: a) identical to the SR-insensitive efficient frontier (i.e. no cost at all), b)
penalized at the bottom only (i.e. a cost for high risk-aversion investors only), c) penalized at
the top only (i.e. a cost for low risk-aversion investors only), d) totally different from the SR-
insensitive efficient frontier (i.e. a cost for every investor). In other words, if portfolio ratings
increase (resp. decreases) with the expected return along the traditional efficient frontier, the
SRI cost arises first at the bottom (resp. at the top) of the frontier. The results are the same in
clearly matters in the potential cost of investing responsibly, this cost being zero in some
cases. We strongly believe that this finding could help portfolio managers of SRI funds.
As the calculations in our paper are very general, we believe it could find other applications in
the asset management industry, notably for portfolio selection with asset liquidity constraints.
However, one limitation of our study is that it assumes expected returns to be independent
from responsible ratings: further research could focus on modelling the impact of an SRI
constraint in the mean-variance optimization when expected returns and volatilities depend on
responsible ratings.
The author thanks Marie Brire, Benjamin Lorent, Valrie Mignon, Kim Oosterlinck, Hugues
Pirotte, Ombretta Signori, Ariane Szafarz and Kokou Topeglo for their helpful comments.
References
Barracchini, C., 2007. An ethical investments evaluation for portfolio selection., Journal of
Business Ethics and Organization Studies 9, 1239-2685.
Basak, G., Jagannathan, R., Sun, G., 2002. A direct test of mean-variance efficiency of a
portfolio. Journal of Economic Dynamics and Control 26, 1195-1215.
Beal, D., Goyen, M., Philips, P., 2005. Why do we invest ethically? Journal of Investing 14,
66-77.
Best, M., Grauer, R., 1990. The efficient set mathematics when mean-variance problems are
subject to general linear constraints. Journal of Economics and Business 42, 105-120.
de Roon, F., Nijman T., 2001. Testing for mean-variance spanning: a survey. Journal of
Empirical Finance 8, 111-156.
Dorfleitner, G., Leidl, M., Reeder J., 2009. Theory of social returns in portfolio choice with
application to microfinance. Working Paper, University of Regensburg.
Drut, B., 2009. Nice guys with cold feet: the cost of responsible investing in the bond
markets, Working Paper No 09-34, Centre Emile Bernheim, Universit Libre de Bruxelles.
European Sustainable Investment Forum, 2008. European SRI Study. European Sustainable
Investment Forum, Paris. Available at: http://www.eurosif.org/publications/sri_studies .
Farmen, T., Van Der Wijst, N., 2005. A cautionary note on the pricing of ethics. Journal of
Investing 14, 53-56.
Geczy, C., Stambaugh, R., Levin, D., 2006. Investing in socially responsible mutual funds.
Working Paper, Wharton School of the University of Pennsylvania.
Landier, A., Nair V., 2009. Investing For Change. Oxford University Press, Oxford.
Lintner, J., 1965. The valuation of risky assets and the selection of risky investment in stock
portfolios and capital budgets. Review of Economics and Statistics 47, 13-37.
Lo, A., 2008. Hedge Funds: An Analytic Perspective. Princeton University Press, Princeton.
Renneboog, L., Ter Horst, J., Zhang, C., 2008. Socially responsible investments: institutional
aspects, performance, and investor behaviour. Journal of Banking and Finance 32, 1723-1742.
Scholtens, B., 2009. Measuring sustainability performance of investments; the case of Dutch
bond funds. Paper prepared for the PRI Academic Conference 2009. Available at
http://www.unpri.org/files/Scholtens_PRI2009.pdf .
Social Investment Forum, 2007. Report on socially responsible investing trends in the United
States. Available at:
http://www.socialinvest.org/resources/pubs/documents/FINALExecSummary_2007_SIF_Tre
nds_wlinks.pdf .
Sharpe, W., 1965. Capital asset prices: a theory of market equilibrium under conditions of
risk. Journal of Finance 19, 425-442.
Sharpe, W., 2007. Expected utility asset allocation, Financial Analysts Journal 63, 18-30.
Following Best and Grauer (1990), in the standard mean-variance case, the weights vector
that solves the problem is:
1 L 1 1 L ' 1 P
Z ( P L )
L' L O
1
L ' 1 L
1 1
PP P 'Z L ' 1 P (( P ' 1 P )(L ' 1 L ) (L ' 1 P ) 2 )
L ' 1 L O
1 1
V P2 Z ' 6Z 1 2 (( P ' 1 P )(L ' 1 L ) (L ' 1 P ) 2 )
L ' 1 L O
1
The corresponding responsible rating I P is also a linear function of :
O
1 1
Ip I 'Z L ' 1 I (( P ' 1 I )(L ' 1 L ) (L ' 1 P )(L ' 1 I )
L ' 1 L O
1
As the expected return is also a linear function of , it is possible to express the responsible
O
rating as a linear function of the expected return:
( P ' 1 P )(L ' 1 I ) (L ' 1 P )( P ' 1 I ) ( P ' 1 I )(L ' 1 L ) (L ' 1 P )(L ' 1 I )
Ip P
(L ' 1 L )( P ' 1 P ) (L ' 1 P ) 2 (L ' 1 L )( P ' 1 P ) (L ' 1 P ) 2
p
The mean-variance optimization with the linear constraint on the portfolio rating has the same
portfolio solutions as the standard mean-variance optimization, while the constraint is not
binding. This is verified if the portfolio rating of the efficient portfolio from the traditional
mean-variance optimization is above the threshold I 0 . Two cases have to be distinguished:
In the mean-variance plan, this corner portfolio is the one for which the portfolio rating is I 0 .
To compute the portion of the efficient frontier for which the linear constraint on the portfolio
rating is binding, we apply the results of Best and Grauer (1990):
1
Pp J0 J1
O
1
V p2 J2 J1
O2
With
( P ' 1 I )(L ' 1 I ) (L ' 1 P )(I ' 1 I ) (L ' 1 P )(L ' 1 I ) (L ' 1 L )( P ' 1 I )
J0 I0
(L ' I ) (L ' L )(I ' I )
1 2 1 1
(L ' 1 I ) 2 (L ' 1 L )(I ' 1 I )
Appendix 3
Following Best and Grauer (1990), in the traditional mean-variance case, the portfolio
solutions stand as:
1 1
Z ( P rL ) > @
O
1
Zr 1 >L ' 1
( P rL ) @
O
1
The responsible rating of the efficient portfolios is a linear function of :
O
1
Ip I 'Z I * Z r I * >(I I *L )' 1
( P rL ) @
O
1
As the expected return P P is also a linear function of , it is possible to express the
O
responsible rating I P of the efficient portfolios as a linear function of the expected return P P :
Pp r
Ip I 'Z I * Z r I * >(I I *L )' 1
( P rL ) @
( P rL )' 1 ( P rL )
Appendix 4
The mean-variance optimization with the linear constraint on the portfolio rating has the same
portfolio solutions as the standard mean-variance optimization while the constraint is not
binding. This is verified if the portfolio rating of the efficient portfolio from the traditional
mean-variance optimization is above the threshold I 0 . Two cases have to be distinguished:
Case G 1* ! 0
In this case, the portfolio rating increases linearly with the expected return in the traditional
mean-variance optimization. The minimal portfolio rating is I * . If I * ! I 0 , the constraint on
the portfolio rating is inactive and the solutions are the same as in the traditional mean-
variance optimization: the Capital Market Line is not modified. If I * I 0 , the constraint on
the portfolio rating is binding at the bottom of the efficient frontier and inactive at the top.
Indeed, the constraint is binding for O ! O*0 with:
(I I *L )' 1 ( P rL )
O*0
I0 I *
The risk aversion parameter O*0 corresponds to the portfolio with the expected return:
1
E 0* r ( P rL )' 1 ( P rL )
O*0
And the expected variance:
1
V0* 2
( P rL )' 1 ( P rL )
O*0
To compute the portion of the efficient frontier for which the linear constraint on the portfolio
rating is binding, we apply the results of Best and Grauer (1990):
(I I *L )' 1 ( P rL ) 1 ((I I *L )' 1 ( P rL )) 2
Pp r (I 0 I *) (( P rL )' 1 ( P rL ) )
(I I * L )' (I I *L )
1
O (I I *L )' 1 (I I * L )
Indeed, when the constraint on the portfolio rating is binding, the Capital Market Line is no
longer a linear function in the mean-standard deviation plan but a hyperbola.
Case G 1* 0
In this case, the portfolio rating I p decreases linearly with the expected return P p in the
Appendix 5 Descriptive statistics for the EMBI + indices in US dollars, January 1994 to
October 2009
Ann. Std. Sharpe
Ann. Mean Skewness Kurtosis Maximum Minimum
Dev. Ratio
ARGENTINA 4.94% 28.84% 0.03 -1.07 9.37 33.80% -43.90%
BRAZIL 15.03% 21.40% 0.51 -0.62 8.22 26.47% -27.17%
BULGARIA 15.10% 20.26% 0.55 -0.98 13.96 25.77% -36.38%
ECUADOR 16.19% 33.51% 0.36 -1.47 10.10 28.29% -55.78%
MEXICO 9.83% 11.48% 0.50 -0.78 7.72 12.84% -14.59%
PANAMA 15.31% 20.73% 0.54 0.26 8.18 28.88% -22.63%
PERU 14.88% 22.17% 0.49 -0.58 10.50 34.50% -29.93%
PHILIPPINES 7.82% 11.91% 0.32 -2.27 13.41 7.75% -20.44%
RUSSIA 18.27% 34.05% 0.42 -1.92 20.27 35.63% -72.18%
VENEZUELA 13.79% 23.07% 0.42 -0.84 12.84 34.05% -39.13%
(%/year) (%/year)
82 21.38 24.03
84 22.10 25.25
86 22.98 27.17
88 23.86 29.48
The return on risk-free assets, if such assets truly exist, is very low today and will
undoubtedly remain so for a long period of time. Therefore, if investment is made only in
risk-free assets or moderately risky assets, savings for retirement will procure low purchasing
power when it comes time to retire and will be insufficient for reaching any consumption
targets (such as maintaining a certain lifestyle).
If wealth is invested in risky assets, the expectations of return are higher; however, there is a
risk factor: in some adverse configurations, purchasing power could be perceived as
catastrophically low.
Risky investing is not an appropriate way to make up for insufficient savings 1. When risk is
rewarded, risk-taking can increase the likelihood of a reaching a minimum purchasing power
target in retirement but at the price of an exposure to adverse circumstances, which risk
measures such as VaR or CVaR accurately capture.
In the specific case of saving for retirement, the individuals concerned have, however, a way
to mitigate the consequences of an unfavourable configuration for return on investment in
risky assets: supplementing inadequate purchasing power with labour income. In practice,
such income can be secured by putting off retirement or by getting a job, even if part-time,
during the first few years of retirement.
In this research paper, we will simultaneously model, with conventional utility and expected
utility functions, the labour supply at the time of retirement and the portfolio allocation choice
for retirement savings. We examine the importance of labour supply flexibility and the impact
of such flexibility on asset allocation. We find that that the portion allocated to risky assets
can be substantially increased.
This result is consistent with the work done on labour supply flexibility and portfolio choice
(Bodie, Merton and Samuelson, 1992) and generalizes it for different levels of risk tolerance
(or aversion).
In light of these results, we can at last give thought to the identification of pension fund
commitments, or objectives.
1
As Zvi Bodie has shown, expressions such as I cannot afford not to invest in risky assets should be banished.
Retirement can be defined as a period in life in which an individual no longer receives income
from his or her profession or labour. This individual's consumption needs will therefore have
to be covered by other funds: either transfers or labour income earned prior to retirement that
was not consumed immediately and was therefore saved. Saving is therefore a key factor in
having funds available and consuming during retirement.
One characteristic of saving for retirement is the length of the time horizon (delayed
purchasing power). Usually retirement lasts 20 years and follows forty years of work. On
average, a period of approximately thirty years elapses between the time savings are amassed
and when they are used for consumption.
This length varies depending on the age of the working individual of interest: for a young
person entering the workforce, this period is almost fifty years a half-century. For an
individual about to enter retirement, it would be ten years or so.
There are two principal schemes for saving for retirement: an individual format and an
institutional format, which we will refer to generically as pension funds. Under the individual
format, the saver (assisted by his or her advisers) has primary responsibility for asset
allocation. Under the institutional format, the fund itself is responsible for asset allocation.
This does not prevent that at the end of the day it is usually the saver who is impacted by the
consequences of the choices made, with one major exception: defined benefit retirement plans
guaranteed by a sponsor who is often the individual's employer. In this case, risk is ultimately
borne by the sponsor (except in the event of bankruptcy) and must be managed within the set
of risks to which he is exposed.
All methods combined, savings for retirement must be substantial. The targeted goal is often
defined as a ratio 50% to 70% of the benefit received to the labour income in the last years
of employment or sometimes the benefit received to the average of labour income over the
working life.
With an actual rate of return (after tax 2) on investment near zero, which is optimistic today for
risk-free savings, approximately one-third of all labour income must be set aside for savings,
with a desirable replacement rate of two-thirds.
Table 1
This assessment is made using a two-period model, assuming that savings is amassed mid-
career and its fruits spent midway through the retirement period. In practice, the model should
be fine-tuned to take account of the characteristics of the labour income time profile and
mortality tables. It is worth bearing in mind that the orders of magnitude obtained are
nonetheless significant.
2
Pension funds and pension accounts usually work under tax neutrality: contributions are deductible from the
income tax base and pensions when retrieved are added to it. If income tax rates do not differ between the two
periods, tax does not impact real return. With other saving channels, tax generally eats into real returns, the more
so inflation is high (Maillard, 2011b).
In this section, we assume that savings have been accumulated and that the individual has, at
moment in time 0, wealth or capital, W0 , that he or she must invest in one way or another (or
that must be invested on their behalf).
The time horizon to retirement is T. We assume that are efficient mechanisms to transform
wealth, WT , secured on this horizon, into annuities 3, and that this final capital reflects the
degree of achievement of this person's goals.
If there is a risk-free investment between dates 0 and T, and if the initial wealth is invested in
this risk-free asset, the capital secured is known with certainty. However, most often, the
initial wealth will be invested, as least in part, in risky assets and the capital secured will be
exposed to risk.
Pension funds have often obligations (in the case of defined benefit plans) or explicit or
implicit objectives of paying specified amounts, defined nominally or in purchasing power,
i.e. WT *. If savings is not managed by an institution, the individual may also think in terms of
a target.
WT t WT *
There are two possible cases. Risk-free investment (nominal or corrected for inflation) can
produce a return, r f , sufficient to comply with commitments or reach targets.
3
The hypothesis is therefore formulated that pensions are not exposed to risk once liquidated, which
undoubtedly satisfies the desires of most retirees.
The question of asset allocation is straightforward in that case. It can consist of investing a
portion of wealth, e WT * , in risk-free assets, and the remainder in high-risk assets. The
rf T
proceeds from investment in this portion could be disbursed as a pension bonus or used to
reimburse contributions by the retirement plan's sponsor.
The first example is perhaps not the most typical, especially in the early 2010s. Often it is
impossible to meet commitments or to reach targets by means of a risk-free asset.
WT *
rf T
W0 e
Complying with commitments or reaching targets is no longer certain. How then should be
question of asset allocation be formulated?
Should the probability of complying with commitments be maximised, or, by the same token,
should the chances of default of the fund due to its obligations be minimised?
Max((Pr(WT ! WT *))
Formalisation of this type could create more room for risky assets (the probability of
complying with obligations is zero if investments are made solely in risk-free assets). But if
we go deeper, it could lead to renouncing all upside in excess of the WT * limit and hence to
selling puts at this threshold on the portfolio of risky assets. Going beyond that, the optimum
solution would be found in investing in a binary option backed by a portfolio of risky assets,
paying 0 with a low probability, and exactly W T * with the highest possible probability.
The drawback of formalising by minimising the probability of default is clear at this stage:
missing targets is not punished in a manner specific to the degree of failure reflecting the
shortfall between the target and actual performance.
3.1.2. Modelling
Wealth can be invested in risk-free assets, if one assumes they exist, or it can be invested in
an optimally-managed risk portfolio 4 (instantly optimising the Sharpe ratio) if adopting a
dynamic management style 5. Here is average annual expectation of return over the risk
portfolio period, is annualised volatility and t is the Sharpe ratio.
P rf
t
V
The eventual value of the wealth, if a portion, , is (continually) invested in the risk portfolio,
is:
1 2 2
r f D ( P r f ) 2 D V T DV T H
WT W0 e
where is a random variable with zero mean and unitary standard deviation. In the interests of
simplicity, we will use a Gaussian risk distribution, in particular because of the distant
horizon.
r * r f
1
H t H* t DV T
DV 2
Pr(WT ! W *T ) 1 ) (H *)
r* is the risk-free rate of return that must be secured in order to reach the target with a 100%
investment in risk-free assets.
4
The portfolio of risky assets is not necessarily the market portfolio. Room is left for active portfolio
management to improve the performance of the risk portfolio.
5
We are speaking of "lite" dynamic allocation in the sense that the parameters are deemed constant over the
period (no predictability). Its results will differ very little from that of static allocation, with sufficient risk
aversion (Maillard, 2011). The choice of dynamic allocation allows an analytic treatment of the optimization
problem.
If it is greater, there is a funding gap at the effective risk free rate and the difference
represents a shortfall in annualized returns. The threshold therefore decreases with the portion
of risky assets. The theoretical optimum is found, for an infinite investment in the risk
portfolio, with leverage over the risk-free asset.
The quid pro quo is obviously risk, which can be assessed using conventional measurements,
the standard deviation of final wealth or Value-at-Risk (VaR) or Conditional Value-at-Risk
(CvaR, aka Expected Shortfall).
As an illustration, the chart below provides a representation of the parallel change in the
probability of reaching a target and VaR and CVaR at a threshold of 99% in proportion to the
investment, with an investment term of 20 years, a risk free rate of 2%, a credit spread of 4%,
a annualised volatility of the risk portfolio of 20% and a funding gap of 2% per year.
Chart 1
1
0,8
0,6
0,4
0,2
0
0% 50% 100% 150% 200% 250% 300% 350% 400%
-0,2
-0,4
-0,6
Share of risky assets
Given the limitations of reasoning in terms of targets, it is logical to place the asset allocation
question within a framework of optimisation.
MaxE (U (WT ))
As to the form of the utility function, we have opted for a function with the feature of constant
relative risk aversion (CRRA). In the catalogue of standard utility functions, we have rejected
the quadratic function (with which utility decreases with consumption after a certain
threshold), and the constant absolute risk aversion (CARA) function. With this last class of
functions, the optimal proportion of risky assets decreases with the initial wealth of the savers,
which is counter-intuitive (See Annex 4). We do not retain a hyperbolic absolute risk aversion
(HARA) function due to the difficulty in identifying irreducible consumption and
distinguishing if from the "targets" described above. However, we will use the results related
to optimisation and describe how to adjust our results if a hyperbolic risk aversion function
were used (Annex 4).
C 1J
U (C )
1 J
The range usually deemed realistic for risk aversion is along the lines of 2 to 7/10.
P rf
D
JV 2
To take into account labour supply flexibility, the utility function is augmented to:
C 1J ( L L)1J
U (C , L) b
1 J 1 J
C is the consumption for the period, L is the labour supplied, L is the maximum amount of
work that is possible to provide and L - L is therefore leisure.
The first term is the conventional utility function (CRRA) with constant relative aversion
equal to . The second term represents the contribution of leisure to total utility, with a
weighting dependent on coefficient b.
Using the same exponent for consumption and for leisure firstly cross-references a
conventional utility function of the constant elasticity of substitution (CES) type. The
elasticity of substitution between consumption and leisure is:
1 1
s
1 (1 J ) J
As relative aversion to risk is generally greater than 1, the elasticity of substitution is confined
in a range 0 to 1, which is reasonable.
Furthermore, using the same exponent provides an analytical solution to the problem of
optimising utility, at the moment of interest, under a budget constraint.
The funds available for spending are in fact made up of accumulated wealth, W, and labour
income. If w represents the labour compensation rate, optimization is written:
MaxU (W wL, L)
Entering this value for labour supply into the utility function gives (see Annex 3):
6
The labor supply derived from this formula can be negative, which is tolerable: if performance of the risky
assets is very strong, the individual can move forward (if rules permit) the age at which he or she can receive his
or her pension.
The function V(W) is a HARA-type function, apart from the fact that the term is constant,
representing minimum consumption, is normally negative in this type of function. Here the
constant term is positive and represents the maximum value of supplemental labour income
that the retiree can secure by sacrificing all his or her leisure.
The second phase entails maximising V(W). We can use the results of Bajeux, Jordan and
Portait (2003) when maximising HARA utility. In this case, the optimal allocation is the
following combination:
- a dynamic portfolio optimally combining risky assets and risk-free assets, the same as
that resulting from optimising a CRRA.
- a short position in the risk free asset corresponding to the present value, at a risk free
rate, of the maximum labour income that the retiree can receive, or w Le
rf T
- a long position in the portfolio combining the risk portfolio and the risk-free asset
using dynamic management.
The proportion of the risky assets will change over time with the value of the portfolio. At the
outset, it is:
rf T rf T
W0 w Le w Le
D D 1
W0 W0
The portion invested in risky assets is increased by the ratio of the present value of potential
earnings from future work to savings accumulated for retirement. As accumulated savings
usually grows with the age of the saver, while the earnings from potential future work are
independent, the optimal portion of the risky assets therefore decreases with age.
The impact of flexibility on the risk asset portion is not immaterial. To demonstrate this point,
let's take an individual mid-way through his or her working life and for whom accumulated
savings represents four times the labour income (20 years times 20% the savings rate),
assumed to be constant. If this person can envisage working the equivalent of two years
It is finally possible to compute the welfare gains of labour supply flexibility (see Annex 3)
by comparing the expected utility with flexibility to the expected utility without (L =0). Part
of those gains stem from the ability to invest more in risky assets that labour supply flexibility
provides.
We have shown, using a simple model without being unrealistic, that labour supply flexibility
at the time of retirement can improve economic welfare, both directly and indirectly, by
making it possible to invest in riskier assets and to capture the rewards of risk.
In fact, it happens that working during retirement or deferring retirement procures additional
wealth at no risk or, in any event, with little risk (but not without pain). This assumes that
there are good assurances against potential joblessness and, above all, against the inability to
work. The optimisation question ultimately translates into optimisation with the constraint of
a fixed investment in a given asset.
As to pension funds, and especially in the case where they manage most of the retirement
savings of their principals, we would recommend promoting flexibility on the liquidation age.
The ideal would be to offer la carte allocation, taking account of risk aversion on the one
hand and, on the other, the opportunity and willingness to get an old-age job or to defer
retirement.
Amenc, Nol, Felix Goltz, Lionel Martellini and Vincent Milhau, 2010, New Frontiers in
Benchmarking and Asset Liability Management, Working Paper, Edhec Risk Institute,
September 2010
Bajeux-Besnainou, Isabelle, James V. Jordan and Roland Portait, 2003, "Dynamic Asset
Allocation for Bonds, Stocks and Cash", Journal of Business, Apr 2003, Vol. 76, Issue 2
Bodie, Z., R. C. Merton and W. F. Samuelson, 1992, Labor Supply Flexibility and Portfolio
Choice in a Life Cycle Model, Journal of Economic Dynamics and Control, 16, 427-449
Bodie, Z., 2001, "Retirement Investing": A New Approach, Working Paper 2001-03, Boston
University School of Management
Maillard, Didier, 2011a, Dynamic versus Static Asset Allocation: From Theory (halfway) to
Practice, Working Papers Series N 1942901, SSRN, March 2011
Maillard, Didier, 2011b, Tax and Investment Return, Working Papers Series N 1968985,
SSRN, September 2011
1 2 2
r f D ( P r f ) 2 D V T DV T H
WT W0 e
t WT *
1 2 2 WT *
r f D ( P r f ) 2 D V T DV T H t ln W
0
WT * 1 2 2
ln r f D ( P r f ) D V T
W0 2 r * r f 1
Ht t DV T H*
DV T DV 2
Pr(WT ! W *T ) 1 ) (H *)
Annex 2
dW
(1 D )r f dt D ( Pdt Vdz ) >r f D ( P r f ) dt DVdz
@
W
Using ,W
VOHPPDand integrating leads to:
1 2 2
d ln W r D ( P r f ) 2 D V dt DVdz
Assuming the random process is Gaussian, and designing a zero mean unitary variance
random variable
1 2 2 1 2 2
r f D ( P r f ) 2 D V T DV T H r f D ( P r f ) 2 D V T DV T H
W (T ) W (0)e
WT W0 e
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
J
rc r f D ( P r f ) JD 2V 2 / 2 E (rP ) V (rP )
2
P rf
D
JV 2
(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
(W wL)1J ( L L)1J
MaxU (W wL, L) b
1 J 1 J
dU
w(W wL) J b( L L) J 0
dL
1 1
w J (W wL) b J ( L L)
1 1
1
1
1
L b J w J b J
L w JW
1 1
J J
b w
L 1 1
L 1 1
W
1 1
J J J J
b w b w
1 1 1 1
b J w J b J
b J
W wL 1 1
w L 1 w 1 1 W 1 1
wL 1 1
W
1 1 1 1
J J
J J
J J
J J
b w b w b w b w
1
J
b
W wL 1 1
(W w L)
1
J J
b w
1 1 1
b J w J w J
LL 1 1 1 L 1 1
W 1 1
(W w L)
1 1 1
J J
J J
J J
b w b w b w
3) Utility function
1 1J 1 1J
1 b J w J
U (C , L) (W w L) 1J
b 1 (W w L) 1J
1 J J1 1
1
J J 1
1
J
b w b w
1J 1J
1J 1J
1 J 1 1
U (C , L) b bw J
(W w L)
1J
1 J 1
1
1
1
J 1
1
b J w J b w J
1J
1 J
(W w L)1J 1 1 1 (W w L)1J 1 1
1
U (C , L) b b J w J 1 b b J w J
1 J
1
b J w1 J
1 J
1 1 J
(W w L)1J 1
U (C , L) 1 b J w J V (W )
1 J
1 2 2
r f D ( P r f ) 2 D V T DV T H
WT W 0 w Le
rf T
e
wL
J J J J
b w b w
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
E (U ) (1 J ) 1 W0 > 1J
e
(1J )( r f t 2 / 2J )T
bL
1J
@
We can measure the gains in welfare associated with flexibility by comparing the savings
leading to expected utility with flexibility and the higher savings needed to reach the same
degree of utility without flexibility.
Expressed in monetary terms, the resulting gain in welfare due to flexibility is:
W 0 W0
1 1 J
(1 J ) 1 W 0 > 1J
e
(1J )( r f t 2 / 2J )T
bL
1J
@ 1
(1 J ) 1 1 b J w J W0 w Le f
r T
1J
e
(1J )( r f t 2 / 2J )T
1 1 J
t2 1
u rf A( w) 1 b J w J
2
1J 1J
W 0 e uT bL
1J
A( w) (W0 w Le
rf T
)e uT
1
1J
W0 e uT
0
A( w) (W w Le r f T )e uT bL
1J 1J
The utility functions commonly used are the quadratic function, the constant absolute risk
aversion (CARA) function, the constant relative risk aversion function (CRRA) and the
hyperbolic absolute risk aversion (HARA) function.
Quadratic function
M
U (C ) C C2
2
U ' (C ) 1 MC
Consumption utility decreases for C ! 1 / M , which is in conflict with the commonly accepted
hypothesis that an abundance of wealth is not harmful.
CARA function
U (C ) e aC
Although it is used in the context of optimising a static portfolio (the result cannot be obtained
analytically in the context of dynamic optimisation, but should not differ significantly), you
obtain a result that is relatively counter-intuitive.
In fact, the final portfolio is worth, if is the portion invested in the risk portfolio (with the
notations in the body of the text):
WT
W0 (1 D )e
rf T
De PT V TH
W0 A BD A e
rf T
B e PT V TH
e
rf T
W0
If
maximises this expression, then D '* D * maximises
W '0
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.
C 1J
U (C )
1 J
HARA function
(C C )1J
U (C )
1 J
C is irreducible consumption, below which the question of utility is irrelevant. It should not
be confused with a target such as that discussed in the body of the article.
To maximise the expectation of a HARA function, on the one hand the risk-free asset must
exist and, on the other, placing in this asset exactly what is needed to attain C , the remainder
being placed in a dynamic portfolio identical to that which would result from maximising a
CRRA function with the same risk aversion parameter.
With labour flexibility, optimisation in the context of such a function would yield a solution
whereby it would be necessary to place the difference (algebraic) between what is necessary
to achieve C and the present value of the maximum of labour income in the risk-free asset.
Incorporating Linkers
in a Global Government Bond
Risk Model
Marielle de Jong,
Head of Fixed Income Quantitative Research, Amundi
Naturally, this market innovation calls for a revision of the existing risk models serving
government bond investment strategies. Since the pricing of inflation has been made explicit
in a way via the bond yield differentials called the breakeven inflation rates, it would in
principle be possible to model inflation risk in an explicit way as well. This was not
conceivable before. Nominal bonds are, unlike real bonds, inherently exposed to inflation risk
since their payoff is typically not compensated for the inflation incurred over the time to
maturity, but the price influence of that cannot be made explicit in an easy way without a real
bond comparative.
We explore, in this article, how to build the new inflation dimension into a bond risk
modeling framework. We have not found studies in the literature making such attempt. We
choose to do this in an international setting focusing on the cross-border price dependency
between national bond markets as a whole, rather than between individual financial
instruments. We adopt the International Capital Asset Pricing Model introduced by
Solnik [1974]. This model is a generalization of the Sharpe-Lintner CAPM that was initially
developed for equity instruments within one country, towards an international scale including
other asset classes as well. The essence of the model is that price returns are being
decomposed into a systematic global market risk component and country-specific residual
components.
The model design typically suits an international bond investor who runs a country allocation
strategy. The dichotomous decomposition of risk facilitates the assessment of the tactical
positions split into global directional bets and individual country bets. The term structure in
the bond yields is not considered in this article. The currency risk inherent to purchasing
foreign assets is not considered either, this by making the assumption that all foreign positions
are fully hedged against exchange rate risk. The model is as such an abbreviated version of
Within the modeling setting that we have chosen we uncover an intriguing price correlation
relationship between the bonds, which would not be easy to detect in a national context. In
next section we discuss this pricing phenomenon. In section 3 we formulate the challenge it
represents for modeling the risk. In section 4 we present which modeling approach would suit
best, and section 5 concludes.
We start by exploring the price covariance of the inflation-linked bonds with the nominal
bonds over the recent past. In Exhibit 1 the price correlation matrix is displayed between the
two types of bond traded in the major developed world markets. Correlations are measured
between the time-variation of the nominal bond yields (NBY), the inflation-linked (real) bond
yields (RBY), and the yield spreads, the breakeven inflation rates (BEIR). Six countries have
been retained in this study: Australia, Canada, the Euro Area, Great-Britain, Sweden and the
United States, which correspond to the members in the Barclays Developed World Inflation-
Linked Bond Index that have started issuing linkers before 2002.1
The correlations, as well as the annual volatilities which are displayed on the diagonal, are
measured on weekly bond returns in the seven-to-ten-year maturity range with an average
duration of 7.5 years as calculated by Barclays Capital, over an eight-year period from June
2002 to July 2010. The particularly volatile months between October 2008 and May 2009
have been discarded for reasons discussed separately below. In de Jong [2010] it is verified
that essentially the same numbers are obtained when measuring on monthly rather than on
weekly data over the same observation period.
es
es
es
n
n
ai
ai
ai
at
at
at
ea
ea
ea
rit
rit
rit
St
St
St
lia
lia
lia
en
en
en
Ar
Ar
Ar
-B
-B
-B
da
da
da
ra
ra
ra
d
d
at
at
at
ed
ed
ed
ite
ite
ite
na
na
na
ro
ro
ro
st
st
st
re
re
re
Sw
Sw
Sw
Au
Un
Au
Un
Au
Un
Eu
Eu
Eu
Ca
Ca
Ca
G
G
RBY Australia 4,4%
Canada 0,27 3,1%
Euro Area 0,40 0,45 4,5%
Great-Britain 0,43 0,44 0,67 5,1%
Sweden 0,43 0,37 0,65 0,54 3,5%
United States 0,43 0,51 0,60 0,52 0,51 5,4%
NBY Australia 0,83 0,26 0,49 0,45 0,51 0,46 6,4%
Canada 0,49 0,59 0,59 0,54 0,52 0,66 0,55 4,3%
Euro Area 0,46 0,40 0,79 0,65 0,69 0,59 0,56 0,68 5,2%
Great-Britain 0,52 0,43 0,74 0,83 0,66 0,60 0,59 0,67 0,81 5,2%
Sweden 0,49 0,35 0,72 0,60 0,77 0,56 0,59 0,65 0,82 0,77 5,1%
United States 0,52 0,46 0,61 0,56 0,55 0,80 0,58 0,80 0,76 0,72 0,69 6,6%
BEIR Australia (0,24) 0,13 0,37 0,28 0,36 0,29 0,75 0,36 0,42 0,41 0,44 0,40 3,7%
Canada 0,36(-0,16) 0,32 0,27 0,31 0,35 0,43 0,70 0,48 0,44 0,47 0,57 0,32 3,5%
Euro Area 0,19 0,01(-0,11) 0,12 0,21 0,11 0,24 0,28 0,52 0,28 0,34 0,38 0,18 0,33 3,2%
Great-Britain 0,16 -0,01 0,14(-0,28) 0,21 0,15 0,25 0,24 0,30 0,31 0,30 0,29 0,23 0,30 0,29 3,0%
Sweden 0,31 0,16 0,43 0,37 (0,15) 0,33 0,39 0,45 0,56 0,51 0,74 0,49 0,31 0,41 0,31 0,24 3,2%
United States 0,27 0,06 0,19 0,21 0,22(-0,05) 0,33 0,42 0,45 0,37 0,38 0,57 0,25 0,46 0,48 0,28 0,36 4,0%
Period: June 2002 to July 2010, excluding the crisis period from October 2008 to May 2009.
Data frequency: weekly (Friday to Friday).
The numbers on the diagonal are annual return volatilities (bond duration 7.5 years).
Threshold of significance for the correlation numbers on a 5% confidence level with T=389 weeks is 0.10.
Data source: Barclays Capital. Calculations made by the author.
Note that the correlations are significantly positive, they are more accentuated within markets
between the NBY and RBY, and between the NBY and BEIR (shaded in grey), yet in contrast
they are lass accentuated between the RBY and BEIR (in brackets). The positive cross-border
correlation overall is usual for financial assets. It reveals a global common price movement,
indicating that there is a global systematic risk component in the bond price behavior. There
appears to be a country-specific component as well. The inflation-linked bonds are more
correlated with the nominal bonds within the same country than to those of other countries,
which is intuitive. The same holds for the breakeven rates with respect to nominal yields. The
observation concerning the RBY and BEIR seems odd though. Why would those variables be
systematically less correlated within countries than between?
Cette and de Jong [2008] give an explanation. They make evident that the country-specific
movements of the RBY and BEIR are systematically anti-correlated with each other and this
as a result of local market distortions. Issues like market liquidity and a time-varying aversion
to inflation risk have been raised in the literature as reasons for those distortions; see
Christensen, Dion and Reid [2004] for a survey. As soon as the inflation-linked bond price
The observations we make are schematized in Exhibit 2 below. The covariance between the
real bond yields and the breakeven inflation rates between countries is the net sum of positive
systematic covariance and negative residual covariance (second row in the Exhibit). This is
atypical compared to the more intuitive situation where the residual covariance within
countries is positive (first row). In next section it is discussed how this particular price
behavior can be take into account in an overall bond risk model.
In their most recent paper, Cette and de Jong [2012] investigate the persistence of the global
correlation between the common trends in the RBY and BEIR. This correlation has been
positive during six years from June 2002 to September 2008, has turned significantly negative
over the particularly turbulent period from October 2008 to May 2009, and has then turned
positive again attaining the same correlation level as before. The explanation given by the
authors for the radical sign change during the crisis is that the market distress, and the market-
related issues related with that, had become a global phenomenon. Market distortions lead to
opposite price movements between the RBY and the BEIR, and if they are concerted over the
globe, the opposite movements are being perceived globally as well.
We aim to build a risk model that specifies the overall price covariance structure effectively in
the newly-extended bond universe, containing thus nominal- as well as inflation-linked
government bonds. The challenge is to encapsulate the atypical price covariance that has been
made apparent in previous section. In order to appreciate its practical implications, consider
the following situation. When inflation swaps, i.e. derivative instruments directly priced on
the breakeven inflation rates, are being added to a nominal bond portfolio, the total risk will
increase, due to the strongly positive price correlation between the two instruments. If the
swaps were added to a portfolio invested in inflation-linked bonds, the risk would not
necessarily increase and may even decrease, due to the negative correlation.
Let us quantify this example for the case of British Gilts, taking the relevant risk parameters
from Exhibit 1.2 We read in the Exhibit that a portfolio fully invested in nominal Gilts and
one invested in inflation-linked Gilts both have an annual volatility level of 5%. Adding 100
percent inflation swaps, with an annual volatility of 3%, increases the overall volatility of the
nominal bond portfolio to 7%, that is 7% # 5% 2 3% 2 2 5% 3% 0.3 , while it leaves the
volatility of the inflation-linked bond portfolio unchanged at 5%,
since 5% # 5% 3% 2 5% 3% 0.3 .
2 2
It is not directly obvious how to take account of this situation in a global risk model. We
describe how we proceed in the context of the International Capital Asset Pricing Model, the
I-CAPM of Solnik [1974]. We start by formulizing a global nominal bond risk model in an I-
CAPM framework. Let the nominal bond yield variations in countries i, denoted as 1%< i , be
'NBYi E i
'NBY
F 'NBY H i (1)
so that the covariance structure between the bond yields is specified as:
2
E i E j V 'NBY
'NBY 'NBY
if i z j
cov'NBYi , 'NBY j 'NBY 2 2
(2)
E i E j V 'NBY V H
'NBY
if i j
where E i'NBY are the market betas, the sensitivities of assets i to the global market factor,
V 2
'NBY
is the global market variance and V H2 the country-specific residual variances.
Question is how to incorporate the inflation-linked bonds into this modeling framework. One
approach would be to define a global inflation-linked market factor onto the real bond yields
in an analogous way, denoted as F5BY, and assume a linear relationship with the nominal
market factor, through a parameter . The global NBY variation that is not captured by the
real bond factor is denoted as F%(,5. All residual terms are assumed to be independent of each
other and of the factors:
'NBYi E i F H i
'NBY 'NBY
'RBY E 'RBY F 'RBY K
i i i i (3)
F 'NBY O F 'RBY F 'BEIR
so that the covariance structure between the two bond yields is specified as:
Given this model, the covariance structure of the bond yields with the breakeven inflation
rates can be derived, by applying the definition BEIR = NBY RBY. This is done for the
RBY with the BEIR in equation (5), by which, interestingly, the country-specific anti-
correlation appears. For this aspect the model given in (3) fits the empirically observed price
behavior of the bonds. In next section we test by means of a statistical analysis to what extent
the model adapts to the data more generally.
It appears, and this is the interest of this paper, that a relatively small modification in the
model specification leads to much better results in terms of data fitting. The idea is to add a
global breakeven market factor to the original risk model (1) instead of a real bond factor as
was done in (3). This model is specified in (6) below. A breakeven factor, denoted as F%(,5,
LVGHILQHGRQWRWKHEUHDNHYHQLQIODWLRQUDWHYDULDWLRQV%(,5WKDWDUHGLUHFWO\GHULYHGIURP
the market data, and a linear relationship with the nominal market factor is specified through a
parameter . Again all residual terms are assumed to be independent of each other and of the
factors:
'NBYi E i F H i
'NBY 'NBY
The reader may verify that this model specification embeds the country-specific anti-
correlation between the RBY and BEIR just as well. The reader may verify also that the third
possible combination of bond variables, defined on the real bond yields with the breakeven
rates, does not have this quality. The residual terms of the RBY and of the BEIR would in this
combination be assumed independent of each other, in a standard CAPM approach, which
would exclude from the start the possibility of negative correlation.
In this section we test to what extent the two model specifications, given in (3) and in (6),
adapt to the market data. We do this by carrying out principal components analyses, by which
we retrieve the first two eigenvectors that explain at maximum the bond yield variance. The
results are displayed in the two Exhibits below. The sensitivities to the first two eigenvectors,
denoted as ev1 and ev2 on the axes, are plotted of the twelve yields, that is six NBY and six
RBY in Exhibit 3, and analogous for Exhibit 4.
We look if the eigenvectors we obtain resemble the global market factors that have been
defined in the two respective models. If they match, it tells that the model factors are the ones
with the highest statistical significance. Note that in both models the first factor represents a
Let us first look at Exhibit 3. The fact that all sensitivities to the first eigenvector are positive
means that it indeed represents a worldwide market factor commonly shared by the nominal-
and real yields. This factor is statistically significant, explaining 62% of the variance. Yet the
second eigenvector doesnt match. As can be seen in the Exhibit it represents a factor that
somehow makes a distinction between Canadian, Australian and American bond yield
movements on the one hand versus Euro Area, Swedish and British yield movements on the
other, regardless of the bond type. This geographical division doesnt correspond with the
factor specified in model (3), which makes an explicit division between the two bond types.
Model (3) is for this matter statistically rejected.
EXHIBIT 3
Results of a principal component analysis on the NBY and RBY correlation matrix
2,00
ev2
1,50
RBY Canada
1,00
RBY United States NBY Canada
0,50
NBY Australia
0,00
-2,00 -1,50 -1,00 -0,50 0,00 0,50 1,00 1,50 ev1 2,00
NBY Sweden
RBY Sweden
-1,00
-1,50
-2,00
2,00
ev2 BEIR Euro Area
1,00
BEIR Great-Britain
BEIR Canada
0,50
NBY United States
NBY Euro Area
BEIR Sweden
NBY Canada
0,00
-2,00 -1,50 -1,00 -0,50 0,00 0,50 1,00 1,50 2,00
ev1
NBY Sweden
-0,50 NBY Great-Britain
-1,00
NBY Australia
-1,50
BEIR Australia
-2,00
Looking at Exhibit 4, it can be seen that the first eigenvector again corresponds to a global
market factor. It is statistically significant as well explaining 52% of the variance. Note that in
this case the second eigenvector matches. The factor reflects a systematically different
movement between the nominal yields on the one hand versus the breakeven rates, with the
exception of Australia, on the other. This factor explains an additional 10% of the total
variance, which gives support for adopting the model (6).
So we find model (6) more in line with how bond prices behave. It is striking that the
relatively small modification we had made in the data presentation, which has no incidence on
its information content, has such a different outcome in the estimation results. We point at two
details in the model specification that are at the origin of this. Firstly, the two models dont
make exactly the same assumptions for the residual terms. In model (3) independence is
assumed between the residual NBY and RBY, and in model (6) between the residual NBY
and BEIR. This has a direct consequence for the volatility levels. The residual variance of the
BEIR is according to model (3):
which is strictly superior to that of the RBY. That doesnt strike with the empirical data, as
can be seen in Exhibit 1. The reader may verify that model (6) inverses this hierarchy: the
country-specific real bond yields are by construction more volatile than the breakeven
counterparts. The fact that this tends to be true in practice may be attributed to the before-
mentioned market distortions agitating the inflation-linked bond prices in particular.
The second seeming detail is that the data presentation indirectly imposes a hierarchy in the
two market factors. In model (3) the first factor defines a global common movement between
the nominal- and real yields, and in model (6) between the nominal yields and breakeven
rates. Those are essentially different factors. With reference to Fishers [1930] seminal
Interest Rate Theory and supposing that this theory applies to long-term sovereign bonds, we
can say that the first factor in (3) is driven by structural shocks that affect the global real
economy. Nominal- and real bonds have this risk source in common in principle. The first
factor in (6) is rather driven by inflation shocks. We may call this the global inflation factor,
with reference to Fisher.
It is an interesting question, yet beyond the scope of this article, whether the inflation shocks
have indeed more impact than structural shocks on sovereign bond prices on a worldwide
scale. The inflation-linked bond markets may, as of today, well be too inexperienced and
illiquid to assert this. What we have shown in this article is that defining a global inflation
factor in the way it is done in model (6) is an effective way to capture the new risk dimension
that the new inflation-linked bonds have introduced.
We point out that a bond investor, who would want to assess the risk of an internationally
invested portfolio using the enhanced modeling framework we propose, is not refined to stay
in the data presentation in which the model was estimated. In other words, although the model
factors are defined onto the nominal yields and the breakeven rates explicitly, the portfolio
may contain inflation-linked bonds as well. In order to assess the portfolio risk, it suffices to
adapt the presentation of the model. This can be done easily by means of a rotation operator
L3 given in equation (8) below. The covariance matrix that is spanned by model (6), denoted
as VNBY,BEIR, can be re-expressed in terms of nominal- and real bond yields, denoted as
VNBY,RBY, by applying the following matrix multiplication:
The original matrix given in (9a) transforms into the matrix in (9b). In order to verify the
overall coherence, the reader may check that if the rotation operator is applied three times
consecutively, the original covariance matrix reappears.
VNBY ,NBY VNBY ,BEIR NBY ,RBY VNBY ,NBY VNBY ,NBY VNBY ,BEIR
V NBY ,BEIR V (9)
V
NBY ,BEIR VBEIR ,BEIR
V
NBY , NBY
V NBY , BEIR
V NBY , NBY
2 V NBY , BEIR
VBEIR , BEIR
(a) (b)
Conclusion
The issuing of inflation-protected securities in countries over the globe has expanded the
investment universe of fixed-income investors and has brought an essentially new dimension
into their performance opportunity set. An investor traditionally holding nominal bonds can
seize this new opportunity by adding inflation swaps, or by adding linkers, or alternatively, by
replacing the entire portfolio by one that contains linkers and inflation swaps only. In
principle the three options are equivalent in terms of risk-and-return profile, however, it is not
trivial to assess the portfolio risk in this new situation in an effective and coherent way.
The bond literature is surprisingly silent on the question. In this article a modeling approach
has been presented which tackles the issue in an international bond allocation setting. The
choice of setting allows dealing with an intriguing price phenomenon that plays on an
international scale. A two-factor linear model has been developed that is set to capture the
international price covariance between the nominal- and inflation-linked bonds. The first
factor may be loosely interpreted as a global inflation factor priced by the bonds. The second
factor divides the nominal yields from the breakeven rates and may for that matter be
interpreted as a real economy factor driven by global structural macroeconomic shocks.
Cette, G., and M. de Jong. The Rocky Ride of Breakeven Inflation Rates. Economics
Bulletin, Vol. 5, No. 30 (2008), pp. 1-8. A more complete version with the same title is
available as a Banque de France working paper, No. 230, 2009.
http://www.banque-france.fr/gb/publications/telechar/ner/dt230.pdf.
__________ Breakeven Inflation Rates and Their Puzzling Correlation Relationships. under
revision with Applied Economics.
Christensen, I., F. Dion, and C. Reid Real Return Bonds, Inflation Expectations and the
Break Even Inflation Rate. Bank of Canada Working Paper, No. 43, 2004.
Grkaynak, R., B. Sack, and J. Wright. The TIPS Yield Curve and Inflation Compensation.
American Economic Journal, Macroeconomics, Vol. 2, No.1 (2010), pp. 70-92.
Market-Implied Inflation
and Growth Rates Adversely
Affected by the Brent
Gilbert Cette,
Associated Professor in Economics, Universit dA ix-Marseille II
Marielle de Jong,
Head of Fixed Income Quantitative Research, Amundi
October 2012
The inf lation and the real yield component deduced from
inf lation-linked and nominal bond prices are adversely affected
by two market effects: price distortions due to certain market-
related events and oil price movements. Their underlying time-
correlation without those effects is stable and positive. Market
data analysis carried out on the worlds major bond markets
gives valuable new insight in the long-debated relationship
between inf lation and growth prospects.
When inflation-linked bonds were introduced on the world bond markets one to two decades
ago, there was positive belief that their pricing would reveal the inflation and growth
expectations of the market participants. In a short version of Fishers (1930) interest rate
theory, the yield of these bonds, the real yield, reflects the economic growth forecast, while
the yield differential (nominal minus real) called the breakeven inflation rate, reflects the
inflation forecast. It has proven difficult though to make such assertions on the market data
that has become available since.
Furthermore, it has proven difficult to learn from the bond data how inflation and economic
growth mutually interact. In our previous articles (Cette and de Jong, 2008, 2013) we had
made an attempt, making apparent that the time-correlation between real-bond yield (RBY)
and breakeven inflation (BEIR) variations is continuously distorted within countries by
market-related events. Observations made within local markets, which is the standard in the
literature on inflation-linked bonds, may therefore be misleading. By taking an international
approach we had been able to separate out the correlation due to country market distortions to
a certain extent, so as to obtain a view on the more fundamentally-driven correlation. It
showed that the correlation measured on a global aggregate scale is positive between RBY
and BEIR, except during the heat of the financial crisis in 2008/2009.
What does this say about the interaction between inflation and growth prospects? We show in
this article that the oil price plays an important role. There is an apparent adverse relation
between breakeven inflation and real yield movements, the former being driven up by an oil
price rise while the latter is pushed down. When eliminating the effect of oil from the bond
prices, the net global correlation between BEIR and RBY rises. In the crisis sub-period in
2008/2009, the oil price was particularly turbulent provoking large adverse movements
Again, taking an international study approach is essential in making the observations. The
influence of oil is easier to detect in global aggregate bond yield variations where the country-
specific effects are diversified away and oil, a common denominator for all economies,
remains. The new test results contribute to the longstanding debate on the relation between
inflation and economic growth prospects.
2. Data
The bond market data has been retrieved from Barclays Capital. The markets, member of the
World Government Inflation-Linked Bond Index (WGILB), which have been issuing
inflation-linked bonds since at least a decade, have been retained. 1 They constitute, in June
2012, of nineteen Inflation-Linked Gilts issued in the United Kingdom, thirty-three Treasury
Inflation-Protected Securities (TIPS) in the United States, twenty-six Obligations
Assimilables du Trsor indexes sur linflation (OATi) in the Euro Area, five Treasury
Indexed Bonds in Australia, six Index-Linked Treasury Bonds in Sweden and six Real Return
Bonds in Canada.
For calculating the breakeven inflation rates, the Barclays Breakeven Comparator indices
have been used, which are nominal bond indices purposely designed to match the
characteristics of the WGILB members. This is to avoid that the breakeven rates which are
simply calculated as the nominal minus the real yield, are being distorted by rotations in the
yield curves. For each market and each bond type, aggregate yields are calculated by Barclays
over all the maturities in the index. We refer to James (2010) for more details on Barclays
calculus.
The crude oil Brent FOB US dollar price series, available via Datastream, is used as the oil
price.
The observation period runs over ten years from July 2002 to June 2012, tests being done on
monthly data, which corresponds to 120 observations. In order to define the crisis months, we
have measured market turbulence by means of the standard deviation of the weekly variation
in the breakeven rates over four weeks over all countries in the dataset. If this measure
1
Japan has not been retained for this reason. This country started issuing inflation-linked bonds in 2004 and has
suspended its program in 2008 until further notice. See http://www.mof.go.jp for press releases by the Ministry
of Finance.
In his seminal book Theory of Interest, Fisher (1930) hypothesized that the two components
of the nominal interest rate, the real rate and the inflation expectation, should be unrelated to
one another, this since they are driven by independent economic factors. In Cette and de Jong
(2008, 2013) we find that the respective bond components, observable since the issuance of
inflation-linked bonds, are not univocal on the matter. Correlations between real yield and
breakeven inflation variations measured locally country by country are close to zero, giving
indication that Fishers hypothesis holds. However, their cross-border correlations are
systematically positive, which indicates that it doesnt.
The deadlock can be broken by separating local and international price effects. Through a
standard regression analysis, we estimate worldwide common bond yield- and country-
specific movements. 2 Ignoring the small cross-correlation terms, the complete correlation
matrix between the 'RBY and 'BEIR over the various countries, given in Figure 1 (a), is
decomposed into a common (b) and an idiosyncratic (c) correlation matrix. This exercise is
carried out over the entire observation period from July 2002 to June 2012, in I, and over the
period barred the crisis months, in II. The eight crisis months have been defined in the data
section.
2
Common movements are obtained, both for RBY and BEIR, by regressing on time-fixed effects. The
idiosyncratic components are the residuals of the regressions. Consequently, the common correlation, in
matrix (b), is identical for all countries and cross-combinations.
es
es
in
in
at
at
ita
ita
ea
ea
St
St
lia
lia
Br
Br
en
en
da
da
Ar
Ar
ra
ra
d
d
ed
ed
at
at
na
na
ite
ite
ro
ro
st
st
re
re
Sw
Sw
Ca
Ca
Un
Un
Au
Eu
Au
Eu
G
G
Australia -0.0 -0.1 0.0 -0.0 0.2** -0.1 -0.4*** 0.1 0.2** 0.1 -0.0 0.1
Canada 0.3*** -0.4*** -0.0 0.0 0.3*** -0.0 0.2** -0.6*** 0.1 0.1 -0.1 0.2**
'BEIR
Euro Area 0.1 -0.3*** -0.5*** -0.3*** 0.2* -0.3*** = -0.1 + 0.2** 0.2*** -0.5*** -0.1 0.2** -0.0
Great Britain 0.0 -0.2* -0.1 -0.3*** 0.3*** -0.2* -0.2*** 0.2*** 0.1 -0.4*** 0.1 0.1
Sweden 0.3*** 0.0 -0.0 0.1 0.0 -0.0 0.0 0.2** -0.0 0.3*** -0.6*** 0.2**
United States 0.1 -0.3*** -0.2* -0.2* 0.2*** -0.3*** 0.2** -0.1 0.1 -0.0 0.5*** -0.6***
*** **
, and *: significant at, respectively, the 1%, 5% and 10% level - Using an asymptotic T-test with T=120.
(a)
(b )
(c )
0
Schematically 0
II Ten-year observation period barred the crisis months from October 2008 to May 2009
es
in
in
at
at
ita
ita
ea
ea
St
St
lia
lia
Br
Br
en
en
da
da
Ar
Ar
ra
ra
d
ed
ed
at
at
na
na
ite
ite
ro
ro
st
st
re
re
Sw
Sw
Ca
Un
Ca
Un
Au
Eu
Au
Eu
G
Australia 0.2*** -0.0 0.2* 0.3*** 0.4*** 0.2** -0.4*** -0.0 0.2* -0.0 0.1 0.2**
Canada 0.5*** -0.4*** 0.1 0.2** 0.3*** 0.1 0.2 -0.6*** 0.1 0.1 0.1 0.0
'BEIR
Euro Area 0.5*** -0.0 -0.3*** 0.2* 0.2** 0.1 = 0.2** + 0.2* 0.4*** -0.6*** 0.1 -0.0 0.1
Great Britain 0.3*** -0.1 0.1 0.0 0.3*** 0.1 -0.3*** 0.3*** 0.2* -0.4*** 0.1 0.1
Sweden 0.4*** 0.0 -0.0 0.2*** -0.0 0.1 0.1 0.2** -0.0 0.1 -0.6*** 0.1
United States 0.4*** -0.3*** 0.0 0.1 0.3*** -0.1 0.3*** -0.3*** 0.1 0.1 0.4*** -0.5***
*** **
, and *: significant at, respectively, the 1%, 5% and 10% level - Using an asymptotic T-test with T=112.
(a)
(b )
(c )
0
Schematically 0
The market distortions are discussed in the finance literature (see Christensen et al., 2004, for
a survey). They are recognised to lead to a price premium; less attention is paid to their
influence on the correlation structure between bonds. A series of articles mention that
liquidity problems on the inflation-linked bond market are the main cause of the price
distortions (see for example Sack and Elsasser, 2004, Shen, 2006, DAmico et al., 2010, and
Grkaynak et al., 2010). Another series of articles points rather at the behaviour of investors.
Hesitance in taking on inflation risk makes prices fluctuate (see for example Hrdahl and
Tristani, 2007, on Euro Area data, Ejsing et al., 2007, or Emmons, 2000, on US data). A few
recent articles recognise both causes and estimate the respective price premiums
simultaneously (see Pflueger and Viceira, 2011, Haubrich et al., 2011, and Christensen and
Gillan, 2012).
The global market distortion in 2008-2009 is discussed in the literature as well. James (2010)
and Campbell et al. (2009) report massive flights to liquidity. Hu and Worah (2009) as well as
Bekaert and Wang (2010) mention that the bankruptcy of Lehman Brothers has added to the
turmoil, for it was the world leader in inflation-secured investment instruments. Pond (2012)
actually mentions that in this period the usual price relations were inverted. Figure 2 below
makes the situation clear. It shows that in the developed countries, the breakeven inflation
level dropped dramatically in late 2008, to renormalize in early 2009, back to pre-crisis levels.
The abnormal BEIR levels stem from a simultaneous decrease in the nominal yields and
increase in the real yields.
In the same Figure, the oil price is displayed which is remarkably synchronised with the
breakeven levels.
4 120
3 100
2 80
1 60
0 40
-1 20
Regarding Figure 2, it seems relevant to take account of the oil price, and decompose the
common correlation between RBY and BEIR (matrix b in Figure 1) further in a Brent-induced
component (b 1 ) and a residual component (b 2 ). The Brent-induced correlation should be
negative. Oil being an important factor of inflation, it should be positively correlated to
LQIODWLRQH[SHFWDWLRQVDQGWKXVWKH%(,5VHHIRUH[DPSOH&KHQDQG'H*UHJRULRet
al., 2007). Meanwhile, the oil price has an opposite impact on the economic activity
HQJHQGHULQJ QHJDWLYH FRUUHODWLRQ ZLWK JURZWK DQG WKXV WKH 5%< VHH %DUVN\ DQd Kilian,
2004, and Cuado and Prez de Gracia, 2003).
The negative impact of oil on the economy may pass through two channels: a production cost
effect (an increase in the production costs decreases the output equilibrium level) and a
Mundell-Tobin effect, which is a behavioural effect (in reaction to an oil price rise households
increase their savings which lowers the output equilibrium level). Ang et al. (2008) find
(weak) evidence of the Mundell-Tobin effect in American bond data.
where y it are the bond yield variations ('RBY and 'BEIR) in country i over month t, x t is a
function of the Brent price variation, E y measures the sensitivities of the two yield variations
to the Brent, It are time dummies, J yt measures the non-Brent common variation and H yit are
the residuals, which are assumed to be identically and independently distributed, and
represent the idiosyncratic variation component of y i .
)RU ERWK 5%< DQG %(,5 WKH PRGHO LV HVWLPDWHG WKURXJK 2UGLQDU\ /HDVW 6TXDUHV RQ WKH
whole period in two steps. First the common- and idiosyncratic variation is split, and then the
common component is split further into a Brent and non-Brent sub-component. Best results
are obtained with a non-linear impact of the oil price. To capture nonlinearity, we mount the
oil price (P) log-returns to the power three, i.e. x t = ln (P t /P t-1 )3. We have deliberately kept
the specification and estimation of the model simple.
The second-step estimation results are given in Table 3. The sensitivities to the oil price have
the intuitive signs as commented above. Interestingly, the total effect of the oil price on the
nominal yield, which is by construction the sum, E 'NBY E 'RBY E 'BEIR , is non-
significantly different from zero. It is perhaps for this reason that there is little discussion in
the literature on the effect of oil on bonds. The introduction of inflation-linked bonds on the
capital markets has made this observable.
Table 3
Estimation results
The common component of y is regressed on the Brent price changes
Explanations in the text
y='BEIR y='RBY
Ey 8.07 -4.95
T-statistic 6.95 -3.90
R2 0.11 0.04
N obs. 120 120
Figure 4
Schematic decomposition of the correlation matrix between 'RBY and 'BEIR
I Entire ten-year observation period
(a) b (c)
> 0.1@ 0
b1 b2
-
> 0.2@ >0.1@ 0
II Ten-year observation period barred the crisis months from October 2008 to May 2009
(a) b (c)
>0.2@ 0
b1
b2
-
> 0.1@ >0.3@ 0
We find the three components constituting the correlation between RBY and BEIR to have
stable signs, yet the total correlation (the sum) to be unstable over time. The net sum depends
on the share of each component, which is time-varying.
5. Conclusion
We have shown that the breakeven inflation and real yields deduced from the developed bond
markets are adversely affected by two factors: price distortions due to market-related events
and oil price movements. Without the influence of those, their correlation is positive. This
finding contributes, we reckon, to a better understanding of the long-debated complex
interrelationship between inflation and economic growth prospects. The effect of oil on bond
prices has become measurable thanks to the emergence of inflation-linked securities on the
markets. The results fit in with macroeconomic theory. An oil price rise drives up inflation
and slows down economic growth.
Barsky, R. and L. Killian (2004): Oil and the Macroeconomy since the 1970s, NBER
Working Paper 10855.
Bekaert G. and X. Wang (2010): Inflation risk and the inflation risk premium, Economic
Policy, 755-806.
Cette, G. and M. de Jong (2008): The Rocky Ride of Breakeven Inflation Rates, Economics
Bulletin, Vol; 5, n 30, 1-8.
Chen, S. (2009): Oil Price Pass-Through into Inflation, Energy Economics 31, 126-133.
Christensen, I., F. Dion and C. Reid (2004): Real Return Bonds, Inflation Expectations and
the Break Even Inflation Rate, Bank of Canada Working Paper, 2004-43.
Christensen, J. and J. Gillan (2012): Could the U.S. Treasury Benefit from Issuing More
TIPS?, Federal Reserve Bank of San Francisco Working Paper, 2011-16.
Cuado, J. and F. Prez de Gracia (2003): Do oil price shocks matter? Evidence for some
European countries , Energy Economics 25, 137-154.
DAmico S., D. H. Kim and M. Wei (2010): Tips from TIPS: The informational content of
Treasury Inflation-Protected Security prices, FEDs finance and economics discussion series,
2010-19.
De Gregorio, J., O. Landerretche and C. Neilson (2007): Another Pass-Through Bites the
Dust? Central Bank of Chile Working Paper n 417.
Ejsing, J., J. A. Garcia and T. Werner (2007): The Term Structure of Euro Area Break Even
Inflation Rates, European Central Bank, Working Paper Series, n 830, November.
Grkaynak, R. S., B. Sack and J. H. Wright (2010): The TIPS Yield Curve and Inflation
Compensation, American Economic Journal, Macroeconomics, 2:1, 70-92.
Hrdahl, P., and O. Tristani (2007): Inflation risk premia in the term structure of interest
rates, BIS Working Papers, n 228, May.
Hu, G. and M. Worah (2009): Why Tips Real Yields moved significantly higher after the
Lehman Bankruptcy, PIMCO, Newport Beach, CA.
Pond, M. (2012): Beta Calculations, Drivers and Uses, Barclays Capital Research
document.
Shen, P. (2006): Liquidity Risk Premia and Breakeven Inflation Rates, Economic Review,
Federal Reserve Bank of Kansas City, Second Quarter, 29-53.
Breakeven Inflation
Rates and their Puzzling
Correlation Relationships
Gilbert Cette,
Associated Professor in Economics, Universit dA ix-Marseille II
Marielle de Jong,
Head of Fixed Income Quantitative Research, Amundi
December 2011
The yield of an inflation-linked bond, informally called the real bond yield (RBY), reflects the
market pricing of the long-term real interest rate in the same manner as the nominal bond
yields (NBY) price the nominal interest rate. The relatively recent issuance of inflation-
protected securities by governments around the world has made it possible, for the first time 1,
to pair up the two bond types and observe the yield differentials, called the breakeven
inflation rates (BEIR). The longstanding idea of decomposing interest rates into two
components, introduced by Irving Fisher in his seminal book named Theory of Interest
(1930), can eventually be tried now the new bond markets are maturing and becoming more
liquid.
Fisher had hypothesized that the two components should be unrelated to one another: the
real interest rate is entirely determined by the real factors in an economy, i.e. the productivity
of capital and the investors time preference, and should thus be unrelated to the inflation
expectation. Many efforts have been undertaken to provide empirical evidence; see Cooray
(2002) for a literature review. In Cette and de Jong (2008), we had made a renewed attempt
with tests on recent bond market data, observing that the Fisher hypothesis seems to hold
country per country yet is definitely rejected in an international context. Tests were based on a
historical correlation matrix measured between the two interest rate components, BEIR and
RBY, across various countries, featuring near zeros (more exactly, unsystematically negative,
nil or positive numbers) on the diagonal, i.e. within countries, and strictly positive numbers
elsewhere, i.e. between countries.
1
The scale in which such bonds are being issued is new, not the concept. According to Shiller (2003) the first
inflation indexed bonds were issued by the Commonwealth of Massachusetts in 1780 during the Revolutionary
war to deal with severe wartime inflation.
total correlation
common correlation
idiosyncratic correlation
0 0
0 0
(a) (b) (c)
The finding is puzzling. The correlation structure, untypical for bonds and casting doubt on
the Fisher hypothesis, pulls all analyses traditionally made in a national perspective into
scrutiny.
In our previous article we had shown the correlation structure to be stable over time up to
mid-2008. In this article we study what happened after, in the current financial crisis, up to
mid-2010, during which bonds have been in great turmoil, as documented by many such as
Campbell et al. (2009). Our main findings are the following. While the idiosyncratic
correlation structure remained unchanged in the turmoil, the common correlation became
negative, resulting in a negative correlation matrix overall with significantly negative
coefficients on the diagonal. We explain this change by a worldwide lack of liquidity in the
real bond markets affecting the prices, and we observe that, as soon as this simultaneous
liquidity problem resolved in 2009, the bond prices settle back into the same regime as given
in Figure 1. Our international study approach remains original in the literature to our
knowledge, many other articles focusing on national indexed bond markets only. It proves
essential in gaining insight in the effects behind the interest rate component movements.
Section 2 discusses the data issues, section 3 and 4 give an analysis of the bond correlations
respectively before and since the crisis, and section 5 concludes.
The data has been retrieved from Barclays Capital. Developed countries issuing inflation-
linked bonds since at least a decade have been retained. 2 It covers the Inflation-Linked Gilts
issued in the United Kingdom, the Treasury Inflation-Protected Securities (TIPS) in the
United States, les Obligations Assimilables du Trsor indexes sur linflation (OATi) in the
Euro Area, the Treasury Indexed Bonds in Australia, the Index-Linked Treasury Bonds in
Sweden and the Real Return Bonds in Canada. Generic bond yields have been calculated by
Barclays Capital per interval of maturity dates. The seven-to-ten-year term-to-maturity
interval with a bond-duration close to 7.5 years has been selected for this study, since it is by
and large the most liquid category. We refer to Barclays Global Inflation-Linked Products
edited by James (2010) for more details on their calculus.
The observation period, from mid-2002 to mid-2010 has been divided in three in order to
separate out the period of great market turbulence in 2008-2009. There is a pre-crisis period,
from July 2002 to September 2008, a turbulent period, from October 2008 to May 2009, and a
post-turbulent period, from June 2009 to June 2010. In order to set the cut-off dates we have
measured market turbulence by means of the standard deviation of the weekly variation in the
breakeven inflation rates over four weeks and over all countries in the dataset. If this measure
exceeds two times its historical average, the market is deemed turbulent.
The correlation matrix measured in Cette and de Jong (2008) between the major bond markets
over a six-year period from 2002 to 2008 is reprinted in Table 1.
2
Japan has not been retained for this reason.
RBY
***
: significant at the 1% level (critical value at 0.13); **: significant at the 5% level (critical value at 0.11);
*
: significant at the 10% level (critical value at 0.09) - Using an asymptotic T-test with T=325.
A dual phenomenon can be observed. The correlations are nil, or more precisely they are
unsystematically negative, nil or positive, as discussed in the introduction, within countries
(on the diagonal), while in contrast they are strictly positive between countries (the cross
terms). This is puzzling. In practical terms, it means that the yield variation of an American
TIPS, to take an example, is uncorrelated with the breakeven inflation movements in the US
(bottom right number in the matrix). It indicates that the bond price is insensitive to inflation
concerns, which is in effect the raison dtre of the security. Yet why would its yield correlate
with breakeven movements registered in other countries (the numbers in the rightmost
column)? The correlations do not strike with theory either. The Fisher hypothesis seems to
hold within countries in effect a joint test of zero correlation is not rejected even on a 10 %
error level-, however it is rejected, on a 1 % significance level, across countries.
3
On a monthly data frequency results are very similar.
The data has been retrieved from Barclays Capital. Developed countries issuing inflation-
linked bonds since at least a decade have been retained. 2 It covers the Inflation-Linked Gilts
issued in the United Kingdom, the Treasury Inflation-Protected Securities (TIPS) in the
United States, les Obligations Assimilables du Trsor indexes sur linflation (OATi) in the
Euro Area, the Treasury Indexed Bonds in Australia, the Index-Linked Treasury Bonds in
Sweden and the Real Return Bonds in Canada. Generic bond yields have been calculated by
Barclays Capital per interval of maturity dates. The seven-to-ten-year term-to-maturity
interval with a bond-duration close to 7.5 years has been selected for this study, since it is by
and large the most liquid category. We refer to Barclays Global Inflation-Linked Products
edited by James (2010) for more details on their calculus.
The observation period, from mid-2002 to mid-2010 has been divided in three in order to
separate out the period of great market turbulence in 2008-2009. There is a pre-crisis period,
from July 2002 to September 2008, a turbulent period, from October 2008 to May 2009, and a
post-turbulent period, from June 2009 to June 2010. In order to set the cut-off dates we have
measured market turbulence by means of the standard deviation of the weekly variation in the
breakeven inflation rates over four weeks and over all countries in the dataset. If this measure
exceeds two times its historical average, the market is deemed turbulent.
The correlation matrix measured in Cette and de Jong (2008) between the major bond markets
over a six-year period from 2002 to 2008 is reprinted in Table 1.
2
Japan has not been retained for this reason.
RBY
***
: significant at the 1% level (critical value at 0.13); **: significant at the 5% level (critical value at 0.11);
*
: significant at the 10% level (critical value at 0.09) - Using an asymptotic T-test with T=325.
A dual phenomenon can be observed. The correlations are nil, or more precisely they are
unsystematically negative, nil or positive, as discussed in the introduction, within countries
(on the diagonal), while in contrast they are strictly positive between countries (the cross
terms). This is puzzling. In practical terms, it means that the yield variation of an American
TIPS, to take an example, is uncorrelated with the breakeven inflation movements in the US
(bottom right number in the matrix). It indicates that the bond price is insensitive to inflation
concerns, which is in effect the raison dtre of the security. Yet why would its yield correlate
with breakeven movements registered in other countries (the numbers in the rightmost
column)? The correlations do not strike with theory either. The Fisher hypothesis seems to
hold within countries in effect a joint test of zero correlation is not rejected even on a 10 %
error level-, however it is rejected, on a 1 % significance level, across countries.
3
On a monthly data frequency results are very similar.
(i) The common global component of the BEIRs tends to move in the same direction as
the common global component of the RBYs, resulting in a positive common
correlation matrix;
(ii) The idiosyncratic country-specific component of the BEIRs tends to move against the
idiosyncratic components of the RBYs within the same country, yet have no statistical
relation with those of other countries, resulting in a diagonal negative correlation
matrix.
We interpret the two effects separately. The negative idiosyncratic correlation can be directly
associated with certain time-varying market distortions that are being mentioned in the
literature; see Christensen et al. (2004) for a survey. As soon as the real bond price moves due
to such country-specific market-related events, which are typically not mirrored in the
nominal bond price, the breakeven rate mechanically moves in opposite direction. Those
events appear to be sufficiently recurrent to provoke a systematic idiosyncratic negative
correlation over time. Our observation complements the literature that analyses a price
premium on real bonds relative to nominal bonds resulting from those market distortions.
Abrupt market liquidity problems are mentioned to be the main cause of the price distortions,
for example by Craig (2003), Sack and Elsasser (2004), Shen (2006), DAmico et al. (2009),
Campbell et al., (2009) as well as Grkaynak et al. (2010). Grkaynak et al. (2010) relate the
liquidity premium, which is being observed in the inflation-linked bond prices compared to
the nominal bond prices, to the particularly low trading volumes for inflation-linked bonds. 5
They explain in a regression analysis the time-variability of the liquidity premium on TIPS
4
The detailed decomposition is available upon request from the authors.
5
Grkaynak et al. (2010) report that in the US, the TIPS market expressed as a share of total Treasury trading
represented about 0.5% in 1999 and 2% in 2006.
0.3
'RBY
0.2
0.1
'NBY
0.0
-0.3 -0.2 -0.1 0.0 0.1 0.2 0.3
-0.1
-0.3
We investigate what happened during the current crisis period. The correlation matrix
between breakeven inflation and real bonds measured between September 2008 and May
2009 is given in Table 2. Over this period, the correlations are (i) strongly negative within
countries (on the diagonal), and (ii) weakly negative between countries (the cross terms). The
test of zero correlation is integrally rejected on a 1% significance level. Despite the admittedly
reduced reliability of the tests in non-stationary times, it may be concluded that the Fisher
hypothesis does not hold, even within countries.
RBY
***
: significant at the 1% level (critical value at 0.39); **: significant at the 5% level (critical value at 0.33);
*
: significant at the 10% level (critical value at 0.28) - Using an asymptotic T-test with T=35.
The same elementary matrix decomposition produces the schema as displayed in Figure 3. 6
Note that the sole difference with the pre-crisis period lies in (b): the correlation between the
common global BEIR and RBY movements turns negative. The sign change is informative. It
is, to us, caused by the same issues of liquidity and risk attitude mentioned above. The
worldwide market distortions have provoked a negative common correlation between the
BEIR and the RBY, in the same way as they do on a national scale in normal times. In the
heat of the crisis, the inflation-linked bond markets contracted in all countries simultaneously.
Considering the relatively small trading volumes on these markets, the price shocks the
inflation-linked bonds incurred provoked a negative correlation with the breakeven inflation
rates in all countries. Any global price shock in the real bonds that is typically not registered
in the nominal bond markets leads to a mechanic opposite movement in the global breakeven
inflation.
6
The complete decomposition is omitted but can be obtained from the authors upon request.
total
correlation
common correlation
idiosyncratic correlation
0
0
(a) (b) (c)
This interpretation fits in with the finance literature reporting on the crisis events: there was a
massive flight to liquidity. James (2010) wrote: the extreme deleveraging phase that
engulfed almost all financial markets included the majority of off-benchmark investors in
inflation-linked bonds being stopped out of their positions. The bankruptcy of Lehman
Brothers added to the turmoil for it was the world leader on inflation-secured instruments (see
Hu and Worah, 2009, or Bekaert and Wang, 2010). Simultaneously in the economic literature,
the near-meltdown of the financial sector was seen as the start of an extended low-growth
period, with lower inflation than expected before the crisis. The relevance of inflation
issuance by inflation-linked bonds decreased, reducing its demand. The price fall of
commodities -in particular petrol- was reinforcing this view.
The flight to the mature nominal bond markets, which penalized TIPS demand and increased
their risk premium in 2008, as show Grkaynak, Sack and Wright (2010), was worldwide, its
factors being themselves worldwide, as mentioned before. It is shown in Figure 4 that not
only in the US, but in all developed world countries, the BEIR decreased dramatically from in
the last quarter of 2008, to normalize after in the first half of 2009, back to pre-crisis levels.
The abnormal BEIR levels in 2008 and 2009 stem from a decrease in the nominal rates as
well as an increase in the real rates. These global co-movements, that are a good illustration of
the generalized flight-to-liquidity behaviour, explain the common negative co-movements of
the BEIR and the RBY, which explain on its turn why the common correlation matrix exhibits
negative numbers during this short period.
From June 2009 when markets calmed down, the correlation matrix settled back into the same
regime as before (see Table 3), and its decomposition matches with Figure 1. This shows, or
gives strong indication, that the untypical correlation structure between BEIR and RBYs is
nevertheless robust.
RBY
***
: significant at the 1% level (critical value at 0.31); **: significant at the 5% level (critical value at 0.26);
*
: significant at the 10% level (critical value at 0.22) - Using an asymptotic T-test with T=56.
5. Conclusion
The events on the capital markets during the present crisis provide new insight in the price
covariance structure of bonds. The two components of the nominal bond yield, the breakeven
inflation and the real bond yield, have the propensity to be positively correlated, yet are
pushed into a negative correlation relationship by certain market events. As long as those
market distortions are local, the net result is near-zero correlation within countries; when they
become global, as was the case in the heat of the crisis, the correlation between real bond
yields and breakeven rates turns negative worldwide. Those empirical findings have been
shown to be robust over the pre- and post-crisis period.
This untypical correlation behaviour has been left uncommented in the literature, the reason
for that being that studies on inflation-linked bonds are traditionally made in a national
context, while an international analysis is necessary to reveal the underlying effects. We
consider the demonstration of the correlation relationships an important contribution to the
The results are a contribution to the economic literature as well, the more that they are not in
line with conventional theory of Fisher (1930) postulating zero correlation between the two
interest rate components. The discrepancy between macroeconomic theory and financial
practice has been indirectly reported by central bankers, e.g. in Bernankes (2004) speech
What Policymakers Can Learn from Asset Prices. The issuance of inflation-linked bonds had
in part been motivated by the expectation that the observed breakeven inflation would in some
way reflect the credibility granted to Central Banks regarding their control on inflation.
Bernanke reckons the volatility of the breakeven inflation too high, to the extent that the
market data remains as it stands of very limited use for policymaking purposes.
The authors thank Hubert Kempf and Jean-Paul Renne for their valuable comments.
References
Bernanke, B., (2004): What Policymakers Can Learn from Asset Prices, The Federal
Reserve Board.
Bekaert G. and X. Wang (2010): Inflation risk and the inflation risk premium, Economic
Policy, 2010, pp. 755-806.
Cette, G. and M. de Jong (2008): The Rocky Ride of Breakeven Inflation Rates, Economics
Bulletin, Vol; 5, n 30, pp. 1-8.
Christensen, I., F. Dion and C. Reid (2004): Real Return Bonds, Inflation Expectations and
the Break Even Inflation Rate, Bank of Canada Working Paper, 2004-43.
Cooray, A., (2002): The Fisher Effect: a Review of the Literature, Research paper n 206,
department of economics, Macquarie University.
Ct, A., J. Jacob, J. Nelmes and M. Whittingham (1996): Inflation Expectations and Real
Return Bonds, Bank of Canada Review, Summer, pp. 41-53.
Craig, B. (2003): Why are TIIS Yields so High? The case of the Missing Inflation-Risk
Premium, Federal Reserve Bank of Cleveland Economic Commentary.
DAmico S., D. H. Kim and M. Wei (2009): Tips from TIPS: The informational content of
Treasury Inflation-Protected Security prices, mimeo, December, 29.
Ejsing, J., J. A. Garcia and T. Werner (2007): The Term Structure of Euro Area Break Even
Inflation Rates, European Central Bank, Working Paper Series, n 830, November.
Evans, M. (1998): Real Rates, Expected Inflation and Inflation Risk Premia, Journal of
Finance, pp. 187-218.
Grkaynak, R. S., B. Sack and J. H. Wright (2010): The TIPS Yield Curve and Inflation
Compensation, American Economic Journal, Macroeconomics, 2:1, pp. 70-92.
Hu, G. and M. Worah (2009): Why Tips Real Yields moved significantly higher after the
Lehman Bankruptcy, PIMCO, Newport Beach, CA.
Hunter, D. and D. Simon (2005): Are TIPS the real deal? A conditional assessment of their
role in a nominal portfolio, Journal of Banking and Finance, 29 (2005), pp. 347-368.
Pond, M. (2008): Effective duration of linkers and beta, in Barclays Capital Research:
Global Inflation-Linked Products, February.
Shen, P. (2006): Liquidity Risk Premia and Breakeven Inflation Rates, Economic Review,
Federal Reserve Bank of Kansas City, Second Quarter, pp. 29-53.
At May 2014
Much ado about not much: the ECBs call DOISY Nicolas, 2014-05
for a negative rate DRUT Bastien
The good, the bad and the ugly: DOISY Nicolas 2014-04
deflationary risks in America and Europe
China: the big banks are not defenceless BEN ABDALLAH Marc-Ali 2014-04
against rising credit risk
Emerging debt markets: time for reason BEN ABDALLAH Marc-Ali 2014-03
Will 2014 be the year for more M&A LEOPOLD Marie-Hlne 2014-02
in the pharmaceutical sector?
The Fed breathes new life into the markets ITHURBIDE Philippe 2013-10
The link between credit spreads and bond BERTONCINI Sergio 2013-09
yields: is this time different?
Equities: spring is in the air, but the winter MIJOT Eric 2013-06
blues still need shaking
The hunt for yield: where to find spreads? BERTONCINI Sergio, 2013-06
DRUT Bastien
What does the future hold for the pound DRUT Bastien 2013-04
sterling?
Equity markets peak: what comes next? MIJOT Eric, WANE Ibra 2013-04
Dutch mortgage debt: what are the risks? PERRIER Tristan 2013-03
What are the risk factors for portfolios ITHURBIDE Philippe 2013-02
in a risk on mode?
2013 and beyond: what are the likely ITHURBIDE Philippe 2013-01
scenarios and which allocations to make?
Loose monetary policies for the long term BOROWSKI Didier 2013-01
Local debt attractiveness should not be only BEN ABDALLAH Marc-Ali 2012-12
assessed through currency appreciation
The auto sector: the death of the mid-price LABIA Frdric 2012-12
range
Italy and Spain: a tale of two situations and ITHURBIDE Philippe 2012-10
two trajectories
Flows: markets more reckless than investors MIJOT Eric, WANE Ibra 2012-10
Can the ECB become like the Fed or the BOROWSKI Didier 2012-09
BoE?
Global economy: what are the risks and ROGER Florian 2012-04
dynamics of inflation?
The ECBs balance sheet: hit by the crisis AINOUZ Valentine 2012-02
Forex markets: the euros fate will play BOROWSKI Didier 2012-01
out in 2012
After the full defensive play of 2011, be ready WANE Ibra 2012-01
to add a touch of emerging play for 2012
Euro zone is looking for the last resort investor BOROWSKI Didier 2011-11
China: dont we have to worry about banks? BEN ABDALLAH Marc-Ali 2011-11
Spain: the euro area situation is not enabling VARTANESYAN Sosi 2011-07
the country to fully get over the crisis
Credit: there is still value in High Yield bonds BERTONCINI Sergio 2011-07
Equities: are high margins a cause MIJOT Eric, WANE Ibra 2011-06
for concern?
Could the ECB derail growth in the Eurozone? ITHURBIDE Philippe 2011-05
Why is the euro rising and how far can it go? BOROWSKI Didier 2011-05
Japan: The economic and market outlook HEGARTY James, KONO 2011-04
following the earthquake Masanaga, OHNI Shizuko,
YOSHINO Akio
The ECB on the verge to hike rates, the BOROWSKI Didier, 2011-03
Fed No! What impact on the yield curve ITHURBIDE Philippe
and on the euro?
Excellent start to 2011 for the credit market BERTONCINI Sergio 2011-02
The bullish cycle on the equity markets is MIJOT Eric, WANE Ibra 2011-01
not over yet
DP-01-2014 Will the Real Janet Yellen Stand Up? ITHURBIDE Philippe Mar-14
2013 and beyond what are the likely scenarios ITHURBIDE Philippe Jan-13
and which allocations to make
Pascal BLANQU
Deputy Chief Executive Officer
Head of Institutional Investors and Third Party Distributors
Group Chief Investment Officer
Philippe ITHURBIDE
Global Head of Research, Strategy and Analysis
About Amundi
Amundi ranks first in Europe1 and ninth worldwide1 in the
asset management industry with AUM of over 800 billion
worldwide2.
1. Source IPE Top 400 asset managers active in the European marketplace
published in June 2013, based on figures as at December 2012. Interviews
with asset management companies on their assets as at end-December
2012 (open-end funds, dedicated funds, mandates).
2. Amundi Group figures as of 31 March 2014.
research-center.amundi.com