INVESTMENT & PENSIONS EUROPE

SUMMER 2011

EDHEC-Risk
Institute
Research Insights
EDHEC-Risk Institute
Research Insights

Introduction Noël Amenc Contents

I
t is my privilege to introduce the third EDHEC- have only the objective of lowering risk, rather than Is there a risk/return trade-off across
Risk Institute Research Insights supplement to aiming to optimise the risk/reward ratio, minimum stocks?  2
Investment & Pensions Europe. We are again variance portfolio optimisation leads to a pronounced
delighted to be teaming up with IPE to provide infor- concentration in low volatility stocks at the expense
Felix Goltz
mation on research-based solutions to some of the key of exploiting correlation properties. While mini- EDHEC-Risk Institute
challenges facing institutional investors today. mum variance portfolios have been shown to lead to Dev Sahoo
In the first article in the present supplement, ‘Is relatively poor performance, they may be suitable for
there a risk/return trade-off across stocks? An answer investors who wish to load up on low-risk or ‘defen- EDHEC-Risk Institute
from a long-horizon perspective’, Felix Goltz and Dev sive’ stocks.

I
Sahoo address one of the key questions in modern A post-crisis perspective on
finance from both an academic and practitioner n an article drawn from the Advanced Modelling
perspective: are investors rewarded for investing in for Alternative Investments research chair at diversification for risk management 3
high-risk stocks by enjoying higher expected returns? EDHEC-Risk Institute, supported by the Prime Noël Amenc
By adopting a long-horizon standpoint, our research Brokerage Group at Newedge, Lionel Martellini
provides an unambiguously affirmative answer to examines the question of optimal hedge fund alloca-
EDHEC Business School
this question: the trade-off between risk and return tion. This article discusses an application of improved Felix Goltz
does indeed exist over long horizons as postulated by estimators for higher-order comoment parameters in EDHEC-Risk Institute
financial theory. the context of hedge fund portfolio optimisation. In
The second article discusses what is probably the recent research we have found that the use of these Stoyan Stoyanov
single most important subject to come out of the enhanced estimates generates considerable benefits EDHEC-Risk Institute – Asia
financial crisis: risk management. As Stoyan Stoyanov for investors in hedge funds.
explains, risk management is about maximising the In the final article in the supplement we analyse
probability of achieving certain objectives at the dynamic core-satellite strategies with exposure to Efficient indices and efficient relative
investment horizon while staying within a risk budget. value and momentum strategies. In this research, return benchmarks 5
Diversification, hedging, and insurance can be relied produced as part of the Core-Satellite and ETF Invest-
on to make optimal use of risk budgets. These three ment research chair in partnership with Amundi
Noël Amenc
techniques involve different aspects of risk manage- ETF, we find that these investment strategies alone EDHEC Business School
ment, but they are complementary techniques rather could achieve higher returns but are exposed to high
than competing ones. extreme risk because they consist of equity portfolios
In the article on ‘Efficient indices and bench- that are concentrated in the sectors with the high- Advantages and shortcomings of
marks’, we look at the difference between a reference est value or momentum exposure. Combining these minimum variance portfolios 6
index and a custom benchmark and explain how this strategies with the dynamic core-satellite approach, Felix Goltz
distinction leads to different approaches to passive however, improves portfolio returns while also keep-
investment. We also introduce EDHEC-Risk Indices ing downside risk in check. EDHEC-Risk Institute
& Benchmarks’ efficient relative return benchmark We wish you an informative and enjoyable read of
approach, which allows investors to benefit from the supplement and look forward to continuing this
the performance of efficient diversification while editorial partnership with IPE. Our mutual objective
Optimal hedge fund allocation with
continuing to rely on the popularity and simplicity of with this supplement is to provide academic insights improved estimates for coskewness
traditional cap-weighted indices. that will genuinely contribute to improving institu- and cokurtosis parameters 7
Felix Goltz also looks in a subsequent article tional investment practices.
at the advantages and shortcomings of minimum Lionel Martellini
variance portfolios, and how the shortcomings can Noël Amenc, Professor of Finance, EDHEC Business EDHEC Business School
be addressed. Since minimum variance portfolios School, and Director, EDHEC-Risk Institute

Value and momentum effects across

exchange-traded funds  8
Elie Charbit
EDHEC Business School
Jean-René Giraud
Koris International/
EDHEC-Risk Institute
Felix Goltz
EDHEC-Risk Institute
Lin Tang
EDHEC-Risk Institute

EDHEC-Risk Institute Research Insights is published as a supplement to Investment & Pensions Europe
IPE International Publishers Ltd, 320 Great Guildford House, 30 Great Guildford Street, London SE1 0HS
Tel: +44(0)20 7261 0666, Fax: +44(0)20 7928 3332, Web site: www.ipe.com, ISSN 1369-3727
Investment & Pensions Europe is published monthly by IPE International Publishers Ltd. No part of this publication may be reproduced in any form without the prior permission of the
publishers. Printed by Hastings Printing Company, Drury Lane, St Leonards-on-Sea, East Sussex TN38 9BJ, UK.

2011 SUMMER INVESTMENT&PENSIONS EUROPE

2 | EDHEC-Risk Institute Research Insights

Is there a risk/return trade-off

across stocks?
An answer from a long-horizon perspective
T Felix Goltz, Head of Applied Research, EDHEC-Risk Institute
heory and – perhaps more importantly
– financial common sense suggest that
there should be a trade-off between a
stock’s riskiness and its expected returns. On
Dev Sahoo, Research Engineer, EDHEC-Risk Institute
the one hand, standard asset pricing models
suggest that systematic risk should be posi- 1. Risk/return profile for portfolios using long and short horizon
tively rewarded – ie, stocks with higher betas
should earn a higher expected return (see Ross’s
Arbitrage Pricing Theory, 1976). Subsequently, 25% 35%

research has underlined the explanatory power

Returns and standard error bounds

Returns and standard error bounds

20% 30%

of stock-specific or so-called idiosyncratic risk

15% 25%
for expected returns (Merton, 1987). Taken
together, these results suggest that total volatil- 10% 20%

ity, which is the model-free sum of systematic

5% 15%
volatility explained by a factor model, and
idiosyncratic volatility, should also be positively 0% 10%

rewarded (Martellini, 2008). -5% 5%

In contrast to this consensus regarding the
existence of an unambiguously positive risk- -10%
0% 5% 10% 15% 20% 25% 30%
0%
0% 5% 10% 15% 20% 25% 30%
return relationship from a theoretical perspec- Risk – idiosyncratic volatility Risk – idiosyncratic volatility

tive, a number of older as well as more recent

papers have reported a number of puzzling or, The graphs show the arithmetic average return and average risk of decile portfolios built on idiosyncratic volatility (relative to Fama-French Factor exposures). The
at least, contrasted findings from an empirical values plotted are the average values over all cross sections and are annualised. The bounds of the one-standard-deviation error are shown along the return axes, along
with the average values. The left-hand graph replicates the short horizon results of Ang et al (2006) while the right-hand graph uses a longer horizon of 24 months. The
perspective. First, the ‘low beta anomaly’ stipu- period of analysis runs from July 1963 to December 2009.
lates that the relationship between systematic
risk as measured by a stock beta and return
is much flatter than predicted by the Capital tive seems to be the natural approach since the that the short-horizon underperformance of
Asset Pricing Model (see early papers by Black, theoretical predictions of standard asset pricing high-volatility stocks can be attributed to the
1972; Black, Jensen and Scholes, 1972) and models relate to the relationship between a short-term return reversal effect. The cap-
sometimes even inverted (paper by Haugen and stock’s risk estimate and expected return on weighting scheme employed to weight stocks
Heins, 1975). More recently, Ang, Hodrick, Xing that stock across many varying market condi- within the portfolio further increases the
and Zhang (2006, 2009) have drawn new atten- tions. This can only be assessed by looking at a exposure to short-term return reversals as it
tion to these results with a focus on the specific stock performance over long horizons. A similar overweights the past winners. Using equally-
risk component, finding that high idiosyncratic approach is taken by Bandi and Perron (2008) weighted portfolios (which are not sensitive
volatility stocks have had “abysmally low in recent work on the long-term risk/return to past returns) and holding the portfolios for
returns” in longer US samples and in interna- relationship, with a focus on the time-series longer horizons avoids short-term reversal
tional markets. This result is now widely known perspective. In a related effort, from a cross- effects and provides a clear positive risk/return
as the ‘idiosyncratic volatility puzzle’. Yet other sectional perspective, Bandi et al (2010) find relationship across all the portfolios.
papers have documented a rather flat or even empirical support for an approximate long-run In order to show the horizon effect more
negative relationship between total (as opposed version of the CAPM, where betas and returns clearly we show the geometric mean returns of
to specific) volatility and expected return, an are both measured over long horizons. the high volatility and low volatility portfolios
anomaly that some call the ‘total volatility Below we report results obtained using a over different horizons in figure 2. The low-
puzzle’ (Haugen and Baker, 2008; Blitz and Van broad cross-section of US stocks over the period volatility portfolio yields higher return over a
Vliet, 2007; Baker, Bradley and Wurgler, 2011). from July 1963 to December 2009. First, we one-month horizon (ie, the risk/return relation-
Several attempts have been made to explain replicate the short-horizon findings in the early ship appears to be negative). But holding the
these puzzling empirical results. A number of literature on the idiosyncratic volatility puzzle. stocks for longer horizons shows the opposite
recent papers have questioned the robustness To study the effect of idiosyncratic volatility relationship (high-volatility portfolios have
of Ang et al’s (2006, 2009) results. Among other on long-horizon expected returns, we use a higher returns and lower-volatility portfolios
concerns, the findings are not robust to changes simple trading strategy similar to that used by have lower returns).
in data frequency, portfolio formation period, to Jegadeesh and Titman (1993). This trading Taking the risk/return relationship beyond
the screening out of illiquid stocks (Bali and Cak- strategy sorts stocks into portfolios every month the mean-variance setting, theoretical models
ici, 2008), or to adjusting for short-term return by their volatility and holds the portfolios over have also shown that investors are willing
reversals (Huang et al, 2010). Several other longer horizons of up to three years to allow to accept lower expected returns and higher
authors have changed the short-term measure expected return differences to materialise1. volatility compared to the mean-variance
of volatility in Ang et al (2006) with conditional From the short-horizon results, we can see benchmark in exchange for higher skewness and
measures estimated from returns over longer that the high volatility portfolio has negative lower kurtosis of returns. High skewness and
calibration periods and found a positive relation- returns over the following month. As Figure low kurtosis have been shown to be associated
ship (Fu, 2009; Brockmann and Schutte, 2007). 1 shows, the finding of a negative relation with lower expected returns in theory (see, for
In order to understand the risk-return between risk and return can be largely attrib- example, Barberis and Huang, 2004, Mitton
relationship further, recent research at EDHEC- uted to the portfolios of high-volatility stocks. and Vorkink, 2007, among many others). The
Risk Institute adopts a long-horizon perspec- The finding of a negative relation does not hold intuition behind this result is that investors like
tive. Rather than looking at return realisations when comparing returns across the first six to hold portfolios with positive skew and low
over a monthly horizon to proxy for expected portfolios. This is similar to the findings of Ang kurtosis. In terms of idiosyncratic and total risk
returns, we look at longer investment horizons et al. It is also well known (Huang et al 2010) measures, Boyer, Mitton and Vorkink (2010)
beyond one year, which are arguably closer and Conrad, Dittmar and Ghysels (2008) also
to horizons relevant for a typical institutional 1 Multiple portfolios are held at the same time and only those that ar- provide converging empirical evidence that indi-
equity investor. Taking a long-horizon perspec- rive at their horizon are rebalanced in a given month. vidual stocks’ skewness and kurtosis are indeed

INVESTMENT&PENSIONS EUROPE SUMMER 2011

 EDHEC-Risk Institute Research Insights | 3

2. Geometric mean returns for multi-portfolio analysis Bandi, F. M., R. Garcia, A. Lioui and B. Perron. 2010, A
long-horizon perspective on the cross-section of expected
Geometric average returns 1 month 3 months 12 months 24 months 36 months returns, working paper.
Idiosyncratic volatility: Low 14.21% 14.10% 13.34% 11.92% 12.07% Barberis, N., and M. Huang. 2007. Stocks as lotteries:
High 12.63% 15.89% 21.89% 21.63% 24.09% The implications of probability weighting for security
The table shows the results using geometric averaging of portfolio returns over the whole analysis period. The annualised returns of the high and low portfolios are prices. NBER Working Papers 12936, National Bureau of
provided for both risk measures. The period of analysis runs from July 1963 to December 2009. Economic Research.
Black, F. 1972. Capital Market Equilibrium with Restricted

3. Multivariate regression results using three risk measures Borrowing. Journal of Business 45 (3): 444–454.
Black, F., M. C. Jensen and M. Scholes. 1972. The Capital
Idiosyncratic volatility Idiosyncratic skewness Idiosyncratic kurtosis R-squared Asset Pricing Model: Some Empirical Tests. Studies in the
Slope coefficient 0.1428 –0.0065 0.0002 6.45% Theory of Capital Markets. Michael C. Jensen, ed. New
t-stat 2.00 –3.98 1.55 York: Praeger: 79–121.
Each month we run stock-level regressions (similar to a Fama-MacBeth stock regression) using the stock returns as the dependent variable and the idiosyncratic/ Blitz, D. C., and P. Van Vliet. 2007. The volatility effect:
total risks (volatility, skewness, and kurtosis) as the independent variables. The stock returns used are the average monthly returns for the next 24 months and the risk Lower risk without lower return. Journal of Portfolio
measures use the historical daily data for the previous 12 months. This table shows the average values of the coefficients of regression and R-squared values over all of
Management 34 (1): 102–13.
the cross-sections. Newey-West is used to correct the t-stats. The period of analysis runs from July 1963 to December 2009.
Boyer, B., T. Mitton, and K. Vorkink. 2010. Expected
idiosyncratic skewness. Review of Financial Studies 23 (1):
positively related to future returns. the case for a negative relationship is not only 169–202.
To assess the impact of higher-order risk contrary to common sense and theory but also Brockman, P., and M. Schutte. 2007. Is idiosyncratic vola-
measures along with volatility we performed a weak empirically given the opposing evidence tility priced? The international evidence. Working paper,
multivariate regression analysis. This assess- in empirical papers. Our findings suggest that University of Missouri-Columbia.
ment is potentially important to take into the ambiguous nature of research into the risk/ Conrad, J., R. F. Dittmar and E. Ghysels. 2008. Ex ante
account any link between volatility, skewness return trade-off may be accounted for in part by skewness and expected stock returns. Working paper.
and kurtosis. Also, using such analysis we can the horizon used in that research. Our results Fu, F. 2009. Idiosyncratic risk and the cross-section of
test the joint impact of all of these risk meas- provide evidence that, although there may well expected stock returns. Journal of Financial Economics 91
ures. We run a monthly regression by using be short-term anomalies of higher risk not (1): 24–37.
volatility, skewness and kurtosis as the inde- leading to higher expected returns, the trade-off Haugen, R. A., and A. J. Heins. 1975. Risk and the Rate
pendent variables. Figure 3 shows the average between risk and return plays out over longer of Return on Financial Assets: Some Old Wine in New
slope coefficient estimate and R-squared, over horizons much as is posited by financial theory. Bottles. Journal of Financial and Quantitative Analysis 10
all the cross sections, as well as the autocorrela- (5): 775–784.
tion-adjusted t-statistics. References Haugen, R. A., and N. L. Baker. 2008. Case Closed. The
The regression results confirm the strong Amenc, N., F. Goltz, L. Martellini and D. Sahoo. 2011. handbook of portfolio construction: Contemporary applica-
positive risk/return relationship for volatil- Is There a Risk/Return Tradeoff across Stocks? An Answer tions of Markowitz techniques, John B. Guerard Jr ed.
ity and skewness2. The effect of kurtosis is from a Long-Horizon Perspective, EDHEC-Risk Institute Forthcoming.
insignificant when used along with volatility Publication Huang, W., Q. Liu, S. G. Rhee and L. Zhang. 2010. Return
and skewness. On the whole, these multivariate Ang, A., R. J. Hodrick, Y. Xing and X. Zhang. 2006. The reversals, idiosyncratic risk, and expected returns. Review
results suggest that the greatest effects of risk cross-section of volatility and expected returns. Journal of of Financial Studies 23 (1): 147–68.
on expected stock returns stem from volatility Finance 61 (1): 259–99. Jegadeesh, N., and S. Titman. 1993. Returns to buying
and skewness. Ang, A., R. J. Hodrick, Y. Xing and X. Zhang. 2009. High winners and selling losers: Implications for stock market
Although the results of past research into idiosyncratic volatility and low returns: International efficiency. Journal of Finance 48 (1): 65–91.
the cross-sectional relationship between idio­ and further US evidence. Journal of Financial Economics Martellini, L. 2008. Toward the Design of Better Equity
syncratic/total risk and expected stock returns Elsevier 91(1): 1–23. Benchmarks: Rehabilitating the Tangency Portfolio from
are puzzling, the results are not universal and Baker, M. P., B. Bradley and J. Wurgler. 2011. Bench- Modern Portfolio Theory. Journal of Portfolio Management
the puzzle exists only as a short-term effect marks as limits to arbitrage: Understanding the low volatil- 34 (4): 34–41.
that depends on how we go about measuring ity anomaly. Financial Analysts Journal 67 (1): 1–15. Merton, R. C. 1987. A simple model of capital market equi-
volatility and its effects. Various authors have Bali, T. G., and N. Cakici. 2004. Value at Risk and expected librium with incomplete information. Journal of Finance 42
shown that the risk-return relation is posi- stock returns. Financial Analysts Journal 60 (2): 57–73. (3): 483–510.
tive, for example using value-at-risk instead of Bali, T., and N. Cakici. 2008. Idiosyncratic volatility and Mitton, T., and K. Vorkink. 2007. Equilibrium Underdi-
volatility (Bali et al, 2004) and after adjusting the cross-section of expected returns? Journal of Financial versification and the Preference for Skewness. Review of
for reversal effects (Huang et al, 2010). Overall, and Quantitative Analysis 43: 29–58. Financial Studies 20: 1255–1288.
Bandi, F. M., and B. Perron. 2008. Long-run risk-return Ross, S. 1976. The arbitrage theory of capital asset pricing.
2 For skewness, we expect a negative relationship since negatively
skewed stocks are riskier. trade-offs. Journal of Econometrics 143 (2): 349–74. Journal of Economic Theory 13 (3): 341–60.

A post-crisis perspective on
diversification for risk management
Noël Amenc, Director, EDHEC-Risk Institute cisms reveal common misconceptions not only
about the benefits and limitations of diversifica-
Felix Goltz, Head of Applied Research, EDHEC-Risk Institute tion, but also about its relationship with hedging
and insurance. In this article, our goal is to
Stoyan Stoyanov, Head of Research, EDHEC-Risk Institute – review diversification as a general method and
Asia, Singapore also hedging and insurance, sometimes incor-
rectly regarded as competing techniques.

T
he global financial crisis of 2008 has shifted minimum funding ratio or maximum drawdown Diversification: advantages and
the attention of investors to risk manage- constraint. Since the global crisis, there has been disadvantages
ment. From the standpoint of long-term a special focus on management of extreme risks Diversification is one of the most widely used
investors, the goal of risk management is to motivated by a genuine interest from industry. concepts in modern finance. Although the
maximise the probability of achieving long-term Critical reviews of some industry practices idea behind it has long existed, a scientifically
objectives while satisfying short-term constraints. blamed the negative impact of the crisis on consistent framework for diversification, MPT,
Although different institutions have different inadequate diversification largely caused by was first posited by Markowitz (1952). He
short-term constraints, a unifying feature is the shortcomings of modern portfolio theory introduced variance as a proxy for risk and
that they usually concern the downside – eg, a (MPT). Although technically justified, the criti- formulated a portfolio construction technique •

2011 SUMMER INVESTMENT&PENSIONS EUROPE

4 | EDHEC-Risk Institute Research Insights
•
by means of an optimisation problem – com- 1. A comparison of simulated average annual on asset value that can accommodate, for
bine risky assets in such a way as to minimise example, maximum drawdown and rolling
variance at each level of expected return. The return distribution and their PSP compo- performance floors. Simple insurance strategies
resulting set of portfolios describes the effi- nents with a 10% maximum drawdown floor such as CPPI and OBPI arise as dynamic opti-
cient frontier. Diversification – international mal strategies for investors subject to particular
diversification, sector and style diversification, explicit or implicit floor constraints respectively.
and so on – has since become the pillar of many PSP Dynamic strategy Dynamic fund separation in an ALM context
15 15
investment philosophies. It has also become a µ = 8.46% µ = 6.72% suggests constructing a performance-seeking
very important risk management technique, so SP = 13.24% SP = 0.16% portfolio (PSP) and a building block of safe
10 10
much so that it is often considered, erroneously, assets dedicated to hedging the stream of
as being synonymous with risk management. liabilities.1 The allocation to the two building
It is a common misconception to regard 5 5 blocks is generally not constant and depends on
diversification as a method of risk reduction, the market state. Thus, in a market downturn,
especially in the context of extreme risk man- 0 0 the closer the portfolio
c value gets to the floor
0 20 40 0 20 40
agement. In fact, according to MPT diversifica- constraint, the higher the allocation to the safe
tion is related to risk reduction as much as it building block becomes, limiting the potential
is to improving performance and, therefore, it 15 15 for further losses. See Martellini and Milhau
is most effective when it is used to extract risk µ = 12.18% µ = 8.19% (2010) for a more general discussion.
premia. A key insight of MPT is that the notion SP = 7.08% SP = 0.06% Conceptually, the functional separation
10 10
of efficient portfolios depends critically on the of the building blocks provides a key insight:
joint behaviour of asset returns. diversification, hedging, and insurance are
5 5
This insight identifies one limitation of the complementary techniques. The first two are
method – diversification is less effective when responsible for the optimal construction of
0 0
asset returns are more highly correlated and, 0 20 40 0 20 40 the PSP and the safe building blocks and the
empirically, correlations increase in times of method of insurance guarantees that the floor
market crashes. In the extreme case of 100% constraint is satisfied through the dynamic and
correlation between stocks, holding a diversi- The bottom set is obtained after a 50% improvement of the Sharpe ratio of the PSP compo- state-dependent weights of the building blocks.
fied portfolio is as efficient as holding only a few nent. The investment horizon is 10 years. Unlike diversification, however, insurance
stocks. In such an environment, if we were to always comes at a cost. The cost can materialise
rebalance the global minimum variance portfolio, In the presence of a risk-free asset, Tobin in different ways depending on implementation:
which is the lowest risk portfolio of the efficient (1958) argued that risk-averse investors should as an implicit opportunity cost, if the implemen-
frontier, it would be composed of only one stock. hold portfolios of only two funds – the risk-free tation is through dynamic trading (eg, CPPI),
Amenc, Goltz and Stoyanov (2011) emphasise asset and a fund of risky assets characterised or explicitly as the price of a derivative overlay
that the importance of joint behaviour for the by the property of having the highest Sharpe (eg, OBPI). The top two plots in figure 1 provide
effectiveness of diversification extends beyond ratio. The degree of risk aversion influences an illustration with a dynamic strategy with a
mean-variance analysis. Diversification opportu- only the relative weight of the two funds, not 10% maximum drawdown floor constraint. The
nities are in fact determined by the multi-dimen- their composition. The implication for portfolio benefit of the maximum drawdown constraint
sional dependence structure of asset returns. construction is that lower-risk portfolios are results in a much lower shortfall probability
The choice of risk measure influences how these best obtained by increasing the weight of the (SP) of getting a negative return and the implicit
opportunities are translated into actual alloca- risk-free asset rather than by re-optimising opportunity cost of insurance materialises as a
tions but if no opportunities are present, then the the fund of risky assets and, as a consequence, lower expected return of the dynamic strategy
functional form of the risk measure is irrelevant. global minimum variance portfolios are inef- compared to that of the PSP component.
As a consequence, switching to a downside risk ficient in the presence of a risk-free asset. Concerning the PSP component itself, the
measure may improve portfolio construction in This arrangement is the result of a so-called common approach is to adopt a stock market
general but will not provide better loss protection two-fund separation theorem and can be index: a cap-weighted portfolio that is concen-
in big market crashes. regarded as a simple form of hedging. Amenc et trated and highly inefficient.2 Therefore, we
A central point regarding choosing MPT al (2010a) point out that the two-fund separa- can improve the performance of the dynamic
versus alternative methods as a framework for tion implies that any risk-averse investor can strategy by improving the efficiency of the PSP
implementing diversification is the classic trade- construct portfolios in two steps: (a) build component. Amenc et al (2010b) demonstrate
off between model sophistication and estimation the maximum Sharpe ratio (MSR) portfolio that optimal diversification based on a robust
complexity. As a consequence, theoretical ben- designed to extract risk premia from risky assets methodology can consistently improve the
efits may not materialise because of difficulties and (b) depending on the degree of risk-aver- Sharpe ratio of S&P 500 by more than 50%.
with empirical estimation. In applied research, sion, partially hedge the risk present in the MSR The bottom set of plots in figure 1 is
model enhancement should balance both theo- portfolio by allocating a fraction of the capital to produced after a realistic improvement of 50%
retical and empirical concerns – see Martellini the risk-free asset. In the second step, the risks of the Sharpe ratio of the PSP component.
and Ziemann (2010) for a discussion. in the MSR portfolio are partially hedged, which The expected return of the dynamic strategy
Finally, being a method for efficient extrac- limits the downside of the return distribution. increases and almost matches the expected
tion of risk premia, diversification is designed to There is, however, a caveat. Along with the left return of the PSP component before the
work over long horizons rather than in specific tail, this technique scales down the right tail improvement, nearly compensating for the cost
market conditions. This insight alone renders as well. As a consequence, limited drawdown of insurance. This illustration underlines an
diversification useless in times of market down- comes at the cost of lower upside potential. important practical benefit of carefully imple-
turns. Furthermore, perhaps more importantly, Because of the static nature of hedging, mented optimal diversification: it can partially
it underlines the significance of a robust quanti- the allocation to the risk-free asset does not offset the implicit cost of insurance.
tative methodology for proper implementation depend on the state of the market and a specific
of diversification – see Amenc et al (2010b). drawdown limit can be achieved only ex-post. Conclusion
Misunderstanding the limitations of diver- Although in the presence of liabilities the two- In a broad context, risk management is about
sification can mislead investors into conclud- fund separation implies that the risk-free asset maximising the probability of achieving certain
ing that, since it did not protect them from is not cash but a portfolio designed to hedge the objectives at the investment horizon while stay-
big losses in 2008, diversification is a useless liabilities, the same disadvantage holds – see ing within a risk budget. Diversification, hedging
concept in general when it was never meant to see Amenc et al (2010a) for the asset-liability and insurance can be relied on to make optimal
provide loss protection in the first place. management (ALM) context). use of risk budgets. These three techniques
involve different aspects of risk management,
Hedging: fund separation and risk Insurance: dynamic risk but they are complementary techniques rather
reduction management than competing ones.
Diversification is unreliable in highly correlated Drawdown, being a dynamic characteristic, is Diversification provides investors with the
markets and it is not an efficient technique for best managed through dynamic portfolio theory. best reward per unit of risk through a smart
loss control in the short term. To limit losses Posited by Merton (1969), the theory presents combination of individual assets. It is designed to
in market downturns, investors should identify the most natural form of risk management,
1 In the presence of a stochastic opportunity set, there are also hedging
techniques that can complement it and offset generalising substantially the static portfolio demands that represent another component, see Martellini and Milhau
its shortcomings. One potential technique is selection model of MPT. The theory has been (2010).
hedging. extended with absolute or relative constraints 2 See, for example, Amenc et al (2006) for an extensive empirical study.

INVESTMENT&PENSIONS EUROPE SUMMER 2011

 EDHEC-Risk Institute Research Insights | 5

work in the long run across different market con- to control the downside of the return distribution perspective on diversification for risk management. EDHEC
ditions and is, therefore, helpless in such specific while preserving access to the upside through the Risk-Institute publication (May).
conditions as severe market downturns. Since the performance-seeking portfolio. In this context, Amenc, N., L. Martellini, F. Goltz and V. Milhau. 2010a.
purpose of diversification is efficient extraction of a well-diversified portfolio is a building block New frontiers in benchmarking and liability-driven invest-
risk premia, it is most effective in the construc- of crucial importance. A carefully designed ing. EDHEC Risk-Institute publication (September).
tion of performance-seeking portfolios. performance-seeking portfolio with an improved Markowitz, H. M. 1952. Portfolio selection. Journal of
Hedging can be combined with diversification Sharpe ratio resulting from good diversification Finance 7 (1): 77–91.
to reduce risks that cannot be diversified away. can reduce the implicit cost of insurance. Martellini, L., and V. Milhau. 2010. Hedging versus insur-
Hedging is achieved through a portfolio of safe ance: Long-term investing with short-term constraints.
assets, or simply through cash, which is another References Working paper, EDHEC-Risk Institute.
dedicated building block. A non-diversifiable Amenc, N., F. Goltz and V. Le Sourd. 2006. Assessing the Martellini, L., and V. Ziemann. 2010. Improved estimates
risk that can be handled in this way is the risk of quality of stock market indices. EDHEC-Risk Institute of higher-order comoments and implications for portfolio
a large drawdown. Publication (September). selection. Review of Financial Studies 23 (4): 1467–1502.
Insurance, unlike diversification and hedging, Amenc, N., F. Goltz, L. Martellini and P. Retkowsky. Merton, R. C. 1969. Lifetime portfolio selection under
combines the building blocks optimally to comply 2010b. Efficient indexation: An alternative to cap-weighted uncertainty: The continuous time case. Review of Economics
with the corresponding risk budgets. So downside indices. EDHEC-Risk Institute Publication (January). and Statistics 51: 247–57.
risk control is best achieved through dynamic Forthcoming in Journal of Investment Management. Tobin, J. 1958. Liquidity preference as behavior towards
asset allocation. This technique makes it possible Amenc, N., F. Goltz and S. Stoyanov. 2011. A post-crisis risk. Review of Economic Studies 25: 65–86.

Efficient indices and efficient

choices in terms of risk and reward. Since it is
not being used as an external reference, a custom
benchmark will be judged less on its transpar-
ency or its relative simplicity than on its capacity

relative return benchmarks

to enable investors to achieve their diversification
objectives, notably with regard to an external
reference represented by a cap-weighted index.
The Efficient Relative Return Benchmark
methodology enables institutional investors to
Noël Amenc, Professor of Finance, EDHEC Business School, benefit from the latest progress in the area of
diversification in order to avail themselves of
Director, EDHEC-Risk Institute and Chairman, EDHEC Risk customised benchmarks that are representa-
Indices & Benchmarks tive not only of their choice of absolute risks
(geographic, sector, style, etc) but also of their

T
relative risks by implementing a particularly
he words ‘index’ and ‘benchmark’ are with a factor model, or uses accounting attributes innovative and efficient process for managing
often used indiscriminately in practice to define the size of a company and de facto its tracking error with respect to a market index.
even though they are two a priori very position in an index, or creates a link between This relative return approach allows
different concepts: the risk and return of a stock, all of these investors to limit the risk of eventual under-
• A reference index is a portfolio that should methodological choices are more or less relevant performance when market conditions do not
represent the performance of a given segment of depending on the period chosen. That is why we allow efficient indices to outperform (which
the market, so the focus is on representativeness; have always considered that the evaluation of an is the case in speculative bubbles when diver-
• A custom benchmark is a portfolio that alternative weighting scheme for an index can sification is not useful and momentum is the
should represent the fair reward expected in only be carried out over a long period; that glob- best investment) and obviously given the fact
exchange for risk exposures that an investor is ally this evaluation could not concern the ability that, as with the vast majority of alternative
willing to accept, so the focus is on efficiency. of diversification to reduce portfolios’ risks (see forms of indices, there can be moments when
For most investors this distinction may Amenc, Goltz and Stoyanov, 2011) but instead the in-sample estimation, through significant
be semantic, but it leads clearly to different involves obtaining greater efficiency in the invest- deformation of the structures (eg, correlation),
approaches to passive investment. ment over long periods – ie, a better return for loses its out-of-sample robustness. The relative
For example, an index that is constructed each unit of risk. This serious approach to the return benchmark approach represents a choice
differently to a cap-weighted index will always performance of alternative forms of indices will of implementation of the efficiency concept that
be considered a substitute for the latter, so it probably lead investors to diversify the alterna- is more modest, and less high-performance, but
seems normal that investors would expect this tive forms of investment. As an example, it is also less risky.
new reference index to have the same level of interesting to observe that minimum volatility Ultimately, this efficient relative return
transparency, and perhaps the same level of and efficient indices do not have the same outper- benchmark offering allows investors to benefit
popularity, as the previous one. In the end, what formance in relation to cap-weighted indices in from the performance of efficient diversification
determines the success of a new reference index different market conditions. while continuing to rely on the popularity and
will be as much its financial characteristics as its A custom benchmark does not necessarily simplicity of traditional cap-weighted indices
‘popularity’, not only with investors but also with aim to replace an index because the objective for their global asset allocation and also for their
consultants. in using it relates to the implementation of a communication to all their stakeholders.
Naturally, implementation of a new form of passive investment strategy. The goal of the
reference index is not risk-free. All rebalanc- custom benchmark is not to serve as an external References
ing schemes, with the notable exception of reference for the investment but to be a genu- Amenc, N., F. Goltz, L. Martellini and P. Retkowsky.
cap-weighting or equal-weighting, assume a ine representation of the investor’s inter- or 2010. Efficient Indexation: An Alternative to Cap-Weighted
certain level of out-of-sample stability in the intra-class allocation choices. Ultimately, it is Indices. EDHEC Risk-Institute publication (January).
structures that led to the in-sample estimation not so much the ‘popularity’ of a benchmark that Amenc, N., F. Goltz, and S. Stoyanov. 2011. A post-crisis
of the parameters. Whether one tries to reduce will lead to its success but its customisation and perspective on diversification for risk management. EDHEC
the dimensions of a variance-covariance matrix appropriateness to reflect the investor’s strategic Risk-Institute publication (May).

1. Risk-adjusted performance and relative risk of relative return benchmarks and indices: US large-cap universe (500 stocks)
Return Volatility Sharpe Excess Tracking Information Worst tracking Worst relative
ratio return error ratio error (95% confidence) return (95% confidence)
Relative return efficient benchmark 3% TE 11.0% 15.4% 0.37 1.4% 2.3% 0.63 3.8% –3.0%
Efficient index 12.2% 14.8% 0.46 2.6% 4.2% 0.63 7.6% –6.1%
Equal-weighted index 11.8% 16.3% 0.39 2.2% 4.8% 0.46 9.3% –7.7%
S&P 500 cap-weighted (CRSP) 9.6% 15.5% 0.27 0.0% 0.0% 0.00 0.0% 0.)%
This table shows performance statistics computed based on weekly total returns data from January 1959 to December 2010. Worst tracking error and relative returns refer to the 5th percentile of most extreme values observed over the entire
period for a rolling one-year window when assessing this at the end of each quarter. The relative return benchmarks use a target tracking error level of 3% and aims at reliably controlling the extreme tracking error. Data for the EDHEC-Risk
Relative Return Efficient Benchmark Series is used. The efficient index is based on EDHEC-Risk’s Efficient Index long-term US data. The equal-weighted index is rebalanced quarterly using the same constituents. The cap-weighted index for the
S&P 500 universe is computed by CRSP.

2011 SUMMER INVESTMENT&PENSIONS EUROPE

6 | EDHEC-Risk Institute Research Insights

Advantages and shortcomings of

minimum variance portfolios
report that a typical minimum variance portfo-
Felix Goltz, Head of Applied Research, EDHEC-Risk Institute lio invests 47% in the utility sector while the cor-
responding weight in the market cap-weighted

S
portfolio is 9% and the corresponding weight in
cientific diversification is based on reach- poor performance. DeMiguel, Garlappi and the equally-weighted portfolio is 15%.
ing a high risk/return objective through Uppal (2009) for example evaluate a range Such concentration may correspond to the
portfolio construction techniques. It of minimum variance portfolios across seven requirements of investors who – for whatever
corresponds to the desire of investors to posi- empirical datasets, and they find that none is reason – want to bet on low volatility stocks.
tion their portfolios on the efficient frontier of consistently better than a simple equal-weight- Whether or not such a bet is promising depends
Modern Portfolio Theory. Approaches based ing rule in terms of Sharpe ratio. on the properties of such stocks. For example, it
on scientific diversification are increasingly In addition to such empirical findings, there is sometimes argued that the highest volatility
being used to construct equity core portfolios in are two main interrogations with minimum stocks come with lower returns in a short-term
institutional investment management, with the variance portfolios. From an ex-ante perspec- perspective (Ang et al, 2006), though this find-
aim of gaining an optimal risk-reward profile tive, minimum variance portfolios are not ing has been shown to lack robustness and does
from exposure to equities. optimal portfolios. They will be dominated by a not hold for long holding periods where holding
In practice, to obtain a decent proxy for combination of the risk/reward optimal portfo- high volatility stocks is actually rewarded with
efficient portfolios, one needs to use care- lio (tangency portfolio) with cash. In principle, higher returns (see Fu, 2009, and Huang et al,
ful risk and return parameter estimates. In investors should only care about designing this 2010, and the article, “Is there a risk/return
essence, practical approaches to equity portfolio tangency portfolio, using cash holdings, if they trade-off across stocks?” on page 2 of this sup-
construction based on scientific diversification wish to reduce their portfolio’s volatility to a plement). Researchers have also studied the
make different choices regarding the challenge lower level. This tangency portfolio will only general risk properties of low volatility stocks
of risk and return estimation. coincide with the minimum variance portfolios and have found that low volatility stocks – while
The minimum variance portfolio is a remark- if one is ready to assume that expected returns they have low risk in terms of volatility – may
able portfolio that provides the lowest possible of all assets are identical, clearly a strong and display high extreme risks (Boyer, Mitton and
portfolio volatility. This means that the only rather unrealistic assumption. Vorking, 2010; Chen, Hong and Stein, 2001) or
optimisation inputs required are correlations From an ex-post perspective, minimum vari- unfavourable exposures to shocks in aggregate
and volatilities. Since the estimation risk inher- ance portfolios are typically heavily concen- market volatility (Barinov, 2010).

I
ent in expected returns is well known, the fact trated in the assets with the lowest volatility.
that the minimum volatility portfolio relies only The high concentration in GMV portfolios is a rrespective of whether low-volatility stocks
on risk parameters is an appealing feature (see widely recognised issue. Clarke, De Silva and are attractive or unattractive, it is clear that
for example Amenc and Martellini, 2002, among Thorley (2011) note that their long-only mini- a minimum variance strategy leads to poorly
many others who have made this argument). mum variance “portfolio averages about 120 diversified portfolios and does not fully exploit
To be sure, estimating risk parameters is also a long securities, ie, about 12% of the 1,000-secu- correlations. In the end, for investors that have
serious challenge, with issues such as the ‘curse rity investable set”. Likewise, DeMiguel, made the decision to move away from market
of dimensionality’ when dealing with a large Garlappi, Nogales and Uppal (2009) note that cap-weighted portfolios, one can reasonably
number of assets and the ‘curse of non-station- “shortsale-constrained minimum-variance question whether replacing the concentration
arity’ of risk in the stock market. However, ever portfolios [….] tend to assign a weight differ- in the largest capitalisation stocks inherent in
since Markowitz published his theory of ‘port- ent from zero to only a few of the assets”. This cap-weighting with concentration in the lowest
folio selection’ in the 1950s, constant progress concentration can be seen in figure 1, where volatility stocks addresses their concerns.
has been made on dealing with these challenges we sort categories of stocks according to their Researchers have recognised this limitation
and today we dispose of a rich set of tools that volatility and analyse the weights allocated to of minimum volatility portfolio construction,
allows risk estimates to be improved, including each category in a minimum volatility portfolio and have proposed various ways to remedy the
sophisticated factor models as well as dynamic and in a cap-weighted portfolio. concentration of optimised portfolios in low-
risk models. In equity portfolio construction, such volatility stocks.
Despite the reasonable idea of avoiding concentration in low-volatility stocks leads to a The most straightforward solution to any
expected returns estimation, minimum variance pronounced sector bias towards utility stocks. concentration problem is to impose weight con-
portfolios have been shown to lead to relatively Chan, Karceski and Lakonishok (1999, Table 4) straints. Imposing lower and/or upper bounds
on weights provides quite rigid constraints
which leave reduced room for optimisation, but
1. Concentration of minimum volatility portfolio in low-volatility stocks can help to obtain more reasonable portfolios.
Recently, more flexible weight constraints have
60% been proposed by DeMiguel, Garlappi, Nogales,
and Uppal (2009), who use so-called ‘norm
Average weight allocated to stocks

constraints’. Such constraints put limits on the

overall amount of concentration in the portfolio
(eg, on the sum of squares of portfolio weights)
40% rather than limiting the weight of each stock in
Minimum volatility the portfolio, thus leaving more room for the
portfolio optimiser while avoiding concentration overall.
An alternative to the use of weight constraints
is to avoid making a difference between stocks
20% based on their volatilities. Christofferson et al
Cap-weighted (2010) minimise volatility with the assump-
tion that volatilities are identical across stocks.
portfolio Hence the only difference across stocks that the
optimiser then takes into account are differ-
0% ences in correlations. The minimum volatility
Low volatility 2 3 4 High volatility portfolio under this assumption will thus be
unaffected by a concentration in low volatility
Each month, stocks in the S&P 500 universe are sorted into quintiles by their past volatility (12 months). The graph shows the average weight allocated to the stocks stocks, at the cost of an implicit assumption that
belonging to each of these quintiles for two different portfolios: minimum volatility portfolio and cap-weighted portfolio. The analysis period runs from January 1959 all volatilities are equal.
until December 2008 (weekly returns).
A different way of avoiding concentration in

INVESTMENT&PENSIONS EUROPE SUMMER 2011

 EDHEC-Risk Institute Research Insights | 7

low volatility stocks is to penalise these stocks in work on portfolio construction is that, while Finance 61 (1): 259–99.
the portfolio optimisation. Amenc et al (2010) portfolio returns are a simple weighted average Boyer, B., T. Mitton and K. Vorkink. 2010. Expected
construct efficient indices and benchmarks by of component returns, the portfolio volatility idiosyncratic skewness. Review of Financial Studies 23 (1):
maximising the Sharpe ratio, rather than mini- is not just the weighted average of the stock’s 169–202.
mising volatility. The approach effectively penal- volatilities. For a given level of return, one can Chan, L. K. C., Jason Karceski and Josef Lakonishok.
ises low-risk stocks through assuming a low lower the risk by intelligently combining stocks 1999. On Portfolio Optimization: Forecasting Covariances
expected return. This penalty on the expected according to their correlations. This diversifica- and Choosing the Risk Model, Review of Financial Studies,
returns side counterbalances the attractiveness tion principle is not what drives minimum vari- 12 (5): 937–974.
of low-risk stocks from a risk perspective. To ance portfolio construction. Since minimum Chen, J., H. Hong and J. C. Stein. 2001. Forecasting
ensure parsimony and robustness, they group variance portfolios have only the objective of crashes: trading volume, past returns, and conditional
stocks by their total downside risk (in particular lowering risk, rather than aiming to optimise skewness in stock prices. Journal of Financial Economics 61
a stock’s semi-deviation, which incorporates the risk/reward ratio, minimum variance (3): 345–81.
higher moments) and distinguish stocks based portfolio optimisation leads to a pronounced Christoffersen, Peter F., Vihang Errunza, Kris Jacobs
on their riskiness only across groups rather than concentration in low-volatility stocks at the and Xisong Jin. 2010. Is the Potential for International
on a stock-by-stock basis. expense of exploiting correlation properties. Diversification Disappearing?, working paper.
What the abovementioned approaches have Such portfolios are therefore suitable for inves- Clarke, R., H. de Silva and S. Thorley. 2011. Minimum-
in common and what distinguishes them from tors who wish to load on low-risk or ‘defensive’ Variance Portfolio Composition. Journal of Portfolio
a pure minimum variance approach is that they stocks, while alternative approaches may be Management, 37 (2): 31–45.
avoid concentration in low-risk stocks and try relevant for investors who want to manage DeMiguel, Victor, Lorenzo Garlappi, Javier Nogales and
to exploit more fully the information available risk and reward properties through combining Raman Uppal. 2009. Optimal versus Naive Diversification:
on correlations in the relevant equity universe. both low-risk and high-risk stocks in a broadly How Inefficient Is the 1/N Portfolio Strategy? Review of
Using the covariance matrix solely to minimise diversified portfolio. Financial Studies 22 (5): 1915–1953.
volatility tends to result in concentrated port- DeMiguel, Victor, Lorenzo Garlappi, Javier Nogales and
folios, dominated by the low-volatility entries References Raman Uppal. 2009. A Generalized Approach to Portfolio
on the diagonal of the covariance matrix. Amenc, N., and L. Martellini. 2002. Portfolio optimiza- Optimization: Improving Performance By Constraining
Penalising the low-volatility stocks that add no tion and hedge fund style allocation decisions. Journal of Portfolio Norms. Management Science, forthcoming.
diversification benefit either directly or indi- Alternative Investments, 5 (2): 7–20. Fu, F. 2009. Idiosyncratic risk and the cross-section of
rectly through constraints and implicit assump- Amenc, N., F. Goltz, L. Martellini and P. Retkowsky. expected stock returns. Journal of Financial Economics 91
tions is meant to exploit better the covariance 2010. Efficient Indexation: An Alternative to Cap-Weighted (1): 24–37.
(off-diagonal) entries. In fact the important Indices, EDHEC-Risk Institute Publication. Huang, W., Q. Liu, S. G. Rhee and L. Zhang. 2010a. Return
insight that carries through from Markowitz’s Ang, A., R. J. Hodrick, Y. Xing and X. Zhang. 2006. The reversals, idiosyncratic risk, and expected returns. Review of
early work on efficient diversification to recent cross-section of volatility and expected returns. Journal of Financial Studies 23 (1): 147–68.

Optimal hedge fund allocation with

improved estimates for coskewness
and cokurtosis parameters
Lionel Martellini, Professor of Finance, EDHEC Business School who possesses quadratic preferences. For exam-
ple, it can be shown, through a statistical model
and Scientific Director, EDHEC-Risk Institute integrating fatter tails than those of the normal
distribution, that minimising the second order

G
moment (volatility) can be accompanied by a
iven that hedge fund returns are not tors are used that portfolio selection with significant increase in extreme risks (Sornette et
distributed in a Gaussian manner, higher-order moments is consistently superior al, 2000).
in the classic bell curve distribution to mean-variance analysis from an out-of- This finding is confirmed in Amin and Kat
around the mean, mean-variance optimisation sample perspective. (2003), where the authors present empiri-
techniques, which would be sub-optimal and cal evidence that low volatility is generally
impact negatively on the investor’s welfare, Diversification benefits obtained at the cost of lower skewness and
need to be replaced by optimisation proce- One of the principle reasons why asset owners higher kurtosis. As a result, as stressed in
dures that incorporate higher-order moments generally are willing to include hedge funds in Cremers et al (2005), in the presence of asym-
and comoments. As such, optimal portfolio their portfolios is that they expect to achieve metric and/or fat-tailed return distribution
decisions relating to hedge fund style alloca- diversification benefits with respect to other functions, the use of mean-variance analysis
tion require estimates not only for covariance existing investment possibilities. can potentially lead to a significant loss of util-
parameters, but also for coskewness and Many academics (see for example Terhaar et ity for investors.
cokurtosis parameters. This is a consider- al, 2002) have stressed that mixing hedge funds
able challenge that significantly augments the with traditional assets leads to a reduction in Extending portfolio optimisation
dimensionality issue that already exists with the volatility of the traditional portfolio. If they techniques
mean-variance analysis. wish to capitalise fully on the benefits of diver- As a consequence of the shortcomings of
In a recent research paper that is part sification in a top-down approach, investors or mean-variance optimisation, many attempts
of the Advanced Modelling for Alternative (funds of hedge funds) managers must be able have been made to account for the specific risk
Investments research chair at EDHEC-Risk to rely on robust techniques for optimising features of hedge funds in a better way and to
Institute, supported by the Prime Brokerage portfolios that include hedge funds. Standard extend portfolio optimisation techniques in
Group at Newedge, we present an application mean-variance portfolio selection techniques order to account for the presence of fat-tailed
of enhanced estimators for higher-order como- are known to suffer from a number of short- distributions, mostly by introducing some
ment parameters, introduced by Martellini and comings, and the problems are exacerbated risk objective (eg, value at risk as in Favre
Ziemann (2010), in the context of hedge fund in the presence of hedge funds. First, because and Galeano, 2002, or conditional value at
portfolio optimisation. We find that using these hedge fund returns are not normally distrib- risk as in De Souza and Gokcan, 2004, and
improved estimates leads to a considerable uted (see for example Brooks and Kat, 2002), a Agarwal and Naik, 2004), that is more general
improvement for investors in hedge funds. We mean-variance optimisation would be severely than volatility, integrating the presence of
also find that it is only when enhanced estima- ill-adapted, except in the case of an investor non-trivial higher moments in asset returns. •

2011 SUMMER INVESTMENT&PENSIONS EUROPE

8 | EDHEC-Risk Institute Research Insights
•
In the presence of non-normally distrib- ance analysis from an out-of-sample perspec- References
uted asset returns, optimal portfolio selec- tive. More specifically, we construct portfolios Agarwal, V., and N. Naik. 2004. Risks and Portfolio Deci-
tion techniques require not only estimates based on various hedge fund style indices using sions Involving Hedge Funds. Review of Financial Studies,
for variance-covariance parameters, but also a 4th order approximation of expected CARA 17 (1): 63–98.
estimates for higher-order moments and como- utility, using shrinkage estimators so as to Amin, G., and H. Kat. 2003. Stocks, bonds and hedge funds:
ments of the return distribution. However, the alleviate the concern over robustness of purely not a free lunch! Journal of Portfolio Management, 29 (4):
need to estimate coskewness and cokurtosis sample-based estimates. We find that the use 113–120.
parameters considerably exacerbates the of improved estimators leads to substantial Brooks, C., and H. Kat. 2002. The statistical properties of
dimensionality problem, which is already a increases in the investor’s utility as compared hedge fund returns and their implications for investors.
serious concern in the context of covariance to using sample estimators. On the other hand, Journal of Alternative Investments, 5, (2): 26–44.
matrix estimation. This concern is particularly we find that extending the objective function Cremers, J. H., M. Kritzman and S. Page. 2005. Optimal
acute in the hedge fund universe, where data is to encompass higher-order moments of hedge hedge fund allocations: do higher moments matter? Journal
scarce, with a short history and low frequency, fund return distribution can lead to value being of Portfolio Management, 31 (3): 70-81.
and where a number of performance biases destroyed, as opposed to added, when sample- DeSouza, C., and S. Gokcan. 2004. Allocation methodolo-
are present (see for example Fung and Hsieh, based estimators are used. In our research gies and customizing hedge fund multi-manager multi-strat-
1997, 2000, 2002). In this context, given the paper, we introduce the improved estimators egy products. Journal of Alternative Investments, 6 (4): 7–21.
dramatic increase in dimensionality involved, for hedge fund return covariance, coskewness Elton, E., and M. Gruber. 1973. Estimating the dependence
one might wonder whether portfolio selection and cokurtosis parameters and then present structure of share prices – Implications for portfolio selec-
techniques that rely on higher-order moments our empirical analysis. tion. Journal of Finance, 28 (5): 1203–1232.
can efficiently be implemented at all in realistic Favre, L., and J.-A. Galeano. 2002. Mean-modified Value-
situations. Conclusion at-Risk optimization with hedge funds. Journal of Alterna-
In a recent paper, Martellini and Ziemann This article discusses an application of the tive Investments, 5 (2): 2–21.
(2010) shed some light on this question by improved estimators for higher-order como- Fung, W., and D. A. Hsieh. 1997. Empirical characteristics
introducing improved estimators for the ment parameters, introduced by Martellini of dynamic trading strategies: the case of hedge funds.
coskewness and cokurtosis parameters. They and Ziemann (2010), in the context of hedge Review of Financial Studies, 10 (2): 275–302.
extend to the skewness and kurtosis dimen- fund portfolio optimisation. We have found Fung, W., and D. A. Hsieh. 2000. Performance character-
sions several improved estimates that had been in recent research that the use of these istics of hedge funds and commodity funds: natural versus
proposed for the covariance matrix, including enhanced estimates generates considerable spurious biases. Journal of Financial and Quantitative
most notably the factor-based approach (Sharpe, improvements for investors in hedge funds. Analysis, 35 (3): 291–307.
1963), the constant correlation approach (Elton We also find that it is only when improved Fung, W., and D. A. Hsieh. 2002. Benchmark of hedge fund
and Gruber, 1973) and the statistical shrinkage estimators are used that portfolio selection performance, information content and measurement biases.
approach (Ledoit and Wolf, 2004). In an empiri- with higher-order moments is consistently Financial Analysts Journal, 58 (1): 22–34.
cal analysis based on US large-cap stock returns, superior to mean-variance analysis from an Ledoit, O., and M. Wolf. 2003. Improved estimation of the
Martellini and Ziemann (2010) subsequently out-of-sample perspective. Our results have covariance matrix of stock returns with an application to
find that the use of these enhanced estimates important potential implications for hedge portfolio selection. Journal of Empirical Finance, 10 (5):
generates a significant improvement in inves- fund investors and fund of hedge funds man- 603–621.
tors’ welfare. agers who routinely use portfolio optimisation Martellini, L., and V. Ziemann. 2010. Improved estimates
In our new research, we complement these that incorporates higher moments without of higher-order comoments and implications for portfolio
results by providing the first application of a formal analysis of the induced increase in selection. Review of Financial Studies, 23 (4): 1467–1502.
improved estimators for higher-order como- parameter uncertainty and related lack of Sharpe, W. 1963. A simplified model for portfolio analysis.
ment parameters in the context of hedge fund robustness of the results. Management Science, 9 (2): 277–293.
portfolio optimisation. We find that the use of Sornette, D., J.V. Andersen and P. Simonetti. 2000. Portfo-
these enhanced estimates generates a significant The research from which this article was drawn lio theory for “fat tails”. International Journal of Theoretical
improvement for investors in hedge funds. We was produced as part of the research chair on and Applied Finance, 3 (3): 523–535.
also find that it is only when improved esti- Advanced Modelling for Alternative Investments Terhaar, K., R. Staub and B. Singer. 2003. An appropriate
mators are used that portfolio selection with at EDHEC-Risk Institute sponsored by the policy allocation for alternative investments. Journal of
higher-order moments dominates mean-vari- Prime Brokerage Group at Newedge. Portfolio Management, 29 (3): 101–110.

Value and momentum effects

across exchange-traded funds
Combining trading strategies with risk management
Elie Charbit, EDHEC Business School; Jean-René Giraud, CEO, Koris International and
Research Associate, EDHEC-Risk Institute; Felix Goltz, Head of Applied Research, EDHEC-
Risk Institute; and Lin Tang, Research Engineer, EDHEC-Risk Institute

I
nvestors are willing to take on risk only if returns in the future. There is ample evidence in ment medium, to apply value or momentum
they are compensated with greater expected the academic literature on these two strategies (eg, trading strategies across sectors in a dynamic
reward. There are many types of equity Jegadeesh and Timan, 1993; Graham and Dodd, core-satellite (DCS) portfolio to assess the risk-
exposure that can lead to risk premia. Value and 1934; and Fama and French, 1992). control benefits of the DCS portfolios.
momentum are among the most robust return Although exposure to value and momentum The next section describes our method and
drivers in the cross-section of expected equity effects is expected to yield attractive perfor- data choices. Section three discusses our findings.
returns. Value effects often refer to the fact that mance over the long run, in moving away from A final section summarises our conclusions.
stocks with low price/earnings ratios or high the market factor and trying to exploit value
dividend yields tend to outperform stocks with or momentum effects, investors’ portfolios Data and methodology
high price/earnings ratios or low dividend yields, tend to become more concentrated, increasing In this section, we describe both the data we use
while momentum effects usually refer to the fact drawdown risk. In this article, we use exchange- to build our portfolios and the dynamic core-
that stocks with high returns in the past yield high traded funds (ETFs), which are a liquid invest- satellite strategy.

INVESTMENT&PENSIONS EUROPE SUMMER 2011

 EDHEC-Risk Institute Research Insights | 9

All of our data is on a monthly basis and 1. Risk and return statistics est shortfall probability. This finding implies
covers the period from 31 January 1989 to 31 that the extreme losses of the value strategy
December 2009. Our investment universe is lim- Return Volatility Sharpe ratio are larger but less frequent than those of the
ited to Europe equities, in particular the STOXX STOXX 600 8.37% 16.43% 0.39 momentum strategy. The higher downside risk
Europe 600 and its sector sub-indices. Value portfolio 10.62% 20.24% 0.43 of the strategies (value and momentum) are
Momentum portfolio 10.16% 17.38% 0.47 also shown by the one-month 99% value at risk,
Methodology From 31 January 1989 to 31 December 2009 which is 15% for value and 14% for momentum
The value factor is computed from the 15 but 12.6% for the STOXX 600. As for the trailing
STOXX Europe 600 sector sub-indices we returns, the value strategy has consistently
initially select. We compute the aggregate book- Though historically there is a higher risk/ higher extreme losses in both the short term
to-market (BM) ratio of each index and then return ratio for value and momentum portfo- and the long term than the broad market index.
rank these BM ratios from highest to lowest. lios, by design, such strategies are more highly The momentum strategy, conversely, has
Every month, we go long the five sectors with concentrated, as they are built on a handful of higher short-term but lower long-term extreme
the highest BM ratios in the previous month. sectors (five, in our case). The result is greater drawdown than the broad market index. The
We can therefore create a long-only equally- exposure to downside risk (see figure 2). The Calmar ratios3 for these three portfolios are
weighted value portfolio. roughly the same, indicating that the premia
We then build a momentum portfolio. To do for bearing one additional percentage point of
so, we calculate the cumulative returns over the “On the whole, the value strategy is drawdown risk are similar for all three portfo-
previous 12 months up to two months earlier exposed to greater downside risk as a lios. On the whole, the value strategy is exposed
(as in much of the literature, observations of the to greater downside risk as a result of its more
most recent returns are discarded to prevent result of its more highly concentrated highly concentrated portfolios. The momentum
short-term reversal effects). Once we have all
the sectors’ 12-2 month cumulative returns, we
portfolios. The momentum strategy is strategy is also riskier in the short term, but,
for the sample we study here, it is less prone to
rank them from highest to lowest and go long also riskier in the short term, but, for the drawdown over the longer term.
the five sectors exhibiting the highest cumula-
tive returns in every month.
sample we study here, it is less prone to Our finding implies that though value and
momentum offer attractive returns, portfolio
We apply the dynamic core-satellite (DCS) drawdown over the longer term” construction should also take into account the
framework developed by Amenc et al (2004) downside risks they are exposed to. Now we
to build the risk-control strategy with either maximum drawdown of the value strategy is explore how the DCS approach could reduce the
the long-only value strategy or the long-only nearly two-thirds of peak portfolio wealth. The downside risk exposure.
momentum strategy as a performance-seeking maximum drawdown of the momentum strategy Figure 3 shows the cumulative returns of the
satellite portfolio 1 (see Amenc et al, 2010). The is also greater than that of the STOXX 600. The value DCS (bold blue line). To better understand
core portfolio in each case is a euro cash invest- shortfall probability is indicated as the prob- the results, we also show the core, the satellite,
ment comparable to a money market ETF. ability of experiencing a maximum drawdown the floor and the goal.
The DCS approach will combine the core that breaches the 10% limit. Although it has the It is clear that the DCS value strategy per-
and satellite portfolios such that we generate largest maximum drawdown, the value strategy, forms smoothly throughout the entire period.
a participation in the upside potential of the interestingly, has the lowest shortfall probabil- It reduces the fluctuation of the value satellite
satellite (ie, the momentum or value strategy’s ity, whereas the momentum strategy, despite its and limits downside risk. This relatively high
returns) while ensuring that the investment smaller maximum drawdown, has the high- return compared to the core portfolio suggests
value respects a floor level which in particular
limits the downside risk at a maximum level
of 10%. Our examples show that dynamic asset
allocation techniques make it possible to better 2. Summary of downside risk exposure for satellite and DCS portfolios
address investor concerns over drawdown risk.
The DCS portfolio is constructed by first Maximum Shortfall 99% VAR 3-month 12-month Calmar
specifying a maximum drawdown floor equal to drawdown probability over a month trailing return trailing return ratio
10% and a performance cap (investment goal) 1st percentile 1st percentile
set at the wealth achieved by compounding STOXX 600 54.34% 43.75% 12.57% –21.91% –42.06% 0.15
twice the cash rate over the 20-year period2. The Value portfolio 65.13% 36.25% 15.07% –25.88% –49.99% 0.16
maximum allocation to the satellite is set at 50%. Momentum portfolio 54.98% 50.00% 14.08% –24.12% –39.76% 0.18
Now we turn to the analysis of the performance From 31 January 1989 to 31 December 2009
of these portfolios

Access to value and momentum 3. Evolution of the value DCS and its parameters
premia with downside risk control
Amenc et al (2010) conclude, with a different
dataset and over a different time period, that Core Value satellite DCS value Floor Goal Satellite allocation (right axis)
the DCS can offer better returns and at the same 900% 50%
time limit downside risk. We now look into
whether the DCS could be used to gain access to 800%
value and momentum premia. Keeping down-
side risk under control is, of course, even more 40%
700%
important in strategies in which the investment
universe is reduced to concentrate the port-
folio in stocks with high exposure to value or 600%
momentum.
30%
Figure 1 summarises the performances 500%
of value, momentum and the market index
portfolio. We find higher returns and also higher
volatilities for both value and momentum 400%
portfolios compared to the market index STOXX 20%
600. Value and momentum portfolios however 300%
generated higher Sharpe ratios in our sample.
200%
1 For comparison purposes, the core is the bond ETF on the PIBOR rate 10%
and the satellite is the ETF on the STOXX Europe 600 index.
2 This cash rate used is the France PIBOR one-month interest rate from
100%
1989 to 1999 and the EONIA from 2000 to 2009. In this case, the goal is
700% of the initial wealth at the end of the 20-year horizon.
2 The Calmar ratio, which is calculated by dividing the annual return by
0% 0%
the maximum drawdown (Young, 1991), indicates the premium for bear-
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
ing one additional percentage point of drawdown risk. A higher Calmar
ratio implies better downside risk-adjusted performance.

2011 SUMMER INVESTMENT&PENSIONS EUROPE

10 | EDHEC-Risk Institute Research Insights

4. Evolution of the momentum DCS and its parameters mean that for bearing an additional percentage
of drawdown risk, DCS strategies could achieve
higher returns. In addition, DCS portfolios are
Core Value satellite DCS value Floor Goal Satellite allocation (right axis) less risky in the long term; in particular, the 1st
900% 50% percentile of 12-month trailing returns is lower.

800% Conclusion
In this article, we analyse dynamic core-
40% satellite strategies with exposure to the value
700% and momentum strategies. We find that these
investment strategies alone could achieve
600% higher returns but are exposed to high extreme
30%
risk because they consist of equity portfolios
that are concentrated in the sectors with the
500%
highest value or momentum exposure. Com-
bining these strategies with the DCS approach,
400% however, dopes portfolio returns and, at the
20% same time, keeps downside risk in check. In
300% addition, by comparing DCS approach with
the ex-post fixed-mix strategies, we find that
DCS portfolios outperform the static strate-
200%
10% gies as the DCS approach allows access to the
upside potential of the satellite portfolios while
100% fixed-mix strategies forfeit the upside by limit-
ing the majority of the allocation to the cash.
0% 0% Exchange-traded funds on sectors rather than
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 on stocks can be used to put these strategies
into effect; ETFs would also greatly facilitate
the shifts – required by dynamic strategies –
from core to satellite.
however that the value DCS may help investors Dynamic versus static
gain access to the value premium and, at the We also build the fixed-mix portfolio with This research was conducted as part of the
same time, limit the huge drawdown that may cash and the satellite portfolios (ie, value and second year of research for the Core-Satellite and
otherwise afflict the value portfolio. momentum portfolios). From a risk manage- ETF Investment research chair at EDHEC-Risk
Similarly, figure 4 shows the cumulative ment point of view, we set the constant weight Institute, in partnership with Amundi ETF
returns of the DCS momentum strategy. As of the fixed-mix strategy so that the historical
usual, the momentum DCS (bold blue line) is maximum drawdown (MDD) is equal to the References
unaffected by market downturns but profits preservation objective of the simple DCS, which Amenc, N., F. Goltz and A. Grigoriu. 2010. Risk control
from upturns. results in 10% maximum peak-to-valley draw- through dynamic core-satellite portfolios of ETFs: Applica-
Now we show the risk/return measures of down. Thus, by authorising the fixed-mix strat- tions to absolute return funds and tactical asset allocation.
the DCS portfolios in one table (figure 5). egy with value and momentum satellites to lose Journal of Alternative Investments 13 (2): 47–57.
Compared to the results shown in figure up to 10% historically, the fixed-mix portfolios Amenc, N., P. Malaise and L. Martellini. 2004. Revisit-
2, we find that DCS approaches significantly have to allocate more than 90% to cash. Figure ing core-satellite investing – A dynamic model of relative
reduce the extreme risk exposures while main- 6 summarises the comparison of the risk/ risk management. Journal of Portfolio Management 31 (1):
taining relatively high returns. The volatilities return profiles between static fixed-mix strate- 64–75.
drop to about 6% from 20%. Both DCS portfolios gies and DCS approaches. Though respecting Fama, E., and K. French. 1992. The cross-section of
respect the 10% limit on maximum drawdown the 10% MMD limit, DCS portfolios in general expected stock returns. Journal of Finance 47 (2): 427–65.
and have a one-month 99% VaR of less than have higher volatilities because the allocation Graham, B., and D. Dodd. 1934. Security analysis. New
5%. The first percentiles of trailing returns are to the satellite could go up to 50% of the entire York: McGraw Hill.
also significantly reduced. In addition, both the portfolio as long as the total wealth is above the Jegadeesh, N., and S. Titman. 1993. Returns to buying
short-term and the long-term trailing returns floor. Hence, in the short term, DCS strate- winners and selling losers: Implications for stock market
are reduced to around –7 to –6%. Higher Calmar gies are more risky. However, DCS strategies efficiency. Journal of Finance 48: 65–91.
ratios mean that, for about the same drawdown could also deliver higher returns compared to Young, T. 1991. Calmar ratio: A smoother tool. Futures
risk, portfolios achieve higher returns. the fixed-mix strategies. Higher Calmar ratios Magazine (1 October).

5. Summary of risk/return measures of DCS portfolios

Return Volatility Sharpe ratio* Maximum Shortfall 99% VAR 3-month 12-month Calmar
drawdown probability over a month trailing return trailing return ratio
1st percentile 1st percentile
DCS with value satellite 7.05% 6.41% 0.79 9.76% 0 4.69% –7.42% –6.24% 0.72
DCS with momentum satellite 7.70% 6.24% 0.91 9.25% 0 4.09% –5.97% –6.81% 0.83
* The annual risk-free rate is assumed to be 2%. From 31 January 1989 to 31 December 2009

6. Comparison between static and dynamic allocations

Return Volatility Sharpe ratio* Maximum 99% VAR 3-month 12-month Calmar
drawdown over a month trailing return trailing return ratio
1st percentile 1st percentile
DCS allocation with value satellite 7.05% 6.41% 0.79 9.76% 4.69% –7.42% –6.24% 0.72
Static allocation with value satellite 5.09% 2.47% 1.25 10.00% 1.95% –3.32% –6.46% 0.51
DCS allocation with momentum satellite 7.70% 6.41% 0.91 9.25% 4.09% –5.97% –6.81% 0.83
Static allocation with momentum satellite 5.31% 3.39% 0.98 10.00% 2.82% –4.73% –7.68% 0.53
* The annual risk-free rate is assumed to be 2%. From 31 January 1989 to 31 December 2009

INVESTMENT&PENSIONS EUROPE SUMMER 2011