New Frontiers in Benchmarking and Liability-Driven Investing - Presentation: Noël Amenc, Director, EDHEC Risk Institute, EDHEC Business School

An EDHEC-Risk Institute Publication

New Frontiers in
Benchmarking and
Liability-Driven Investing
September 2010

Institute

This publication has benefitted from research conducted as part of numerous EDHEC-Risk Institute research programmes and
chairs, notably the "Asset-Liability Management and Institutional Investment Management" research chair in partnership with
BNP Paribas Investment Partners and the "Dynamic Allocation Models and New Forms of Target-Date Funds" research chair in
partnership with UFG-LFP.

Printed in France, September 2010. Copyright© EDHEC 2010.
2 The opinions expressed in this study are those of the authors and do not necessarily reflect those of EDHEC Business School.
The authors can be contacted at research@edhec-risk.com.

New Frontiers in Benchmarking and Liability-Driven Investing - September 2010

Table of Contents

Abstract ................................................................................................................... 5

Introduction ........................................................................................................... 7

1. Asset Allocation and Portfolio Construction Decisions
in the Optimal Design of the Performance-Seeking Portfolio ....................... 11

2. Asset Allocation and Portfolio Construction Decisions
in the Optimal Design of the Liability-Hedging Portfolio ............................. 21

3. Dynamic Allocation Decisions to the Performance-Seeking
and Liability-Hedging Portfolios ....................................................................... 29

Conclusion .............................................................................................................. 43

Appendix ................................................................................................................. 45

References .............................................................................................................. 51

About EDHEC-Risk Institute ............................................................................... 57

EDHEC-Risk Institute Publications and Position Papers (2007-2010) ......... 61

An EDHEC-Risk Institute Publication 3


About the Authors

Noël Amenc is professor of finance and director of EDHEC-Risk Institute.
He has a masters in economics and a PhD in finance and has conducted active
research in the fields of quantitative equity management, portfolio performance
analysis, and active asset allocation, resulting in numerous academic and
practitioner articles and books. He is a member of the editorial board of the
Journal of Portfolio Management, associate editor of the Journal of Alternative
Investments and a member of the scientific advisory council of the AMF (French
financial regulatory authority).

Lionel Martellini is professor of finance at EDHEC Business School and
scientific director of EDHEC-Risk Institute. He has graduate degrees in
economics, statistics, and mathematics, as well as a PhD in finance from
the University of California at Berkeley. Lionel is a member of the editorial
board of the Journal of Portfolio Management and the Journal of Alternative
Investments. An expert in quantitative asset management and derivatives
valuation, Lionel has published widely in academic and practitioner journals
and has co-authored textbooks on alternative investment strategies and
fixed-income securities.

Felix Goltz is head of applied research at EDHEC-Risk Institute. He does
research in empirical finance and asset allocation, with a focus on alternative
investments and indexing strategies. His work has appeared in various
international academic and practitioner journals and handbooks. He obtained
a PhD in finance from the University of Nice Sophia-Antipolis after studying
economics and business administration at the University of Bayreuth and
EDHEC Business School.

Vincent Milhau holds master's degrees in statistics (ENSAE) and financial
mathematics (Université Paris VII), as well as a PhD in finance (Université
de Nice-Sophia Antipolis). In 2006, he joined EDHEC-Risk Institute, where
he is currently a senior research engineer. His research focus is on portfolio
selection problems and continuous-time asset-pricing models.

4 An EDHEC-Risk Institute Publication

Abstract



Abstract

Meeting the challenges of modern
investment practice involves the design
of novel forms of investment solutions, as
opposed to investment products, customised
to meet investors' long-term objectives
while respecting the short-term (regulatory
or otherwise) constraints they have to face.
We argue in this paper that such new forms
of investment solutions should rely on the
use of improved performance-seeking
and liability-hedging building-block
portfolios, as well as on the use of improved
dynamic allocation strategies. Although
each of the ingredients discussed in this
paper may already be found separately in
existing investment products, we suggest
that it is only by putting the pieces of
the puzzle together, and by combining
the underlying sources of expertise and
added value that the asset management
industry will satisfactorily address investors'
needs.


2. xxxxxxxxxxxxxxxxxx
Introduction



Introduction

Asset management is justified as an industry approaches to optimal spending of investors’
by the capacity of adding value through the risk budgets.
design of investment solutions that match
investors' needs. For more than fifty years, Academic research has provided very
the industry has in fact focused mostly useful guidance to the ways asset
on security selection as a single source allocation and portfolio construction
of added value. This focus has somewhat decisions should be analysed so as to best
distracted the industry from another key improve investors’ welfare. In brief, the
source of added value, namely, portfolio “fund separation theorems” that lie at the
construction and asset allocation decisions. core of modern portfolio theory advocate
In the face of recent crises, and given the separate management of performance
intrinsic difficulty of delivering added value and risk-control objectives. In the
through security selection decisions alone, context of asset allocation decisions with
the relevance of the old paradigm has been consumption/liability objectives, it can be
questioned with heightened intensity, and shown that the suitable expression of the
a new paradigm is starting to emerge. fund separation theorem provides rational
support for liability-driven investment
1 - More generally, there
are other forms of hedging
In a nutshell, the new paradigm recognises (LDI) techniques that have recently been
demand that allow investors that the art and science of portfolio promoted by a number of investment
to neutralise the impact
of unexpected changes in management consists of constructing banks and asset management firms. These
risk factors affecting the
opportunity set or the wealth
dedicated portfolio solutions, as opposed solutions involve, on the one hand, the
process. This is discussed in to one-size-fits-all investment products, so design of a customised liability-hedging
more detail later in this paper,
where we cover life-cycle as to reach the return objectives defined portfolio (LHP), the sole purpose of which
investment strategies.
by the investor, while respecting the is to hedge away as effectively as possible
investor's constraints expressed in terms the impact of unexpected changes in risk
of (absolute or relative) risk budgets. In factors affecting liability values (most
this broader context, asset allocation and notably interest rate and inflation risks),
portfolio construction decisions appear and, on the other hand, the design of a
as the main source of added value by performance-seeking portfolio (PSP), whose
the investment industry, with security raison d’être is to provide investors an
selection being a third-order problem. optimal risk/return trade-off.1
As argued throughout this paper, asset
allocation and portfolio construction One of the implications of this LDI
decisions are intimately related to risk paradigm is that one should distinguish
management. In the end, the quintessence two different levels of asset allocation
of investment management is essentially decisions: allocation decisions involved in
about finding optimal ways to spend risk the design of the performance-seeking or
budgets that investors are reluctantly the liability-hedging portfolio (design of
willing to set, with a focus on allowing better building blocks, or BBBs), and asset
the greatest possible access to performance allocation decisions involved in the optimal
potential while respecting such risk budgets. split between the PSP and the LHP (design
Risk diversification, risk hedging, and risk of advanced asset allocation decisions, or
insurance will be shown to be three useful AAAs). Each level of analysis involves its



Introduction

own challenges and difficulties, and while
the LDI paradigm is now widely adopted
in the institutional world, very few market
participants adopt an implementation
approach of the paradigm that is fully
consistent with the state-of-the-art of
academic research.

Our ambition in this paper is to describe the
most advanced forms of LDI strategies. Our
focus is to provide not so much a thorough
and rigorous treatment of all technical
questions related to asset allocation and
portfolio construction as a holistic overview
of the key conceptual challenges involved.
We address both questions (BBB and AAA)
in this paper. More specifically, we first
focus here on how to construct efficient
performance-seeking and liability-hedging
portfolios, and then move on to provide
information on how to allocate optimally
to these two building blocks once they have
been designed.

In the next section, we present the
challenges related to asset allocation and
portfolio construction decisions within
the PSP. We then discuss the challenges
related to asset allocation and portfolio
construction decisions within the LHP. The
last section provides an introduction to
optimal allocation to the PSP and the LHP
for a long-term investor facing short-term
constraints, once these two key building
blocks have been properly designed. A few
concluding thoughts can be found in a
final section. Further technical details are
relegated to an appendix.



Introduction


1. Asset Allocation and Portfolio
Construction Decisions in
the Optimal Design of the
Performance-Seeking Portfolio



1. Asset Allocation and Portfolio Construction
Decisions in the Optimal Design of the

Modern portfolio theory again provides In this context, it is straightforward to
some useful guidance to the optimal design show by standard arguments that the only
of a PSP that would best suit investors’ efficient portfolio composed with risky
needs. More precisely, the prescription is assets is the maximum Sharpe ratio portfolio,
that the PSP should be obtained as the also known as the tangency portfolio.
result of a portfolio optimisation procedure Appendix A.1 provides more details.
aiming at generating the highest risk-reward
ratio. Finally, the Sharpe ratio reads (where we
further let e be a vector of ones of size
Portfolio optimisation is a straightforward N):
procedure, at least in principle. In a
mean-variance setting, for example, the
prescription consists of generating a
maximum Sharpe ratio (MSR) portfolio And the optimal portfolio is given by:
based on expected return, volatility, and
pairwise correlation parameters for all assets
to be included in the portfolio, a procedure
which can even be handled analytically in
the absence of portfolio constraints.
(1)
More precisely, consider a simple
mean-variance problem: This is a two-fund separation theorem,
1 2 which gives the allocation to the MSR
max μp − γσ p performance-seeking portfolio (PSP), with
w 2
the rest invested in cash, as well as the
Here, the control variable is a vector w composition of the MSR performance-
of optimal weight allocated to various seeking portfolio.
risky assets, µp is the portfolio expected
return, and σp the portfolio volatility. We In practice, investors end up holding more
further assume that the investor is facing or less imperfect proxies for the truly
the following investment opportunity set: optimal performance-seeking portfolio, if
a riskless bond paying the risk-free rate r, only because of the presence of parameter
and a set of N risky assets with expected uncertainty, which makes it impossible to
return vector µ (of size N) and covariance obtain a perfect estimate for the maximum
matrix Σ (of size NxN), all assumed constant Sharpe ratio portfolio. If we let λ be the
so far. Sharpe ratio of the (generally inefficient)
PSP actually held by the investor, and σ
With these notations, the portfolio expected be its volatility, we obtain the following
return and volatility respectively are given optimal allocation strategy:
by:
(2)




Hence, the allocation to the performance- which is typically delegated to professional
seeking portfolio is a function of two money managers, the portfolio construction
objective parameters, the PSP volatility step. On the other hand, when the MSR
and the PSP Sharpe ratio, and one subjective proxies are obtained for each asset class,
parameter, the investor’s risk aversion. The an optimal allocation to the various asset
optimal allocation to the PSP is inversely classes is eventually generated so as to
proportional to the investor’s risk-aversion. generate the maximum Sharpe ratio at the
If risk aversion rises to infinity, the investor global portfolio level. This step is called
holds only the risk-free asset, as should the asset allocation step, and it is typically
be expected. For finite risk aversion, the handled by a centralised decision maker
allocation to the PSP is inversely proportional (e.g., a pension fund CIO) with or without
to the PSP volatility, and it is proportional the help of specialised consultants, as
to the PSP Sharpe ratio. As a result, if the opposed to being delegated to decentralised
Sharpe ratio of the PSP is increased, one asset managers. We discuss both of these
can invest more in risky assets. Hence, steps in what follows.
risk management is not only about risk
reduction; it is also about performance
enhancement through a better expenditure 1.1. Portfolio Construction Step:
of investors’ risk budgets. We revisit this Designing Efficient Benchmarks
point later in the paper. In the absence of active views, the default
option consists of using market-cap-
The expression (1) is useful because, in weighted indices as proxies for the asset
principle, it provides a straightforward class MSR portfolio. Academic research,
expression for the optimal portfolio however, has found that such indices were
starting from a set of N risky assets. In the likely to be severely inefficient portfolios
presence of a realistically large number N (Haugen and Baker 1991; Grinold 1992;
of securities, the curse of dimensionality, Amenc, Goltz, and Le Sourd 2006). In sum,
however, makes it practically impossible for market-cap-weighted indices are not good
investors to implement such direct one-step choices as investment benchmarks because
portfolio optimisation decisions involving they are poorly diversified portfolios. In
all individual components of the asset fact, capitalisation weighting tends to
mix. The standard alternative approach lead to exceedingly high concentration in
widely adopted in investment practice relatively few stocks. As a result of their
consists instead of first grouping individual lack of diversification, cap-weighted indices
securities in various asset classes by various have empirically been found to be highly
dimensions, e.g., country, sector, and/or inefficient portfolios that do not provide
style within the equity universe, or country, investors fair rewards for the risks they
maturity, and credit rating within the bond take. For the same reason, they have been
universe, and subsequently generating the found to be dominated by equally-weighted
optimal portfolio through a two-stage benchmarks (De Miguel, Garlappi, and Uppal
process. On the one hand, investable proxies 2009); equally-weighted benchmarks use
are generated for maximum Sharpe ratio a naive weighting scheme that ensures
(MSR) portfolios within each asset class in that they are well diversified, but they
the investment universe. We call this step, are optimal in the mean-variance sense



if and only if all securities have identical expected return estimates should include
expected returns and volatilities and all systematic risk measures, idiosyncratic risk
pairs of correlation are identical. measures, and downside risk measures.

In what follows, we analyse in some detail 1.1.1. Robust Estimators for
a number of alternatives based on practical Covariance Parameters
implementation of modern portfolio theory In practice, successful implementation of
that have been suggested to generate more a theoretical model relies not only on its
efficient proxies for the MSR portfolio in conceptual grounds but also on the reliability
the equity or fixed-income investment of the input to the model. In mean-variance
universes (see exhibit 1). (MV) optimisation the results will depend
greatly on the quality of the parameter
Modern portfolio theory was born with the estimates: the covariance matrix and the
efficient frontier analysis of Markowitz in expected returns of assets.
1952. Unfortunately, early applications of
the technique, based on naïve estimates of Several improved estimates for the
the input parameters, have been of little use covariance matrix have been proposed,
2 - Another key challenge
is the presence of
because they lead to unreasonable portfolio including most notably the factor-based
non-stationary risk allocations. approach (Sharpe 1963), the constant
parameters, which can
be accounted for with correlation approach (Elton and Gruber
conditional factor models
capturing time-dependencies
We explain below first how to help bridge 1973), and the statistical shrinkage
(e.g., GARCH-type models) the gap between portfolio theory and approach (Ledoit and Wolf 2004).
and state-dependencies (e.g.,
Markov regime-switching portfolio construction by showing how to In addition, Jagannathan and Ma (2003)
models) in risk parameter
estimates.
generate enhanced parameter estimates find that imposing (non-short selling)
and to improve the quality of the portfolio constraints on the weights in the
optimisation outputs (optimal portfolio optimisation programme improves the
weights). We begin by focusing on enhanced risk-adjusted out-of-sample performance
covariance parameter estimates and in a manner that is similar to some of the
explain how to meet the main challenge aforementioned improved covariance matrix
of sample risk reduction.2 Against this estimators.
backdrop, we present the state-of-the
art methods of reducing the problem In these papers, the focus was on testing
of dimensionality and estimating the the out-of-sample performance of global
covariance matrix with multi-factor models. minimum variance (GMV) portfolios, as
We then turn to expected return estimation. opposed to the MSR portfolios (also known
We argue that statistical methods are not as tangency portfolios), given that there is
likely to generate any robust expected a consensus that purely statistical estimates
return estimates, which suggests that of expected returns are not robust enough
economic models such as the capital asset to be used, a point we return to later in this
pricing model (CAPM) and the arbitrage paper when we look at expected return
pricing theory (APT) should instead estimation.
be used to estimate expected returns.
Finally, we present evidence that proxies for




Exhibit 1: Inefficiency of cap-weighted benchmarks and the quest for an efficient proxy
for the true tangency portfolio.

The key problem in covariance matrix (excess) returns, and εt the Nx1 vector
estimation is the curse of dimensionality; containing the zero mean uncorrelated
when a large number of stocks are considered, residuals εi t.The covariance matrix for the
the number of parameters to estimate grows asset returns, implied by a factor model,
exponentially, where the majority of them is given by:
are pairwise correlations. Ω = β ⋅ ΣF ⋅ β T + Σε
Therefore, at the estimation stage, the where ΣF is the KxK covariance matrix of
challenge is to reduce the number of the risk factors and Σε an NxN covariance
factors that come into play. In general, a matrix of the residuals corresponding to
multifactor model decomposes the (excess) each asset.
return (in excess of the risk-free asset) of an
asset into its expected rewards for exposure Although the factor-based estimator is
to the “true” risk factors as follows: expected to allow for a reasonable trade-
K
off between sample risk and model risk,
ri t = αi t + ∑ βi , j t ⋅F j t + εi t there is still the problem of choosing the
j =1
“right” factor model. One popular approach
or in matrix form for all N assets: attempts to rely as little as possible on
rt = αt + βt Ft + εt strong theoretical assumptions by using
principal component analysis (PCA) to
where βt is a NxK matrix containing determine the underlying risk factors
the sensitivities of each asset i to the from the data. The PCA method is based
corresponding j-th factor movements; rt is on a spectral decomposition of the sample
the vector of the N assets’ (excess) returns, covariance matrix and its goal is to use only
Ft a vector containing the K risk factors’ a few linear combinations of the original




stochastic variables—combinations that higher-order moments and co-moments of
will constitute the set of (unobservable) the return distribution. This is a formidable
factors—to explain covariance structures. challenge that greatly exacerbates the
dimensionality problem already present
Bengtsson and Holst (2002) and Fujiwara with mean-variance analysis. In a recent
et al. (2006) motivate the use of PCA in a paper, Martellini and Ziemann (2010)
similar way, extracting principal components extend the existing literature, which has
to estimate expected correlation within focused mostly on the covariance matrix,
MV portfolio optimisation. The latter find by introducing improved estimators for the
that the realised risk-return of portfolios co-skewness and co-kurtosis parameters.
based on the PCA method outperforms On the one hand, they find that the use
the portfolio based on a single index and of these enhanced estimates generates
that the optimisation gives a practically a significant improvement in investor
reasonable asset allocation. On the whole, welfare. On the other hand, they find that
the main strength of the PCA approach at when the number of constituents in the
this point is that it leads to “letting the portfolios is large (more than twenty, for
data talk” and having them tell us what the example), the increase in sample risk arising
underlying risk factors that govern most of from the need to estimate higher-order
the variability of the assets at each point in co-moments far outweighs the benefits
time are. This strongly contrasts with having of considering a more general portfolio
to rely on the assumption that a particular optimisation procedure. When portfolios
factor model is the true pricing model and with large numbers of assets are optimised,
reduces the specification risk embedded maximising the Sharpe ratio leads to better
in the factor-based approach, all while out-of-sample results than does maximising
keeping the sample risk reduction. a return-to-VaR ratio. It does so even when
portfolio performance is assessed with
The question of determining the appropriate measures that rely on VaR rather than on
number of factors to structure the volatility to adjust for risk. Similar arguments
correlation matrix is critical for the risk hold for other extreme risk measures such
estimation when PCA is used as a factor as CVaR. In the end, using extreme risk
model. Several options, some on greater measures in portfolios with large numbers
theoretical grounds than others, have been of assets leads to a formidable estimation
proposed to answer this question. problem, and empirical results suggest that
it is sensible to stay with the mean-variance
As a final note, we need to recognise approach, in which reliable input estimates
that the discussion is so far cast in a can be derived.
mean-variance setting, which can, in
principle, be rationalised only for normally 1.1.2. Robust Estimators for Expected
distributed asset returns. In the presence Returns
of non-normally distributed asset returns, Although it appears that risk parameters
optimal portfolio selection techniques can be estimated with a fair degree of
require estimates for variance-covariance accuracy, expected returns are difficult to
parameters, along with estimates for obtain with a reasonable estimation error




(Merton 1980). What makes the problem Taken together, these findings suggest
worse is that optimisation techniques are that total risk, a model-free quantity given
very sensitive to differences in expected by the sum of systematic and specific risk,
returns, so portfolio optimisers usually should be positively related to expected
allocate the largest fraction of capital to return. Most commonly, total risk is the
the asset class for which estimation error in volatility of a stock’s returns. Martellini
the expected returns is the largest (Britten- (2008) has investigated the portfolio
Jones 1999; Michaud 1998). implications of these findings and found
that tangency portfolios constructed on the
In view of the difficulty of using sample- assumption that the cross-section of excess
based expected return estimates in a expected returns could be approximated
portfolio optimisation context, a reasonable by the cross-section of volatility posted
alternative is to use some risk estimate better out-of-sample risk-adjusted
as a proxy for excess expected returns.3 performance than their market-cap-
This approach is based on the most basic weighted counterparts.
principle in finance, i.e., the natural
relationship between risk and reward. In More generally, recent research suggests
3 - This discussion focuses on
estimating the fair neutral
fact, standard asset pricing theories such that the cross-section of expected returns
reward for holding risky as the APT imply that expected returns might best be explained by risk indicators
assets. If one has access to
active views on expected should be positively related to systematic taking into account higher-order moments.
returns, one may use a
disciplined approach (e.g., the
volatility, as measured through a factor Theoretical models have shown that, in
Black-Litterman model) to model that summarises individual stock exchange for higher skewness and lower
combine the active views and
the neutral estimates. return exposure to a number of rewarded kurtosis of returns, investors are willing to
4 - For a similar conclusion
from a behavioural
risk factors. accept expected returns lower (and volatility
perspective, see Barberis and higher) than those of the mean-variance
Huang (2001).
More recently, several papers have focused benchmark (Rubinstein 1973; Krauz and
on the explanatory power of idiosyncratic Litzenberger 1976). More specifically,
rather than systematic risk for the cross- skewness and kurtosis in individual stock
section of expected returns. In particular, returns (as opposed to the skewness and
Malkiel and Xu (2006), extending an kurtosis of aggregate portfolios) have been
insight from Merton (1987), show that shown to matter in several papers. High
an inability to hold the market portfolio, skewness is associated with lower expected
whatever the cause, will force investors returns in several studies (Barberis and
to deal, to some degree, with total risk Huang 2004; Brunnermeier, Gollier, and
in addition to market risk, so firms with Parker 2007; Mitton and Vorkink 2007). The
larger firm-specific variances require intuition behind this result is that investors
higher average returns to compensate like to hold positively skewed portfolios.
investors for holding imperfectly diversified The highest skewness is achieved by
portfolios.4 That stocks with high concentrating portfolios in a small number
idiosyncratic risk earn higher returns has of stocks that themselves have positively
also been confirmed in a number of recent skewed returns. Thus investors tend to be
empirical studies. underdiversified and drive up the price of
stocks with high positive skewness, which in




turn reduces their future expected returns. that efficient equity benchmarks designed
Stocks with negative skewness are relatively on the basis of robust estimates for risk and
unattractive and thus have low prices and expected return parameters substantially
high returns. The preference for kurtosis is in outperform in terms of risk-adjusted
the sense that investors like low kurtosis and performance the market-cap-weighted
thus expected returns should be positively indices often used as default options for
related to kurtosis. Two studies provide investment benchmarks in spite of their
empirical evidence that individual stocks’ well-documented lack of efficiency (Haugen
skewness and kurtosis are indeed related to and Baker 1991; Grinold 1992).
future returns (Boyer, Mitton, and Vorkink
2010; Conrad, Dittmar, and Ghysels 2008). Exhibit 2, borrowed from Amenc et al. (2010),
An alternative to direct consideration shows summary performance statistics for
of the higher moments of returns is to an efficient index constructed in keeping
use a risk measure that aggregates the with the aforementioned principles. For
dimensions of risk. In this respect, Bali the average return, volatility, and Sharpe
and Cakici (2004) show that future returns ratio, we report differences with respect
on stocks are positively related to their to cap-weighting and assess whether this
Value-at-Risk and Estrada (2000) and Chen, difference is statistically significant.
Chen, and Chen (2009) show that there is
a relationship between downside risk and Exhibit 2 shows that the efficient weighting
expected returns. of index constituents leads to higher average
returns, lower volatility, and higher Sharpe
1.1.3. Implications for Benchmark ratios. All these differences are statistically
Portfolio Construction significant at the 10% level, whereas the
Once careful estimates for risk and return difference in Sharpe ratios is significant
parameters have been obtained, one may even at the 0.1% level. Given the data, it is
then design efficient proxies for asset class highly unlikely that the unobservable true
benchmarks with attractive risk/return performance of efficient weighting was
profiles. For example Amenc et al. (2010) find not different from that of capitalisation

Exhibit 2: Risk and return characteristics for the efficient index
Index Ann. average Ann. standard Sharpe ratio Information Tracking
return deviation (compounded) ratio error
(compounded)
Efficient index 11.63% 14.65% 0.41 0.52 4.65%
Cap-weighted 9.23% 15.20% 0.24 0.00 0.00%
Difference
2.40% -0.55% 0.17 - -
(efficient minus cap-weighted)
p-value for difference 0.14% 6.04% 0.04% - -
The table shows risk and return statistics portfolios constructed with the same set of constituents as the cap-weighted index.
Rebalancing is quarterly subject to an optimal control of portfolio turnover (by setting the reoptimisation threshold to 50%).
Portfolios are constructed by maximising the Sharpe ratio given an expected return estimate and a covariance estimate. The
expected return estimate is set to the median total risk of stocks in the same decile when sorting by total risk. An implicit factor
model for stock returns is used to estimate the covariance matrix. Weight constraints are set so that each stock's weight is between
1/2N and 2/N, where N is the number of index constituents. P-values for differences are computed using the paired t-test for the
average, the F-test for volatility, and a Jobson-Korkie test for the Sharpe ratio. The results are based on weekly return data from
01/1959 to 12/2008.




weighting. Economically, the performance a factor model approach (Ledoit and Wolf
difference is pronounced, as the Sharpe 2003; Ledoit and Wolf 2004).
ratio increases by about 70%.

1.2. Asset Allocation Step: Putting
the Efficient Benchmarks Together
After efficient benchmarks have been
designed for various asset classes, these
building blocks can be assembled in a second
step, the asset allocation step, to build a
well-designed multiclass performance-
seeking portfolio. Although the methods
we have discussed so far can, in principle,
be applied in both contexts, a number of
key differences should be emphasised.

In the asset allocation context, the number
of constituents is small, and using time-
and state-dependent covariance matrix
estimates is reasonable; nonetheless, these
estimates do not necessarily improve the
situation in portfolio construction contexts,
in which the number of constituents is
large. Similarly, although it is not, in
general, feasible to optimise the portfolio
with higher-order moments in a portfolio
construction context, in which the number
of constituents is typically large, it is
reasonable to go beyond mean-variance
analysis in an asset allocation context,
in which the number of constituents is
limited.

Furthermore, in asset allocation, the universe
is not homogeneous, which has implications
for expected returns and covariance
estimation. In terms of covariance matrix,
it will not prove easy to obtain a universal
factor model for the entire investment
universe. In this context, it is arguably better
to use statistical shrinkage towards, say, the
constant correlation model, than to take


2. Asset Allocation and Portfolio
Construction Decisions in the
Optimal Design of the
Liability-Hedging Portfolio




Risk diversification is only one possible of cash outflows to be paid, the real value
form of risk management, focusing merely of which is known today, but for which the
on achieving the best risk/return trade-off nominal value is typically matched with an
regardless of investment objectives and inflation index. It is possible, in theory, to
constraints. On the other hand, one should construct a portfolio of assets whose future
recognise that diversification is simply cash flows will be identical to this structure
not the appropriate tool when it comes to of commitments. Doing so would involve
protecting long-term liability needs. purchasing—assuming that securities of
that kind exist on the market—inflation-
One key academic insight from the linked zero-coupon bonds with a maturity
pioneering work of Robert Merton in the corresponding to the dates on which the
nineteen-seventies is that the presence of monthly pension instalments are paid out,
state variables impacting the asset return with amounts that are proportional to the
and/or wealth process will lead to the amount of real commitments.
introduction of dedicated hedging demands,
in addition to cash and optimally diversified Although this technique has the advantage
PSP (which is still needed). of simplicity and, in theory, allows perfect
risk management, it has a number of
In particular, it is clear that the risk factors limitations and implementation poses
impacting pension liability values should be several challenges. In particular, finding
hedged rather than diversified away. Two of bond portfolios with the proper duration is
these factors, interest rate risk and inflation hardly feasible, especially in the corporate
risk, stand out. Although constructing bond segment.
interest rate and inflation hedging
benchmarks might seem straightforward The conflicts of interest between issuers and
compared to constructing performance- investors about the duration of corporate
seeking benchmarks, some challenges bonds is known as the duration problem.
remain, which we discuss now. Each bond investor has in mind a specific
time horizon, and there is no reason to
expect that these needs correspond to the
2.1. Towards the Design of Improved optimal financing plan of the issuers. In
Interest Rate Risk Benchmarks fact, the duration structure of outstanding
A first approach to the design of the bonds reflects the preferences of the issuers
LHP, called cash-flow matching, involves in their aim to minimise the cost of capital.
ensuring a perfect static match between This minimisation is fundamentally opposed
the cash flows from the portfolio of assets to the interest of the investors, who usually
and the commitments in the liabilities. Let try to maximise their returns. Although
us assume, for example, that a pension fund as such a part of the suitability problem
has a commitment to pay out a monthly mentioned above, the duration mismatch in
pension to a retired person. Leaving aside the corporate bond market is of primordial
the complexity relating to the uncertain life importance to investors. Pension funds have
expectancy of the retiree, the structure of some fixed nominal liabilities originating
the liabilities is defined simply as a series from their defined-benefit plans. Given this




long-term perspective, long-term bonds liability-matching portfolios, the sole
are a much better hedge than short-term purpose of which is to hedge away as
debt. Issuers of such bonds therefore have effectively as possible the impact of
to pay only a small yield premium—even unexpected changes in the risk factors—
though they are more volatile. In contrast, most notably inflation—affecting liability
for short-term investors with no fixed time values. A variety of cash instruments
horizon in mind such investments are far (Treasury inflation protected securities, or
less attractive. The duration of the indices TIPS) as well as dedicated OTC derivatives
is nonetheless a result of the sell-side of (such as inflation swaps) are used to
corporate bonds—so no investor should hold achieve a customised exposure to consumer
just this benchmark duration. Hence, many price inflation. One outstanding problem,
corporate bond indices are not well suited however, is that such solutions generate
to serving as benchmarks for corporate very modest performance given that real
bond investors. returns on inflation-protected securities,
negatively impacted by the presence of
More worrisome perhaps is that the a significant inflation risk premium, are
characteristics of corporate bond indices usually very low. In this context, it has
can change over time. So, efforts to design been argued that some other asset classes,
stable corporate bond indices optimised such as stocks, real estate, or commodities,
in an attempt not only to maximise their could provide useful inflation protection,
risk-adjusted performance but also to especially when long-term horizons are
display a (quasi-) constant duration and considered, at a cost lower than that of
allocation by rating class over time are investing in TIPS.
required.
Empirical evidence suggests that there is
in fact a negative relationship between
2.2. Towards the Design of Improved expected stock returns and expected
Inflation-Hedging Benchmarks inflation, which is consistent with the
A recent surge in worldwide inflation has intuition that higher inflation depresses
increased the need for investors to hedge economic activity and thus depresses stock
against unexpected changes in prices. returns. On the other hand, higher future
Inflation hedging has in fact become a inflation leads to higher dividends and
concern of critical importance for private thus higher returns on stocks, so equity
investors, who consider inflation a direct investments should offer significant
threat to their purchasing power, as well as inflation protection over long horizons (as
for pension funds, which must make pension it happens, several recent empirical studies
payments often indexed to consumer prices [Boudoukh and Richardson 1993; Schotman
or wages. and Schweizer 2000] have confirmed that
equities provide a good hedge against
In this context, novel forms of institutional inflation over the long term). This property
investment solutions have been promoted is particularly appealing for long-term
by asset managers and investment banks, investors such as pension funds, which need
focusing on the design of customised to match price increases at the horizon but




not on a monthly basis. Obviously, different Commodity prices, in particular, are believed
kinds of stocks offer different inflation- to be leading indicators of inflation in that
hedging benefits, and it is in fact possible they are quick to respond to economy-wide
to select stocks or sectors on the basis of shocks to demand. Commodity prices are
their ability to hedge against inflation. usually set in highly competitive auction
For example, utilities and infrastructure markets and consequently tend to be more
typically have revenues highly correlated flexible than prices in general. In addition,
with inflation, and as a result they tend recent inflation is fuelled heavily by
to provide better-than-average inflation increases in commodity prices, in particular
protection. So it seems possible to select in agriculture, minerals, and energy. In the
stocks or sectors on the basis of their same vein, commercial and residential
ability to hedge against inflation (hedging real estate provide at least a partial hedge
demand), as opposed to selecting them as a against inflation, and portfolios that
function of their outperformance potential include real estate realise an increase in
(speculative demand). In this context, one inflation hedgeability, especially over longer
can envision selecting stocks or sectors in an horizons.
attempt to maximise the inflation-hedging
property of equity-based inflation-hedging Exhibit 3 (taken from Amenc, Martellini,
solutions. The analysis typically involves two and Ziemann [2009], a paper to which we
separate phases, selection and optimisation. refer for further details on the calibration
The goal of the selection phase is to select of the VAR and VECM models), which
the set of stocks likely to exhibit the most displays a set of estimated term structure
attractive inflation-hedging properties. In of correlation coefficients between asset
the second phase, a portfolio of selected returns and inflation-linked liability returns.
stocks will be formed in such a way as to It confirms that various asset classes have
optimise the expected inflation-hedging different inflation-hedging properties over
benefits. various horizons; inflation-hedging capacity
increases in tandem with the horizon for
Going beyond the equity universe, similar stocks, bonds, and real estate.
inflation-hedging properties are expected
for bond returns. Indeed, bond yields may As a consequence of the aforementioned
be decomposed into a real yield and an findings, it is tempting to investigate
expected inflation component. Since whether novel liability-hedging investments
expected and realised inflation move can be designed to decrease the cost to the
together over the long term, a positive investor of inflation insurance. In particular,
long-term correlation between bond returns it is possible to construct different versions
and changes in inflation is expected. In the of the inflation-hedging portfolio to assess
short-term, however, expected inflation the impact of introducing investment classes
may deviate from actual realised inflation, such as equities, commodities, and real
leading to low or negative correlations in estate in addition to inflation-linked bonds.
the short term. It has also been recently Amenc, Martellini, and Ziemann (2009)
argued that alternative forms of investment have shown that the increased expected
offer attractive inflation-hedging benefits. return generated by adding asset classes




Exhibit 3: Term structure of correlation coefficients between different asset returns and inflation-linked liability returns for various
horizons

Dotted lines correspond to implied correlations estimated from a vector auto regressive (VAR) model, whereas solid lines correspond
to implied correlations estimated from a vector error correction model (VECM).
Source: Amenc, Martellini, and Ziemann (2009).

with good long-term inflation-hedging and hence enhance the performance of
properties allows pension fund sponsors to the inflation-hedging portfolio.
maintain contributions and lower exposure
to downside risk. As a final note, we analyse in the next
section situations in which the separation
Other advanced solutions may involve of the performance-seeking and liability-
hedging a particular segment of inflation hedging portfolios does not apply in a
distribution, with an expected focus on strict sense. More specifically, we consider
hedging large, as opposed to moderate, situations in which there are multiple and
inflation shocks, in an attempt to reduce equally attractive candidates for the PSP
yet again the costs of inflation hedging and LHP.




2.3.Performance-Seeking Portfolios If an investor were given a choice of two
with Attractive Liability- /Inflation- (or more) performance-seeking portfolios
Hedging Properties with identical risk/reward ratios but distinct
As previously mentioned, asset pricing liability-hedging properties he would
theory is grounded on separation theorems obviously favour the performance portfolio
that state that risk and performance are with the most attractive liability-hedging
two conflicting objectives best managed properties.
separately. According to this paradigm,
performance generation is obtained first In fact, this hypothetical question would
through optimal exposure to rewarded not arise if the efficient frontier is strictly
risk factors to alleviate the burden on concave, which would ensure the existence
contributions, while hedging against and uniqueness of the maximum risk/
unexpected shocks that impact current reward ratio portfolio, as is the case in the
value of (assets and) liabilities is accounted standard mean-variance paradigm with
for by a separate dedicated portfolio. perfect information.
It has been shown, however, that when a
This clear separation of performance and general, not strictly convex risk measure is
hedging portfolios is very useful and has used the efficient frontier may not be strictly
a number of important implications not concave, and as a result the max reward/
only for portfolio construction and asset risk portfolio may not be unique (Stoyanov,
allocation techniques but also for the Rachev, and Fabozzi 2007). A similar result
organisational structure of the institutional would hold for a mean-variance objective
investor. in the absence of perfect information
regarding the risk/return parameters. See
exhibit 4 for an illustration. As a result, in

Exhibit 4: Multiple true or estimated tangency portfolios




most realistic situations, if given a choice
of seemingly attractive portfolio-seeking
portfolios, it would be rational for an
investor to select the performance portfolio
with the most attractive liability-hedging
properties, and this would not conflict with
the fund separation theorem. Conversely, if
an investor had a choice of liability-hedging
portfolios with equally attractive inflation-
hedging benefits, the investor would tend
to favour the hedging portfolio with the
most attractive risk/reward ratio.


3. Dynamic Allocation Decisions
to the Performance-Seeking and
Liability-Hedging Portfolios



3. Dynamic Allocation Decisions to the
Performance-Seeking and Liability-Hedging
Portfolios

Assuming that reasonable proxies for already the case in equation (2).6 The
performance-seeking and liability-hedging allocation to the “safe” building block is
portfolios have been designed using an increasing function of the beta β of
some of the aforementioned methods, liability portfolio with respect to the
we must still determine the optimal liability-hedging portfolio. If there is an
strategy to allocate assets to those two asset portfolio that perfectly matches
building blocks. Several novel paradigms, the liability value, then the beta is 1, and
which we describe next, are reshaping an infinitely risk-averse investor fully
our approaches to long-term investment allocates to the LHP. This is consistent with
decisions for long-term investors facing the intuition that for an investor facing
liability commitments and short-term liability commitments the LHP, as opposed
performance constraints. to cash, is the true risk-free asset.

Equation (2) is the solution to a
3.1. Accounting for the Presence of static optimisation problem, and the
Investors’ Liability Commitments: corresponding strategy is (by design) of
The Liability-Driven Investment the buy-and-hold kind. Equation (3) is
5 - See appendix A.2 for
details.
Paradigm the solution to a dynamic optimisation
6 - Risk aversion is not As explained earlier in this paper, investors problem, as shall be evidenced by the
observable, and not even
well-defined for institutions, with consumption/liability objectives presence of an explicit time dependency in
but it should be treated as an
implicit parameter that can
need to invest in two distinct portfolios, the expression for the optimal allocation
be inferred from a given risk in addition to cash: one performance- strategy. The corresponding strategy is a
budget, often expressed in
terms of expected shortfall seeking portfolio and one liability-hedging fixed-mix strategy, where, in principle,
with respect to liabilities.
7 - For more details, see
portfolio, construction methods for which constant trading occurs to rebalance the
Martellini and Milhau (2009). have been discussed in previous sections. portfolio allocation back to the constant
target. The fund separation theorem is
Formally, under the assumption of a expressed here under the assumption of a
constant opportunity set,5 we obtain constant opportunity set. In later sections,
the following expression of the fund we shall relax this assumption and analyse
separation theorem in the intertemporal how the allocation is impacted by the
context when trading is possible between introduction of time variation to the
current date and investment horizon: expected return and volatility of the PSP.
λ ⎛ 1⎞
wt* = PSP + ⎜ 1− ⎟ β LHP (3)
γσ ⎝ γ⎠ Exhibit 5 shows the distribution of the
funding ratio, defined as asset value
This expression is similar to that in divided by liability value, at horizon. The
equation (2), extended to the asset/ initial funding ratio is assumed to be
liability management setting. As appears 100%; the horizon is 11.32 years (taken to
from equation (3), the allocation to the be the duration of liabilities of a Dutch
“risky” building block is still an increasing defined-benefit pension fund.7 On the
function of the PSP Sharpe ratio λ and a left-hand side the Sharpe ratio of the
decreasing function of the investor’s risk PSP is assumed to be 0.24, whereas it is
aversion γ and PSP volatility σ, as was assumed to have improved by 50% to



Portfolios

0.36 on the right-hand side. The idea 3.2. Accounting for the Presence
here is to capture possible improvements of Investors’ Long-Term Objectives:
to the performance-seeking Sharpe ratio The Life-Cycle Investment Paradigm
that would result from the use of efficient Although it may be acceptable to assume
rather than cap-weighted benchmarks, as a constant opportunity set when investors
discussed in previous sections. have a short-term horizon, a long horizon,
typical of most investors’ problems, makes
The expected funding ratio has increased it necessary to go beyond Markowitz
significantly. Volatility increases as well, static portfolio selection analysis. The
but this is mostly the result of higher next important step after Markowitz
dispersion on the upside. Regarding (1952) is Merton (1969, 1971), who
downside risk, the shortfall probability takes portfolio construction techniques
(formally defined as the probability that beyond the static setting and shows how
the funding ratio ends up below 100%) to use dynamic programming to solve
decreases from 19.23% to 11.97% when dynamic portfolio optimisation problems.
the Sharpe ratio of the PSP portfolio In terms of industry implications, the
rises from 0.24 to 0.36. On the whole, development of dynamic asset pricing
the substantial improvement in the theory has led to the emergence of
distribution of the funding ratio at horizon improved investment solutions that take
is the result of two effects. On the one into account the changing nature of
hand, if the Sharpe ratio of the PSP rises, investment opportunities. These novel
the expected value of the funding ratio forms of investment are broadly referred
improves, with associated benefits from an to as life-cycle investing strategies.
ALM perspective. On the other hand, the
allocation to the PSP has also increased. Current forms of life-cycle investing
Overall, we obtain that improving the are sometimes grossly sub-optimal.
risk/reward ratio is a key component in For example, a popular asset allocation
meeting investors’ long-term objectives. strategy for managing equity risk on

Exhibit 5: Distribution of funding ratio with inefficient versus efficient PSP

The initial funding ratio is assumed to be 100%; the horizon is 11.32 years (taken to be the duration of liabilities of a Dutch defined-
benefit pension fund). On the left-hand side the Sharpe ratio of the PSP is assumed to be 0.24, whereas it is assumed to have
improved by 50% to 0.36 on the right-hand side.



Portfolios

behalf of a private investor in the context in these variables have an impact on
of a defined-contribution pension plan is portfolio risk and performance (through
known as deterministic life-cycle investing. changes in interest rates and risk premium
In the early stages, when the retirement process parameters), which should be
date is far away, the contributions managed optimally (Detemple and
are invested entirely in equities. Then, Rindisbacher 2009). In fact, one can show
beginning on a predetermined date (ten that the dynamic asset allocation problem
years, say) before retirement, the assets involves an optimal hedging of the state
are switched gradually to bonds at some variables impacting the risk-free rate and
pre-defined rate (10% a year, say). By the the risk-premium processes. Except for
date of retirement, all the assets are held a myopic investor, the optimal dynamic
in bonds.8 This is somewhat reminiscent solution does not consist of merely taking
of the rule of thumb put forward by the static solution and updating it with
Shiller (2005), advocating a percentage time-varying parameters; one should also
allocation to equity given by 100 minus introduce dedicated hedging portfolios.
the investor’s age in years.
For example, when interest rates are
8 - In fact, the rationale
behind the strategy is to
While deterministic life-cycle investing is stochastic, cash is no longer a risk-free
reduce the impact on the a simple strategy popular with investment asset; the risk-free asset is instead a bond
pension of a catastrophic
fall in the stock market just managers and consultants, and it is widely with a term to maturity matching the
before the plan member
retires and to hedge the
used by defined-contribution pension investor’s horizon. It is hardly surprising,
interest-rate risk inherent to providers, there is no evidence that it is an then, that the optimal allocation decision
the pension-related liability
value. As we will argue later, optimal strategy in a rational sense, and in the presence of interest rate risk
holding no exposure to
equity is a very trivial way
we argue below that these strategies are involves an additional building block, the
of managing equity risk, and very imperfect proxies for truly optimal bond with a term to maturity matching
one that keeps investors
from enjoying equity upside stochastic life-cycle investing strategies. the investor’s horizon, in addition to cash
potential.
In general, the presence of risk factors and the highest risk/reward performance-
can impact the opportunity set (risk and seeking portfolio (a three-fund rather
return parameters), and they can have a than two-fund separation theorem). As
direct impact on the wealth process for another illustration, let us assume that
non-self-financed portfolios when inflows the expected return on most risky assets
and outflows of cash are taking place. is, for example, negatively impacted by
increases in oil prices. To compensate
3.2.1. Accounting for Risk Factors for the deterioration of the investment
Impacting the Investment Opportunity opportunity set in the event of a sharp
Set increase in oil prices, the investor will
A large body of empirical research has benefit from holding a long position in a
shown that interest and inflation rates, portfolio optimised to exhibit the highest
as well as expected return, volatility, and possible correlation with oil prices.
correlation parameters are stochastically
time-varying, as a function of key state Kim and Omberg (1996), for example, have
variables that describe the state of the analysed a model including a stochastic
business cycle. Unexpected changes equity risk premium with a mean-



Portfolios

reverting component. In this model, one than when it is constant (σλ=0);
can show that the optimal allocation • The hedging demand disappears if there 
involves not only a deterministic decrease is no equity risk premium risk (σλ=0), or if
of the allocation to equity as the investor the risk exists but cannot be hedged away
gets closer to the time horizon, which (ρλS=0).
is consistent with standard target • The investment in stock decreases when 
date fund practice, but also a state- approaching horizon T; this is consistent
dependent component, suggesting with the prescriptions of target date
that the allocation to equity should be funds.
increased (respectively, decreased) when,
as measured by such proxies as dividend We also confirm that there is one
yields or price-earnings ratios, equity has additional state-dependent factor: if/
become cheap and decreased when it has when equity prices are low (high), and
become expensive. therefore expected return is high (low),
one should allocate more (less) to stocks,
We obtain the following expression for the regardless of horizon.
optimal allocation strategy, assuming for
simplicity that an equity benchmark is the 3.2.2. Accounting for Risk Factors
only risky asset so that the PSP is 100% Impacting the Wealth Process
invested in that benchmark (see appendix As noted earlier, the presence of risk
A.3 for details): factors can also have a direct impact on
the wealth process, even in the case of
a constant opportunity set. In fact, most
portfolios are not self-financed portfolios,
because of the presence of outflows of
cash (consumption, liability payments)
Here we let ρλS be the correlation of and/or inflows of cash (endowment,
the Sharpe ratio of the equity index, contribution, income, and so on). For
denoted by λS, and the equity index example, labour income risk will have
return; a negative correlation means that an impact on optimal portfolio decisions
high realised return periods tend to be and will legitimise the introduction of
followed by low expected return periods, a dedicated hedging demand. More
which is supported by empirical evidence. generally, the present value of liability
Moreover, σλ is the volatility of the equity and endowment flows are impacted by
Sharpe ratio process and σS is the volatility state variables (interest rate risk, inflation
on the stock index. risk, income risk, etc.), the impact of which
must be hedged away.
The hedging demand against
unexpected changes in the PSP Sharpe A pension fund, for example, should hold
ratio has the following properties (for ρλ S<0 a long position in a liability-hedging
and γ>1): portfolio to hedge away the implicit short
• The investor with γ>1 holds more stocks position in liability flows. Conversely, a
when equity Sharpe ratio is mean-reverting sovereign wealth fund from an oil-rich


New Frontiers in Benchmarking and Liability-Driven Investing - Presentation: Noël Amenc, Director, EDHEC Risk Institute, EDHEC Business School

New Frontiers in Benchmarking and Liability-Driven Investing - Presentation: Noël Amenc, Director, EDHEC Risk Institute, EDHEC Business School

Empfohlen

Empfohlen

Weitere ähnliche Inhalte

Ähnlich wie New Frontiers in Benchmarking and Liability-Driven Investing - Presentation: Noël Amenc, Director, EDHEC Risk Institute, EDHEC Business School

Ähnlich wie New Frontiers in Benchmarking and Liability-Driven Investing - Presentation: Noël Amenc, Director, EDHEC Risk Institute, EDHEC Business School (20)

Mehr von Investments Network marcus evans

Mehr von Investments Network marcus evans (20)

Kürzlich hochgeladen

Kürzlich hochgeladen (20)

New Frontiers in Benchmarking and Liability-Driven Investing - Presentation: Noël Amenc, Director, EDHEC Risk Institute, EDHEC Business School