This article appeared in a journal published by Elsevier.

The attached
copy is furnished to the author for internal non-commercial research
and education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/authorsrights
 Author's personal copy

Economic Modelling 40 (2014) 369–381

Contents lists available at ScienceDirect

Economic Modelling
journal homepage: www.elsevier.com/locate/ecmod

Optimal and investable portfolios: An empirical analysis with scenario

optimization algorithms under crisis market prospects
Mazin A.M. Al Janabi
P.O. Box 15551, College of Business & Economics, Department of Economics and Finance, United Arab Emirates University (UAEU), Maqam Male Campus, Building H3, Office 1061,
Al-Ain, United Arab Emirates

a r t i c l e i n f o a b s t r a c t

Available online 8 December 2013 This paper develops scenario optimization algorithms for the assessment of investable financial portfolios
under crisis market outlooks. To this end, this research study examines from portfolio managers' standpoint
JEL classification: the performance of optimum and investable portfolios subject to applying meaningful financial and opera-
C10 tional constraints as a result of a financial turmoil. Specifically, the paper tests a number of alternative sce-
C13 narios considering both long-only and long and short-sales positions subject to minimizing the Liquidity-
G20
Adjusted Value-at-Risk (LVaR) and various financial and operational constraints such as target expected
G28
return, portfolio trading volume, close-out periods and portfolio weights. Robust optimization algorithms
Keywords: to set coherent asset allocations for investment management industries in emerging markets and particu-
Emerging markets larly in Gulf Cooperation Council (GCC) financial markets are developed. The results show that the obtained
Financial engineering investable portfolios lie off the efficient frontier, but that long-only portfolios appear to lie much closer to
Financial risk management the frontier than portfolios including both long and short-sales positions. The proposed optimization algo-
GCC financial markets rithms can be useful in developing enterprise-wide portfolio management models in light of the aftermaths
Liquidity-Adjusted Value-at-Risk of the most-recent financial crisis. The developed methodology and risk optimization algorithms can aid in
Optimization
advancing portfolio management practices in emerging markets and predominantly in the wake of the lat-
Portfolio management
Stress testing
est credit crunch.
© 2013 Elsevier B.V. All rights reserved.

1. Introduction While common risk technique such as Value-at-Risk (VaR) and

As a result of the recent financial market shocks, capital-market
corporations, institutional investors and portfolio managers are
reconsidering specific issues and focusing on: 1) How to not only
incorporate risk and reward tradeoffs using modern portfolio theory,
but also plan for unexpected market shocks; and 2) the resulting
effects of these shocks on the asset management business and its
impact on asset allocation and the construction of robust investable
portfolios. To this end, prominent financial institutions are linking
their downside portfolio risk with the return on capital and integrat-
ing market liquidity risk into their assessments in an effort to obtain
better understanding of embedded-risk and expected return. As a
result, optimization of the capital deployed–rather than just a single
view of risk exposures–and its application for optimizing the
asset allocation structures of investable market portfolios has be-
come the new role of risk management.1
E-mail addresses: mazinaljanabi@gmail.com, m.aljanabi@uaeu.ac.ae.
1
In this paper the concept of investable market portfolios refers to rational portfolios that
are contingent on meaningful financial and operational constraints. In this sense, investable
market portfolios are not located on the efficient frontiers as defined by Markowitz (1952),
and instead have logical and well-structured long/short asset allocation proportions.
probability of default are still employed, they fail to anticipate systemic
changes in the structure of financial markets. These techniques
assume that volatility of the market and correlations among as-
sets change slowly or not at all; they are not designed to handle
systemic negative changes caused by jumps in the availability of
liquidity or jumps in market values (Scholes and Kimner, 2010).
One other critique that can be leveled against the VaR method
is that it does not explicitly consider portfolios' asset liquidity
risk during the unwinding (close-out) period. In fact, typical VaR
models assess the worst change in the mark-to-market portfolio
value over a given time horizon but do not account for the actual
risk of liquidation. Indeed, neglecting asset liquidity risk can lead to
an underestimation of the overall market risk and misapplication of
capital cushion for the safety and soundness of financial institutions
(Al Janabi, 2011a,b).
In this backdrop and to address the above shortcomings, the goals and
challenges in this paper are to develop robust scenario optimization-
algorithms for the assessment of investable financial portfolios under cri-
sis market prospects. To this end, this paper examines from portfolio
managers' perspective the performance of investable structured portfoli-
os within a Liquidity-Adjusted Value-at-Risk (LVaR) framework, subject
to the application of meaningful operational and financial constraints,

0264-9993/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.econmod.2013.11.021
 Author's personal copy
370 M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381
particularly in the wake of the aftermaths of most-recent global financial
crisis.2
The rationality behind introducing LVaR as an effective portfolio
management tool is because it complies with real-life trading situations,
where traders can liquidate (or re-balance) small portions of their port-
folios on a daily basis according to prevailing market liquidity condi-
tions. To this end, an LVaR approach is introduced to allocate financial
assets by minimizing LVaR subject to enforcing meaningful operational
and financial constraints that are based on fundamental asset manage-
ment considerations and practices, such as: a) the target expected
return of the investable portfolio; b) total trading volume of the invest-
able portfolio; c) monetary asset allocation of each asset class; d) port-
folio managers' choices of pure long positions or a combination of
long/short trading positions; and e) the unwinding or close-out liquid-
ity horizons of each asset-class.
In a nutshell, the primary motivation of this research is to set
advanced portfolio management optimization techniques (drawn
from rational and meaningful financial investment considerations)
that can be applied to investable portfolios in emerging markets, such
as in the context of the Gulf Cooperation Council (GCC) stock markets.
As such, this research study and the obtained empirical results can
contribute to the existing body of knowledge and extend current
optimization-techniques' literatures related to the assessment of
investable financial portfolios. Specifically this paper provides general-
ized scenario optimization-algorithm foundation that is theoretically
appealing while capturing the essential aspects of optimal and invest-
able financial assets and risk-capital allocations under difficult and unfa-
vorable market circumstances. Essentially, the proposed scenario
optimization-algorithms can be useful in developing enterprise-wide
portfolio management models that financial entities may consider in
assessing coherent risk-capital allocations and can offer practical tools
to portfolio managers. As such, the portfolio modeling techniques and
the achieved empirical results can have many uses and applications
for portfolio managers and can have relevant practical implications
that will benefit several end-users, such as: institutional investors, port-
folio managers, mutual-fund industry and other financial institutions in
the GCC region as well as other emerging financial markets.
The remainder of the paper is organized as follows. The following
section discusses relevant literatures reviews and highlights specific ob-
jectives of this paper. This is followed by Section 3 in which the quanti-
tative infrastructure of a non-linear dynamic risk-function and robust
scenario optimization-algorithms are described. In Section 4 we analyze
and interpret empirical results and discuss the simulation results of
optimal and investable portfolios. Summary and concluding remarks
are drawn in the final section. Full set of empirical testing and simula-
tion results of optimal and investable portfolios are included in
Appendix A.
2. Literature review and objectives
Portfolio management has different meanings in different contexts
and the measurement of risk/return within a portfolio optimization
2
Indeed, there are other relevant studies that have tackled the issues of liquidity and as-
set pricing but not necessarily within the context of trading portfolios. For the sake of brev-
ity, we provide a succinct narrative of some of the proposed models, detailed as follow:
Within the VaR framework, Jarrow and Subramanian (1997) provide a market impact
model of liquidity by considering the optimal liquidation of an investment portfolio over
a fixed horizon. Bangia et al. (1999) approach the liquidity risk from another angle and
provide a model of VaR adjusted for what they call exogenous liquidity—defined as com-
mon to all market players and unaffected by the actions of any one participant. It com-
prises such execution costs as order processing costs and adverse selection costs
resulting in a given bid-ask spread faced by investors in the market. In a different vein,
Almgren and Chriss (1999) present a concrete framework for deriving the optimal execu-
tion strategy using a mean–variance approach, and show a specific calculation method. For
other relevant literature on liquidity, asset pricing and portfolio choice and diversification
one can refer as well to Takahashi and Alexander (2002); Amihud et al. (2005); Cochrane
(2005) and Meucci (2009), among others.
setting is probably of more importance today than it has ever been dur-
ing modern financial history. In recent years, the growth of trading ac-
tivities and frequent occurrences of financial market upheavals has
highlighted the necessity for market participants to develop reliable as-
sessment methods and algorithms for portfolio management. Histori-
cally, Markowitz's (1952) classical mean–variance analysis was
groundbreaking as it provided the framework within which optimal
portfolio allocations are still examined today. Probably the most com-
monly associated measure of risk within this context, and one that has
been in operation for the longest time, is the standard deviation. Over
time a number of extensions have been suggested. For example, a
natural replacement is a GARCH (Bollerslev, 1986; Engle, 1982)
time-varying variance measure that allows for volatility clustering
which leads to the regularly observed leptokurtic nature of return
distributions.
In fact, one of the basic problems of applied finance is the optimal
selection of assets, with the aim of maximizing future returns and
constraining risk by appropriate measures. Undeniably, the portfolio
mean–variance analysis approach, pioneered by Markowitz (1952), is
one of the cornerstones of modern portfolio management and has
served as the standard procedure for constructing portfolios. Albeit
Markowitz's mean–variance portfolio optimization methodology is a
landmark in the development of modern investment theory, there
are no risk measures universally adopted in financial applications
(Al Janabi, 2013). In his classical mean–variance analysis Markowitz
(1952) described the theoretical framework for modern portfolio
theory and the creation of efficient portfolios under the notion of max-
imizing expected return subject to some risk constraints. In this frame-
work, risk is defined in terms of the standard deviation of each asset,
which implies that the probability of negative returns, as the probability
of positive returns, is weighted in the same way by the portfolio manag-
er. As a result, the solution to the Markowitz theoretical models revolves
around the portfolio weights, or the percentage of asset allocation that
can be invested in each security (Al Janabi, 2013; Markowitz, 1952).
To this end, Markowitz (1952) demonstrated that, for a given levels
of risk, one can recognize certain groups of assets that maximize expect-
ed return. Accordingly, for asset-allocation purposes, portfolio man-
agers should choose portfolios located along the efficient frontier.3 As
such, Markowitz (1952) considered these optimum portfolios as ‘effi-
cient’ and referred to a continuum of such portfolios in dimensions of
expected return and standard deviation as the efficient frontier
(Al Janabi, 2013; Markowitz, 1952).
Nonetheless, optimized portfolios do not normally perform as well in
practice as one would expect from theory. For example, they are often
outperformed by simple allocation strategies such as the equally weight-
ed portfolio (Jobson and Korkie, 1981) or the global minimum variance
portfolio (Jorion, 1991). Portfolio weights are often not stable over time
but change significantly each time the portfolio is re-optimized, leading
to unnecessary turnover and increased transaction costs. Moreover,
these portfolios typically present extreme holdings (“corner solutions”)
in a few securities while other securities have close to zero weight
(Al Janabi, 2013; Jobson and Korkie, 1981; Jorion, 1991).
It is well documented (Michaud, 1989) that mean–variance opti-
mizers, if left to their own devices, can sometimes lead to unintuitive
portfolios with extreme positions in asset classes. In a portfolio optimiza-
tion context, assets with large expected returns and low standard
3
Indeed, institutional investors manage their strategic asset mix over time to achieve fa-
vorable returns subject to various uncertainties, financial, operational and regulatory con-
straints, and other requirements. It has been demonstrated in a number of studies (Blake
et al., 1999) that the mix of various classes of assets is a critical factor affecting the perfor-
mance of institutional investors' diversified funds. While the asset-allocation decision is
clearly important for multiple sectors portfolios, the literature is sparse in terms of under-
standing the process by which active investment managers allocate assets across the spec-
trum of securities and of analyzing their ability to fine-tune the portfolio's asset-allocation
from a fund's strategic benchmark position in an attempt to capture active returns.
 Author's personal copy
M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381
deviations will be overweighted and conversely, assets with low expect-
ed returns and high standard deviations will be underweighted. There-
fore, large estimation errors in expected returns and/or variances/
covariances will introduce errors in the optimized portfolio weights
(Fabozzi et al., 2006). As a result, these “optimized” portfolios are not
necessarily well diversified and exposed to unnecessary ex-post risk
(Michaud, 1989). The reason for these phenomena is not a sign that
mean–variance optimization does not work but rather that the modern
portfolio theory framework is very sensitive to small changes in the
inputs. Simply put, the “optimized” portfolio, in many instances, is not
optimal at all (Al Janabi, 2013; Michaud, 1989).
More recently, focus has shifted from measuring general volatility to
tail risk with the popularity of Value-at-Risk (VaR). This simple and
intuitive measure is widely used in practice, but fails the basic mathemat-
ical requirements of rationality (Kaplanski and Kroll, 2002) and coherence
(Artzner et al., 1999), implying that the VaR of a portfolio can be greater
than that of the individual assets. To this end, Conditional-VaR (CVaR),
which measures the average loss conditional on returns dropping below
the VaR level, is widely documented to be superior to VaR measure. It in-
corporates more information (Agarwal and Naik, 2004), satisfies all the
desirable properties for a coherent risk measure (Acerbi and Tasche,
2002; Rockafellar and Uryasev, 2002) and is much easier to use in optimi-
zation techniques (Rockafellar and Uryasev, 2000, 2002). Other studies
examined portfolio optimization based on variance, VaR and CVaR and
under normality assumption (see for instance, Artzner et al., 1999;
Rockafellar and Uryasev, 2002; Ho et al., 2008), and found that the
resulting risk exposures are equivalent under the three different optimi-
zation procedures. It is, however, well known that asset returns exhibit
negative skewness and positive excess kurtosis [evidence was first pro-
vided on this by Mandelbrot (1963) but is still a topic of investigation;
see Poon and Granger (2005) for an excellent review], with the three
optimization procedures generating very different results under these
conditions (Agarwal and Naik, 2004; Gaivoronski and Pflug, 2005). Like-
wise, Ho et al. (2008) show that CVaR is superior in terms of percentage
reduction in downside risk, with this effect more prevalent over their se-
lected ‘credit crunch’ period of analysis. In a similar vein, recently Cain and
Zurbruegg (2010) propose a technique that involves switching between
risk measures in different market environments, to capture the well-
documented dynamic nature of risk within a portfolio optimization set-
ting. In-sample results show categorically that switching between various
measures, such as CVaR, time-varying (GARCH) variances and simple
standard deviations, can lead to a better performance than using any sin-
gle measure. This result, along with the mixed evidence for a single opti-
mal measure, suggests that there may be specific times over which these
different risk measures are optimal.
Given the fact that literatures on modern portfolio management
have been relatively inconclusive and providing mixed results, the aim
of this paper is to examine realistic optimization-algorithms that can al-
locate assets under crisis market circumstances and produce investable
portfolios (subject to applying meaningful operational and financial
constraints) predominantly in light of the aftershocks of the most-
recent global financial crisis. Indeed, for more than five decades a
wide body of knowledge has been accumulated about the performance,
strengths, and weaknesses of the classical mean–variance approach
when applied to equity portfolios (Al Janabi, 2013; Jobson and Korkie,
1981; Jorion, 1991; Markowitz, 1952). However, much less is known
about robust portfolio optimization techniques in emerging equity
markets and particularly under austere and crisis market conditions.
Likewise, and given the fact that mean–variance optimizers have funda-
mental financial shortcomings (which could often lead to financially
worthless ‘optimal’ portfolios) and since asset-allocation decision is
the most fundamental issue facing portfolio managers who invest
across multiple asset classes, this research paper examines:
1) How to structure investable market portfolios and to not only inte-
grate risk and reward tradeoffs using portfolio theory, but also
371
allocate assets under the assumption of different asset unwinding
(or close-out) horizons and implementing different trading scenari-
os and strategies (of pure long positions and/or combinations of
long/short-sales trading positions).
2) How to plan for market shocks and financial crisis and how to assess
the resulting impacts of these shocks on the performance of asset
management firms in determining the optimal risk-capital cushion
for portfolio managers.
Set against this background, in this work we develop a model for
optimizing portfolio risk-return with LVaR constraints using realistic
operational and financial scenarios and conduct a simulation study on
optimizing equity portfolios of the six GCC stock markets. The simula-
tion study shows that the optimization algorithm, which is based on
quadratic programming techniques, is very stable and efficient in han-
dling different liquidity horizons (close-out periods), overall portfolio
trading/investment volume, the different choices of long/short-sales
positions, and disparate correlation factors during market stress
periods. Moreover, the approach can tackle large number of equity
securities and rational asset management scenarios as well as other
operational constraints. Indeed, the LVaR's risk management constraints
(reduced to linear constraints) can be used in various portfolio manage-
ment optimization applications to bound percentiles of loss distributions.
As such, in this research paper the aim is to look at scenario optimi-
zation problem from a different realistic operational angle. In view of
that, the optimization-algorithms are formulated by finding the set of
structured investible portfolios that maximize risk-adjusted-returns,
with risk-tolerance, and regulatory restrictions are constrained accord-
ing to the requirements of the institutional investors. Moreover, a
unique feature of this study that warrants particular emphasis is the
superiority of the robust risk-adjusted-returns benchmark for “non-
normal” returns that will be employed. As such, the focus in this work
is on the forecast of risk-adjusted-return measure, rather than on ex-
pected returns for two reasons: first, several studies have analyzed the
forecasts of expected returns in the context of mean–variance optimiza-
tion (see for instance, Best and Grauer, 1991). The common opinion is
that expected returns are not easy to forecast, and that the optimization
process is very sensitive to these variations. Second, there exists a gen-
eral notion that risk-adjusted-returns, in a wide sense, are simpler to
assess than expected returns from historical data (Al Janabi, 2013;
Best and Grauer, 1991). The principal impetus of this paper is to provide
unique coverage of investable portfolios along with tactical asset-
allocations in GCC stock market setting, where little is known about
how dynamic asset-allocations can be implemented.
3. Research design and methodological approach
Set against this background and to address the above shortcomings
in the classical mean–variance approach and to provide proper answers
to the research questions as posed above, the proposed risk-engine and
robust scenario optimization-algorithms proceed as follows.
3.1. Methodological Step (1)
In this first step we attempt to define a non-linear dynamic risk-
function (i.e. objective function) of multivariable. For this purpose, the
non-linear dynamic risk-function can be defined as a vector of: 1) Mon-
etary investment in each asset-class; 2) close-out periods or unwinding
periods of each asset-class and the overall close-out periods of invest-
able portfolios; 3) overall trading volume of investible portfolios;
4) constrained asset-allocation proportions according to contemporary fi-
nancial market regulations and subject to the imposition of rational and
meaningful operational and financial boundaries; 5) downside risk con-
straints so that additional risk resulting from any non-normality and illiq-
uid assets may be used to estimate the characteristics of investable
portfolio. This enables a much more generalized framework to be

372
developed, with the distributional assumption most appropriate to the
type of financial assets to be employed, and which can be of crucial
value for more accurate market risk assessment during market stress pe-
riods and particularly when liquidity dries-up; 6) correlation coefficients
among all asset classes; 7) expected returns of investable portfolios;
8) confidence level of estimated parameters under different scenarios
and market settings; and 9) portfolio managers' choices of pure long
asset positions or a combination of long/short-sales trading asset
positions.
In this backdrop, the objective risk-function is defined and explained
along these lines:Portfolio trading risk in the presence of multiple risk
factors is determined by the combined effect of individual risks. The
magnitude of total risk is determined not only by the magnitudes of
individual risk factors but also by their correlations. In fact, portfolio
effects are crucial in risk management not only for large diversified
portfolios but also for individual instruments that depend on several
risk factors (Al Janabi, 2013). For multiple assets, the LVaR of a portfolio
of financial assets is a function of the individual risk of each asset, hold-
ing period (close-out or liquidation horizon) and the correlation factors
[ρi,j] between the returns on the individual securities, detailed as
follows4:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uX qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u n X n
LVaRP ¼ t LVaRi LVaR j ρi; j ¼ jLVaRjT jρjjLVaRj:
i¼1 j¼1
In fact, Eq. (1) is a generalized formula for the calculation of LVaRP for
any portfolio regardless of the number of securities. It should be noted
that the second term of this formula is presented in terms of matrix-
algebra techniques—a useful form to avoid mathematical complexity,
as more and more securities are added. This approach can simplify the
algorithmic programming process and can permit a straightforward in-
corporation of short-sales positions in the market risk assessment
process.
Indeed, Eq. (1) considers adverse price impact liquidity risk through-
out the close-out period. As a result, the assumption of a given close-out
horizon for orderly liquidation inevitably implies that assets' liquidation
occurs during the holding period. Accordingly, scaling the holding peri-
od to account for orderly liquidation can be justified if one allows the
assets to be liquidated throughout the close-out period. In order to per-
form the calculation of LVaR under accurate illiquid market circum-
stances, we can define the following throughout the close-out period:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð2t i þ 1Þðt i þ 1Þ
LVaRi ¼ VaRi
6t i
where VaRi = Value at Risk of asset i under liquid market conditions;
LVaRi = Value at Risk of asset i under illiquid market conditions;
and ti = number of trading days required for orderly liquidation of
asset i. Moreover, the latter equation indicates that LVaR i N VaRi ,
however when the required number of days to liquidate the entire
asset is one trading day, then we have a special case in which
LVaRi = VaRi. Consequently, the difference between LVaRi − VaRi
should equal to the residual market risk due to the illiquidity of asseti
under adverse market conditions. Furthermore, it is important to note
that Eq. (2) can be used for the calculation of LVaR for any time horizon
subject to applying a constraint on the total LVaR figure. In fact, the
overall LVaR figure should not exceed at any setting the nominal expo-
sure, or in other words the total trading volume of the portfolio.5
4
The mathematical and optimization algorithms are largely drawn from Al Janabi
(2012, 2013) research papers.
5
For further details on the mathematical derivation and rational usefulness of Eq. (2)
we refer the readers to Al Janabi (2012, 2013) papers.
Author's personal copy
M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381
As a matter of fact, the number of liquidation days or close-out time
(ti) necessary to liquidate the entire position of asset i fully is related to
the choice of the liquidity threshold; however the size of this threshold
is likely to change under severe market conditions. Effectively, a linear liq-
uidation procedure of assets is assumed in Eq. (2), i.e. selling equal parts of
each asset every day till the last trading day (ti), where the entire asset is
sold. The above model is more appropriate for daily trading circumstances
where traders can unwind part of their positions on a daily basis.
Indeed, the choice of liquidation horizon can be estimated from the
total trading position size and daily trading volume that can be un-
wound into the market without significantly disrupting market prices;
and in actual practices it is generally estimated as:
t i ¼ total tradingposition size of asseti =daily tradingvolume of asseti : ð3Þ
As such, the close out time (ti) is the time required to bring the posi-
tions to a state where the financial entity can make no further loss from
the trading positions. It is the time taken to either sell the long positions
or alternatively the time required to buy securities in case of short posi-
tions. In real practices the daily trading volume of any trading asset is
estimated as the average volume over some period of time, generally a
month of trading activities. In effect, the daily trading volume of assets
can be regarded as the average daily volume or the volume that can be
ð1Þ
unwound under severe crisis periods.6 The trading volume in a crisis
period can be roughly approximated as the average daily trading
volume less a number of standard deviations. Albeit this alternative ap-
proach is quite simple, it is still relatively objective. Moreover, it is reason-
ably easy to gather the required data to perform the necessary liquidation
scenarios.
In this backdrop, we can now define VaRi for single asset positions.
By and large, the absolute value of VaRi for any single financial asset
can be defined in monetary terms as:
 
VaRi ¼ ðμ i −α  σ i Þ½Asset i  Fxi ≈jα  σ i ½Asset i  Fxi j ð4Þ
where μi is the expected return of asseti, α is the confidence level (or in
other words, the standard normal variant at confidence level α) and σi is
the forecasted standard deviation (or conditional volatility) of the
returns of asseti. While the term Asseti denotes the mark-to-market
monetary value, and Fxi indicates the unit foreign exchange rate of
asseti. Without a loss of generality, we can assume that the expected
value of daily returns μi is close to zero. As such, though Eq. (4) includes
some simplifying assumptions, yet it is routinely used by researchers
and practitioners in financial markets for the estimation of VaR for any
ð2Þ
single trading position.
3.2. Methodological Step (2)
In the second stage we define the corresponding robust scenario
optimization-algorithm which is based on quadratic programming
techniques, subject to applying meaningful financial and operational
meaningful constraints. Furthermore, in the development of portfolio
investment policy there are many types of constraints that can be con-
sidered as an integral part of the optimization process. These constraints
are drawn from rational financial investment considerations and can be
used in various applications to bound percentiles of loss distributions;
and it can include constraints on the asset classes that are allowable
and concentration limits on investments. Moreover, in making the
asset-allocation decision, consideration must be given to any risk-based
capital requirements. For this objective, the scenario optimization-
6
Thus, if trading volume is low because of a “one-way market,” in that most people are
seeking to sell rather than to buy, then (ti) can rise substantially (Saunders and Cornett,
2008).
 Author's personal copy
M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381
algorithms can be defined as a minimization algorithmic process of the
dynamic objective risk-function and as follows:
1) The minimization process is attained here by minimizing the objec-
tive risk-function while requiring minimum expected returns sub-
ject to imposing several rational and meaningful financial and
operational constraints;
2) As a minimum the bounding limits may include the following:
a) The target investable portfolio's expected return; b) total volume
of the investable portfolio; c) monetary asset-allocation of each
asset-class; d) portfolio managers' choices of pure long positions or
a combination of long/short-sales trading positions; and e) close-
out or unwinding liquidity horizons of each asset-class.
Essentially, our proposed optimization algorithm is a robust
enhancement to the classical Markowitz mean–variance approach,
where the original risk measure, variance, is replaced by LVaR algorithm.
The task is attained here by minimizing LVaR algorithm, while requiring a
minimum expected return subject to applying meaningful financial and
operational constraints. Thus, by considering different expected returns,
we can generate an optimal LVaR frontier. Alternatively, we can also max-
imize expected returns while not allowing for large risks. For the purpose
of this study, the optimization problem is formulated as follows.
From Eq. (2) we can define liquidation horizon factor (LHFi) for each
trading asset as:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ð2t i þ 1Þðt i þ 1Þ
LH F i ¼ ≥1:0; ∀i i ¼ 1; 2; …; n:
6t i
After substituting Eq. (5) into Eq. (1), we can now compute the min-
imum portfolio downside-risk by solving the following nonlinear qua-
dratic algorithm (objective function) under adverse and illiquid market
situations:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uX qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u n X n
Minimize : LVaRP ¼ t LVaRi LVaR j ρi; j ¼ jLVaRjT jρjjLVaRj: ð6Þ
i¼1 j¼1
The above objective function can be minimized conditional on satis-
fying the following operational and financial budget constraints as spec-
ified by the portfolio manager:
X
n val of 95% (or 97.5% with “one-tailed” loss side) and several liquidation
Ri xi ¼ RP ; li ≤xi ≤ui i ¼ 1; 2; …; n
i¼1
X
n 17/10/2004–22/05/2009, are assembled for the purpose of assessing
xi ¼ 1:0; li ≤xi ≤ui i ¼ 1; 2; …; n
i¼1
X
n
Vi ¼ VP i ¼ 1; 2; …; n
i¼1
jLHF j≥1:0; ∀i i ¼ 1; 2; …; n :
Here RP and VP denote the target portfolio mean expected return and
total portfolio volume respectively, and xi the weight or percentage
asset allocation for each asset. The values li and ui, i = 1, 2, …, n, denote
the lower and upper constraints for the portfolio weights xi. If we choose
li = 0, i = 1, 2, …, n, then we have the situation where no short-sales
are allowed. Moreover, |LHF| indicates an (n × 1) liquidity risk factor
vector for all i = 1, 2, …, n.
Now the portfolio manager can specify different liquidity horizons and
correlation factors and calculate the necessary amount of LVaR to sustain
trading operations without subjecting the financial entity to insolvency
matters. In fact, the rationality behind imposing the above constraints is
to comply with contemporary regulations which enforce capital cushion
requirements on investment companies, proportional to their LVaR and
373
economic-capital, besides other operational limits (for instance, volume
trading limits, the choices of long/short-sales positions).
3.3. Methodological Step (3)
In this final phase, we validate and compare the output results of
investable portfolios obtained in Step (2) with the optimal portfolios de-
fined in Step (1). To this end, we attempt to rerun the optimization-
engine until a new convergence to meaningful investable portfolios is
attained. At this stage, new investable portfolios with coherent asset-
allocation structures, that satisfy the boundary conditions defined in
Step (2), are accomplished accordingly.7
4. Analysis and interpretation of empirical results
In essence, the non-linear objective risk-function and robust scenar-
io optimization-algorithms discussed above can be programmed (using
quadratic programming techniques) and implemented without a com-
mensurate increase in computational power. Moreover, time-series
datasets required to estimate the dynamic risk-function and robust sce-
nario optimization-algorithm parameters under adverse and crisis mar-
ket outlooks are readily available from the various historical dataset
providers, such as the Thomson Reuters DataStream platform. To this
end, in this research paper, databases of daily price returns of the six
GCC stock markets' main indicators (indices) are implemented. The
ð5Þ total number of stock market indices that is applied in this work is
nine indices; seven local indices for the six GCC stock markets (includ-
ing two stock market indices for the UAE markets) and two general
benchmark indices, detailed as follows: DFM General Index (Dubai
Financial Market General Index, Country: United Arab Emirates);
ADSM Index (Abu Dhabi Stock Market Index, Country: United Arab
Emirates); BA All Share Index (All Share Stock Market Index, Country:
Bahrain); KSE General Index (Stock Exchange General Index, Country:
Kuwait); MSM30 Index (Muscat Stock Market Index, Country: Oman);
DSM20 Index (Doha Stock Market General Index, Country: Qatar); SE
All Share Index (All Share Stock Market Index, Country: Saudi Arabia);
Shuaa GCC Index (GCC Stock Markets Benchmark Index, Provider:
Shuaa Capital in UAE); and Shuaa Arab Index (Arab Stock Markets
Benchmark Index, Provider: Shuaa Capital in UAE).
For this particular research study we have chosen a confidence inter-
ð7Þ time horizons (unwinding periods) to compute LVaR. Historical data-
base (of more than six years) of daily closing index levels, for the period
ð8Þ optimal and investable portfolios.8 In effect, the selected time-series
datasets fall within the period of the most sever part of the sub-prime
financial crunch. As a result, we estimated the asset return distribution
dynamic and conditional volatility using a GARCH-M model and tested
ð9Þ
for normality as indicated in Exhibit (1). Furthermore, the cross-
correlations between risk factors are obtained from daily logarithmic
ð10Þ returns and are summarized in Exhibit (2).
In this backdrop, the empirical optimization process is based on the
definition of LVaR as the minimum possible loss over a specified time
horizon within a given confidence level. The iterative-optimization
algorithm solves the problem by finding the market positions that min-
imize the loss, subject to satisfying the requirements of all constraints
within their boundary values. Furthermore, in all cases the liquidation
7
Other than the method used in this paper, there are also the multi-group model as
well as multi-index model that can similarly be used to find optimal portfolios using iter-
ative algorithms (see for instance, Xia et al., 2000). Moreover, for an excellent review on
modern portfolio theory and other relevant portfolio optimization techniques, we refer
the readers to Elton and Gruber (1997) and Elton et al. (1977).
8
The historical database of daily index levels is drawn from Reuters 3000 Xtra Hosted
Terminal Platform and Thomson Reuters Datastream datasets.
 Author's personal copy
374 M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381
Fig. 1. Optimal portfolios with 10-day LVaR technique (case of long positions under crisis market conditions).
horizons (close-out periods) specified in Tables 1–4 are applied all the
way through the optimization route. For the sake of restraining the sce-
nario optimization algorithm and thereafter its analysis, a volume trad-
ing limit of 10 million AED is assumed as a constraint—that is, the asset
management entity must preserve a maximum overall market value of
diverse equities of no more than 10 million AED (of long-only posi-
tions). As such, Fig. 1 provides evidence of the empirical LVaR optimal
frontier (with 10-day liquidation horizons and under crisis market con-
ditions) defined using a 97.5% confidence level. As indicated above, the
optimal portfolio selection is performed by imposing a short-sales pro-
hibition constraint for the different equity assets. On the other hand, op-
timal portfolios cannot always be attained in the day-to-day real-world
portfolio management operations and, hence, portfolio managers
should establish proactive “investable portfolios” under more realistic
and restricted budget constraints, detailed as follows:
• The total trading volume (of long equity trading positions) for any
investable portfolio is 10 million AED.
• For any investable portfolio, the asset allocation for long equity trad-
ing position varies from 0% to 50%.
• All liquidity horizons (close-out periods) for all equities are kept con-
stant and in accordance with the values indicated in Tables 1–4.
• Volatilities under crisis market notions are estimated as the maximum
historical-simulation events with the highest downside-risk. These
conditional volatilities are kept constant throughout the optimization
process and in accordance with the estimated values indicated in
Tables 1–4.
In this particular case the asset allocation weights are allowed to take
only positive values, however, since arbitrarily high or low positive
weights could have no financial sense, we determined to introduce
lower and upper boundaries for the weights and in accordance with rea-
sonable trading practices. Furthermore, for comparison purposes and
since the endeavor in this work is to minimize LVaR under crisis market
prospects subject to specific expected returns and total portfolio volume,
we decided to plot LVaR versus expected returns and not the reverse, as
is commonly done in the various modern portfolio management litera-
ture. Accordingly, it is appealing to note that the four benchmark invest-
able portfolios (investable portfolios [1], [2], [3] and [4]) as indicated in
Fig. 2 are located in some way close to the optimal (efficient) frontier,
however definitely not on the efficient frontier as the classical mean–
variance theory would suggest.9 This is because financial and operational
real-world investment considerations make it unlikely that an invest-
ment portfolio will behave exactly as theory predicts and mainly under
crisis market perspectives. Imperfections such as restriction on long trad-
ing positions, total trading volume and liquidation horizons make it un-
likely to create an optimal equity trading portfolio. Thus, the portfolio
manager should apply active strategies in order to earn excess returns
and particularly under adverse market perceptions. These considerations
are particularly important for individual portfolio managers who may
spread their purely long trading positions across a few financial assets.
In order to illustrate the composition of investable portfolios [1], [2],
[3] and [4], Tables 1–4 emphasize the asset allocation weights for all
equity assets along with their expected returns, close-out periods and
sensitivity factors. Similarly, the four tables depict the minimum re-
quired LVaR to support and maintain the operations of these portfolios
as well as the ratio of LVaR/volume under the assumption of three dif-
ferent correlation factors.10 In this way, portfolio managers can employ
appropriate downside-risk measures that allow them to take prompt
decisions which could produce a risk budget lower than a specific target.
9
It is particularly interesting to note that incorporating the above realistic and restricted
budget constraints—as a result of the aftermaths of a severe financial crunch—into the
portfolio manager's operational process has caused that investable portfolio [2], for in-
stance, tends to be positioned on the 10-day optimal frontier. It is also important to men-
tion that the liquidity horizons of all trading assets that constitute portfolio [2] are much
less than 10-days as indicated in Table 2. Thus, it seems that by incorporating the liquidity
holding periods and other operational constraints into the constrained optimization func-
tion, portfolio managers could apply active trading strategies in order to earn excess
returns that could be, one way or another, beyond of what the classical mean–variance
efficient-frontier dictates. Furthermore, for the sake of comparison with long/short-sales
optimization cases, our empirical results have shown that none of the long/short-sales
portfolios have demonstrated similar features given that none of these portfolios are locat-
ed on the 10-day optimal frontiers [we refer the readers to the second scenario optimiza-
tion case study of long/short-sales positions and Fig. 4]. As such, the above conclusions of
long-only positions cannot be applied to the long/short-sales optimization cases.
10
After running several minimization and investable portfolio simulation studies, our em-
pirical results indicate by and large that under crisis market conditions the actual obtained in-
vestable portfolios can expect to realize greater downside-risk of approximately 6.0 times
more than the results under normal market conditions. Furthermore, and in relation to the
simulated stress-testing scenarios, it is possible to assess the probability of occurrence of the-
se rare, though possible, events. In fact, we have used approximately 1200 trading days in our
risk-engine evaluation and it is evident that the severest crisis events occurred within a prob-
ability of 0.083% (this means, there is only 1 chance in a 1200 or in other words one day in
every 1200 trading days for such an event to occur within a 97.5% level of confidence).
 Author's personal copy

M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381 375

Fig. 2. Optimal and investable portfolios with LVaR technique (case of long positions under crisis market conditions).

Fig. 3. Optimal portfolios with 10-day LVaR technique (case of long and short-sales positions under crisis market conditions).

In this line of reasoning, under adverse market conditions for instance,
LVaR is calculated by implementing downside volatilities (i.e. maximum
negative asset returns throughout the sampling period). Thus, this anal-
ysis and the implemented LVaR algorithm permit one to determine the
asymmetric aspect of risk and investable portfolios under crisis market
notions and, as such, it is a substantial improvement to the classical
mean–variance approach. As a result, it seems that by incorporating
the liquidity holding periods and other operational constraints into
the constrained optimization function, portfolio managers could apply
active trading strategies in order to earn excess returns that could be,
one way or another, quite different of what the classical mean–variance
efficient-frontier indicates.
In the second scenario optimization case study we considered
different combinations of long and short-sales positions (that is,
disallowing both pure long positions and borrowing constraints).11
Similar to the previous case, the empirical optimization process is
based on the definition of LVaR as the minimum possible loss over a spe-
cific time horizon and within a given confidence level. The iterative-
optimization modeling algorithm solves the problem by finding the
market positions that minimize the loss, subject to satisfying that all
constraints are located within their boundary values. Furthermore, in
all cases the liquidation horizons (as indicated in Tables 5–8) are
11
Albeit current operational platform in GCC financial markets does not permit short-
sales of assets, and in anticipation that this restriction might be relaxed in the future, we
perform an analysis for long/short-sales positions with the objective of determining struc-
tured investable portfolios.
 Author's personal copy
376 M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381
assumed constant throughout the optimization process. For the sake of
restricting the optimization algorithm and thereafter its analysis, a trad-
ing volume limit of 10 million AED is assumed as a constraint—that is
the equity trading entity must keep a maximum market value of differ-
ent equities of no more than 10 million AED (between long and short-
sales positions). As such, Fig. 3 provides evidence of the empirical
LVaR optimal frontier (under 10-day liquidation horizons and for crisis
market conditions) defined using a 97.5% confidence level. As men-
tioned above, the optimal portfolio selection is performed by relaxing
the short-sales constraint for the different equity assets. On the other
hand, optimal portfolios cannot always be attained (e.g. short-sales
without realistic lower boundaries on xi) in the day-to-day real-world
portfolio management operations and, hence, the portfolio manager
should establish proactive investable portfolios under realistic and
restricted budget constraints, detailed as follows:
• The total trading volume (between long and short-sales equity
trading positions) of any investable portfolio is 10 million AED.
• For any investable portfolio, the asset allocation for long equity
trading position varies from 10% to 100%.
• For any investable portfolio, the asset allocation for short-sales equity
trading position varies from −10% to −60%.
• All liquidity horizons (close-out periods) for all equities are kept
constant and in accordance with the values indicated in Tables 5–8.
• Volatilities under crisis market notions are estimated as the maximum
historical-simulation events with the highest downside-risk. These
conditional volatilities are kept constant throughout the optimization
process and in accordance with the estimated values indicated in
Tables 5–8.
Now the asset allocation weights are permitted to contain negative
or positive quantities, however, since arbitrarily high or low percent-
ages have no financial logic, we determined to introduce lower and
upper boundaries for the weights and in accordance with reasonable
trading and asset management practices. In view of that, and on the
contrary to the previous optimization case of long-only positions, it is
interesting to note that the four benchmark investable portfolios
(investable portfolios [1], [2], [3] and [4]) are not positioned near the
optimal frontier as depicted in Fig. 4. This is because financial and oper-
ational real-world investment concerns make it unlikely that an invest-
ment portfolio will perform precisely as theory expects and above all
under crisis market perspectives. Imperfections such as restriction on
long and short-sales trading positions, total trading volume and liquida-
tion horizons make it unlikely to obtain an optimal equity trading port-
folio.12 Thus, the portfolio manager should employ active strategies in
order to earn excess returns and particularly under adverse market per-
ceptions. These considerations are especially relevant for individual
12
Indeed, one other reason that can be added is a result of the small cross-correlation fac-
tors that we have observed in the entire GCC stock markets. In fact, and in accordance with
previous studies on other emerging financial markets, such as Mexico (see for instance, Al
Janabi, 2007), the author found that short-selling usually decreases LVaR figures and, hence,
lower the maximum LVaR trading budget limits (i.e. risk budgeting). Indeed, in the case of
the GCC stock markets our empirical results contradict some of the previous studies on other
emerging markets, such as the case of Mexico. In essence, the above observed phenomena in
this paper (that is, long-only portfolios appear to lie much closer to the frontier than portfo-
lios including both long and short positions) can be explained by the nature of the diminutive
correlation factors that we have witnessed in the entire GCC stock markets. As such, these ti-
ny correlation factors have led to grand diversification benefit for long-only equity trading
positions and vice-versa for short-sales positions. Furthermore, in contrast to previous re-
search on other emerging financial markets, it is important to note here that in this work
we have considered a more realistic model for the incorporation of daily conditional volatility
under severe market condition, which takes into account the maximum negative returns
(losses) that are identified in the historical time series for all nine stock market indices. More-
over, our risk algorithm model is outfitted with additional features that can produce a clear-
cut risk assessment and empirical simulation results under crisis-driven correlation factors
that cannot be obtained with the plain assumption of normality.
portfolio managers who may spread their long/short-sales trading posi-
tions across a few financial securities.
In order to illustrate the composition of investable portfolios [1], [2],
[3] and [4], Tables 5–8 highlight the asset allocation weights for all equity
assets along with their expected returns, close-out periods and sensitiv-
ity factors. Similarly, the four tables depict the minimum required LVaR
to support the operations of these particular portfolios as well as the
ratio of LVaR/volume under the assumption of three different correlation
factors. In this way, portfolio managers should employ appropriate
downside-risk measures which would allow them to take swift decisions
which could produce a risk budget lower than a specific target. Thus, this
analysis is substantially a generalized enhancement to the classical
Markowitz's (1952) analysis since it permits one to determine the asym-
metric aspect of risk and investable portfolios under crisis market
notions.
It is worthwhile to note that the empirical results of the alternative
LVaR method and its constrained portfolio optimization parameters
acknowledge that some of the standard Markowitz's assumptions are
being violated and this can influence the outcomes with respect to effi-
ciency estimates and particularly under crisis market circumstances.
Indeed, this remark is correct once we consider constrained versus
non-constrained optimization techniques, the stylized variance/covari-
ance matrix, close-out horizons, total trading volume, expected return,
portfolio managers' choices of long-only position or a combination of
long/short positions, and so forth. Furthermore, this observation may
provoke some issues to portfolio managers with respect to the actual
application of LVaR for the management of real investable portfolios.
As stated throughout this research paper, we consider that LVaR can
generate better investable portfolios for portfolio managers since it con-
siders no-nonsense trading situations where traders can unwind small
parts of their holdings on a daily basis. Furthermore, LVaR method can
integrate downside-risk under unfavorable market circumstances so
that any added risk resulting from any non-normality and illiquid assets
may be used to assess the total LVaR of the portfolio. This allows a much
more generalized framework to be established, with the distributional
assumption most suitable to the category of financial assets to be
employed, and which can be of vital value for more precise market
risk assessment during market stress periods and mainly when liquidity
dries up; correlation factors switching signs; and the integration of non-
normal distribution of assets' returns into the risk measurement pro-
cess. By clarifying and resoling the above issues and concerns, we are
hopeful that academics and practitioners can focus their attention on
the positive features of LVaR and its role in assessing better investable
portfolios so that to ease any concern that the empirical results are in-
significant and less meaningful because of the input parameters
employed in the constrained optimization process.
Finally, it is important to emphasis that other researchers such as
Al Janabi (2013) has investigated and applied similar optimization algo-
rithm for the GCC financial markets under normal market conditions
along with some asset liquidity stress-testing analysis. Indeed, his em-
pirical analysis was directed to investigate an optimization case study
of only long/short-sales trading positions under normal market settings.
While some of his general optimization elements and parameters are
adapted herein, this work builds on Al Janabi (2013) research paper
and differs in the sense that the optimization algorithm was carried
out under severe crisis market circumstances as a result of a financial
crunch (e.g. the most recent financial turmoil). Furthermore, in this
paper our empirical analysis was performed for investable portfolios
that have both long and short-sales trading positions (that is,
disallowing both pure long positions and borrowing constraints) or for
those portfolios that consists merely of pure long positions. Indeed,
our optimization algorithm and the newly obtained empirical results
have shown that portfolio managers can obtain financially meaningful
investable portfolios (under crisis market settings) and demonstrated
new interesting market-microstructures' patterns (e.g. the impact of
close-out periods, overall trading volume, and expected returns, on
 Author's personal copy
M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381
the optimization-algorithm process and particularly for the optimiza-
tion case of long-only trading positions). Indeed, these newly observed
patterns cannot be attained by using the classical Markowitz's mean–
variance approach. Moreover, the new market-microstructures' pat-
terns and the drawn conclusions were not apparent in Al Janabi
(2013) research study.
5. Summary and concluding remarks
This paper develops scenario optimization algorithms by which
investable financial portfolios during crisis periods can be evaluated.
Specifically, the paper examines a number of alternative scenarios
considering both long-only and long and short positions subject to
minimizing the Liquidity-Adjusted Value-at-Risk (LVaR) and various fi-
nancial and operational constraints such as target expected return,
portfolio trading volume, close-out periods and portfolio weights.
The results show that the obtained investable portfolios lie off the
efficient frontier, but that long-only portfolios appear to lie much
closer to the frontier than portfolios including both long and short
positions.
More specifically, in this research paper a non-linear risk-function
and robust optimization technique for the assessment of illiquidity of
pure long trading position is incorporated. To this end, the liquidity
technique that is applied in this work is more suitable for real-world
trading and asset management practices since it considers selling
small fractions of the trading assets on a daily basis. Furthermore, this
liquidity model can be implemented for the entire portfolio or for
each individual security within the equity trading portfolio. Indeed,
the developed methodology and risk assessment algorithms can aid in
advancing portfolio management risk management practices in emerg-
ing markets and above all in the wake of the latest credit crunch and the
succeeding financial upheavals.
The empirical results for the GCC zone confirm that in almost all
tests, there are strong asymmetric characteristics in the distribution of
daily returns of the selected stock market indices. The appealing result
of this research study endorses the necessity of combining LVaR estima-
tions with other techniques such as stress-testing (under crisis market
circumstances) and scenario analysis (as a consequence of a financial
crunch) to obtain systematic descriptions of other remaining risks
(such as, fat-tails in the probability distribution) that cannot be
disclosed with the simple assumption of normality.
Indeed, our suggested modeling technique is in-line with the recom-
mendation of previous studies (see, for example, Neftci, 2000). In es-
sence, a number of authors have argued that many asset distributions
have “fat tails” and that the assumption of normal distribution can def-
initely underestimate the risk of extreme losses. As such, our proposed
portfolio risk-engine is a combination of a closed-form parametric
LVaR along with a stress-testing approach that is based on historical
simulation of real severe upheavals in the GCC markets. As a matter of
fact, our empirical results indicate generally that under crisis market
conditions the actual obtained investable portfolios can expect to realize
greater downside-risk of approximately 6.0 times more than the results
under normal market conditions.
In addition, our proposed risk-algorithm has shown that portfolio
managers can obtain financially meaningful investable portfolios and
demonstrated interesting market-microstructures' patterns (e.g. the im-
pact of close-out periods, overall trading volume, and expected returns,
on the optimization-algorithm process) which cannot be attained by
using the classical Markowitz's mean–variance approach. In view of
that, the obtained investable benchmark portfolios are noticeably located
far away from the optimal frontiers and particularly for combinations of
long and short-sales positions. While, it seems that this optimization phe-
nomenon could not be attained for long and short-sales trading position,
however for long-only trading positions it seems that it is possible to get
closer to the optimal frontier and synchronize to a certain degree the per-
formance of both optimal and investable portfolios. For this reason, the
377
empirically obtained benchmark investable portfolios of pure long posi-
tions, under the notion of a crisis event, are noticeably located near the
optimal frontier, nevertheless not on the efficient frontier as the classical
mean–variance theory might suggest. This is due to the fact that financial
and operational real-world investment considerations make it unlikely
that an investable portfolio will behave exactly as theory predicts. Imper-
fections such as restriction on long trading positions, total trading volume
and liquidation horizons make it unlikely to create an optimal equity port-
folio. As such, the portfolio manager should apply active strategies in
order to earn excess returns and particularly under adverse market per-
ceptions. Thus, it seems that incorporating the liquidity holding periods
and other operational constraints into the constrained optimization func-
tion, portfolio managers could employ active trading and investment
strategies so as to capture additional returns that could be, one way or
another, quite different of what the classical mean–variance efficient
frontier indicates.13 In a nutshell, the obtained empirical results can
have several practical applications and could aid in overcoming some
of the shortcomings of conventional VaR and the classical mean–
variance approach, especially in light of the aftermaths of the latest
financial crunch.
In conclusion, the presented methodology and empirical results of
this research paper can have important practical uses and applications
for financial institutions, risk managers, portfolio managers, financial
regulators and policymakers operating in the GCC and other emerging
markets, and particularly in the wake of the most recent financial crisis.
For instance, the proposed modeling technique and simulation algo-
rithms can be used by risk managers and portfolio managers for the as-
sessment of appropriate asset allocations of different structured
investable portfolios. Finally, we believe that the proposed risk-engine
and optimization-algorithm have the potential of producing realistic
risk-return profiles and may improve real-world understanding of em-
bedded risks and asymmetric microstructure patterns and could poten-
tially create better investable portfolios for portfolio managers in the
GCC zone and other developing economies.
Acknowledgment
The author would like to thank two anonymous reviewers and the
special issue guest editor (Prof. Duc Khuong Nguyen, Ph.D.) for their con-
structive inputs throughout the review process. This work has benefited
from a financial support (competitive grant for summer research pro-
jects) from the College of Business and Economics (CBE), United Arab
Emirates University, Al-Ain, UAE. The usual disclaimer applies.
13
Furthermore, our empirical result indicated that incorporating realistic and restricted
budget constraints—as a result of the aftermaths of a severe financial crunch—into the port-
folio manager's operational process can cause that some investable portfolios (particularly for
the case of long-only trading positions) to be positioned on the highest liquidity-horizon op-
timal frontier (i.e. 10-days holding period) although the actual liquidity horizons of all trading
assets that constitute these portfolios are much less than the liquidity-horizon of the optimal
frontier. Thus, it seems incorporating the liquidity holding periods and other operational con-
straints into the constrained optimization function, portfolio managers could apply active
trading strategies in order to earn excess returns that could be, one way or another, beyond
of what the classical mean-variance efficient-frontier indicates. The above phenomena was
observed merely for pure long position, however for long/short-sales optimization cases
our empirical results have shown that none of the long/short-sales portfolios have demon-
strated similar features. This is because none of the resulting structured portfolios are located
on the efficient frontier. In fact, our optimization algorithm and the obtained empirical results
have shown that portfolio managers can determine meaningful investable portfolios (under
crisis market settings). In addition, the attained results have revealed new interesting
market-microstructures' patterns. These patterns (e.g. the effects of close-out periods, overall
trading volume, expected returns, etc.) have indeed important impact on the optimization-
algorithm process and particularly for the optimization case of long-only trading positions.
Indeed, these newly observed patterns cannot be achieved using the conventional
Markowitz's mean-variance technique. Furthermore, the new market-microstructures' pat-
terns and the drawn conclusions were not evident in Al Janabi (2013) research paper.
 Author's personal copy

378 M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381

Fig. 4. Optimal and investable portfolios with LVaR technique (case of long and short positions under crisis market conditions).

Appendix A. Simulation results and empirical analysis of optimal and investable portfolios

Exhibit 1
Risk function dataset: Conditional volatility, expected returns and test for non-normality.

Stock market Daily volatility Daily volatility Arithmetic Expected Sensitivity (beta) Skewness Kurtosis Jarque–Bera
indices (normal market)* (severe market)* mean return* factor (JB) Test

DFM General Index 1.81% 12.16% 0.12% 0.14% 0.58 0.01 7.86 955**
ADSM Index 1.32% 7.08% 0.07% 0.07% 0.40 0.12 7.26 734**
BA All Share Index 0.58% 3.77% 0.05% 0.04% 0.06 0.43 10.24 2142**
KSE General Index 0.71% 3.74% 0.09% 0.08% 0.14 −0.18 8.38 1173**
MSM30 Index 0.79% 8.70% 0.12% 0.10% 0.10 −0.57 18.40 9617**
DSM20 Index 1.48% 8.07% 0.06% 0.07% 0.31 −0.11 5.59 273**
SE All Share Index 1.86% 11.03% 0.03% 0.01% 0.98 −0.97 8.47 1361**
Shuaa GCC Index 1.30% 8.10% 0.06% 0.08% 1.05 −0.66 14.00 4949**
Shuaa Arab Index 1.15% 7.57% 0.07% 0.10% 1.00 −0.61 13.79 4758**

Notes: 1) Single asterisk (*) denotes estimation of conditional volatility and expected return using GARCH-M model.
2) Double asterisks (**) denote statistical significance at the 0.01 level.

Exhibit 2
Risk function dataset: Cross-correlation matrix of stock market indices.

DFM General ADSM BA All Share KSE General MSM30 DSM20 SE All Share Shuaa GCC Shuaa Arab
Index Index Index Index Index Index Index Index Index

DFM General Index 100%

ADSM Index 56% 100%
BA All Share Index 12% 8% 100%
KSE General Index 17% 16% 12% 100%
MSM30 Index 12% 17% 11% 11% 100%
DSM20 Index 18% 23% 12% 12% 20% 100%
SE All Share Index 20% 20% 7% 16% 11% 10% 100%
Shuaa GCC Index 37% 35% 13% 19% 13% 26% 62% 100%
Shuaa Arab Index 39% 36% 12% 24% 15% 26% 60% 93% 100%
 Author's personal copy

M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381 379

Table 1
Composition of investable market portfolio [1]. Case analysis of long positions under crisis market conditions.

Market index Volatility (crisis market) Liquidation period (in days) Market value in AED Asset allocation per market

DFM General Index 12.16% 2.0 1,000,000 10%

ADSM Index 7.08% 3.0 0 0%
BA All Share Index 3.77% 4.0 1,000,000 10%
KSE General Index 3.74% 3.0 0 0%
MSM30 Index 8.70% 4.0 3,000,000 30%
DSM20 Index 8.07% 3.0 5,000,000 50%
SE All Share Index 11.03% 2.0 0 0%

Correlation factor LVaR (crisis market) LVaR/volume Diversification benefit Expected return

ρ = Empirical 1,445,912 14.5% ρ = Emp. Vs. ρ = 1 0.080%

ρ=1 2,096,756 21.0% 650,844 Sensitivity factor
ρ=0 1,268,633 12.7% 45.01% 0.250

Table 2
Composition of investable market portfolio [2]. Case analysis of long positions under crisis market conditions.

Market index Volatility (crisis market) Liquidation period (in days) Market value in AED Asset allocation per market

DFM General Index 12.16% 2.0 3,000,000 30%

ADSM Index 7.08% 3.0 2,000,000 20%
BA All Share Index 3.77% 4.0 500,000 5%
KSE General Index 3.74% 3.0 500,000 5%
MSM30 Index 8.70% 4.0 1,000,000 10%
DSM20 Index 8.07% 3.0 1,000,000 10%
SE All Share Index 11.03% 2.0 2,000,000 20%

Correlation factor LVaR (crisis market) LVaR/volume Diversification benefit Expected return

ρ = Empirical 1,405,342 14.1% ρ = Emp. Vs. ρ = 1 0.081%

ρ=1 2,199,765 22.0% 794,423 Sensitivity factor
ρ=0 1,065,504 10.7% 56.53% 0.500

Table 3
Composition of investable market portfolio [3]. Case analysis of long positions under crisis market conditions.

Market index Volatility (crisis market) Liquidation period (in days) Market value in AED Asset allocation per market

DFM General Index 12.16% 2.0 4,000,000 40%

ADSM Index 7.08% 3.0 2,000,000 20%
BA All Share Index 3.77% 4.0 0 0%
KSE General Index 3.74% 3.0 0 0%
MSM30 Index 8.70% 4.0 0 0%
DSM20 Index 8.07% 3.0 0 0%
SE All Share Index 11.03% 2.0 4,000,000 40%

Correlation factor LVaR (crisis market) LVaR/volume Diversification benefit Expected return

ρ = Empirical 1,811,704 18.1% ρ = Emp. Vs. ρ = 1 0.076%

ρ=1 2,427,181 24.3% 615,477 Sensitivity factor
ρ=0 1,510,229 15.1% 33.97% 0.702

Table 4
Composition of investable market portfolio [4]. Case analysis of long positions under crisis market conditions.

Market index Volatility (crisis market) Liquidation period (in days) Market value in AED Asset allocation per market

DFM General Index 12.16% 2.0 0 0%

ADSM Index 7.08% 3.0 4,000,000 40%
BA All Share Index 3.77% 4.0 0 0%
KSE General Index 3.74% 3.0 0 0%
MSM30 Index 8.70% 4.0 0 0%
DSM20 Index 8.07% 3.0 2,000,000 20%
SE All Share Index 11.03% 2.0 4,000,000 40%

Correlation factor LVaR (crisis market) LVaR/volume Diversification benefit Expected return

ρ = Empirical 1,457,139 14.6% ρ = Emp. Vs. ρ = 1 0.054%

ρ=1 2,095,600 21.0% 638,461 Sensitivity factor
ρ=0 1,278,460 12.8% 43.82% 0.613
 Author's personal copy

380 M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381

Table 5
Composition of investable market portfolio [1]. Case analysis of long and short-sales positions under crisis market conditions.

Market index Volatility (crisis market) Liquidation period (in days) Market value in AED Asset allocation per market

DFM General Index 12.16% 2.0 (2,000,000) −20%

ADSM Index 7.08% 3.0 6,000,000 60%
BA All Share Index 3.77% 4.0 6,000,000 60%
KSE General Index 3.74% 3.0 (2,500,000) −25%
MSM30 Index 8.70% 4.0 7,000,000 70%
DSM20 Index 8.07% 3.0 (1,000,000) −10%
SE All Share Index 11.03% 2.0 (3,500,000) −35%

Correlation factor LVaR (crisis market) LVaR/volume Diversification benefit Expected return

ρ = Empirical 2,134,605 21.3% ρ = Emp. Vs. ρ = 1 0.091%

ρ=1 1,504,198 15.0% (630,407) Sensitivity factor
ρ=0 2,328,469 23.3% −29.53% −0.180

Table 6
Composition of investable market portfolio [2]. Case analysis of long and short-sales positions under crisis market conditions.

Market index Volatility (crisis market) Liquidation period (in days) Market value in AED Asset allocation per market

DFM General Index 12.16% 2.0 (2,000,000) −20%

ADSM Index 7.08% 3.0 2,500,000 25%
BA All Share Index 3.77% 4.0 6,000,000 60%
KSE General Index 3.74% 3.0 3,000,000 30%
MSM30 Index 8.70% 4.0 (3,000,000) −30%
DSM20 Index 8.07% 3.0 (3,000,000) −30%
SE All Share Index 11.03% 2.0 6,500,000 65%

Correlation factor LVaR (crisis market) LVaR/volume Diversification benefit Expected return

ρ = Empirical 1,946,431 19.5% ρ = Emp. Vs. ρ = 1 0.020%

ρ=1 1,080,993 10.8% (865,438) Sensitivity factor
ρ=0 2,097,413 21.0% −44.46% 0.570

Table 7
Composition of investable market portfolio [3]. Case analysis of long and short-sales positions under crisis market conditions.

Market index Volatility (crisis market) Liquidation period (in days) Market value in AED Asset allocation per market

DFM General Index 12.16% 2.0 3,500,000 35%

ADSM Index 7.08% 3.0 2,500,000 25%
BA All Share Index 3.77% 4.0 2,500,000 25%
KSE General Index 3.74% 3.0 (2,000,000) −20%
MSM30 Index 8.70% 4.0 1,000,000 10%
DSM20 Index 8.07% 3.0 (1,000,000) −10%
SE All Share Index 11.03% 2.0 3,500,000 35%

Correlation factor LVaR (crisis market) LVaR/volume Diversification benefit Expected return

ρ = Empirical 1,712,911 17.1% ρ = Emp. Vs. ρ = 1 0.073%

ρ=1 2,364,562 23.6% 651,651 Sensitivity factor
ρ=0 1,429,737 14.3% 38.04% 0.609

Table 8
Composition of investable market portfolio [4]. Case analysis of long and short-sales positions under crisis market conditions.

Market index Volatility (crisis market) Liquidation period (in days) Market value in AED Asset allocation per market

DFM General Index 12.16% 2.0 8,500,000 85%

ADSM Index 7.08% 3.0 (5,500,000) −55%
BA All Share Index 3.77% 4.0 5,000,000 50%
KSE General Index 3.74% 3.0 (2,000,000) −20%
MSM30 Index 8.70% 4.0 (6,000,000) −60%
DSM20 Index 8.07% 3.0 4,500,000 45%
SE All Share Index 11.03% 2.0 5,500,000 55%

Correlation factor LVaR (crisis market) LVaR/volume Diversification benefit Expected return

ρ = Empirical 2,972,997 29.7% ρ = Emp. Vs. ρ = 1 0.043%

ρ=1 2,503,323 25.0% (469,674) Sensitivity factor
ρ=0 3,359,789 33.6% −15.80% 0.891
 Author's personal copy
M.A.M. Al Janabi / Economic Modelling 40 (2014) 369–381
References
Acerbi, C., Tasche, D., 2002. On the coherence of expected shortfall. J. Bank. Financ. 26 (7),
1487–1503.
Agarwal, V., Naik, N.Y., 2004. Risks and portfolio decisions involving hedge funds. Rev.
Financ. Stud. 17 (1), 63–98.
Al Janabi, M.A.M., 2007. Risk analysis, reporting and control of equity exposure: via-
ble applications to the Mexican financial market. J. Deriv. Hedge Funds 13 (1),
33–58.
Al Janabi, M.A.M., 2011a. A generalized theoretical modeling approach for the assessment
of economic capital under asset market liquidity risk constraints. Serv. Ind. J. 31 (13 &
14), 2193–2221.
Al Janabi, M.A.M., 2011b. Dynamic equity asset allocation with liquidity-adjusted market
risk criterion: appraisal of efficient and coherent portfolios. J. Asset Manage. 12 (6),
378–394.
Al Janabi, M.A.M., 2012. Optimal commodity asset allocation with a coherent market risk
modeling. Rev. Financ. Econ. 21 (3), 131–140.
Al Janabi, M.A.M., 2013. Optimal and coherent economic-capital structures: evidence
from long and short-sales trading positions under illiquid market perspectives. Ann.
Oper. Res. 205 (1), 109–139.
Almgren, R., Chriss, N., 1999. Optimal execution of portfolio transaction. Working Paper.
Department of Mathematics, University of Chicago.
Amihud, Y., Mendelson, H., Pedersen, L.H., 2005. Liquidity and asset prices. Found. Trends
Financ. 1 (4), 269–364.
Artzner, P., Delbaen, F., Eber, J.-m, Heath, D., 1999. Coherent measures of risk. Math.
Financ. 9 (3), 203–228.
Bangia, A., Diebold, F., Schuermann, T., Stroughair, J., 1999. Modeling liquidity risk with
implications for traditional market risk measurement and management. Working
Paper. The Wharton School, University of Pennsylvania.
Best, M.J., Grauer, R.R., 1991. On the sensitivity of mean–variance-efficient portfolios to
changes in asset means: some analytical and computational results. Rev. Financ.
Stud. 4, 315–342.
Blake, D., Lehmann, B., Timmerman, A., 1999. Asset allocation dynamics and pension fund
performance. J. Bus. 72, 429–461.
Bollerslev, T., 1986. Generalized autoregressive conditional heteroskedasticity. J. Econ. 31
(1), 307–327.
Cain, B., Zurbruegg, R., 2010. Can switching between risk measures lead to better portfolio
optimization? J. Asset Manage. 10 (6), 358–369.
Cochrane, J.H., 2005. Asset Pricing. Princeton University Press, Princeton, New Jersey.
Elton, E., Gruber, M.J., 1997. Modern portfolio theory, 1950 to date. J. Bank. Financ. 21
(11–12), 1743–1759.
Elton, E., Gruber, M.J., Padberg, M.W., 1977. Simple rules for optimal portfolio selection:
the multi group case. J. Financ. Quant. Anal. 12 (3), 329–345.
Engle, R.F., 1982. Autoregressive conditional heteroskedasticity. Econometrica 50 (1),
987–1008.
Fabozzi, F.J., Focardi, S., Kolm, P., 2006. Incorporating trading strategies in the Black–
Litterman framework. J. Trading 28–37 (Spring Issue).
Gaivoronski, A., Pflug, G., 2005. Value-at-risk in portfolio optimization: properties and
computational approach. J. Risk 7 (2), 1–31.
381
Ho, L.-c, Cadle, J., Theobald, M., 2008. Portfolio selection in an expected shortfall frame-
work during the recent ‘credit crunch’ period. J. Asset Manage. 9 (2), 121–137.
Jarrow, R., Subramanian, A., 1997. Mopping up liquidity. Risk 10 (12), 170–173.
Jobson, J.D., Korkie, B.M., 1981. Putting Markowitz theory to work. J. Portf. Manag. 7,
70–74.
Jorion, P., 1991. Bayesian and CAPM estimators of the means: implications for portfolio
selection. J. Bank. Financ. 15, 717–727.
Kaplanski, G., Kroll, Y., 2002. VaR risk measures versus traditional risk measures: an
analysis and survey. J. Risk 4 (3), 1–27.
Mandelbrot, B., 1963. The variation of certain speculative prices. J. Bus. 36, 394–419.
Markowitz, H., 1952. Portfolio selection. J. Financ. 7 (1), 77–91.
Meucci, A., 2009. Managing diversification. Risk 22 (5), 74–79.
Michaud, R.O., 1989. The Markowitz optimization enigma: is ‘optimized’ optimal? Financ.
Anal. J. 45 (1), 31–42.
Neftci, Salih F., 2000. Value at risk calculations, extreme events and tail estimations.
J. Deriv. 23–37 (Spring Issue).
Poon, S.-H., Granger, C., 2005. Practical issues in forecasting volatility. Financ. Anal. J. 61
(1), 45–56.
Rockafellar, R.T., Uryasev, S., 2000. Optimization of conditional value-at-risk. J. Risk 2 (3),
21–41.
Rockafellar, R.T., Uryasev, S., 2002. Conditional value-at-risk for general loss distributions.
J. Bank. Financ. 26 (7), 1443–1471.
Saunders, A., Cornett, M., 2008. Financial Institutions Management, 6th edition. McGraw
Hill International Edition, Singapore.
Scholes, M., Kimner, T., 2010. Evolving from quantitative risk management to a high-
performance risk management analytic framework. White Paper. SAS Institute Inc.
Takahashi, D., Alexander, S., 2002. Illiquid alternative asset fund modeling. J. Portf. Manag.
28 (2), 90–100.
Xia, Y., Liu, B., Wang, S., Lai, K.K., 2000. A model for portfolio selection with order of
expected returns. Comput. Oper. Res. 27 (5), 409–422.
Mazin A.M. Al Janabi earned his Ph.D. from the University of London, UK, and currently is a
fulltime Professor of Finance & Banking at the College of Business and Economics (CBE),
United Arab Emirates University, Al-Ain, UAE. He has several years of real world experi-
ence in financial markets and the banking sectors where he held a number of senior posi-
tions, such as Head of Trading of Financial Derivative Products, Head of Trading Risk
Management, Director of Asset and Liability Management and Director of Global Market
Risk Management. He has written extensively, in leading scholarly journals, on finance &
banking and contemporary topics in trading, market and credit risk management as well
as on strategic asset allocation and modern portfolio management. His research interests
and consulting activities address practitioners and regulatory issues in finance & banking,
financial risk management, derivative securities and portfolio management. His research
has appeared in the: International Review of Financial Analysis, Annals of Operations
Research, Review of Financial Economics, Journal of Asset Management, Service Industries
Journal, Studies in Economics and Finance, Emerging Markets Finance and Trade, Review of
Middle East Economics and Finance, Journal of Emerging Market Finance, Middle East Devel-
opment Journal, Journal of Economic Research, Journal of Risk Finance, Journal of Financial
Regulation and Compliance, Journal of Banking Regulation, Journal of Derivatives & Hedge
Funds, and Annals of Nuclear Energy, among others.