Portfolio Construction by Using Different Risk Models:

A Comparison among Diverse Economic Scenarios

Abstract
We aim to construct portfolios by employing different risk models and compare their performance in order
to understand their appropriateness for effective portfolio management for investors. Mean variance (MV),
semi variance (SV), mean absolute deviation (MaD) and conditional value at risk (CVaR) are considered
as risk measures. The price data were extracted from the Pakistan stock exchange, Bombay stock
exchange and Dhaka stock exchange under diverse economic conditions such as crisis, recovery and
growth. We take the average of GDP of the selected period of each country as a cut-off point to make
three economic scenarios. We use 40 stocks from the Pakistan stock exchange, 92 stocks from the
Bombay stock exchange and 30 stocks from the Dhaka stock exchange. We compute optimal weights
using global minimum variance portfolio (GMVP) for all stocks to construct optimal portfolios and analyze
the data by using MV, SV, MaD and CVaR models for each subperiod. We find that CVaR (95%) gives
better results in each scenario for all three countries and performance of portfolios is inconsistent in
different scenarios.
Keywords: 
mean variance; semi variance; mean absolute deviations; conditional value at risk; stock
market; portfolio construction

1. Introduction
Portfolio construction is considered to be one of the main concerns in securities investment as the
future returns are uncertain. Portfolio selection is an ex-ant decision procedure. Despite the fact that risk
is inevitable in every aspect of life, undertaking strategies to minimize risk has become an essential part
of life. In finance, the risk is the possibility of deviation identified with the inconstancy of future returns
(Artzner et al. 1999). The role of proper risk management of a portfolio of financial assets or securities
has been recognized in the literature. Thus, the concerns relate back to the era of  Markowitz’s (1959)
portfolio selection model which is the foundational framework of portfolio selection with optimal risk-return
trade-off. In the last decade, the development of trading activities has provided new methods of
underscoring the necessity for market participants to manage the risk.
Markowitz (1952) establishes a quantitative paradigm, leading to the first ever mean risk models to
incorporate different classes of assets into a portfolio. He further suggests historical variance as a
measure of risk. Moreover, Markowitz utilizes this in a more computational model by measuring portfolio
risk by the covariance framework identified with the individual asset returns. On occasion, variance is
criticized as a measure of risk, most commonly for its deliberate nature and for the irregularity with
aphoristic models for decision (semi variance, mean absolute deviation and stochastic predominance).
This motivates the development of alternative risk measures, and Markowitz’s mean variance (MV) model
worked as the premise for numerous other models which utilize its essential assumptions (Bodie et al.
2011; Elton et al. 2007).
Efficient risk measurement strategies punish attractive (upside) deviations from the mean same way
as they punish undesirable (drawback) deviations. Owing to this criticism, Markowitz (1959) proposes the
semi variance as a risk estimation method. The semi variance gives a measure of the variance of the
returns that remains below the normal return. As such, rational investors seem more responsive towards
downside deviations in relation to upside deviations. Hence, SV is a preferred risk indicator (Chow and
Denning 1994; Grootveld and Hallerbach 1999; Markowitz et al. 1993). Konno and Yamazaki (1991)
recommend the mean absolute deviation (MaD) model as another risk evaluation measure. Unlike, the
mean variance (MV) model, which is a quadratic model, the MaD model is a linear programming model
that generally can handle portfolios of fewer assets, and it gives more efficient results than the MV model
(Liu and Gao 2006). The conditional value at risk (CVaR) model (also identifies as a mean excess loss,
tail VAR) was proposed by Rockafellar and Uryasev (2000). It can be described as the conditional loss
estimation over that sum α at a predefined likelihood level β (Lim et al. 2011). This technique fulfills all
premises of coherent risk.
The approaches adopted by portfolio managers for optimizing portfolios can be classified into two
broad groups. First, managers opted for an active strategy attempt to outperform the market by
generating portfolios that are designed in a more unique manner than the index. The goal is to achieve a
higher return whilst assuming lower or same risk as that of market, or to assume lower risk of the higher
or same returns as that of the market. However, index tracking is a passive management strategy.
Tracking an index of the stock market involves generating a portfolio that simulates the index’s
performance. The portfolio generated is referred to as a tracking portfolio, and the tracked index is
sometimes referred to as a benchmark (García et al. 2018). However, passive managers tacitly presume
that it is not possible to outperform the risk–return combo of the market. The solution to the issue is
therefore distinct from that applied by Markowitz’s MV model, where the construction of the portfolio is
determined by an optimization model method. However, managers using passive strategies tend to lower
the level of adjustments to their portfolios and perform tasks only to redesign them in order to comply with
the index’s composition. Portfolios created using the MV model or index tracking vary considerably. The
global portfolio risk is mitigated in the mean variance model, while only the unsystematic risk portion is
mitigated in the index tracking method.
A well-developed stock market ensures a strong economy. Stock markets are volatile in nature and
highly affected by the cost of borrowing. As the cost of borrowing rises, businesses become unstable and
incapable of investing funds further, which in turn has a direct impact on their profit margins. The decline
in the level of profit also causes decline in stock prices, making it difficult for investors to find attractive
investments in a risky stock market environment. The stock market comes as a medium to facilitate the
investors and borrowers by providing a platform to exchange and mobilize funds to different economic
sectors (Jaaman et al. 2013). This study examines the performance of the above-mentioned portfolios
covering the stock markets of three fast-growing emerging markets of South Asia i.e., Pakistan, India and
Bangladesh.
Byrne and Lee (2004) carried out research to affirm the validity of different risk measures in
portfolio construction and made a comparison of parameters. They propose that all risk measures perform
in a diversified manner and have their own merits and demerits. So, the model selection depends not on
the model’s theoretical merits over the other but on the attitude of individual towards risk. This may be the
reason for not using alternative risk measures frequently. The mean variance that is used in portfolio
optimization models may be due to the difficulty in deciding the appropriate risk measure and continuous
comparison of different risks measures until a common risk measure is brought into action. To outsource
this, comparing the portfolio holdings generated by diverse risk measures, instead of the risk return trade-
off, is utilized. The results show that the portfolio structures generated by diverse risk measures vary from
one another. They propose that choosing a model depends highly on the risk attitude rather than the
merits of the model. Having multiple risk modeling alternatives in portfolio optimization, causes various
concerns to arise about their legitimacy, utilization and applicability. Specifically, a question arises
regarding the barriers of the alternative models with respect to the basic presumption of Markowitz and
Todd’s (1987) classical model. Comparatively an analysis is made among the classic Markowitz’s model
and other models that employ Markowitz and Todd’s (1987) classic model formulation. There are plenty
of unresolved questions related to the selection of different risk models by different investors for risk and
return estimation. Do each of these models perform equally well in each different economic scenario? If
one model yields better in growth days, does it yield the same results in declining days?
Stock market crises cause investors to be unable to recover their original investments in the stock
market, real estate business, and gold and currency market. Government and private sector investment
schemes dictate their own behaviors. Moreover, inconsistent policies and strict regulations add fuel to fire.
Later on, in money-related crisis, investors have taken a couple of fundamental measures in order to
achieve maximum possible return. These measures have driven theorists to give cautious to managing
the risk. With these necessities it has become a need of every investor to assess the performance of
portfolios under different risk models in different economic scenarios. South Asian emerging markets are
an important region to study. For instance, Hunjra et al. (2020) studied profitability of momentum and
contrarian tactics incorporating sample of South Asian countries. Given the economic and financial growth
in the region over the last two decades (South Asian Financial Market Review 2010), it is an important
topic for research. It is of great significance for investors in emerging South Asian countries (Pakistan,
Bangladesh and India) to make suitable investment decisions on the performance of portfolios. In this
study, analysis of efficient portfolios is conducted by utilizing four different risk measures; the mean
variance (MV), semi variance (SV), mean absolute deviation (MaD) and conditional value at risk (CVaR),
under three different scenarios of the economy (crisis, recovery, growth). Further comparison of results is
performed to understand their relative performance as effective portfolio management technique.
To the best of our knowledge, this study is the first in the context of emerging South Asian countries
which compares the portfolio performance constructed by using various risk measures under three
different economic scenarios (crises, recovery and growth). The contribution of the paper is two-fold. (i)
The outcome of the current study helps to identify if the same risk assessment is applicable in all three
substituents or a different risk measure is required for a different sub period. (ii) It provides insights on
asset choice, management and the use of an optimal risk model to estimate expected risk and return. The
implications of our findings may be beneficial for the investors, investment decision makers and policy
makers not only in Pakistan, Bangladesh and India but also those in other emerging economies.
The remainder of the paper is organized as follows. Section 2 offers past studies findings and
hypotheses development; Section 3 presents the data, variables and methodology used for the empirical
analysis. Section 4 shows the empirical results. Section 5 provides recommendations and concludes the
paper.

2. Literature Review
Portfolio optimization is a key idea in valuation theories. Markowitz (1952) started additional work
for building up a scientific way to deal with multi asset portfolios. Markowitz’s (1952) classical model
utilizes the relationship of mean and variance of the returns. Markowitz’s (1952) expected variance of
returns (EV) rule inferred an adequate diversification which relies upon the quantity and diversification of
securities crosswise over assets with a specific end goal to reduce variance. Generally, firms in a similar
industry, as a rule, demonstrate high covariance among themselves as they respond like economic
occasions which prompt to a higher risk presentation. Thus, an investor prefers a MV effective portfolio
when a given risk level gives the maximum expected return (Campbell et al. 1997). With a specific end
goal to decrease or to eradicate the unsystematic risk, the equity allocation must demonstrate an optimal
allocation of uncorrelated stocks.
Artzner et al. (1999) introduce the idea of a coherence asset of desired properties for risk measures
associated with the tail of distribution. CVaR is the coherent risk indicator, also called expected shortfall
by few authors (Konno et al. 2002; Rockafellar and Uryasev 2000). Despite the amount of studies on
risk measures and average risk models, there is still controversy as to which risk measure is the most
suitable one. In addition, it continued to challenge the theoretical validity of the mean risk models.
Regardless of the number of studies on risk indicators, the topic of which risk measure performs well is
still questionable. The Markowitz efficient frontiers (Markowitz 1952) explain how investors select their
portfolios maximizing return for a specified amount of risk as measured by the return’s variance. A key
understanding of this work is that assets should not be chosen exclusively in light of only their exceptional
attributes of return and variance. Market specialists need to consider the connection of the portfolio
having the same return with lower risk than a portfolio that overlooked these connections (Elton et al.
1997).
Since the early work by Markowitz (1952), optimization models aimed at managing the exchange
between risk and return in finance have been studied extensively. According to  Filippi et al. (2020),
shortfall and quantile risk measures are receiving increasing attention where CVaR is probably the most
common of such indicators. Ogryczak and Sliwinski’s (2010) study on CVaR finds that the classical
Markowitz’s model utilizes the variance as the risk measures by providing solution to quadratic
optimization issue. Some other studies presented a few option risks measures which are computationally
alluring (for discrete random factors) for better understanding of straight programming (SP) issues
(Konno and Yamazaki 1991; Rockafellar and Uryasev 2000). The SP reasonability is essential for
applications to genuine financial choices where well-developed portfolios need to meet various limitations
and consider exchange costs. A range of SP-utilizable risk measures have been introduced in the
portfolio construction, albeit the vast majority of them are identified with the outright Lorenz bend and
accordingly, the CVaR measures. Ogryczak and Sliwinski (2010) demonstrate that all risk measures are
utilized as a part of the SP feasible portfolio improvement models can be taken from the setback
criteria. Gluzicka (2010) conducts a research on CVaR and Gini’s Mean Difference (GMD) models. GMD
is analyzed for different measures of risk such as: variance, semi variance, VaR or CVaR. Different
measures of risk are centered on different properties of distribution of returns. For example, the variance
measures the deviation of return and the VaR or CVaR measures the possible loss. Gluzicka (2010)
studies risk models with two measures (mean and variance) applied for different risk portfolios. He finds
that the CVaR and SV models provide better estimates than MV model. When making comparisons of
outcomes from all models it can be mentioned that the models with two measures of risk most often
provide better outcomes than the models with only one risk measure (Gluzicka 2010).
Vidovic (2011) conducts a study by characterizing various risk indicators in portfolio composition on
seven European stock markets. He identifies that extreme kurtosis and skewness exist in stock returns.
Consequently, portfolios employed stocks with high kurtosis and stocks with depreciative
skewness. Banihashemi and Navidi (2017) employee CVaR as a risk measure using various confidence
level. They compare two measures of risk such as VaR and CVaR to identify the best one for portfolio
optimization. They reveal that CVaR is more accurate than VaR. Further, they indicate that high
confidence levels are more accurate than low levels. Jaaman et al. (2013) conducted a study and
compared the portfolio construction and performance based on different risk measures using Malaysian
Stock returns data in three diverse economic criterions. The study shows multiple optimal portfolios’
construction and performances for the three subperiods. However, among the risk models studied, CVaR
model gave the peak portfolio skewness and is the most optimal portfolio optimization model. Based on
the above discussion, it can be said that CVaR model significantly impacts the portfolio performance and
this performance is consistent in all scenarios. We can hypothesize it as:
Hypothesis 1 (H1). 
Conditional value at risk (CVAR) is significantly related to portfolio performance.
Hypothesis 2 (H2). 
Performance of conditional value at risk (CVAR) is consistent in each economic scenario.
It is evident from the literature that portfolios have different risk criteria so they cannot be easily
compared. The reason behind this is that portfolios have different construction, so their performance is
also different from one another. This fact has also been proved by Moazeni et al. (2014). All portfolios
have different risk criteria, construction and also performance. Soleimani and Govindan (2014) identify
that supply chain risk shows better presentations with mean-CVaR models. Baghdadabad (2015)
conducts a study to assess the risk accustomed performance of global mutual assets utilizing the risk
measurement created by the mean absolute deviation (MaD).
Henriques and Neves (2019) present portfolio optimization framework that considers investment
decision under various assumptions of risk. Their findings reflect the risk and return trade-off. In addition,
they argue that investors who are less sensitive to risk find the formulation based on mean absolute
deviation (MaD) more attractive as it generally offers lower returns variance. According to the study
of Baghdadabad (2015), MaD is one of the important determinants to evaluate the portfolio performance.
Moreover, portfolio managers can use the results anywhere in the global stock market. Based on their
findings, we hypothesize as follows:
Hypothesis 3 (H3). 
Mean absolute deviation (MaD) is significantly related to portfolio performance.
Hypothesis 4 (H4). 
Performance of mean absolute deviation (MaD) is consistent in each economic scenario.
As discussed in the literature, different risk measures have different structures, so their performance
also differs from each other. In order to test the performance of different risk measures with the objective
of increasing the returns while limiting risk, Gökgöz and Atmaca (2017) examine the different risk
measures including mean-variance and semi variance for development in power sectors. According to
them, these two models could be associated with all sectors to make efficient points of confinement,
diversified portfolios and utility limits. They find that different optimal portfolios offer different return levels
to investors with different risk avoidance levels. Recent studies have seen a growing use of the portfolio
optimization approach centered on risk/diversification. Mean-diversification efficient frontiers are taken for
different diversification measures and portfolios are identified that optimize the risk and return trade-off
(Oyenubi 2020). Oyenubi (2020) further claims that these portfolios are better options than the portfolio
that is designed to be optimal for mean variance, as they are sensitive to returns. Various measures
support the returns while limiting the related market risks. We develop the following hypotheses for our
study.
Hypothesis 5 (H5). 
Mean variance (MV) is significantly related to portfolio performance.
Hypothesis 6 (H6). 
Semi variance (SV) is significantly related to portfolio performance.
Hypothesis 7 (H7). 
Performance of mean variance (MV) is consistent in each economic scenario.
Hypothesis 8 (H8). 
Performance of semi variance (SV) is consistent in each economic scenario.

3. Methodology
Our sample included 40 stocks from Pakistan stock exchange, 92 Stocks from Bombay stock
exchange and 30 stocks from Dhaka stock exchange. Summary statistics of all stocks of Pakistan,
Bombay and Dhaka stock exchange is given in Table A1, Table A2 and Table A3. Based on the
availability of complete data, the sample period of our study was from 2003 to 2015. Further, the study
period was also of particular significance as it covers before, during and after the financial crisis 2008–09.
The global financial tsunami of 2008 has been recognized as one of the century’s greatest catastrophes.
It has posed many challenges to world social and economic development. However, studies revealed
mixed outcomes in terms of impact of GFC on South Asian stock markets (Long et al. 2012; Rastogi
2014). Thus, we are also interested in knowing how the emerging South Asian stock markets performs
before, during and after the crisis. The data were comprised of monthly closing prices for every security
from January 2003 to the end of December 2015. Only those stocks that were indexed before 2003 and
whose complete data were accessible were included. Stocks were extracted from the national stock index
which incorporates top performing stocks. Then, we computed a natural logarithm of monthly returns for
every stock an individual considered. We employed the global minimum variance portfolio (GMVP)
method for optimal portfolio composition (Engle and Kelly 2012). We computed the global minimum
variance portfolio weights by computing covariance matrix of return series. We calculated optimal weights
for all our stocks under each scenario for Pakistan stock exchange, Dhaka stock exchange and Bombay
stock exchange. We analyzed the data by using MV, SV, MaD and CVaR models for each subperiod.
We collected data of the annual GDP from the World Development Indicator (WDI) of selected
countries from 2003 to 2015. Annual GDP was converted into monthly GDP by using Eviews. Conversion
of annual data is widely accepted and has been applied in large number of studies such as  Tang and
Chua (2012) and Shahbaz et al. (2014). Following Jaaman et al. (2013), for each country, average of
GDP for the selected period was calculated and taken as the cut-off point to make three scenarios; crisis,
recovery and growth. The crisis, recovery and growth periods were different for different countries as
GDP varied significantly, depending on a variety of factors such as the scale of the economy, the degree
of trade openness and the potential for countercyclical policies (Keat 2009). For Pakistan, data from 2003
to 2005 were used as crisis, 2006 to 2011 as recovery and 2012 to 2015 as growth period. For
Bangladesh, data from 2006 to 2009 were used as decline, 2010 to 2013 as recovery and 2014 to 2015
as growth period. For India, data from 2004 to 2007 were used as decline, 2008 to 2011 as recovery and
2012 to 2015 as growth period. Each risk measure was applied on a selected country under three
different economic scenarios (crises, recovery and growth). As the studies completed in various time
periods probably offer various compositions of optimal portfolios (Prakash et al. 2003). The result for
each case is presented below in next section.
The models used for estimation are as follows:
The MV model calculates the variation below and above the mean. It not only penalizes the
downward deviations but upstream deviations too (Markowitz 1952).
𝑉=∑𝑛𝑖=1∑𝑛𝑗=1𝜎𝑖𝑗𝑥𝑖𝑥𝑗�=∑�=1�∑�=1��������

(1)
where 𝜎𝑖𝑗��� is the covariance between assets i and j, 𝑥𝑗�� is the amount invested in asset j.
 Markowitz (1959) proposes semi variance (SV) which centers only on the deviations below the
mean. This model penalizes only the downstream deviations not the upward deviations. The semi
variance model is presented as:
𝑆𝑉=1𝑇∑𝜏𝑖=1(𝑚𝑎𝑥[0,𝐸(𝑅)−𝑅𝑝𝑡])2,��=1�∑�=1�(���[0,�(�)−���])2,

(2)
where t is used as the number of periods, 𝑅𝑝𝑡��� is the portfolio return at
period t, 𝐸(𝑅𝑝𝑡)�(���) is the mean return.
The mean absolute deviation (MaD) model is proposed by Konno and Yamazaki (1991). In this
model absolute deviation is used as risk measure to replace variance.
𝑦𝑡+∑𝑛𝑗=1(𝑟𝑗𝑡−𝑟𝑗)𝑥𝑗 ≥ 0,  𝑡= 1, …, 𝑇��+∑�=1�(���−��)�� ≥ 0,  �= 1, …, �

(3)
𝑦𝑡−∑𝑛𝑗=1(𝑗𝑡−𝑟𝑗)𝑥𝑗 ≥ 0,  𝑡= 1, …, 𝑇��−∑�=1�(��−��)�� ≥ 0,  �= 1, …, �

(4)
where 𝑦𝑡�� is the continuous variable that represents the deviations among the mean return of the
portfolio as well as return of portfolio at time t, T is the number of period, 𝑟𝑗𝑡��� is the realization of
random variable 𝑟𝑗�� during period t, 𝑟𝑗�� is the expected return of asset 𝑗� per period, 𝜌� is a
parameter indicating the minimal rate of return required by an investor and 𝑥𝑗�� is the amount invested
in asset 𝑗�.
The conditional value at risk (CVaR) model is presented by Rockafellar and Uryasev (2000) and is
also called the mean excess loss, mean shortfall or tail VaR. The CVaR is defined as the conditional
expectation of loss beyond that amount α at a given probability level β (Lim et al. 2011).
CVaR=𝛼+1𝑇(1−𝛽)∑′𝑡=1𝑧𝑡,CVaR=�+1�(1−�)∑�=1′��,

(5)
where 𝛼� is the minimal amount of loss, 𝛽� is the probability that the loss will not
surpass 𝛼�, 𝑇� is the number of period, 𝑧𝑡�� is the variable, 𝑟𝑗𝑡��� is the realization of random
variable 𝑅𝑗�� during period t, 𝑥𝑗�� is the amount invested in asset 𝑗� and 𝑟𝑗�� is the expected return
of asset 𝑗�. In this study, we followed this framework with the selected risk confidence levels β = 0.99
and β = 0.95. Investors seldom experience a loss exceeding VaR(X) when β is 0.99 as compared with
0.95. CVaR (0.99) means investors have higher downstream risk aversion than CVaR (0.95).

4. Results
Table 1, Table 2, Table 3, Table 4, Table 5, Table 6, Table 7, Table 8 and Table 9 report optimal
portfolio compositions for Pakistan (Table 1, Table 2 and Table 3), Bangladesh (Table 4, Table
5 and Table 6) and India (Table 7, Table 8 and Table 9) under economic status of crisis, recovery and
growth.
Table 1. Optimal portfolio compositions for economic crisis period (2003–2005).

Table 2. Optimal portfolio compositions for economic recovery period (2006–2011).

Table 4. Optimal portfolio compositions for economic crisis period (2006–2009).

Table 5. Optimal portfolio compositions for economic recovery period (2010–2013).

Table 6. Optimal portfolio compositions for economic growth period (2014–2015).

Table 7. Optimal portfolio compositions for economic crisis period (2004–2007).

Table 8. Optimal portfolio compositions for economic recovery (2008–2011).

Table 9. Optimal portfolio compositions for economic growth period (2012–2015).

Markowitz (1952) describes that finding the right mix of stocks (risky assets) in terms of expected
return and variance of return, that is, an optimal portfolio in terms of yield and risk, is the greatest
challenge for an investor. A simple principle for the optimal stock composition is that the highest returning
portfolio is not automatically the lowest risk (variance) portfolio. We use four risk measures in every
scenario and their results are quite different from each other. These models are utilized to develop ideal
portfolio from 40 accessible firms listed on the Pakistani stock exchange.
Choosing a model depends highly on the risk attitude of an investor. However, rather than being risk
averse or a risk taker, he/she may choose an optimal portfolio with low risk and high returns. Table 1,
presenting optimal portfolio composition for economic crises, explains that MCB make up the maximum
percentage in optimal portfolios respectively for all four measures. Furthermore, it explains that MCB
invested the most where investors invest 36.41% of funds in MCB if he/she chooses to employ the MV
model, 36.79% funds are invested for SV model, 22.17% if MaD model is used, 40.60% of funds are used
for employing CVaR (95%) and 9.89% of funds are invested when using CVaR (99%) model.
According to Pakistan’s annual GDP recovery rate, this scenario is built where four risk models
yields distinctive optimal portfolios for economic period of recovery. It is found that when investors are
willing to take more risk, either the expected return of a portfolio increases, or a risk averse investor is
able to decrease the variance in trade for a lower expected return. From Table 2 it is assumed that during
economic recovery, PPL makes up the highest proportion of optimal portfolio. If an investor opts for an
MV model, he/she invests up to 20.49% in aforesaid security. He/she invests 5.14% of funds when using
an SV model, 20.95% for MaD, 25.27% for CVaR (95%) and 13.81% of investment for the CVaR (99%)
model.
It is implied that ample diversification is based on the quantity of shares, and volatility is minimized
by diversification across sectors and industries. Companies in the same sector typically display high
covariance between themselves in general as they respond similarly to economic events that lead to
greater risk exposure. Consequently, when a prescribed level of risk earns the maximum possible
expected return, an investor carries an MV-efficient portfolio (Campbell et al. 1997). A financial specialist
can decrease the risk of his investment by diversifying it over various stocks. From Table 3, it is examined
that during the economic growth period, OGDCL invested the most where it is deduced that an investor
puts 8.22% of funds in OGDCL if he/she chooses to opt for the MV model, 6.68% if SV model is
employed, 12.21% for MaD model, 67.52% if CVaR (95%) model is chosen and 12.01% for CVaR (99%).
It can also be examined that CVaR (99%) provides negative returns for all securities except OGDCL
during this period.
If we summarize the results for the Pakistan stock exchange during the periods of crises, recovery
and growth, it can be said that MCB made the highest proportion in optimal portfolio in the crisis period
and in the case of recovery and growth, the highest proportion lies with PPL and OGDCL, respectively.
The results are similar for all types of risk measures. It shows that these securities are able to attract
more investment in any scenario.
Literature stats that portfolio theory affects the management of money and other assets worldwide.
From Table 4, it is concluded that during economic crises in Dhaka Stock Exchange, Titas and Dist. Co.,
Ltd. remains the riskiest security in the optimal portfolio for all risk measures. It makes up the highest
proportion in optimal portfolio. Investors invest 45.65% of funds in Titas and Dist. Co., Ltd. if they opt for
the MV model, 17.95% investments when using the SV model, 21.92% for employing the MaD Model,
3.25% funds are invested in the case of CVaR (95%) and 6.03% funds are invested if investors choose
the CVaR (99%) model.
By analyzing all the optimal portfolios from Table 5, it is stated that during the time of economic
recovery, Pharmaceuticals Ltd. comprises the highest risk. By considering the fact that high risk leads to
high return, investors tends to invest 4.54% of their funds into this particular security while they choose
the MV risk model. This proportion is 1.54% for using the SV model, 86.36% for the MaD model and
19.26% for CVaR (95%). It can also be examined that CVaR (99%) provides negative returns for all
securities during this period.
This research reveals that the multiperiod issue can be resolved as a series of single-period issues
under many sets of rational assumptions. However, if only one period is analyzed, the optimal portfolio
will be different from that which is chosen. The distinction occurs because in the multiperiod case, the
relevant utility function is a derived that takes multiple periods into account. Independence of returns
between periods is one principle behind multiperiod portfolio research. In the last decade, there has been
considerable research showing that mean returns and variances are correlated over time and are
components of variables that are easily noticeable (Fama and French 1989; Campbell and Shiller
1988).
In Table 6, optimal portfolios are created for economic period of growth for Bangladesh market. This
table shows that Pubali Bank comprises the highest risk. In order to receive high returns, investors are
willing to invest in risky securities. It can be deduced that investors tend to invest 17.93% of their funds
into this particular security while they choose the MV risk model. This proportion is 49.21% for using the
SV model, 212% for employing the CVaR (95%) model and 4.96% funds are invested if investors opt for
the CVaR (99%) model for optimal portfolio selection. However, if they choose the MaD model, the
highest proportion is made by Jamuna Oil Com. Ltd.
From the above discussion, it can be concluded that for each of the scenario, top risky securities
continue to change. Investors choose securities for investment by keeping the nature of economic
condition in mind. They also choose securities depending on their investing attitude. Risk averse investors
choose less risky securities and hence get low profits/returns. Whereas, risk takers invest in risky
securities to get high returns. They prefer high returns, so they are more likely to choose the optimal
portfolio. Diversification can keep them safe from risks so they may choose different securities in single
scenario.
Table 7 presents results for the Bombay stock exchange about portfolio composition in the time of
economic crises. The results reveal that investors invest 45.23% of their funds in Titan Company Ltd. if
they choose the MV model, 46.44% if they choose the SV model, 35.79% investment is for the MaD
model and this percentage is about 52.32% in the case of choosing CVaR (95%). However, if they
choose CVaR (99%), they invest 9.07% of their funds in HDFC Bank Ltd.
The four risk models build distinctive ideal portfolio creations for the economic period of recovery.
From Table 8 it is deduced that during the time of economic recovery, Tata Global Beverages Ltd. gives
up the highest risk level in the optimal portfolios for all risk indicators. This particular security is the riskiest
so the returns for investing in this security under different risk measures are also high. In order to receive
high returns, investors invest 20.44% of their funds in Tata Global Beverages Ltd. if they choose the MV
model, 44.86% is invested for the SV model, 19.64% for using the MaD model, 64.53% of funds are
allocated if the CVaR (95%) model is selected and 13.23% for CVaR (99%).
For Bombay stock exchange, these four risk measures again give distinctive portfolios creations for
each of the scenario. From Table 9 it is deduced that during the time of economic growth, Asian Paints
Ltd. remains the riskiest security in the optimal portfolios for all risk measures except when the CVaR
(95%) and CVaR (99%) models are employed. For these two models, the riskiest security is State Bank of
India. From the above results, it is concluded that with the change of economic conditions, investor’s
investment decisions also change. In the decline, recovery and growth phase, their motive is to invest in
those securities which generate more returns. They also prefer to invest in more than one security for
minimizing risks. For this purpose, they are more inclined towards choosing the optimal portfolio.
From the above tables, it is evident that the top risky securities, in different scenarios, keeps on
changing. It is not necessary that a security which is risky in the period of crisis stays risky in recovery
and growth period. So, it totally depends on the investor’s attitude, portfolio type and nature of economy
(or economic scenario) that which the security (or group of securities) is to be selected for the portfolio.
The risks associated with security also impact the decision. Similarly, risk measures also have direct
influence on choosing the optimal structure as each model does not perform equally well in each
economic scenario. In addition, a model performing better in crises phases does not necessarily yield the
same results in recovery or growth period.
The summary statistics of the optimal portfolios in three different economic sub periods of Pakistan,
Bangladesh and India are shown in Appendix A. For Pakistan, the average rate of return calculated for
the period is 9.31%, which is the average risk-free rate from the period 2003 to 2015. This average rate of
return is not achieved for all models in each scenario. This implies all portfolios developed are not ready
to produce returns at any rate identical to the risk-free return. CVaR (95%) gives the best results for all
optimal portfolios in each scenario of Pakistan. The results about mean returns are highest during the
growth time period (2012–2015) for portfolios created by the CVaR (95%), MV, MaD and SV models.
However, CVaR (99%) yields low or negative returns. From risk perspectives, as appeared by the
portfolios’ variance and skewness during the economic crisis investors bring more risk. Model CVaR
(95%) provides the highest portfolio skewness during recovery period. Positive skewness is desirable as it
decreases the probability of getting extreme negative returns. Results are consistent with each scenario.
Hence, the CVaR (95%) is the most appropriate risk measure to control downside risk. Comparison of
CVaR (95%) and CVaR (99%) is also shown in Appendix B.
For Bangladesh, the average required rate of return calculated for the period is 5.75%. Again, this
average rate of return is not achieved for all models in each scenario. During the recovery period (2010–
2013), mean returns are highest for all portfolios created by each of these four risk models. However, the
portfolios developed by the CVaR (95%) model provides the maximum returns. Results with respect to
skewness indicates that the CVaR (95%) model provides the highest skewness during the economic
growth period. Results are consistent as model CVaR (95%) yields outstanding results in each scenario
for the Dhaka stock exchange. These findings suggest that the MV, SV and MAD models are unable to
develop optimal portfolios with maximum returns in each scenario. Thus, The CVaR (95%) model is the
most suitable portfolio optimization model for all three economic scenarios.
For India, the minimum required rate of return for the period is 7.89%, which is the average risk-free
rate from the period 2004 to 2015. Again, this average rate of return is not achieved for all models, but
crisis and growth periods yield return more than their average in specific scenario. Results of the Bombay
stock exchange show that mean returns are highest during the crises time period (2004–2007) for all
portfolios produced by these four risk models. Portfolios that are built by utilizing CVaR (95%)
demonstrate the highest returns. From risk perspectives, as appeared by the portfolios’ variances and
skewness during the economic crisis’ economic specialists bring about more risks. Portfolios that are built
by utilizing the CVaR (95%) model also gives the highest skewness during periods of economic crisis. For
the Indian market the results are also consistent because CVaR (95%) gives outstanding results in each
economic scenario but the trend is different from the country perspective. Similar to the other two
countries, CVaR (99%) does not perform likewise for this market but CVaR (95%) yields outstanding
results.
The merits and demerits of each risk measure are evident by the results. H1, H3, H5 and H6 are
accepted because each of the risk measure has its impact on the portfolio performance. However, their
performance is not consistent in each period, so H4, H7 and H8 are not supported. H2 is also supported
because the performance of CVaR is consistent in each scenario for all three countries. This study is
evident the prior studies done by Jaaman et al. (2013) that variance is not a proper risk measure for all
markets and for all economic scenarios. The CVaR is fundamentally the downward risk that particularly
concentrates on the descending deviations under the mean return. CVaR is a superior coordinating
choice for economic specialist’s recognition that is risk averse.

5. Conclusions
This study concludes that for Pakistan, all risk measures perform better in the growth phase, for
Bangladesh they perform better in the recovery phase and for India they show better performance in the
crises phase. We found that the CVaR model gave a better result in each scenario. Thereby, investors
are recommended to use this model for their portfolio construction. Public and private investors are
required to re-evaluate their investment portfolio to improve their investment skills. The contribution of the
paper is two-fold. (i) The outcome of the current study helps to identify if the same risk assessment is
applicable in all three substituent or different risk measure is required for a different sub period. (ii) It
provides an insight on asset choice, management and the use of an optimal risk model to estimate
expected risk and return. The implications of our findings would be beneficial for the investors, investment
decision makers and policy makers not only in Pakistan, Bangladesh and India, but also those in other
developing countries. Our study has some limitations. Firstly, it only covered three developing countries. It
will be of great importance to conduct a study to make a comparison between developing and developed
economies so that it can further validate the consistencies of these risk models. Secondly, this study
focuses on four models (MV, SV, MaD, CVaR), future studies may test the applicability of other risk
models such as the mean Gini (MG) model, second degree stochastic dominance (SSD) and other
quadratic models. Furthermore, future research may back-test risk measures such as CVaR as
conducted by Amendola and Candila (2017). Moreover, our study only considered GDP to identify
different economic scenarios, and therefore some other criteria may be used to identify the economic
scenario. Researchers can use other macroeconomic indicators such as interest rates, inflation, gross
national product (GNP) and foreign direct investment (FDI) instead of GDP as criteria to identify different
economic scenario.

Author Contributions
Conceptualization, U.S., M.H., S.C. and A.I.H.; methodology, A.I.H., M.H. and U.S.; software, U.S.
and M.H.; validation, S.M.A. and S.C.; formal analysis, U.S., M.H. and A.I.H.; resources, S.M.A. and S.C.;
data curation, S.M.A.; writing—original draft preparation, U.S., M.H. and A.I.H.; writing—review and
editing, S.C. and S.M.A.; supervision, A.I.H. and S.M.A.; project administration, S.C. and A.I.H. All
authors have read and agreed to the published version of the manuscript.
Funding
This research received no external funding.

Conflicts of Interest
The authors declare no conflict of interest.

Appendix A
Table A1. Summary Statistics of Optimal Portfolios.

Table A2. Summary Statistics of Pakistan Stock Exchange.

Table A3. Summary Statistics of Bombay Stock Exchange.

Table A4. Summary Statistics of Dhaka Stock Exchange.

Appendix B
Table A5. Comparison of CVaR (95%) and CVaR (99%).

Appendix B presents the comparison of CVaR (95%) and CVaR (99%) by providing values of
mean, variance, t-stat and p-values. CVaR (95%) has higher mean and smaller variances as compared to
CVaR (99%) for all three stock exchanges under each scenario. The values of t-stat are significant for the
Pakistan stock exchange, Dhaka stock exchange and Bombay stock exchange under each scenario.
However, values of t-stat are slightly higher in crisis period for all three stock exchanges as compared to
recovery and growth period. Results indicates that CVaR (95%) is a better measure of risk as compared
to CVaR (99%).
References
1. Amendola, Alessandra, and Vincenzo Candila. 2017. Comparing multivariate volatility forecasts by direct
and indirect approaches. Journal of Risk 19: 6. [Google Scholar] [CrossRef]
2. Artzner, Philippe, Freddy Delbaen, Jean-Marc Eber, and David Heath. 1999. Coherent measures of
risk. Mathematical Finance 9: 203–28. [Google Scholar] [CrossRef]
3. Baghdadabad, Mohammad Reza Tavakoli. 2015. An empirical analysis of funds’ alternative measures in
the mean absolute deviation (MaD) framework. International Journal of Emerging Markets 10: 726–46.
[Google Scholar] [CrossRef]
4. Banihashemi, Shokoofeh, and Sarah Navidi. 2017. Portfolio performance evaluation in Mean-CVaR
framework: A comparison with non-parametric methods value at risk in Mean-VaR analysis. Operations
Research Perspectives 4: 21–28. [Google Scholar] [CrossRef]
5. Bodie, Zvi, Alex Kane, and Alan J. Marcus. 2011. Investment and Portfolio Management. Hong Kong:
McGraw-Hill Irwin. [Google Scholar]
6. Byrne, Peter, and Stephen Lee. 2004. Different risk measures: Different portfolio compositions? Journal of
Property Investment & Finance 22: 501–11. [Google Scholar]
7. Campbell, John Y., and Robert J. Shiller. 1988. Stock prices, earnings, and expected dividends. The
Journal of Finance 43: 661–76. [Google Scholar] [CrossRef]
8. Campbell, John Y., John J. Champbell, John W. Campbell, Andrew W. Lo, Andrew W. Lo, and A. Craig
MacKinlay. 1997. The Econometrics of Financial Markets. Princeton: Princeton University Press. [Google
Scholar]
9. Chow, K. Victor, and Karen C. Denning. 1994. On variance and lower partial moment betas the
equivalence of systematic risk measures. Journal of Business Finance & Accounting 21: 231–41. [Google
Scholar]
10. Elton, Edwin J., Martin J. Gruber, and Christopher R. Blake. 1997. Common Factors in Mutual Fund
Returns. NYU Working Paper No. FIN–97–004. New York, NY, USA: American Finance Association.
[Google Scholar]
11. Elton, Edwin J., Martin J. Gruber, and T. Clifton Green. 2007. The impact of mutual fund family
membership on investor risk. Journal of Financial and Quantitative Analysis 42: 257–77. [Google
Scholar] [CrossRef][Green Version]
12. Engle, Robert, and Bryan Kelly. 2012. Dynamic Equicorrelation. Journal of Business and Economic
Statistics 30: 212–28. [Google Scholar] [CrossRef]
13. Fama, Eugene F., and Kenneth R. French. 1989. Business conditions and expected returns on stocks and
bonds. Journal of Financial Economics 25: 23–49. [Google Scholar] [CrossRef]
14. Filippi, Carlo, Gianfranco Guastaroba, and Maria Grazia Speranza. 2020. Conditional value-at-risk beyond
finance: A survey. International Transactions in Operational Research 27: 1277–319. [Google Scholar]
[CrossRef][Green Version]
15. García, Fernando, Francisco Guijarro, and Javier Oliver. 2018. Index tracking optimization with cardinality
constraint: A performance comparison of genetic algorithms and tabu search heuristics. Neural Computing
and Applications 30: 2625–41. [Google Scholar] [CrossRef]
16. Gluzicka, Agata. 2010. Models for choosing optimal investment portfolios with two risk
measures. Methods and Applications of Operational Research 10: 76–99. [Google Scholar]
17. Gökgöz, Fazıl, and Mete Emin Atmaca. 2017. Portfolio optimization under lower partial moments in
emerging electricity markets: Evidence from Turkey. Renewable and Sustainable Energy Reviews 67: 437–
49. [Google Scholar] [CrossRef]
18. Grootveld, Henk, and Winfried Hallerbach. 1999. Variance vs downside risk: Is there really that much
difference? European Journal of Operational Research 114: 304–19. [Google Scholar] [CrossRef]
19. Henriques, Carla Oliveira, and Maria Elisabete Duarte Neves. 2019. A multi-objective interval portfolio
framework for supporting investor’s preferences under different risk assumptions. Journal of the
Operational Research Society 70: 1639–61. [Google Scholar] [CrossRef]
20. Hunjra, Ahmed Imran, Tahar Tayachi, Rashid Mehmood, Sidra Malik, and Zoya Malik. 2020. Impact of
credit risk on momentum and contrarian strategies: Evidence from South Asian markets. Risks 8: 37.
[Google Scholar] [CrossRef][Green Version]
21. Jaaman, Saiful Hafizah, Weng Hoe Lam, and Zaidi Isa. 2013. Risk measures and portfolio construction in
different economic scenarios. Sains Malaysiana 42: 875–80. [Google Scholar]
22. Keat, Heng S. 2009. The global financial crisis: Impact on Asia and policy challenges ahead. In Federal
Reserve Bank of San Francisco Proceedings. San Francisco: Federal Reserve Bank, pp. 267–76. [Google
Scholar]
23. Konno, Hiroshi, and Hiroaki Yamazaki. 1991. Mean-absolute deviation portfolio optimization model and
its applications to Tokyo stock market. Management Science 37: 519–31. [Google Scholar] [CrossRef]
[Green Version]
24. Konno, Hiroshi, Hayato Waki, and Atsushi Yuuki. 2002. Portfolio Optimization under Lower Partial Risk
Measures. Asia-Pacific Financial Markets 9: 127–40. [Google Scholar] [CrossRef]
25. Lim, Andrew E.B., J. George Shanthikumar, and Gah-Yi Vahn. 2011. Conditional value-at-risk in portfolio
optimization: Coherent but fragile. Operations Research Letter 39: 163–71. [Google Scholar] [CrossRef]
26. Liu, Mingming, and Yan Gao. 2006. An algorithm for portfolio selection in a frictional market. Applied
Mathematics and Computation 182: 1629–38. [Google Scholar] [CrossRef]
27. Long, Wen, Nan Li, Huiwen Wang, and Siwei Cheng. 2012. Impact of US financial crisis on different
countries: Based on the method of functional analysis of variance. Procedia Computer Science 9: 1292–98.
[Google Scholar] [CrossRef]
28. Markowitz, Harry. 1952. Portfolio selection. The Journal of Finance 7: 77–91. [Google Scholar]
29. Markowitz, Harry. 1959. Portfolio Selection: Efficient Diversification of Investment. Cowles Foundation
Monograph No. 16. New York: Wiley. [Google Scholar]
30. Markowitz, Harry M., and G. Peter Todd. 1987. Mean-Variance Analysis in Portfolio Choice and Capital
Markets. Hoboken: John Wiley & Sons. [Google Scholar]
31. Markowitz, Harry, Peter Todd, Ganlin Xu, and Yuji Yamane. 1993. Computation of mean-semi variance
efficient sets by the critical line algorithm. Annals of Operations Research 45: 307–17. [Google Scholar]
[CrossRef]
32. Moazeni, Somayeh, Warren B. Powell, and Amir H. Hajimiragha. 2014. Mean-conditional value-at-risk
optimal energy storage operation in the presence of transaction costs. IEEE Transactions on Power
Systems 30: 1222–32. [Google Scholar] [CrossRef]
33. Ogryczak, Włodzimierz, and Tomasz Sliwinski. 2010. Efficient portfolio optimization with conditional
value at risk. Paper presented at International Multi Conference on Computer Science and Information
Technology, Wisla, Poland, October 18–20; pp. 901–8. [Google Scholar]
34. Oyenubi, Adeola. 2020. Optimal Portfolios on Mean-Diversification Efficient Frontiers. In Advanced
Studies of Financial Technologies and Cryptocurrency Markets. Singapore: Springer, pp. 35–63. [Google
Scholar]
35. Prakash, Arun J., Chun-Hao Chang, and Therese E. Pactwa. 2003. Selecting a portfolio with skewness:
Recent evidence from US, European, and Latin American equity markets. Journal of Banking &
Finance 27: 1375–90. [Google Scholar]
36. Rastogi, Shailesh. 2014. The financial crisis of 2008 and stock market volatility–analysis and impact on
emerging economies pre and post crisis. Afro-Asian Journal of Finance and Accounting 4: 443–59.
[Google Scholar] [CrossRef]
37. Rockafellar, R. Tyrrell, and Stanislav Uryasev. 2000. Optimization of conditional value-at-risk. Journal of
Risk 2: 21–42. [Google Scholar] [CrossRef][Green Version]
38. Shahbaz, Muhammad, Mohamed Arouri, and Frédéric Teulon. 2014. Short-and long-run relationships
between natural gas consumption and economic growth: Evidence from Pakistan. Economic Modelling 41:
219–26. [Google Scholar] [CrossRef]
39. Soleimani, Hamed, and Kannan Govindan. 2014. Reverse logistics network design and planning utilizing
conditional value at risk. European Journal of Operational Research 237: 487–97. [Google Scholar]
[CrossRef]
40. South Asian Financial Market Review. 2010. A Publication of the South Asian Federation of Exchanges.
Kathmandu: The South Asian Association for Regional Cooperation (SAARC). [Google Scholar]
41. Tang, Chor Foon, and Soo Y. Chua. 2012. The savings-growth nexus for the Malaysian economy: A view
through rolling sub-samples. Applied Economics 44: 4173–85. [Google Scholar] [CrossRef]
42. Vidovic, Jelena. 2011. Performance of risk measures in Portfolio construction on Central and South-East
European emerging markets. American Journal of Operations Research 1: 236. [Google Scholar]
[CrossRef][Green Version]
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.
© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article
distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license
(http://creativecommons.org/licenses/by/4.0/).

Share and Cite