Journal of the Operational Research Society

ISSN: 0160-5682 (Print) 1476-9360 (Online) Journal homepage: https://www.tandfonline.com/loi/tjor20

A multiobjective interval portfolio framework for

supporting investor’s preferences under different
risk assumptions

Carla Oliveira Henriques & Maria Elisabete Duarte Neves

To cite this article: Carla Oliveira Henriques & Maria Elisabete Duarte Neves (2019): A
multiobjective interval portfolio framework for supporting investor’s preferences under different risk
assumptions, Journal of the Operational Research Society, DOI: 10.1080/01605682.2019.1571004

To link to this article: https://doi.org/10.1080/01605682.2019.1571004

Published online: 02 May 2019.

Submit your article to this journal

Article views: 19

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tjor20
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY
https://doi.org/10.1080/01605682.2019.1571004

A multiobjective interval portfolio framework for supporting investor’s

preferences under different risk assumptions
Carla Oliveira Henriquesa,b and Maria Elisabete Duarte Nevesa,c
a
Polytechnic Institute of Coimbra, ISCAC-Coimbra Business School, Coimbra, Portugal; bINESCC, Coimbra, Portugal;
c
CETRAD-UTAD, Vila Real, Portugal

ABSTRACT ARTICLE HISTORY

This paper is aimed at presenting a multiobjective interval portfolio framework which con- Received 8 August 2017
siders investment decisions under different risk assumptions. New surrogate problems are Revised 19 December 2018
obtained for the mean-absolute deviation risk measure based on the concept of necessary Accepted 14 January 2019
subtraction between interval numbers. A proposal for obtaining the efficient portfolio solu-
KEYWORDS
tions is also suggested, which allows accounting for three types of investment strategies. Portfolio theory;
Indices of robustness have also been computed, which allow assessing the assets which are multiobjective interval
more often selected irrespective of the investment strategy followed, and regardless of the linear portfolio problems;
business cycle contemplated. Results illustrate the trade-off between risk and return, being mean-absolute
also consistent with the type of strategy followed by the investor. Overall, we were able to deviation-risk models
conclude that less prone to risk investors might find the formulation based on the mean-
absolute necessary deviation more appealing since it allows reaching, in general, lower vola- MATHEMATICS SUBJECT
CLASSIFICATION
tility of returns.
(2010 DATABASE)
91G10; 90C29; 65G40

1. Introduction by Markowitz (1968), providing a better risk assess-

ment than the measure originally used, but involv-
In 1952, Markowitz (1952) paved the grounds for
ing a higher computational burden and disclosing
modern portfolio theory, with the application of
variance or standard deviation as a measure of risk. the same type of information.
The classical Markowitz formulation is valid as The Markowitz classical model should thus be
long as the expected return is multivariate normally regarded as an approximation to rather complex
distributed and the investor is risk averter and pre- problems that investors have to face. Real-world
fers lower risk (Papahristodoulou & Dotzauer, portfolios are made up of a large number of assets,
2004). This approach has drawn many criticisms, possibly with very small holdings for some of them,
since it might allow for the choice of a portfolio minimum lot sizes, complexity of management, or
that might be outperformed and because of its com- policy of the asset management companies
plexity (the objective function is quadratic), being (Cesarone, Scozzari, & Tardella, 2011).
very hard to find an optimal portfolio when the The classical Markowitz model has evolved and
number of assets is large. Additionally, even if an some variations therefrom have been comprehen-
optimal solution is obtained, it might be hard to sively studied over the past decade, particularly
implement it in practice, eventually leading the from the computational standpoint. Several attempts
investor to assign his/her budget to a large number have been made to build less complex portfolio
of small blocks of shares, which might be unprofit- selection problems by linearising the quadratic
able, given the transaction costs. Finally, the stand- objective function (see e.g., Mansini, Ogryczak, &
ard Markowitz’ formulation, considers variance as a Speranza, 2003; Sharpe, 1971; Speranza, 1993).
measure of risk volatility, disregarding the fact that Usually the approximation or the decomposition of
quite often the investor’s stance regarding risk is not the covariance matrix are considered (Mitra,
consistent with symmetry or normal distribution. A Kyriakis, Lucas, & Pirbhad, 2003). Alternative risk
small loss might be enough to make an investor not measures for portfolio planning have also been used
prone to risk very unhappy and the opposite might (for an overview on distinct risk estimation meas-
also be true. The semi-variance was also proposed ures usually accounted for in the framework of port-

CONTACT Carla Oliveira Henriques [email protected] Polytechnic Institute of Coimbra, ISCAC, Quinta Agr
ıcola, Bencanta, Coimbra
3040-316, Portugal
Supplemental data for this article can be accessed on the publisher’s website.
ß Operational Research Society 2019
2 C. O. HENRIQUES AND M. E. D. NEVES

folio theory, see eg, Biglova, Ortobelli, Rachev, & his/her wealth among a small number of assets,
Stoyanov, 2004; Ortobelli, Rachev, Stoyanov, tackling the explicit uncertainty of the model mainly
Fabozzi, & Biglova, 2005). in its expected returns and/or risk levels. In fact, the
Although a prevalent number of publications majority of these models focus on a single objective
exists regarding risk measures and mean-risk mod- function, either addressing return or risk concerns,
els, portfolio decisions may also be based on the and do not account for interval uncertainty in the
investor’s expectations regarding return, risk and constraints. Additionally, the measures used for
liquidity characteristics of the assets (Gupta, assessing risk in these studies do not guarantee
Mehlawat, Inuiguchi, & Chandra, 2014). Investors obtaining the smallest deviation possible of the
may be interested in grasping how different assets return of each portfolio from its corresponding
may be combined in order to obtain the aimed interval expected returns.
return, risk and liquidity. In this framework the novelty of our proposal is
Conventional multiobjective models usually given as follows: it reduces the complexity of the
address practical portfolio selection problems in problem from a computational standpoint even
which all coefficients and parameters are a priori when the consideration of the uncertainty involves
given. Nevertheless, in real-world problems, infor- both the objective functions and the constraints of
mation regarding the asset returns, risk and liquidity the model (also contemplating the uncertainty of
is often incomplete, the markets in which the assets the maximum percentage of investment that should
are traded exhibit volatility and experts’ opinions be allocated to each asset in order to guarantee a
might vary. Therefore, besides multiple axes of minimum level of diversification of the portfolio
evaluation, these problems inherently involve selected) being applicable to real problems with a
inexactness and uncertainty issues. Uncertainty han- large number of assets as a starting point; it suggests
dling can be dealt with in various ways, namely by new surrogate problems based on the mean-absolute
means of stochastic, fuzzy and interval program- deviation risk measure and on the concept of neces-
ming techniques. In the stochastic approach the sary subtraction between interval numbers, thus
coefficients are treated as random variables with guaranteeing the smallest deviation of the return of
known probability distributions. In the fuzzy each portfolio from its corresponding interval
approach, the constraints and objective functions expected returns; it allows reflecting different invest-
are regarded as fuzzy sets with known membership ment strategies according to the investor’s preferen-
functions. However, it is not always easy for the ces in the selection of the possibly efficient
decision-maker (DM) to specify these probability portfolios, i.e., a conservative strategy, an aggressive
distributions and membership functions. Thus it strategy and a combined strategy; it proposes indices
would be more realistic to define portfolio parame- of robustness which allow assessing the assets which
ters in terms of intervals rather than crisp numbers. are more frequently selected regardless of the eco-
In the interval approach it is assumed that the nomic business cycle studied and the investment
uncertain values are perturbed simultaneously and strategy followed.
independently within known fixed bounds, being The remaining of this paper is organised as fol-
therefore intuitively preferred by the DM in practice lows. In Section 2 we briefly describe the methodo-
(Oliveira & Antunes, 2007). In this framework, sev- logical approach used. Section 3 describes the main
eral authors have proposed portfolio selection mod- premises regarding data collection. In Section 4 a
els based on the traditional semi-absolute deviation discussion of the illustrative results obtained is pre-
measure of risk, by taking the uncertain returns of sented. Finally, some conclusions are drawn and
assets in financial markets as intervals. The minimax future work developments are suggested.
regret approach based on a regret function has also
been studied in portfolio selection problems in
2. Methodology and assumptions
which the prices of the securities are treated as
interval numbers. Multi-period portfolio selection In the next sections some of the underpinning
models with interval coefficients have also been sug- assumptions regarding multiobjective linear pro-
gested. Other combined approaches have also been gramming (MOLP) models with interval coefficients
proposed where the average return of every asset is are provided, the objective functions and the con-
given as an interval number, and the risk of every straints considered are also described which allow
asset is treated by a probabilistic measure. obtaining two different modelling approaches
Table 1 provides a brief review of portfolio selec- according to different risk measures. Finally, three
tion models with interval coefficients. Usually, the different mathematical models are obtained for each
portfolio selection models therein studied were of these models according to different investor’s
applied assuming that an investor decides to invest standpoints.
Table 1. Portfolio selection problems with interval coefficients.
Interval valued
Brief description coefficients Criteria Constraints Solution approach Reference
Suggests a model for portfolio selec- Objective function – Maximisation of return. The average costs and the costs in the worst The noninferior solutions to such problems are Lai, Wang, Xu,
tion based on the semi-absolute expected return case scenario are less than or equal to the defined based on two order relations Zhu, &
deviation measure of risk, which Constraints – costs average value and the maximal possible value between intervals, and can be found by solv- Fang (2002).
can be transformed into a linear and resources. of the uncertain resources. ing a parametric linear programming prob-
interval programming model. No short selling is allowed. lem.
Applied to six stocks in the Shanghai
Stock Exchange.
Deals with mean-variance portfolio Objective functions - Maximisation of return and Capital budget constraint. Theoretical approach (not applied to a Ida (2003)
selection problems with objective expected return minimisation of risk. real problem).
function coefficients given and covariance.
by intervals.
Proposes a semi-absolute deviation Objective functions - Maximisation of return after Capital budget constraint and imposition of Applies the simplex method. Fang & Wang
model for portfolio selection and expected return and removing transaction costs. maximum risk tolerance level. Data set obtained from Shanghai 30 index. (2006)
introduces the concepts of pes- risk tolerance.
simistic satisfactory index and
optimistic satisfactory index of
interval inequality relations.
Formulates and solves a portfolio Objective function - prices Minimisation of the worst Capital budget constraint. Theoretical approach (not applied to a real Giove, Funari,
selection problem using a min- of the securities. regret, i.e., the maximum problem). & Nardelli
maxregret approach. deviation between the It assumes that the investor minimises the (2006)
return that it could be upper bound of interval regret.
received by investing in
the optimal portfolio and
the obtained return.
Extends the Markowitz’s portfolio Objective functions and Minimization of risk. Capital budget constraint and imposition of Theoretical approach (not applied to a real Gioia (2009)
selection model to the case in constraints – given returns. problem) which makes use of the interval
which the returns of any consid- expected return. algebra tools together with some optimiza-
ered asset are interval-valued. tion techniques to consider the interval as a
whole structure and to compute the interval
of solutions.
Studies portfolio models based on Objective functions – Minmax deviation between Capital budget constraint and minimum returns. Translates the interval problem into parametric Bao, Zhu,
fuzzy interval numbers under expected return. the actual portfolio return programming problems. Theoretical approach & Zhao
minimax rules. and the average degree of (not applied to a real problem). (2010)
portfolio return as stand-
ard of a risk measurement.
Utilizes the concept of interval num- Objective functions - Maximisation of return, mini- Capital budget constraint, maximum number of Applies the weighted sum approach and three Bhattacharyya,
bers in fuzzy set theory to extend expected returns, misation of risk and maxi- assets, minimum short and long term returns, different models have also been proposed by Kar, &
the classical mean–variance port- covariance and central misation of skewness. minimum dividend level, minimum liquidity defining the future financial market optimis- Majumder
folio selection model into mean–- co-moment. level and maximum fraction of capital that tically, pessimistically and in the combined (2011)
variance–skewness model with the Constraints - turn- can be invested in a single selected asset. form.
consideration of transaction costs. over rate. Data set extracted from Bombay stock
exchange.
Consideration of monthly share price data for
sixty months (March 2003–February 2008) of
five companies which are included in
BSE index.
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY

(continued)
3
 4

Table 1. Continued.
Interval valued
Brief description coefficients Criteria Constraints Solution approach Reference
Discusses the uncertain portfolio Objective function – Minimisation of risk. Capital budget constraint, maximum fraction of According to the concept of the mean-absolute Liu (2011)
selection problem where the expected returns. capital that can be invested in a single deviation function, a pair of two-level math-
returns of assets were represented selected asset and minimum return. ematical programming models to are
by interval data by considering obtained in order to calculate the lower and
the concept of mean-abso- upper bounds of the investment return of
lute deviation. the portfolio selection Problem.
Considers a small example of 3 stocks.
Considers an interval portfolio selec- Objective functions and Maximisation of return. Capital budget constraint, imposition of a given Based on the concept of satisfaction index of Jong (2012)
tion problem and constraints – maximum acceptable risk and imposition of interval inequality relation, the interval semi-
introduces an inclusive concept of expected returns. an upper bound on the proportion of the absolute deviation portfolio selection problem
satisfaction index for interval total investment devoted to a risky asset. is converted into two parametric linear pro-
inequality relation. gramming problems.
C. O. HENRIQUES AND M. E. D. NEVES

Applied to a small example with 6 compo-

nential stocks and a risk-less asset
for investment.
Develops a fuzzy portfolio selection Objective functions – Maximisation of return. Capital budget constraint, imposition of a given Converts the fuzzy portfolio selection problem Tan (2012)
model with liquidity constraints expected returns. maximum acceptable risk, imposition of an with liquidity constraints into a linear pro-
where the profit rate, risk rate and Constraints – risk upper bound on the proportion of the total gramming problem with parameters.
turnover ratio of securities are (interval numbers of investment devoted to a risky asset and Applied to a small example from scien-
given as interval fuzzy numbers. standard variance) imposition of a minimum liquidity level. tific literature.
and liquidity.
Discusses a fuzzy multi-period port- Objective function - Maximisation of the terminal Capital budget constraint, and imposition that Converts a fuzzy bi-objective programming Liu, Zhang,
folio optimization problem with expected returns. wealth obtained at the the return, liquidity and diversification degree problem with interval coefficients into a crisp & Zhang
interval coefficients Constraints – expected end of period T. of the portfolio at each period must achieve form nonlinear programming problem and (2013)
returns, risk and liquid- or exceed the given expected interval levels. uses particle swarm optimization for obtain-
ity of securities. Besides, the risk of portfolio at each period ing the solutions.
should be less than the pre-set maximum risk Applied to a small example with 4 stocks
tolerance interval. from the Shanghai Stock Exchange market.
Suggests portfolio optimization mod- Objective functions - Maximisation of return, mini- Capital budget constraints, cardinality con- The weighted-sum method is used to convert Gupta
els with interval coefficients, expected returns, risk misation of risk and maxi- straints, maximal and minimal fraction of cap- the multiobjective problem into the following et al. (2014)
where the expected return, risk and liquidity of assets. misation of liquidity. ital that can be invested in a single asset. single objective optimization problem.
and liquidity of assets are treated Applied to a small example of 5 stocks from
as interval numbers. NSE, Mumbai, India market.
Presents a model which can be used Objective function – Minimisation of risk. Imposition that the number of units of each Develops an algorithm for rebalancing the port- Kumar, Panda,
to create an initial (or first covariance of returns asset bought (sold) by an investor must lie in folio. & Gupta
time) portfolio, as well as the of the assets. an interval, for each asset a single operation Applied to a small example of 4 stocks from (2015)
rebalancing of this (or existing) Constraints - expected is permitted (i.e., either one can buy or sell Bombay Stock Exchange, India.
portfolio over an investment returns and fixed or/ or do nothing), definition of the total number
time horizon. and variable transac- of units present in the portfolio after transac-
tion costs. tion, imposition that the total transaction
cost varies within an interval, imposition that
the money consumed in total transaction
cost should not cross a maximum limit, after
the rebalancing of the portfolio, the monetary
value of rebalanced portfolio must be equal
to the sum of value of the current portfolio
and new cash minus total transaction cost
incurred in trading and capital budget
constraints.
(continued)
Table 1. Continued.
Interval valued
Brief description coefficients Criteria Constraints Solution approach Reference
Establishes a bilinear programming Objective function and Maximisation of return or Capital budget constraints and imposition of a Uses an algorithm for solving portfolio selection Li, Wang,
model for maximising the port- constraints - expected minimisation of risk. maximum risk and minimum return level. problems, which can be transformed & Jin
folio return at a given level of risk returns and risk to classical linear programming problems. (2015)
and meanwhile minimising the of assets. Applied to six stocks shown from
portfolio risks at a given level www.stockstar.com.
of return.
Uses the absolute deviation of return, Objective function and Maximisation of return. Capital budget constraints, imposition of a max- The interval problems are transformed into two Şerban, Costea,
modelled using interval numbers. constraints – expected imum risk level and maximal fraction of cap- linear programming problems, which can be & Ferrara
returns and risk ital that can be invested in a single asset. solved using the simplex algorithm. Applied (2015)
of assets. to 2 assets from Bucharest Stock Exchange.
Proposes an interval mean–average Objective function and Maximisation of the terminal Wealth accumulation constraint, imposition of a Uses a dynamic programming method for find- Zhang (2016)
absolute deviation model for mul- constraints – expected wealth over the whole minimum level for the investment proportion ing an optimal solution. Applied to 30 stocks
tiperiod portfolio selection returns and risk periods of investment. of risk-free asset at period t and cardinality from Shanghai Stock exchange.
of assets. constraints.
Suggests four interval portfolio opti- Objective function and Maximisation of the terminal Problem 1 - imposition of a minimum return The models are transformed into deterministic Liu, Zhang,
mization models with cardinality constraints - expected wealth over the whole level models and after that a dynamic & Wang
constraints for multiperiod port- returns, risk and turn- periods of investment. at period t, imposition of a maximum risk differential evolution algorithm with self- (2016)
folio selection in emerg- over rate level at period t, cardinality constraints, cap- adapting control parameter is proposed to
ing markets. ital budget constraints at period t and impos- solve the transformed
ition of a minimum and maximal fraction of deterministic models.
capital that can be invested in a single asset Applied to 8 stocks from Shanghai
at period t. Stock exchange.
Problem 2 – Problem 1 with the additional
imposition of a minimum level of the turn-
over rate at period t.
Problem 3 – Problem 1 with the imposition
of a minimum diversification level at period t.
Problem 4 – Considers all the constraints.
Suggests a mean semi-absolute devi- Objective function and Maximisation of terminal util- Borrowing constraints, threshold constraints The discrete approximate iteration method is Zhang (2017)
ation interval multiperiod portfolio constraints – expected ities over the whole peri- which limit the amount of capital to be designed to obtain the optimal port-
selection model with transac- returns and risk. ods of investment. invested in each stock and prevent very small folio strategy.
tion costs. investments in any stock and imposition of a
diversification degree of the portfolio meas-
ured by the presented possibilistic entropy.
Establishes a model for portfolio Objective functions - Maximisation of return or Capital budget constraints, a given level of Suggests an algorithm that allows obtaining a Li & Jin
selection associated with interval expected returns, while minimisation of the max- return and definition of a probabilistic solution where the average returns of the (2018)
numbers and interval-typed ran- the risk of every asset imum risk. risk measure. assets satisfy the order relations between
dom variables, where the prob- is treated as a prob- intervals.
abilistic risk measure is flexible for abilistic measure. Applied to 8 stocks from http://quote.ea-
investors with different stmoney.
risk tolerance. com, to 10 stocks from the Tokyo Stock
Exchange and to 20 stocks from Shanghai
Stock Exchange.
Formulates a multiobjective portfolio Objective functions - Maximisation of return and Capital budget constraints and a maximum The uncertainties are addressed by transforming Kumar, Panda,
selection model with interval expected returns and maximisation of liquidity risk level. the multiobjective interval problem into a & Gupta
parameters and interval deci- liquidity. deterministic form. (2018)
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY

sion variables. Constraints – risk. Applied to 30 stocks from Bombay Stock

Exchange-30 (BSE-30).
5
6 C. O. HENRIQUES AND M. E. D. NEVES

2.1. Multiobjective portfolio problems with Having this in mind, in order to account for the
interval coefficients uncertainty handling, the expected return of an asset
should be rather given as an interval number.
Let us assume that investors allocate their wealth
Therefore, the return of the portfolio is expressed
among n assets offering random rate of returns and
as:
that the portfolio selection problem considered is " #
based on a single-period model of investment. Xn   Xn X
n
ri ; ri x i ¼
L U
ri xi ;
L
ri xi ;
U
(3)
i¼1 i¼1 i¼1
2.1.1.Objective functions  
A portfolio is composed of two or more assets rep- where riL ; riU is the interval valued return.
resented by an ordered n-tuple H ¼ (x1, x2, … , xn), Risk. Usually, an investor would rather prefer to
where xi is the proportion of the total funds have the portfolio return as large as possible and at
invested in the i-th asset. the same time with minimum possible dispersion/
Return. The asset’s return or the rate of return of variability. Therefore, Markowitz (1952) suggested
each asset has been obtained through the computa- the variance to quantify the risk of a portfolio.
Although, variance can be used as a risk measure,
tion of the daily logarithmic returns:
one of its main limitations is that it penalises
pt extreme upside (gains) and downside (losses) devia-
rit ¼ lnðpt Þ  lnðpt1 Þ ¼ ln (1)
pt1 tions from the expected return. Thus, when prob-
ability distributions of the asset returns are
where pit is the closing price of the i-th asset during
asymmetric, variance becomes a less appropriate
period t, pit1 is the closing price during period
measure of the risk of a portfolio (Chunhachinda,
t  1, dit is the dividend of the i-th asset during the
Dandapani, Hamid, & Prakash, 1997). In fact, the
period t.
selected portfolio may sacrifice higher expected
The expected value (E[]) of the rate of return, Ri
returns. Markowitz (1968) also suggested the semi-
(i ¼ 1, 2, … , n), is a random variable taking finitely variance which is a downside risk measure, i.e., a
many values, which can also be approximated by measure which only considers the negative devia-
the average derived from historical data, i.e., tions from a reference return level. Its advantage
over variance is that it does not perceive gains as
1X T
ri ¼ E½Ri  ¼ rit : (2) risk; thus, it is a suitable measure of risk when
T t¼1
investors are concerned about portfolio underper-
formance rather than over performance.
In general, the arithmetic mean of past returns is
Nevertheless, the implementation of mean semi-
viewed as a proxy of expected return of an asset
variance portfolio selection models is computation-
and thus it is obtained as a certain value. However,
ally much more complex as compared to mean-vari-
in real world problems, asset prices and the returns
ance portfolio selection models. In 1994, JP Morgan
obtained therefrom are subject to a set of variables
suggested another risk measure also known as
whose behaviour cannot be simply anticipated on
Value-at-Risk (see Longerstaey & Spencer, 1996),
past events (Gupta et al., 2014). Moreover, the use which entails several drawbacks since the Value-at-
of the arithmetic mean of historical returns as the Risk optimisation problem is not convex and it does
expected return, has two major shortcomings. On not allow expressing the benefits of diversification
one hand, if historical data for a long period of time (Cesarone et al., 2011). Other important risk meas-
are considered, the influence of the earlier historical ure is Conditional Value-at-Risk also called Mean
data is the same as that of recent past data. Excess Loss, Mean Shortfall, or Tail VaR
Nevertheless, recent past data of an asset might be (Rockafellar & Uryasev, 2000). A substitute measure
more significant than the earlier historical data. On for risk is the maximisation of the minimum return
the other hand, if the historical data of an asset are (maximum loss) demanded by the investor (Young,
not suitable, due to lack of information, the estima- 1998):
tion of the statistical parameters would not be X
n
adequate. Additionally, if the investor wants to Max min rit xi ; t ¼ 1; :::; T: (4)
assess the impact of distinct economic business i¼1

cycles on the results obtained (accounting for the Young (1998) argued that for given distributions,
best and worst case coefficient scenarios possible), in particular when data is log-normally distributed,
distinct delimitation periods should be used. This or skewed, this type of formulation might be rather
information can be interesting to appraise how the preferable. This author also advocated its use when
assets behave according to bull and bear scenarios, the portfolio optimisation problem involves a large
respectively. number of decision variables (including integer
 JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 7

P
T  n
 Pn
X 
i¼1 ðrit ri Þxi þ i¼1 ðri rit Þxi
variables), or if the investor’s is more risk averse
than it is implied by the classical minimisation of (7)
t¼1
2T
variance. However, in spite of its simplicity this for-
mulation might lead to an infeasible solution if all The mean semi-absolute deviation model reduces
assets yield a negative return. A different approach the number of constraints by half in comparison with
to replace the Markowitz classic formulation is to the mean-absolute deviation model, since it requires T
use the absolute deviation risk function (Konno & linearising constraints while the mean-absolute devi-
Yamazaki, 1991; Mansini & Speranza, 1999; Rudolf, ation model requires 2T linearising constraints.
Wolter, & Zimmermann, 1999). The absolute devi- If the expected returns of assets are given as
ation of a random variable is the expected absolute interval numbers, the expected semi-absolute devi-
value of the difference between the random variable ation of return of the portfolio below the expected
and its mean. The portfolio risk measured as abso- return is an interval number too:
"     Pn  L 
T Pn
lute deviation can be approximated as follows:
  X 
i¼1 rit ri xi þ i¼1 ri rit xi
L
T X 
1X 
n
 ;
 ð it i Þ i :
r r x (5) t¼1
2T
T t¼1  i¼1 

T Pn
   Pn  U  #
X 
i¼1 rit ri xi þ i¼1 ri rit xi
U
Since the expected returns of assets are given as (8)
interval numbers, the expected absolute deviation of t¼1
2T
the return of the portfolio below the expected return
is an interval number too:
2.1.2. Constraints
  Capital budget constraint. Since xi is the proportion
T X  L U  
1X 
n 
 rit  ri ; ri xi  (6) of the total funds invested, the capital budget con-
T t¼1  i¼1 
straint on the assets is expressed as
X
n
Konno and Yamazaki (1991) concluded that if the xi ¼ 1: (9)
return is multivariate normally distributed, the mini- i¼1
misation of the absolute deviation provides similar
Maximum proportion of capital that can be
results to the classical Markowitz formulation. Rudolf
invested. The maximum proportion of capital allo-
et al. (1999) argue that the minimisation of the abso-
cated to the assets in the portfolio depends upon
lute deviation is equivalent to expected utility maxi-
several factors. Since investors differ in their inter-
misation under risk aversion. This formulation has
several advantages since it does not require the estima- pretation of the available information, in order to
tion of the variance-covariance matrix and the solu- achieve a sufficient diversification of investments we
tion is obtainable even if all possible assets yield a impose  the upper bound given within interval,
negative return. This model is also flexible enough to uLi ; uUi , obtaining the following interval constraint:
be reformulated as an Integer Linear Programming X
n  
(ILP) problem incorporating other important features xi  uLi ; uUi ; i ¼ 1; :::; n; (10)
i¼1
(e.g., fixed and variable costs associated with the pur-
chase of assets) and decision variables (Mansini & No short selling is allowed. Short selling occurs
Speranza, 1999), being easily implemented even when when an investor actually does not own an asset but
a large number of assets is at stake. he/she establishes a market position by selling the
Following the work of Konno and Yamazaki asset in anticipation that the price of that asset will
(1991), Speranza (1993) proposed the semi-absolute fall. In this case, the investor is said to have taken a
deviation as an alternative measure to quantify risk, short position. Mathematically, this situation implies
and concluded that by transforming the risk function that the number of assets owned by the investor is
into a linear combination of the mean semi-absolute negative. In portfolio mathematical modelling, short
deviations (i.e., mean deviations below and above the selling is not allowed, i.e., the values of xi are not
portfolio return), a model equivalent to the mean- negative; hence,
absolute deviation model can be obtained, whenever
xi  0 for all i; i ¼ 1; 2; :::; n (11)
the sum of the coefficients of the linear combination
is positive. Finally, they have also showed that this
model is equivalent to the Markowitz model, if the
2.2. Multiobjective portfolio selection problems
returns are normally distributed.
using interval numbers
The expected mean semi-absolute deviation of
return of the portfolio below the expected return is Two interval portfolio optimisation problems can be
given by: obtained assuming different risk assumptions.
8 C. O. HENRIQUES AND M. E. D. NEVES

2.2.1. Mean-absolute deviation model corresponding interval expected returns Problem

  (12) has the following surrogate representation:
T X  L U  
1X  
n
Min  rit  ri ; ri xi  Min E"t ðxÞ; t ¼ 1; :::;
T t¼1  i¼1  # T;
" # Xn X
n
Xn X
n (12) Max riL xi ; riU xi : (17)
Max riL xi ; riU xi i¼1 i¼1
i¼1 i¼1 s:t: xi 2 X; i ¼ 1; :::; n
s:t: xi 2 X; i ¼ 1; :::; n;
n Pn Problem (17) can either be solved by following an
where X ¼ x ¼ ðx1 ; :::; xn ÞT 2 Rn j i¼1 xi ¼ 1;
Pn  L U optimistic (the lower bound of the “necessary” devi-
i¼1 xi  ui ; ui ; i ¼ 1; :::; n; xi  0; i ¼ 1; :::; ng ation) or a pessimistic procedure (the upper bound
is the feasible region. of the “necessary” deviation). The second procedure
If we consider the necessary subtraction operator allows obtaining the smallest distance from the tar-
(see Appendix A for further details), the absolute get used (Inuiguchi & Kume, 1991; Oliveira &
necessary deviation of the return obtained from the Antunes, 2007; Henriques, Luque, Marcenaro-
expected interval returns in each period is Gutierrez, & Lopez-Agudo, 2018).
8 " #
>
>  X n X n X n Xn  X n Xn
>
>  
>
< r L
x i  r it x i ; r U
x i  r it x i  if r U
x i  riL xi  0;
i i
 i
Et ðxÞ ¼ "i¼1 i¼1 i¼1 i¼1
# i¼1 i¼1
>  X n Xn X n Xn  Xn Xn
>
>  
>
>  r it x i  r L
x i ; rit x i  r U
xi  if r U
xi  riL xi 0:
: i¼1
i
i¼1 i¼1 i¼1
i
 i¼1
i
i¼1

8 " #
>
> Xn X n X
n X n X n X n
>
> r L
x  r x ; r U
x  r x if r U
x  riL xi  0;
>
> i i it i i i it i i i
>
>
>
> " i¼1 i¼1
!i¼1 i¼1
!i¼1
# i¼1
< X n X n X n X n X n X n
¼ 0; riU xi  rit xi Ú rit xi  riL xi if riL xi <0< riU xi ; ; (13)
>
>
>
> " i¼1 i¼1 i¼1 i¼1
# i¼1 i¼1
>
> X n Xn X
n X n X n X n
>
>
>
> rit xi  riU xi ; rit xi  riL xi if riL xi  riU xi  0:
:
i¼1 i¼1 i¼1 i¼1 i¼1 i¼1

since Therefore, Problem (12) can be replaced by the

X
n X
n following problem (thus accounting for a pessimistic
riU xi  riL xi  0 stand point regarding risk):
i¼1 i¼1
1X T
is always verified. Min k vt þ ð1kÞuU ;
T t¼1
The necessary regret interval can be obtained as " #
   Lþ  X
n X
n

Et ðxÞ ¼ eL Max riL xi ; riU xi ;

t þ et Ù et þ eU ;
Uþ
t
i¼1 i¼1
 L U Lþ Uþ  X n
et Úet Úet Ú et ; (14) s:t: xi ¼ 1;

i¼1 
t , et ; et
where the deviational variables, eL Lþ Uþ
and xi  uLi ; uUi ; i ¼ 1; :::; n;
U
et are defined in such a way that: X n X
n
rit xi þ eL
t eLþ
t ¼ riL xi ; t ¼ 1; :::; T;
X
n X
n
i¼1 i¼1
rit xi þ eL
t  et ¼
Lþ
riL xi ; (15) X
n X
n
i¼1 i¼1 rit xi þ eU
t e Uþ
t ¼ riU xi ; t ¼ 1; :::; T;
X
n X
n i¼1 i¼1
rit xi þ eU
t  eUþ
t ¼ riU xi ; (16) eU
t þ eUþt  t ; t ¼ 1; :::; T;
i¼1 i¼1 eL
t þ e Lþ
t  t ; t ¼ 1; :::; T;
¼ 0, eL t  u ;
U
where eU t et ¼ 0 and the “ Ù ” and
Uþ Lþ
t et
“Ú” operators are the minimum and maximum val- xi  0; i ¼ 1; :::; n;
ues, respectively. 0  k  1;
If the DM wishes to minimise the necessary devi- vt  0; t ¼ 1; :::; T:
ation of the return of each portfolio from its (18)
 JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 9

2.2.2. Mean semi-absolute deviation model than return. For this reason, the investor seeks to
maximise return and to minimise risk in the worst
"     Pn  L 
X T Pn  case scenario.
i¼1 rit ri xi þ i¼1 ri rit xi
L
Min ; Mean-absolute deviation model
t¼1
2T 0 ! !1
    Pn  U  # X n XT
T Pn
1
X r it r U
x i
þ r r it xi Max @b riL xi –a k vt þ ð1kÞuU A
i¼1 i i¼1 i
; i¼1
T t¼1
2T
t¼1 X
n
" # s:t: xi ¼ 1;
X n X n
i¼1
Max rLi xi ; i xi ;
rU
xi  uLi ; i ¼ 1; :::; n;
i¼1 i¼1
Xn X
n
rit xi þ eL eLþ
¼ riL xi ; t ¼ 1; :::; T;
s:t: xi 2 X; i ¼ 1; :::; n; (19) i¼1
t t
i¼1
Problem (19) can be replaced by/the following X
n X
n
rit xi þ et et ¼
U Uþ
riU xi ; t ¼ 1; :::; T;
problem: i¼1 i¼1
" # eU þ eUþ  t ; t ¼ 1; :::; T;
1X T
1X T t t
Min p;
1
p ;
2 et þ eLþ
L
t  t ; t ¼ 1; :::; T;
T t¼1 t T t¼1 t  t  uU ;
" #
Xn X
n
xi  0; i ¼ 1; :::; n;
Max ri x i ;
L
ri xi ;
U
0  k  1;
i¼1 i¼1
Xn vt  0; t ¼ 1; :::; T;
s:t: xi ¼ 1; (21)

i¼1 
xi  uLi ; uUi ; i ¼ 1; :::; n; (20) where b and a are weight parameters ranging from
X n   zero to one, representing the weights that the
p1t þ rit riL xi  0; t ¼ 1; :::; T; investor assigns to each objective function.
i¼1
X
n   Mean semi-absolute deviation model
p12 þ rit riU xi  0; t ¼ 1; :::; T; !
X n
1 XT
i¼1 Max u riL xi s p2t ;
xi  0; i ¼ 1; :::; n; T
i¼1 t¼1
p1t  0; t ¼ 1; :::; T; X
n
p2t  0; t ¼ 1; :::; T: s:t: xi ¼ 1;
i¼1
xi  uLi ; i ¼ 1; :::; n; (22)
Xn  
2.3. Solution method p12 þ rit riU xi  0; t ¼ 1; :::; T;
i¼1
Several algorithms can be found in scientific litera- xi  0; i ¼ 1; :::; n;
ture to solve MOLP models (see e.g., Antunes, p1t  0; t ¼ 1; :::; T;
Alves, & Cl
ımaco, 2016). If cardinality constraints
were accounted for, then Problems (18) and (20) where u and s are weight parameters ranging from
could be solved by considering e.g., the methods zero to one, representing the weights that the
presented in Pei, Pardalos, Liu, Fan, & Yang (2015), investor assigns to each objective function.
Zhang (2016), Liu, Zhang, & Wang (2016), Pei, Liu,
Fan, Pardalos, & Lu (2019a), Pei, Cheng, Liu, 2.3.2. Aggressive strategy
Pardalos, & Kong (2019b) and Pei, Wang, Fan, The investor aiming for an aggressive strategy is
Pardalos, & Liu (2018). Since Problems (18) and more prone to risk, being more concerned with
(20) can be transformed into MOLP problems, the return than risk. In this case, the investor strives for
weighted-sum method can thus be used to convert the maximisation of return and the minimisation of
these problems into single interval objective risk in the best case scenario. Additionally, it is
optimisation problems (see e.g., Gupta et al., 2014). worth mentioning that under the mean-absolute
We suggest distinct optimisation models for deviation model we only account for the formula-
portfolio selection regarding three types of invest- tion that uses the upper bound of the necessary
ment strategies: conservative, aggressive and com- deviation of the return of each portfolio from its
bined strategies. corresponding interval expected returns, hence guar-
anteeing the smallest deviation possible (and thus
2.3.1. Conservative strategy only bearing in mind a pessimistic stance regarding
The investor aiming for a conservative strategy is risk, irrespective of the strategy of invest-
more risk averse, being more concerned with risk ment followed).
10 C. O. HENRIQUES AND M. E. D. NEVES

Mean-absolute deviation model where q, d and di , i ¼ 1, … n, are indexes of pessim-

! ism ranging in a scale from 0 (agressive strategy) to
X n
1X T
Maxðb ri xi aðk
U
vt þ ð1kÞuU ÞÞ 1 (conservative strategy).
T
i¼1 t¼1 Mean semi-absolute deviation model
X
n !
s:t: xi ¼ 1; X n
1X T
i¼1 Max q u ri xi s L
p2

xi  uUi ; i ¼ 1; :::; n; T t¼1 t

i¼1 !
Xn X
n X n
1X T
rit xi þ eL
t eLþ
t ¼ riL xi ; t ¼ 1; :::; T; þð1–qÞ u ri xi s U
p1

i¼1 i¼1 i¼1

T t¼1 t
X
n X
n X n
rit xi þ eU
t et
Uþ
¼ riU xi ; t ¼ 1; :::; T; s:t: xi ¼ 1;
i¼1 i¼1
 U
i¼1  
eU
t þ eUþ
t  t ; t ¼ 1; :::; T; xi þ ui þ di uLi þ uUi  0; i ¼ 1; :::; n;
et þ eLþ
L
t  t ; t ¼ 1; :::; T; X n  
t  uU ; p1t þ rit riL xi  0; t ¼ 1; :::; T;
xi  0; i ¼ 1; :::; n; i¼1
X
n  
0  k  1; p12 þ rit riU xi  0; t ¼ 1; :::; T;
vt  0; t ¼ 1; :::; T: i¼1
(23) xi  0; i ¼ 1; :::; n;
p1t  0; t ¼ 1; :::; T;
Mean semi-absolute deviation model p2t  0; t ¼ 1; :::; T:
!
Xn
1 XT (26)
Max u riU xi s p1t ;
i¼1
T t¼1
X
n
s:t: xi ¼ 1; 3. Data and assumptions
i¼1
xi  uUi ; i ¼ 1; :::; n; (24) A sample of 10 complete years (about 3600 trading
Xn   days) regarding a portfolio of diversified stocks
p1t þ rit riL xi  0; t ¼ 1; :::; T; from different countries and distinct activity sectors
i¼1 has been selected, using Bloomberg as a source of
xi  0; i ¼ 1; :::; n;
information. The stocks contemplated for the
p1t  0; t ¼ 1; :::; T:
European countries are Portugal Telecom SGPS SA
(PT)1, Energias de Portugal, SA (EDP), Deutsche
2.3.3. Combined strategy Telekom AG (DT), Bayerische Motoren Werke AG
A combined strategy allows for the investor to choose (BMW) and Adidas AG. For the United States of
a more balanced approach regarding risk and return. America we have collected data for Microsoft
Mean-absolute deviation model Corporation, Apple Incorporation, McDonald’s
!! ! Corporation, The Coca-Cola Company and Pfizer
Xn Xn
Maxq b L
ri x i þ ð1–qÞb U
ri xi Incorporation. Regarding the Emergent markets, we
i¼1
! i¼1 have assessed Petrobras, Vale SA, America Movil
1X T
SAB, Sify Technologies Limited and Taiwan semi-
a k vt þ ð1kÞuU
T t¼1 conductor manufacturing company.
X n The selection of stocks for the portfolio analysed
s:t: xi ¼ 1; has been done by taking into account several levels
 U
i¼1   of diversification. Therefore, stocks from 15 compa-
xi þ ui þ di uLi þ uUi  0; i ¼ 1; :::; n; nies have been carefully chosen from distinct geo-
X n X
n
rit xi þ eL graphical locations (seven countries and four
t et ¼ riL xi ; t ¼ 1; :::; T;
Lþ

i¼1 i¼1 continents), facing different “development stages”

X
n X
n
(10 developed and 5 emerging countries), and
rit xi þ eU
t eUþ
t ¼ riU xi ; t ¼ 1; :::; T;
i¼1 i¼1
belonging to differentiated activity sectors and
eU
t þ eUþ
t  t ; t ¼ 1; :::; T; industries (see Table 2).
et þ eLþ
L
t  t ; t ¼ 1; :::; T;
Finally, we have contemplated distinct business
t  uU ; cycles assuming that the investor wants to appraise
xi  0; i ¼ 1; :::; n; the results of adopting different strategies under dif-
0  k  1; ferent economic settings. In our particular case, we
vt  0; t ¼ 1; :::; T: have accounted for the period between an early
(25) stage of the pre-crisis boom until a late stage of the
boom, i.e., from 2001 (the starting year of our data)
 JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 11

Table 2. Industry classification of the stocks selected. that the proportions of the selected assets are dis-
Industry classification Companies tributed evenly among a diversified set of stocks
Communication Services PT belonging to different industries (see Tables
Utilities – Regulated EDP
Communication Services DT 1A–3A). This result is related to the fact that the
Auto Manufacturers – Major BMW maximisation of return without the bounds that
Textile-Apparel Footwear & accessories Adidas
Application Software Microsoft guarantee a certain diversification level leads to the
Computer Hardware Apple investment in a single asset. Nevertheless, it is inter-
Restaurants McDonalds
Beverages – Non-Alcoholic Coca-Cola esting to see that there is a shift of the industries
Drug Manufacturers Pfizer contemplated in the portfolios selected during the
Oil & Gas – Integrated Petrobras
Metals & Mining Vale
different business cycles considered, with Coca-Cola
Communication Services America and McDonalds gaining their presence after the
Communication Services Sify period of crisis, while EDP, a company from the
Semiconductors Taiwan
regulated utility services, only remains in this port-
folio during the period of crisis.
until 8th of August 2007 (the eve of the Lehman
An aggressive strategy aiming for return only,
Brothers collapse), the crisis period as the time hori-
leads to the choice of two blue-chip stocks, but with
zon between the 9th of August of 2007 and the 31th
lower diversification level (see Portfolios 2 under
of December of 2009 and from then on until the lat-
different economic settings). It is worth noting that
est years of our data (4th of April of 2013) a
Apple remains in this portfolio either in the pre-cri-
“recovery” stage (Milesi-Ferretti & Tille, 2011;
sis or in the recovery periods. Under the crisis scen-
Mobarek, Mollah, & Keasey, 2014). Subsequently, by
ario both a communications service (PT) and an oil
taking into account the mean returns obtained in
company (Petrogas) are present in Portfolio 2.
these distinct economic scenarios after removing the
A regulated utility is always selected when the
corresponding outliers, confidence intervals were
investor is strictly seeking risk minimisation
constructed for these different economic conjunc-
(Portfolios 3 and 4 according to distinct business
tures, i.e., pre-crisis, crisis and recovery periods,
cycles), particularly in the pre-crisis period assuming
assuming a confidence level of 95% (see Figures
both strategies of investment regarding risk, with a
1S–3S of the supplementary material). The end-
conservative strategy towards risk in the period of
points of such confidence intervals were then intro-
duced as the upper and lower-bound of the crisis and following an aggressive strategy in the
logarithmic returns (see Figures 4S–6S of the sup- recovery period. From the Portfolios analysed the
plementary material). highest diversification is obtained when an aggres-
Finally, the maximum proportion of capital allo- sive strategy towards risk is followed (see Portfolios
cated to the assets in the portfolio is given within 4 under the economic conjunctures studied).
 The combined strategy (see Portfolios 5, 6 and 7)
the interval uLi ; uUi ¼ ½20%; 50%, in order to
obtain a certain level of investment diversification. assigns a pessimistic index of 0.5 to all the objective
functions and to the threshold of the interval con-
straints. The portfolios selected according to a bal-
4. Discussion of results anced approach regarding risk (see Portfolios 5
Tables 1A–3A (see Appendix B) provide informa- under the distinct business cycles) correspond to the
tion regarding the efficient Portfolios selected highest proportion of investment allocated to stocks
according to distinct risk profiles and different from: metals and mining industry (Vale), communi-
investment strategies (other scenarios could be cation services (America) and regulated utility serv-
explored) through the use of the mean semi-abso- ices (EDP) during the period of pre-crisis;
lute deviation model in the pre-crisis, crisis and restaurant (McDonalds), non-alcoholic beverages
recovery periods, respectively. industry (Coca-Cola) and communication services
Portfolios 1 and 3 are obtained presuming a (America) in the period of crisis; McDonalds, Coca-
more conservative strategy attaining the lower Cola, computers (Apple) and textile-apparel foot-
bound of maximum return and the upper bound of wear and accessories industry (Adidas) in the period
minimum risk, respectively, in a worst case scenario; of recovery. In addition to the companies of
while Portfolios 2 and 4 are computed according to Portfolios 5, a strong risk averter (Portfolios 6)
a more aggressive approach since they seek to com- should select Pfizer (drug manufacturer) in the pre-
pute the upper bound of the maximum return and crisis period; during the period of crisis another
the lower bound of the minimum risk in a best case company from the communication services (DT) is
scenario, respectively. added to the portfolio; in the recovery period
Regarding Portfolio 1 (irrespective of the eco- Adidas is replaced with a company from the semi-
nomic period presumed) it is possible to conclude conductors industry (Taiwan). With this same
12 C. O. HENRIQUES AND M. E. D. NEVES

Figure 1. Risk vs. log return in the mean semi-absolute model (pre-crisis).

Figure 2. Risk vs. log return in the mean semi-absolute model (crisis).

Figure 3. Risk vs. log return in the mean semi-absolute model (recovery).

strategy, a strong risk prone investor (Portfolios 7) portfolios with higher risk also obtain higher return.
should aim the majority of his/her investment at the These results are also consistent with the type of
following stocks: Vale, America, Adidas and strategy followed by the investor with more aggres-
McDonalds during the period of pre-crisis; sive strategies towards risk leading to lower risk and
Petrobras (oil and gas), PT (communication serv- more aggressive strategies towards return leading to
ices), Coca-Cola and Apple through the period of higher return and vice-versa. In this context, it
crisis; BMW (auto manufacturers), Apple and should be mentioned that an aggressive strategy
McDonalds in the period of recovery. leads to a substantial increase of risk, particularly in
Finally, portfolios 8 and 9 reflect a balanced pro- the period of crisis.
pensity to risk but according to strongly aggressive Indices of robustness have also been computed,
and conservative strategies, respectively. which allow assessing the assets which are more
The expected trade-off between risk and return is often selected irrespective of the investment strategy
highlighted in Figures 1–3, illustrating that followed, and which are more frequently chosen
 JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 13

Figure 4. Width of the interval log return with both models (pre-crisis).

Figure 5. Width of the interval log return with both models (crisis).

according to each investment strategy regardless of BMW. On the other hand, if a conservative strategy
the business cycle taken into account. In the first is followed (regardless of risk presumptions), the
case, this index is the ratio between the sum of the investment is mainly channeled to Coca-Cola,
proportion of a given asset in all the portfolios McDonalds, Apple and America. Finally, a com-
obtained and the overall investment possible accord- bined strategy leads to assign the majority of the
ing to all the investment strategies scrutinized2—see investment to McDonalds, Coca-Cola, America,
Table 4A (see Appendix B). In the latter case, this Vale and Apple.
index is the ratio between the sum of the proportion Tables 6A–8A (see Appendix B) present informa-
of a given asset according to the same strategy in all tion regarding the efficient Portfolios selected
business cycles and the overall investment possible according to distinct risk profiles and different
in all business cycles with the same investment strat- investment strategies (once more, other scenarios
egy (e.g., the asset that always remains immune to could also be explored) with the use of the mean-
the effects of business cycles following a conserva- absolute deviation model, in the pre-crisis, crisis
tive strategy towards return is Apple with a value of and recovery periods, respectively.
(0.2 þ 0.2 þ 0.2)/3)  100 ¼ 20%) – see Table 5A Analogously to the previous model formulation,
(Appendix B). Portfolios 10 and 12 are obtained in view of a more
According to Table 4A the three assets which are conservative strategy, attaining the lower bound of
consistently selected no matter of the strategy fol- maximum return and the minimum risk, respect-
lowed are McDonalds, Coca-Cola and America. If ively, under a worst case scenario; while Portfolios
only the recovery period is used, Taiwan (from the 11 and 13 correspond to a more aggressive attitude
semiconductors industry) replaces America in terms since they pursue the computation of the upper
of importance. Finally, Table 5A suggests that the bound of the maximum return and the minimum
three best assets to invest in from the point of view risk under a best case scenario, correspondingly.
of the maximisation of return are Apple, America From the analysis of Tables 6A–8A it can be con-
and Petrobras (either following a conservative or an cluded that Portfolios 10 and 11 are identical to the
aggressive strategy). From the risk minimisation ones obtained with the previous modelling formula-
stand point (either assuming an aggressive or a con- tion (i.e. Portfolios 1 and 2). When the maximisa-
servative stance) McDonalds, America, Vale, Taiwan tion of return takes place the results are always
and Coca-Cola seem to be four good investment similar because the optimisation problems are iden-
options. If a more balanced perspective towards risk tical regarding return. However, if risk minimisation
and return is grasped, the best options are is sought the results are different (because the risk
McDonalds, America, Coca-Cola, Apple and Vale. If measure used is distinct). In this case, it can be
an aggressive strategy is followed (irrespective of established that with this last modelling formulation
risk assumptions), the majority of investment is the volatility of return (herein measured through
placed in Apple, PT, America, McDonalds and the width of the feasible range of variation of
14 C. O. HENRIQUES AND M. E. D. NEVES

Figure 6. Width of the interval log return with both models (recovery).

Figure 7. Risk vs. log return in the mean-absolute model (pre-crisis).

returns in each portfolio) becomes generally reduced another company from the communication services
except in the case of portfolios 12 (in all business (America) is added to the portfolio; in the recovery
cycles) and portfolio 18 in the period of crisis – see period Apple significantly loses its weight. Following
Figures 4–6. These results suggest that the investor this same strategy, a strong risk prone investor
that is looking for lower volatility of returns should (Portfolios 16) should assign the majority of his/her
prefer the mean-absolute modelling approach. investment to the next stocks: Vale, America and
Overall, results also show that by following a EDP during the period of pre-crisis; Coca-Cola,
more aggressive strategy towards risk, the Portfolios McDonalds, America and Petrobras through the
selected become, in general, less diversified with this period of crisis; Apple, McDonalds and Coca-Cola
modelling formulation vis-a-vis the previous one in the period of recovery.
(particularly in the pre-crisis and recovery periods) Finally, portfolios 17 and 18 reveal a balanced
– see Portfolios 4 vs. 13. With this strategy there is propensity to risk but following strongly aggressive
a particular concentration of investment in two and conservative strategies, correspondingly.
stocks from the industry of non-alcoholic beverages The trade-off between risk and return is depicted
(Coca-Cola) and from a regulated utility (EDP) if in Figures 7–9, clearly indicating that portfolios with
the pre-crisis period is sought; from Coca-Cola and higher risk also involve higher return. Once more, it
communication services (DT and America) if the is worth mentioning that an aggressive strategy
crisis period is used; and, finally, during recovery, towards return leads to a substantial increase of
from restaurant services (McDonalds), semiconduc- risk, particularly in the period of crisis.
tors industries (Taiwan) and Coca-Cola. As depicted in Table 9A (see Appendix B) the
The combined strategy (see Portfolios 14, 15 and assets which are more frequently selected regardless
16) allocates a pessimistic index of 0.5 to all the of the strategy followed are Coca-Cola, McDonalds,
objective functions and to the threshold of the inter- America and Apple. Finally, Table 10A (see
val constraints. The portfolios corresponding to a Appendix B) also indicates that the best assets to
balanced approach regarding risk (see Portfolios 14 select from the point of view of the maximisation of
according to all business cycles) render the highest return are Apple, America, McDonalds and Coca-
percentage of investment in stocks from: Coca-Cola, Cola (regardless of the strategy followed). From the
EDP, America and Vale during the period of pre- point of view of risk minimisation (considering all
crisis; Coca-Cola, McDonalds and DT in the period investment strategies), Coca-Cola, McDonalds,
of crisis; Coca-Cola, McDonalds, Apple and Taiwan America and EDP seem to be good investment
in the period of recovery. In addition to the compa- options. With the pursuit of a more balanced per-
nies of Portfolios 14, a strong risk averter spective towards risk and return, the best three
(Portfolios 15) should assign a higher weight to options are Coca-Cola, McDonalds and America.
Vale against America; during the period of crisis When an aggressive strategy is followed (irrespective
 JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 15

Figure 8. Risk vs. log return in the mean-absolute model (crisis).

Figure 9. Risk vs. log return in the mean-absolute model (recovery).

of risk assumptions), the majority of investment is A new solution method for obtaining efficient
placed in Coca-Cola, McDonalds, Apple and portfolios has also been suggested which allows con-
America. On the other hand, if a conservative strat- sidering three types of investment strategies, i.e., a
egy is followed (regardless of risk presumptions), conservative strategy, an aggressive strategy and a
the investment is mainly channeled to Coca-Cola, combined strategy. Additionally, indices of robust-
McDonalds, America and Apple. Finally, a com- ness have also been computed, which allow apprais-
bined strategy leads to assign the majority of the ing the assets which are more frequently chosen
investment to Coca-Cola, McDonalds and America. regardless of the investment strategy followed, and
which are more commonly selected from the point
of view of each investment strategy irrespective of
5. Conclusions
the business cycle considered.
A new modelling framework based on portfolio the- This study contributes to an understanding of the
ory was proposed which accounts for the uncer- implications of investment decisions under different
tainty handling by means of interval coefficients risk assumption (i.e., mean-absolute deviation and
both in the objective functions and in the con- mean semi-absolute deviation). Our results show
straints. Additionally, new surrogate problems were that under a conservative strategy (i.e., with higher
developed for the mean-absolute deviation risk risk concerns) the portfolios obtained are always
measure by using the concept of necessary subtrac- more diversified, leading to less extreme asset alloca-
tion between interval numbers, thus ensuring the tions. Furthermore, our findings highlight the trade-off
smallest deviation of the return of each portfolio between risk and return, indicating that depending on
from its corresponding interval expected returns. the strategy followed by the investor, higher risk also
The usefulness of the approach proposed has generates higher return and vice-versa.
been illustrated across different business cycles, i.e., Overall, results also indicate that the investor that
pre-crisis, crisis and recovery. Then, by considering pursues lower volatility of returns should rather pre-
the mean returns obtained in these distinct eco- fer the mean-absolute modelling approach.
nomic scenarios after removing the corresponding Future work is currently under way in order to
outliers, confidence intervals were computed for encompass other axes of evaluation, such as liquidity
each economic period, assuming a confidence level as well as cardinality constraints, extending this type
of 95%. The boundaries of such confidence intervals of modelling approach for the consideration of
were then given as the upper and lower bounds of multi-period portfolio selection models also includ-
the logarithmic returns instantiated into the models. ing transaction costs. Finally, future research
16 C. O. HENRIQUES AND M. E. D. NEVES

directions should also include out-of-sample tests, In International Conference on Fuzzy Systems and
i.e., tests on future time periods. Knowledge Discovery (pp. 766–775). Berlin, Heidelberg:
Springer.
Gioia, F. (2009). Portfolio selection model with interval
Notes data. Statistica Applicata, 21(1), 1–31.
Giove, S., Funari, S., & Nardelli, C. (2006). An interval
1. On May 29, 2015, the shareholders decided to portfolio selection problem based on regret function.
change the name to PHAROL, SGPS S.A. European Journal of Operational Research, 170(1),
2. For e.g. in the case of EDP in the pre-crisis period it 253–264. doi:10.1016/j.ejor.2004.05.030
corresponds to ((0 þ 0þ0.2 þ 0.158 þ 0.217 þ 0.27 þ Gupta, P., Mehlawat, M. K., Inuiguchi, M., & Chandra, S.
0þ0.035 þ 0.251)/9)  100 ¼ 13%. (2014). Fuzzy portfolio optimization. In Advances in
Hybrid Multi-criteria Methodologies (p. 316). Berlin,
Germany: Springer.
Disclosure statement Henriques, C. O., Luque, M., Marcenaro-Gutierrez, O. D.,
No potential conflict of interest was reported by & Lopez-Agudo, L. A. (2018). A multiobjective interval
the authors. programming model to explore the trade-offs among
different aspects of job satisfaction under different
scenarios. Socio-Economic Planning Sciences, [in press].
Funding doi: 10.1016/j.seps.2018.07.004.
Ida, M. (2003). Portfolio selection problem with interval
This work has been supported by the Portuguese coefficients. Applied Mathematics Letters, 16(5),
Foundation for Science and Technology (FCT) under pro- 709–713. doi:10.1016/S0893-9659(03)00071-5
ject grant UID/Multi/00308/2019 and it was also sup- Inuiguchi, M., & Kume, Y. (1991). Goal programming
ported by: European Structural and Investment Funds in problems with interval coefficients and target intervals.
the FEDER component, through the Operational European Journal of Operational Research, 52(3),
Competitiveness and internationalisation Programme 345–360. doi:10.1016/0377-2217(91)90169-V
(COMPETE 2020) [Project No. 006971 (UID/SOC/ Jong, Y. (2012). Optimization method for interval port-
04011); Funding Reference: POCI-01- 0145-FEDER- folio selection based on satisfaction index of interval
006971]; and national funds, through the FCT – inequality relation. arXiv Preprint arXiv, 1207, 1932.
Portuguese Foundation for Science and Technology under Konno, H., & Yamazaki, H. (1991). Mean-absolute devi-
the project UID/SOC/04011/2013. ation portfolio optimization model and its applications
to Tokyo stock market. Management Science, 37(5),
519–531. doi:10.1287/mnsc.37.5.519
Kumar, P., Panda, G., & Gupta, U. C. (2015). Portfolio
References rebalancing model with transaction costs using interval
optimization. Opsearch, 52(4), 827–860. doi:10.1007/
Antunes, C. H., Alves, M. J., & Cl
ımaco, J. (2016).
s12597-015-0210-0
Multiobjective linear and integer programming. Schweiz: Kumar, P., Panda, G., & Gupta, U. C. (2016). An interval
Springer International Publishing. linear programming approach for portfolio selection
Bao, F., Zhu, P. P., & Zhao, P. B. (2010). Portfolio selec- model. International Journal of Operational Research,
tion problems based on fuzzy interval numbers under 27(1/2), 149–164. doi:10.1504/IJOR.2016.10000053
the Minimax rules. International Journal of Lai, K. K., Wang, S. Y., Xu, J. P., Zhu, S. S., & Fang, Y.
Mathematical Analysis, 4, 2143–2166. (2002). A class of linear interval programming prob-
Bhattacharyya, R., Kar, S., & Majumder, D. D. (2011). lems and its application to portfolio selection. IEEE
Fuzzy mean–variance–skewness portfolio selection Transactions on Fuzzy Systems, 10(6), 698–704. doi:
models by interval analysis. Computers & Mathematics 10.1109/TFUZZ.2002.805902
with Applications, 61(1), 126–137. doi:10.1016/ Li, C., & Jin, J. (2018). A new portfolio selection model
j.camwa.2010.10.039 with interval-typed random variables and the empirical
Biglova, A., Ortobelli, S., Rachev, S. T., & Stoyanov, S. analysis. Soft Computing, 22(3), 905–920. doi:10.1007/
(2004). Different approaches to risk estimation in port- s00500-016-2396-3
folio theory. The Journal of Portfolio Management, Li, C., Wang, D., & Jin, J. (August, 2015). Portfolio selec-
31(1), 103–112. doi:10.3905/jpm.2004.443328 tion models based on interval numbers. In IEEE 11th
Cesarone, F., Scozzari, A., & Tardella, F. (2011). Portfolio International Conference on Natural Computation
selection problems in practice: a comparison between (ICNC), 2015 (pp. 349–353). Zhangjiajie, China: IEEE.
linear and quadratic optimization models. arXiv pre- Liu, S. T. (2011). The mean-absolute deviation portfolio
print arXiv:1105.3594. selection problem with interval-valued returns. Journal
Chinneck, J. W., & Ramadan, K. (2000). Linear program- of Computational and Applied Mathematics, 235(14),
ming with interval coefficients. The Journal of the 4149–4157. doi:10.1016/j.cam.2011.03.008
Operational Research Society, 51(2), 209–220. doi: Liu, Y. J., Zhang, W. G., & Wang, J. B. (2016). Multi-
10.2307/254261 period cardinality constrained portfolio selection mod-
Chunhachinda, P., Dandapani, K., Hamid, S., & Prakash, els with interval coefficients. Annals of Operations
A. J. (1997). Portfolio selection and skewness: Evidence Research, 244(2), 545–569. doi:10.1007/s10479-016-
from international stock markets. Journal of Banking & 2117-4
Finance, 21(2), 143–167. doi:10.1016/S0378- Liu, Y. J., Zhang, W. G., & Zhang, P. (2013). A multi-
4266(96)00032-5 period portfolio selection optimization model by using
Fang, Y., & Wang, S. (2006). September. An interval interval analysis. Economic Modelling, 33, 113–119. doi:
semi-absolute deviation model for portfolio selection. 10.1016/j.econmod.2013.03.006
 JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY 17

Longerstaey, J., & Spencer, M. (1996). RiskMetricsTM, Rockafellar, R. T., & Uryasev, S. (2000). Optimization of
Technical Document. New York, NY: Morgan Guaranty conditional value-at-risk. The Journal of Risk, 2(3),
Trust Company of New York. 21–42. doi:10.21314/JOR.2000.038
Mansini, R., & Speranza, M. G. (1999). Heuristic algo- Rudolf, M., Wolter, H. J., & Zimmermann, H. (1999). A
rithms for the portfolio selection problem with min- linear model for tracking error minimization. Journal
imum transaction lots. European Journal of Operational of Banking & Finance, 23(1), 85–103. doi:10.1016/
Research, 114(2), 219–233. doi:10.1016/S0377- S0378-4266(98)00076-4
2217(98)00252-5 Şerban, F., Costea, A., & Ferrara, M. (2015). Portfolio
Mansini, R., Ogryczak, W., & Speranza, M. G. (2003). LP optimization using interval analysis. Economic compu-
solvable models for portfolio optimization: a classifica- tation and economic cybernetics studies and research.
tion and computational comparison. IMA Journal of ASE Publishing, 48(1), 125–137.
Management Mathematics, 14(3), 187–220. doi:10.1093/ Speranza, M. G. (1993). Linear programming models for
imaman/14.3.187 portfolio optimization. Finance, 14, 107–123.
Markowitz, H. (1952). Portfolio selection. The Journal of Tan, M. Y. (2012). Interval number model for portfolio selection
with liquidity constraints. In Fuzzy engineering and opera-
Finance, 7(1), 77–91.
tions research (pp. 31–39). Berlin, Heidelberg: Springer.
Markowitz, H. M. (1968). Portfolio selection: Efficient
Young, M. R. (1998). A minimax portfolio selection rule
diversification of investments (Vol. 16). New Haven,
with linear programming solution. Management
CT: Yale University Press.
Science, 44(5), 673–683. doi:10.1287/mnsc.44.5.673
Milesi-Ferretti, G. M., & Tille, C. (2011). The great
Zhang, P. (2016). An interval mean–average absolute devi-
retrenchment: international capital flows during the
ation model for multiperiod portfolio selection with risk
global financial crisis. Economic Policy, 26(66), control and cardinality constraints. Soft Computing, 20(3),
289–346. doi:10.1111/j.1468-0327.2011.00263.x 1203–1212. doi:10.1007/s00500-014-1583-3
Mitra, G., Kyriakis, T., Lucas, C., & Pirbhad, M. (2003). Zhang, P. (2017). Multiperiod mean semi-absolute deviation
A review of portfolio planning: Models and systems. In interval portfolio selection with entropy constraints.
Advances in portfolio construction and implementation Journal of Industrial and Management Optimization,
(pp. 1–39). 13(3), 1169–1187. doi:10.3934/jimo.2016067
Mobarek, A., Mollah, S., & Keasey, K. (2014). A cross-
country analysis of herd behavior in Europe. Journal of
International Financial Markets, Institutions and Appendix A
Money, 32, 107–127. doi:10.1016/j.intfin.2014.05.008
Oliveira, C., & Antunes, C. H. (2007). Multiple objective MOLP problems with interval coefficients
linear programming models with interval coefficient- Let the intervals A ¼ [aL, aU] and B ¼ [bL, bU] be given
s–an illustrated overview. European Journal of in the set of real interval numbers. The subtraction of
Operational Research, 181 (3), 1434–1463. doi:10.1016/ two interval numbers is usually defined as:
j.ejor.2005.12.042      
Ortobelli, S., Rachev, S. T., Stoyanov, S., Fabozzi, F. J., & Að–ÞB ¼ aL ; aU ð–Þ bL ; bU ¼ aL –bU ; aU –bL :
Biglova, A. (2005). The proper use of risk measures in (1A)
portfolio theory. International Journal of Theoretical
The addition of two interval number is also generally
and Applied Finance, 8(8), 1107–1133. doi:10.1142/ given as:
S0219024905003402      
Papahristodoulou, C., & Dotzauer, E. (2004). Optimal AðþÞB ¼ aL ; aU ðþÞ bL ; bU ¼ aL þ bL ; aU þ bU :
portfolios using linear programming models. Journal of (2A)
the Operational Research Society, 55(11), 1169–1177.
Let the following interval equation be defined as:
doi:10.1057/palgrave.jors.2601765
Pei, J., Cheng, B., Liu, X., Pardalos, P., & Kong, M. X ðþÞB ¼ A; (3A)
(2019a). Single-machine and parallel-machine serial- where A ¼ [a , a ], B ¼ [b , b ] and X ¼ [x , x ] is the
L U L U L U
batching scheduling problems with position-based interval unknown.
learning effect and linear setup time. Annals of It is also known that X 6¼ A (–) B ¼ [aL – bU, aU –
Operations Research, 272(1–2), 217–241. doi: 10.1007/ L
b ] since:
s10479-017-2481-8  L U U L    1  
Pei, J., Liu, X., Fan, W., Pardalos, P. M., & Lu, S. (2019b). A a –b ; a –b ðþÞ bL ; bU ¼ aL –bU þ bL ; aU –bL þ bU aL ; aU :
hybrid BA-VNS algorithm for coordinated serial-batching (4A)
scheduling with deteriorating jobs, financial budget, and
The necessary condition for existing a solution to this
resource constraint in multiple manufacturers. Omega, 82, interval equation is that w[B]  w[A], since w[X (þ) B]
55–69. doi:10.1016/j.omega.2017.12.003 ¼ w[X] þ w[B] ¼ w[A] and w[X]  0, where w[A] ¼ aU
Pei, J., Pardalos, P., Liu, X., Fan, W., & Yang, S. (2015). – aL, w[B] ¼ bU – bL and w[X] ¼ xU – xL
Serial batching scheduling of deteriorating jobs in a When w[B]  w[A] we have,
two-stage supply chain to minimize the makespan.    
European Journal of Operational Research, 244(1), X ðþÞB ¼ xL þ bL ; xU þ bU ¼ aL ; aU ;
13–25. doi:10.1016/j.ejor.2014.11.034 i:e; xL þ bL ¼ aL ; xU þ bU ¼ aU : (5A)
Pei, J., Wang, X., Fan, W., Pardalos, P., & Liu, X. (2018).
Scheduling step-deteriorating jobs on bounded parallel- Therefore, the solution to the previous equation is
 
batching machines to maximize the total net revenue. X ¼ aL –bL ; aU –bU : (6A)
Journal of the Operational Research Society, 1–18. doi:
10.1080/01605682.2018.1464428
18 C. O. HENRIQUES AND M. E. D. NEVES

h i
From this perspective another version of subtraction where c kj 2 cLkj ; cUkj for all k ¼ 1; :::; p,
between interval numbers can also be defined as A )–( B, h
i
which is called “necessary” subtraction (Inuiguchi & j ¼ 1; :::; n; a ij 2 aLij ; aUij , for all i ¼ 1; :::; m,
Kume, 1991). Thus, when we have w[B]  w[A],  
  j ¼ 1; :::; n and b i 2 bLi ; bUi ; for all i ¼ 1; :::; m.
AÞ–ðB ¼ aL –bL ; aU –bU : (7A) The optimal values with the upper bounds of each
objective function using the widest version of the feasible
Analogously, another version for the addition of two
region (best case scenario in a maximisation problem)
interval numbers given as A)þ(B and called “necessary”
and with the lower bounds of each objective function and
addition is obtained when w[B]  w[A] (Inuiguchi &
the tightest version of the feasible region (worst case
Kume,1991):
  scenario in a maximisation problem) can be obtained by
AÞþðB ¼ aL þ bU ; aU þ bL : (8A) solving the following linear programming problems,
respectively (Chinneck & Ramadan, 2000):
The relation between A (–) B and A )–( B is
X
n
AÞ–ðB
IAð–ÞB: (9A) max ZkU ðxÞ ¼ cUkj xj ;
j¼1
The difference between these two subtraction concepts
X
n
for two overlapped intervals is illustrated in Figure 1A. s:t: : aLij xj  bUi ; i ¼ 1; :::; m;
Now consider, without loss of generality, the following j¼1
multiobjective programming model with interval
xj  0; j ¼ 1; :::; n: (12A)
coefficients:
Xn h i and
max Zk ðxÞ ¼ cLkj ; cUkj xj ; k ¼ 1; :::; p;
j¼1
X
n
max ZkL ðxÞ ¼ cLkj xj ;
n h
X i   j¼1
s:t: : aLij ; aUij xj  bLi ; bUi ; i ¼ 1; :::; m;
j¼1
X
n
s:t: : aUij xj  bLi ; i ¼ 1; :::; m;
xj  0; j ¼ 1; :::; n: (10A) j¼1

where x ¼ ðx1 ; :::; xn ÞT is the vector of variables. xj  0; j ¼ 1; :::; n: (13A)

h i h i
If the coefficients cLkj ; cUkj ; aLij ; aUij and ½bLi ; bUi  are Let models (12A) and (13A) be identified, respectively,
with p ¼ 0 and p ¼ 1 and the optimal solution to
real values, then problem (10A) becomes a MOLP each model be given by xpk , k ¼ 1; :::; p. Then the
problem: optimal values are denoted by:
  
X
n
ZkU ¼ ZkU x0k ; k ¼ 1; :::; p; (14A)
max Z k ðxÞ ¼ c kj xj ; k ¼ 1; :::; p; L
 
j¼1 Zk ¼ Zk xk ; k ¼ 1; :::; p:
L 1
(15A)
X
n
s:t: : a ij xj  b i ; i ¼ 1; :::; m;
j¼1
Appendix B
xj  0; j ¼ 1; :::; n: (11A)
Results
Table 1A. Proportion of assets obtained using different strategies with the mean semi-absolute deviation model (pre-crisis).
Portfolio 1 Portfolio 2 Portfolio 3 Portfolio 4 Portfolio 5 Portfolio 6 Portfolio 7 Portfolio 8 Portfolio 9
Conservative Aggressive Conservative Aggressive Combined Combined Combined Strongly aggressive Strongly conservative
Strategy Highly prone Highly prone Highly averter Highly averter Balanced Strongly averter Strongly prone Balanced Balanced
Risk stance u ¼ 1, s ¼ 0, u ¼ 1, s ¼ 0, u ¼ 0, s ¼ 1, u ¼ 0, s ¼ 1, u ¼ 0.5, s ¼ 0.5, u ¼ 0.3, s ¼ 0.7, u ¼ 0.7, s ¼ 0.3, u ¼ 0.5, s ¼ 0.5, u ¼ 0.5, s ¼ 0.5,
Companies q¼1 q¼0 q¼1 q¼0 q ¼ 0.5 q ¼ 0.5 q ¼ 0.5 q ¼ 0.3 q ¼ 0.7
PT 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
EDP 0.000 0.000 0.200 0.158 0.217 0.270 0.000 0.033 0.290
DT 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
BMW 0.000 0.000 0.000 0.058 0.000 0.000 0.000 0.000 0.000
Adidas 0.200 0.000 0.070 0.000 0.006 0.000 0.125 0.094 0.057
Microsoft 0.000 0.000 0.200 0.089 0.000 0.000 0.000 0.000 0.000
Apple 0.200 0.500 0.000 0.075 0.005 0.066 0.007 0.000 0.047
McDonalds 0.000 0.000 0.000 0.000 0.018 0.000 0.101 0.079 0.000
Coca-Cola 0.000 0.000 0.195 0.040 0.000 0.079 0.000 0.000 0.023
Pfizer 0.000 0.000 0.200 0.324 0.000 0.109 0.000 0.000 0.000
Petrobras 0.200 0.000 0.000 0.000 0.054 0.002 0.067 0.030 0.002
Vale 0.200 0.000 0.135 0.217 0.350 0.311 0.350 0.410 0.290
America 0.200 0.500 0.000 0.000 0.350 0.162 0.350 0.354 0.290
Sify 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Taiwan 0.000 0.000 0.000 0.039 0.000 0.000 0.000 0.000 0.000

Table 2A. Proportion of assets obtained using different strategies with the mean semi-absolute deviation model (crisis).
Portfolio 1 Portfolio 2 Portfolio 3 Portfolio 4 Portfolio 5 Portfolio 6 Portfolio 7 Portfolio 8 Portfolio 9
Conservative Aggressive Conservative Aggressive Combined Combined Combined Strongly aggressive Strongly conservative
Strategy Highly prone Highly prone Highly averter Highly averter Balanced Strongly averter Strongly prone Balanced Balanced
Risk stance u ¼ 1, s ¼ 0, u ¼ 1, s ¼ 0, u ¼ 0, s ¼ 1, u ¼ 0, s ¼ 1, u ¼ 0.5, s ¼ 0.5, u ¼ 0.3, s ¼ 0.7, u ¼ 0.7, s ¼ 0.3, u ¼ 0.5, s ¼ 0.5, u ¼ 0.5, s ¼ 0.5,
Companies q¼1 q¼0 q¼1 q¼0 q ¼ 0.5 q ¼ 0.5 q ¼ 0.5 q ¼ 0.3 q ¼ 0.7
PT 0.000 0.500 0.000 0.342 0.000 0.000 0.267 0.410 0.000
EDP 0.200 0.000 0.111 0.000 0.000 0.000 0.000 0.000 0.000
DT 0.000 0.000 0.200 0.004 0.008 0.253 0.000 0.000 0.010
BMW 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Adidas 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Microsoft 0.000 0.000 0.000 0.048 0.000 0.000 0.000 0.000 0.000
Apple 0.200 0.000 0.089 0.000 0.013 0.000 0.145 0.000 0.231
McDonalds 0.200 0.000 0.200 0.203 0.350 0.350 0.033 0.345 0.290
Coca-Cola 0.200 0.000 0.200 0.000 0.313 0.187 0.205 0.000 0.290
Pfizer 0.000 0.000 0.000 0.013 0.000 0.000 0.000 0.000 0.000
Petrobras 0.000 0.500 0.000 0.000 0.064 0.000 0.350 0.245 0.000
Vale 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
America 0.200 0.000 0.200 0.000 0.253 0.145 0.000 0.000 0.180
Sify 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Taiwan 0.000 0.000 0.000 0.389 0.000 0.066 0.000 0.000 0.000
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY
19
 20

Table 3A. Proportion of assets obtained using different strategies with the mean semi-absolute deviation model (recovery).
Portfolio 1 Portfolio 2 Portfolio 3 Portfolio 4 Portfolio 5 Portfolio 6 Portfolio 7 Portfolio 8 Portfolio 9
Conservative Aggressive Conservative Aggressive Combined Combined Combined Strongly aggressive Strongly conservative
Strategy Highly prone Highly prone Highly averter Highly averter Balanced Strongly averter Strongly prone Balanced Balanced
Risk stance u ¼ 1, s ¼ 0, u ¼ 1, s ¼ 0, u ¼ 0, s ¼ 1, u ¼ 0, s ¼ 1, u ¼ 0.5, s ¼ 0.5, u ¼ 0.3, s ¼ 0.7, u ¼ 0.7, s ¼ 0.3, u ¼ 0.5, s ¼ 0.5, u ¼ 0.5, s ¼ 0.5,
Companies q¼1 q¼0 q¼1 q¼0 q ¼ 0.5 q ¼ 0.5 q ¼ 0.5 q ¼ 0.3 q ¼ 0.7
PT 0.000 0.000 0.000 0.117 0.000 0.000 0.000 0.000 0.000
EDP 0.000 0.000 0.000 0.207 0.000 0.000 0.000 0.000 0.000
DT 0.000 0.000 0.087 0.000 0.000 0.000 0.000 0.000 0.000
BMW 0.000 0.500 0.000 0.044 0.088 0.056 0.350 0.295 0.000
Adidas 0.200 0.000 0.000 0.000 0.115 0.000 0.000 0.000 0.000
Microsoft 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Apple 0.200 0.500 0.000 0.000 0.153 0.156 0.350 0.410 0.290
McDonalds 0.200 0.000 0.200 0.100 0.350 0.350 0.300 0.295 0.290
Coca-Cola 0.200 0.000 0.200 0.000 0.247 0.196 0.000 0.000 0.290
Pfizer 0.000 0.000 0.184 0.000 0.000 0.000 0.000 0.000 0.000
C. O. HENRIQUES AND M. E. D. NEVES

Petrobras 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Vale 0.000 0.000 0.000 0.076 0.048 0.000 0.000 0.000 0.000
America 0.000 0.000 0.129 0.220 0.000 0.000 0.000 0.000 0.000
Sify 0.200 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.021
Taiwan 0.000 0.000 0.200 0.235 0.000 0.243 0.000 0.000 0.109
PT

DT

Sify
EDP

Vale
BMW

Pfizer
Apple
Adidas

Taiwan
America
Microsoft

Petrobras
Coca-Cola
McDonalds
Business cycle

0%
0%
25%
25%
4%
7%
4%
2%
10%
3%
6%
1%
0%
13%
0%
Pre-crisis
absolute deviation model.

5%
0%
11%
0%
13%
0%
15%
22%
8%
1%
0%
0%
5%
3%
17%
Crisis

24%
0%
0%
0%
0%
0%
20%
35%
16%
0%
0%
6%
0%
0%
0%
Recovery

10%
0%
12%
8%
6%
2%
13%
20%
11%
1%
2%
2%
2%
5%
6%
Table 4A. Assets which are more often selected irrespective

Average
of the investment strategy followed with the mean semi-
Table 5A. Assets which are more often selected irrespective of the business cycle contemplated with the mean semi-absolute deviation model.
1 2 3 4 5 6 7 8 9
Portfolio Cons. Aggr. Cons. Aggr. Comb. Comb. Comb. Strongly Aggr. Strongly Cons. Cons. Aggr. Comb. – – –
Strategy Highly prone Highly prone Highly averter Highly averter Balanced Strongly averter Strongly prone Balanced Balanced – – – Risk prone Risk averter Balanced
Risk stance u ¼ 1, s ¼ 0, u ¼ 1, s ¼ 0, u ¼ 0, s ¼ 1, u ¼ 0, s ¼ 1, u ¼ 0.5, s ¼ 0.5, u ¼ 0.3, s ¼ 0.7, u ¼ 0.7, s ¼ 0.3, u ¼ 0.5, s ¼ 0.5, u ¼ 0.5, s ¼ 0.5,
Companies q¼1 q¼0 q¼1 q¼0 q ¼ 0.5 q ¼ 0.5 q ¼ 0.5 q ¼ 0.3 q ¼ 0.7 – – – – – –
PT 0% 17% 0% 15% 0% 0% 9% 14% 0% 0% 15% 3% 9% 5% 5%
EDP 7% 0% 10% 12% 7% 9% 0% 1% 10% 9% 4% 5% 2% 9% 6%
DT 0% 0% 10% 0% 0% 8% 0% 0% 0% 3% 0% 3% 0% 3% 0%
BMW 0% 17% 0% 3% 3% 2% 12% 10% 0% 0% 10% 5% 9% 3% 4%
Adidas 13% 0% 2% 0% 4% 0% 4% 3% 2% 6% 1% 3% 6% 1% 3%
Microsoft 0% 0% 7% 5% 0% 0% 0% 0% 0% 2% 2% 0% 0% 2% 0%
Apple 20% 33% 3% 3% 6% 7% 17% 14% 19% 14% 17% 10% 23% 5% 13%
McDonalds 13% 0% 13% 10% 24% 23% 14% 24% 19% 15% 11% 21% 9% 19% 22%
Coca-Cola 13% 0% 20% 1% 19% 15% 7% 0% 20% 18% 0% 14% 7% 12% 13%
Pfizer 0% 0% 13% 11% 0% 4% 0% 0% 0% 4% 4% 1% 0% 5% 0%
Petrobras 7% 17% 0% 0% 4% 0% 14% 9% 0% 2% 9% 6% 12% 1% 4%
Vale 7% 0% 5% 10% 13% 10% 12% 14% 10% 7% 8% 12% 6% 11% 12%
America 13% 17% 11% 7% 20% 10% 12% 12% 16% 13% 12% 14% 14% 13% 16%
Sify 7% 0% 0% 0% 0% 0% 0% 0% 1% 2% 0% 0% 2% 0% 0%
Taiwan 0% 0% 7% 22% 0% 10% 0% 0% 4% 3% 7% 3% 0% 11% 1%

Table 6A. Proportion of assets obtained using different strategies with the mean-absolute deviation model (pre-crisis).
Portfolio 10 Portfolio 11 Portfolio 12 Portfolio 13 Portfolio 14 Portfolio 15 Portfolio 16 Portfolio 17 Portfolio 18
Conservative Aggressive Conservative Aggressive Combined Combined Combined Strongly aggressive Strongly conservative
Strategy Highly prone Highly prone Highly averter Highly averter Balanced Strongly averter Strongly prone Balanced Balanced
Risk stance b ¼ 1, a ¼ 0, b ¼ 1, a ¼ 0, b ¼ 0, a ¼ 1, b ¼ 0, a ¼ 1, b ¼ 0.5, a ¼ 0.5, b ¼ 0.3, a ¼ 0.7, b ¼ 0.7, a ¼ 0.3, b ¼ 0.5, a ¼ 0.5, b ¼ 0.5, a ¼ 0.5,
Companies q¼1 q¼0 q¼1 q¼0 q ¼ 0.5 q ¼ 0.5 q ¼ 0.5 q ¼ 0.3 q ¼ 0.7
PT 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
EDP 0.000 0.000 0.300 0.351 0.259 0.307 0.273 0.255 0.270
DT 0.000 0.000 0.015 0.000 0.000 0.000 0.000 0.000 0.000
BMW 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Adidas 0.200 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Microsoft 0.000 0.000 0.000 0.059 0.000 0.000 0.000 0.000 0.000
Apple 0.200 0.500 0.020 0.011 0.024 0.039 0.016 0.023 0.029
McDonalds 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Coca-Cola 0.000 0.000 0.300 0.500 0.350 0.350 0.011 0.377 0.290
Pfizer 0.000 0.000 0.107 0.017 0.000 0.033 0.000 0.000 0.012
Petrobras 0.200 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Vale 0.200 0.000 0.135 0.044 0.157 0.134 0.350 0.143 0.176
America 0.200 0.500 0.105 0.000 0.210 0.130 0.350 0.202 0.224
Sify 0.000 0.000 0.018 0.019 0.000 0.008 0.000 0.000 0.000
Taiwan 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY
21
 22

Table 7A. Proportion of assets obtained using different strategies with the mean-absolute deviation model (crisis).
Portfolio 10 Portfolio 11 Portfolio 12 Portfolio 13 Portfolio 14 Portfolio 15 Portfolio 16 Portfolio 17 Portfolio 18
Conservative Aggressive Conservative Aggressive Combined Combined Combined Strongly aggressive Strongly conservative
Strategy Highly prone Highly prone Highly averter Highly averter Balanced Strongly averter Strongly prone Balanced Balanced
Risk stance b ¼ 1, a ¼ 0, b ¼ 1, a ¼ 0, b ¼ 0, a ¼ 1, b ¼ 0, a ¼ 1, b ¼ 0.5, a ¼ 0.5, b ¼ 0.3, a ¼ 0.7, b ¼ 0.7, a ¼ 0.3, b ¼ 0.5, a ¼ 0.5, b ¼ 0.5, a ¼ 0.5,
Companies q¼1 q¼0 q¼1 q¼0 q ¼ 0.5 q ¼ 0.5 q ¼ 0.5 q ¼ 0.3 q ¼ 0.7
PT 0.000 0.500 0.000 0.000 0.000 0.000 0.000 0.000 0.000
EDP 0.200 0.000 0.162 0.000 0.000 0.000 0.014 0.000 0.000
DT 0.000 0.000 0.200 0.296 0.142 0.180 0.000 0.142 0.175
BMW 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Adidas 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Microsoft 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Apple 0.200 0.000 0.015 0.000 0.044 0.019 0.010 0.007 0.060
McDonalds 0.200 0.000 0.200 0.109 0.324 0.263 0.346 0.275 0.290
Coca-Cola 0.200 0.000 0.200 0.333 0.350 0.350 0.350 0.410 0.290
Pfizer 0.000 0.000 0.023 0.000 0.000 0.000 0.000 0.000 0.000
C. O. HENRIQUES AND M. E. D. NEVES

Petrobras 0.000 0.500 0.000 0.000 0.041 0.000 0.131 0.024 0.015
Vale 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
America 0.200 0.000 0.200 0.212 0.099 0.187 0.149 0.142 0.169
Sify 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Taiwan 0.000 0.000 0.000 0.050 0.000 0.000 0.000 0.000 0.000

Table 8A. Proportion of assets obtained using different strategies with the mean-absolute deviation model (recovery).
Portfolio 10 Portfolio 11 Portfolio 12 Portfolio 13 Portfolio 14 Portfolio 15 Portfolio 16 Portfolio 17 Portfolio 18
Conservative Aggressive Conservative Aggressive Combined Combined Combined Strongly aggressive Strongly conservative
Strategy Highly prone Highly prone Highly averter Highly averter Balanced Strongly averter Strongly prone Balanced Balanced
Risk stance b ¼ 1, a ¼ 0, b ¼ 1, a ¼ 0, b ¼ 0, a ¼ 1, b ¼ 0, a ¼ 1, b ¼ 0.5, a ¼ 0.5, b ¼ 0.3, a ¼ 0.7, b ¼ 0.7, a ¼ 0.3, b ¼ 0.5, a ¼ 0.5, b ¼ 0.5, a ¼ 0.5,
Companies q¼1 q¼0 q¼1 q¼0 q ¼ 0.5 q ¼ 0.5 q ¼ 0.5 q ¼ 0.3 q ¼ 0.7
PT 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
EDP 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
DT 0.000 0.000 0.090 0.041 0.021 0.061 0.000 0.000 0.047
BMW 0.000 0.500 0.000 0.000 0.000 0.000 0.017 0.003 0.000
Adidas 0.200 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Microsoft 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
Apple 0.200 0.500 0.000 0.000 0.135 0.004 0.350 0.229 0.135
McDonalds 0.200 0.000 0.200 0.500 0.350 0.350 0.350 0.410 0.290
Coca-Cola 0.200 0.000 0.200 0.166 0.350 0.350 0.194 0.268 0.290
Pfizer 0.000 0.000 0.007 0.000 0.020 0.000 0.000 0.000 0.000
Petrobras 0.000 0.000 0.000 0.018 0.000 0.000 0.000 0.000 0.000
Vale 0.000 0.000 0.103 0.024 0.000 0.000 0.000 0.000 0.000
America 0.000 0.000 0.200 0.076 0.000 0.043 0.000 0.000 0.000
Sify 0.200 0.000 0.000 0.000 0.000 0.000 0.002 0.000 0.000
Taiwan 0.000 0.000 0.200 0.175 0.124 0.193 0.087 0.091 0.238
 PT

DT

Sify
EDP

Vale
BMW

Pfizer
Apple
Adidas

Taiwan
America
Microsoft

Petrobras
Coca-Cola
McDonalds
Business cycle
deviation model.

0%
0%
21%
15%
2%
2%
24%
0%
10%
1%
2%
0%
0%
22%
0%
Pre-crisis

1%
0%
15%
0%
8%
0%
28%
22%
4%
0%
0%
0%
13%
4%
6%
Crisis

12%
2%
4%
1%
0%
0%
22%
29%
17%
0%
2%
6%
3%
0%
0%
Recovery

4%
1%
13%
5%
3%
1%
25%
17%
10%
0%
1%
2%
5%
9%
2%
Average
of the investment strategy followed with the mean-absolute
Table 9A. Assets which are more often selected irrespective

Table 10A. Assets which are more often selected irrespective of the business cycle contemplated with the mean-absolute deviation model.
1 2 3 4 5 6 7 8 9
Portfolio Cons. Aggr. Cons. Aggr. Comb. Comb. Comb. Strongly Aggr. Strongly Cons. Cons. Aggr. Comb. – – –
Strategy Highly prone Highly prone Highly averter Highly averter Balanced Strongly averter Strongly prone Balanced Balanced – – – Risk prone Risk averter Balanced
Risk stance b ¼ 1, a ¼ 0, b ¼ 1, a ¼ 0, b ¼ 0, a ¼ 1, b ¼ 0, a ¼ 1, b ¼ 0.5, a ¼ 0.5, b ¼ 0.3, a ¼ 0.7, b ¼ 0.7, a ¼ 0.3, b ¼ 0.5, a ¼ 0.5, b ¼ 0.5, a ¼ 0.5,
Companies q¼1 q¼0 q¼1 q¼0 q ¼ 0.5 q ¼ 0.5 q ¼ 0.5 q ¼ 0.3 q ¼ 0.7 – – – – – –
PT 0% 17% 0% 0% 0% 0% 0% 0% 0% 0% 6% 0% 6% 0% 0%
EDP 7% 0% 15% 12% 9% 10% 10% 8% 9% 10% 7% 9% 5% 10% 9%
DT 0% 0% 10% 11% 5% 8% 0% 5% 7% 6% 5% 4% 0% 8% 6%
BMW 0% 17% 0% 0% 0% 0% 1% 0% 0% 0% 6% 0% 6% 0% 0%
Adidas 13% 0% 0% 0% 0% 0% 0% 0% 0% 4% 0% 0% 4% 0% 0%
Microsoft 0% 0% 0% 2% 0% 0% 0% 0% 0% 0% 1% 0% 0% 1% 0%
Apple 20% 33% 1% 0% 7% 2% 13% 9% 7% 10% 14% 7% 22% 3% 8%
McDonalds 13% 0% 13% 20% 22% 20% 23% 23% 19% 15% 14% 22% 12% 21% 22%
Coca-Cola 13% 0% 23% 33% 35% 35% 18% 35% 29% 22% 23% 29% 11% 34% 33%
Pfizer 0% 0% 5% 1% 1% 1% 0% 0% 0% 2% 0% 1% 0% 1% 0%
Petrobras 7% 17% 0% 1% 1% 0% 4% 1% 0% 2% 6% 2% 9% 1% 1%
Vale 7% 0% 8% 2% 5% 4% 12% 5% 6% 7% 2% 7% 6% 4% 5%
America 13% 17% 17% 10% 10% 12% 17% 11% 13% 14% 13% 13% 16% 11% 12%
Sify 7% 0% 1% 1% 0% 0% 0% 0% 0% 2% 0% 0% 2% 0% 0%
Taiwan 0% 0% 7% 8% 4% 6% 3% 3% 8% 5% 4% 4% 1% 6% 5%
JOURNAL OF THE OPERATIONAL RESEARCH SOCIETY
23