Expert Systems With Applications 125 (2019) 345–368

Contents lists available at ScienceDirect

Expert Systems With Applications

journal homepage: www.elsevier.com/locate/eswa

Review

A comprehensive review of deterministic models and applications for

mean-variance portfolio optimization
Can B. Kalayci∗, Okkes Ertenlice, Mehmet Anil Akbay
Pamukkale University, Faculty of Engineering, Department of Industrial Engineering, Kinikli, 20160, Denizli, Turkey

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

Article history: Portfolio optimization is the process of determining the best combination of securities and proportions
Received 14 July 2018 with the aim of having less risk and obtaining more profit in an investment. Utilizing covariance as a risk
Revised 6 February 2019
measure, mean-variance portfolio optimization model has brought a revolutionary approach to quantita-
Accepted 7 February 2019
tive finance. Since then, along with the advancements in computational power and algorithmic enhance-
Available online 7 February 2019
ments, a lot of efforts have been made on improving this model by considering real-life conditions and
Keywords: solving model variants with various methodologies tested on various data and performance measures. A
Mean-variance comprehensive literature review of recent and novel papers is crucial to establish a pattern of the past,
Portfolio optimization and to pave the way on future directions. In this paper, a total of 175 papers published in the last two
Literature survey decades are selected within the scope of operations research community and reviewed in detail. Thus, a
Portfolio constraints comprehensive survey on the deterministic models and applications suggested for mean-variance port-
Exact and heuristic algorithms
folio optimization in which several variants of this model as well as additional real-life constraints are
studied. The review classifies the approaches according to exact and approximate attempts and analyzes
the proposed algorithms based on various data and performance indicators in depth. Areas of future re-
search are outlined.
© 2019 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
2. Applied research methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
2.1. Material collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346
2.2. Descriptive analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
2.3. Category selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
2.4. Material evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
3. MVPO model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
3.1. Single-objective MVPO model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
3.2. Multi-objective MVPO model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
3.2.1. Methods for handling multi-objective MVPO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
3.3. Single vs. Multi objective MV model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
3.4. Consideration of real-world constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
4. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
5. Performance measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353
6. Applications on MVPO model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
6.1. Exact solution techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
6.2. Inexact solution techniques. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
6.2.1. Metaheuristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
6.2.2. Machine learning algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360

∗
Corresponding author.
E-mail address: [email protected] (C.B. Kalayci).

https://doi.org/10.1016/j.eswa.2019.02.011
0957-4174/© 2019 Elsevier Ltd. All rights reserved.
346 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

6.3. A peek into the solution approaches for MVPO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360

6.4. Hybridization techniques. . . . . . . . . . ... ...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
6.5. Constraint handling methodology. . . ... ...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
7. Conclusion and discussion . . . . . . . . . . . . . . ... ...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . ... ...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ... ...... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364

1. Introduction a numerical example. Masmoudi and Abdelaziz (2018) focused on

deterministic and stochastic multi-objective programming models
Individual investors, brokers and fund managers invest billions comparing the different assumptions and proposed solutions in
of dollars in various sectors every year. Therefore, proper secu- portfolio optimization. Zhang, Li, and Guo (2018) reviewed vari-
rity selection for financial investment comes into prominence since ous extensions of Markowitz’s MV model such as dynamic, robust,
generating profit throughout all market climates and minimizing fuzzy portfolio optimization with practical factors and pointed out
losses during market downturns are desired. The most common that combined forecasting theory with portfolio selection would
investment strategy is building a portfolio consisting of different be a promising future research direction to deal with uncertainty.
securities in order to spread the risk. Traditional portfolio anal- Aouni, Doumpos, Pérez-Gladish, and Steuer (2018) reviewed multi-
ysis requires the evaluation of return and risk conditions of in- ple criteria decision aid methods for portfolio selection with a fo-
dividual securities and may not provide success due to its sub- cus on exact solution methods on the construction and optimiza-
jective nature. In 1952, performing an analysis of the impact of tion of portfolios as well as on the analysis and the evaluation of
risk, Markowitz presented a revolutionary approach for portfolio specific securities.
theory called Mean-Variance (MV) model (Markowitz, 1952) and The primary purpose of this paper is to deeply analyze the
initiated the era of modern portfolio theory. Using covariance as publications based on deterministic models and applications in
a risk measure is the key of this revolution since it was a spark the MVPO literature. This paper classifies the scientific literature
that triggered quantitative finance. Following this milestone, the presenting a taxonomy of solution techniques on solving MVPO
MV model has been a standard decision-making approach to struc- with the aim of answering the following five research questions:
ture and measure the performance of portfolios (Rubinstein, 2002) (i) Which deterministic models are mostly investigated? (ii) What
that quantitatively focus on the investment alternatives utilizing practical constraints have been introduced for portfolio optimiza-
the covariance between securities based on return-risk trade-off tion? (iii) How have researchers handled problem constraints? (iv)
(Kolm, Tutuncu, & Fabozzi, 2014). Thanks to his pioneering works Which solution techniques have been employed to solve MVPO at
in finance theory, in 1991, Markowitz was awarded the Nobel what rate? (v) Which data sets and performance measures have
Prize for economics. Along with the developments in computa- been used to test the algorithms designed?
tional power, a growing number of researchers from not only fi- Section 2 clarifies the applied research methodology while
nancial fields, but also computer scientists and mathematicians Section 3 is dedicated to the deterministic models for MVPO as
have engaged a great attention to portfolio optimization (PO), ev- well as problem types and additional constraints. Section 4 de-
ident by the vast number of publications in scientific journals to scribes that data while performance measures utilized to test the
deal with mean-variance portfolio optimization (MVPO). New con- proposed algorithms are investigated in Section 5. Section 6 pro-
straints, objectives, and solution approaches have been developed vides a detailed look on the attempts of developing solution ap-
to address shortcomings of the early MV model (Konno & Ya- proaches for MV model while Section 7 concludes the paper and
mazaki, 1991; Rockafellar & Uryasev, 20 0 0; Young, 1998).Thus, a discusses future research directions.
significant number of academic papers has been reached by accu-
mulation of successive additions to MV model. 2. Applied research methodology
Over the last decade, researchers have analyzed the current
trends and future research directions on PO (Table 1). In this Content analysis and description of research methodology
context, Azmi and Tamiz (2010) reviewed lexicographic, weighted, are summarized in four categories: material collection, de-
minmax and fuzzy goal programming models and discussed the is- scriptive analysis, category selection and material evaluation
sues concerning multi-period returns, extended factors and mea- (Govindan, Soleimani, & Kannan, 2015; Özceylan, Kalayci, Güngör,
surement of risk. Metaxiotis and Liagkouras (2012) and Ponsich, & Gupta, 2018).
Jaimes, and Coello (2013) analyzed the current state of re-
search in portfolio optimization with a focus on Multi Ob- 2.1. Material collection
jective Evolutionary Algorithms in which the lack of many
real-life constraints as well as ineffectiveness of Pareto rank- The material for the literature review is detailed in this section.
ing schemes in the presence of many objectives are indicated. The study was conducted by covering the journal papers, book
Mansini, Ogryczak, and Speranza (2014) focused on linear pro- chapters and proceedings written in English language published
gramming solvable models in the portfolio optimization classify- from 1998 to 2018.
ing the models according to decision variables used in the inte- The search terms were identified through several trial and error
gration of real features. Kolm et al. (2014) discussed practical ad- attempts according prior experience of the authors and the key-
vances in MVPO and pointed out new research directions such as words utilized in various papers. A keyword structure (Table 2)
diversification methods and multi-period optimization. Aouni, Co- that initially aims to reach a broad range of search for captur-
lapinto, and La Torre (2014) reviewed the lexicographic, weighted, ing various publications in MVPO literature is carried out. Level
polynomial, stochastic and fuzzy goal programming models and 1 defines the main search context while level 2 indicated vari-
pointed out the lack in developing computerized decision sup- ous alternatives of an author’s approach. Level 3 focused on mean-
port systems to accomplish a helpful tool to facilitate the decision- variance and finally, Fuzzy and Robust keywords are excluded since
making process in portfolio optimization. Doering, Juan, Kizys, Fito, the models dealing with uncertainty and robust portfolio optimiza-
and Calvet (2016) focused on recent contributions of metaheuris- tion are not within the scope of this study with level 4. Utilizing
tics in the sense of an introduction to this topic supported with the ‘‘title, abstract, keywords’’ search in Google Scholar database
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 347

Table 1
Previous reviews on portfolio optimization.

Year Publication Perspective Focus

2010 Azmi and Tamiz (2010) Goal Programming Lexicographic/weighted/minmax/fuzzy goal programming models
2012 Metaxiotis and Liagkouras (2012) Multi-objective optimization Multi-objective Evolutionary Algorithms
2013 Ponsich et al. (2013) Multi-objective optimization Multi-objective Evolutionary Algorithms
2014 Mansini et al. (2014) Linear programming Linear programming solvable portfolio optimization models
2014 Kolm et al. (2014) Practical challenges Practical advances in MVPO
2014 Aouni et al. (2014) Goal programming Lexicographic/weighted/polynomial/stochastic/fuzzy goal programming models
2016 Doering et al. (2016) Computational analysis Metaheuristics
2017 Masmoudi and Abdelaziz (2018) Programming models Deterministic and stochastic multi-objective programming models
2017 Zhang, Li and Guo (2018) Uncertainty Dynamic/robust/fuzzy portfolio optimization with practical factors
2018 Aouni et al. (2018) Multiple criteria decision aid methods Exact methods for portfolio selection possessing multiple criteria
This study Computational analysis Exact and approximate attempts on deterministic models of MVPO

Table 2 spectively. Fig. 2 shows that among the journals, Expert Systems
The proposed keyword combination structure.
with Applications is dominant representing 15% of all published
Level Search terms articles. In Fig. 3 and Fig. 4, it is shown that 88% of proceedings
1 Portfolio are published by IEEE while 95% of book chapters are published by
AND Springer.
2 Selection OR Management OR Optimization
AND
3 Mean-variance 2.3. Category selection
AND NOT
4 Fuzzy OR Robust
The categorization applied in this review is formed based on
different properties of deterministic models and applications on
MVPO. The categories are formed according to the following cri-
which covers more publications than the Web of Science and Sco- teria: models, constraints, constraint handling, solution techniques,
pus databases, initially 34,100 publications are reached with the performance measures and data. The main categories used to
search. In order to obtain relevant publications about determinis- classify the publications are shown in Fig. 5 and are described
tic models and applications of MVPO from an operations research below:
perspective, 175 papers are hand-selected, reviewed, classified and
saved in a spreadsheet for a deeper analysis and evaluation.
2.4. Material evaluation
2.2. Descriptive analysis
To enhance the credibility of the research, during the validation
The publishing trend using the number of publications in a stage, Microsoft Excel spreadsheet and Endnote reference manager
given year is demonstrated in Fig. 1. A total number 175 publi- software are also applied to evaluate, analyze and reduce errors.
cations (20 book chapters, 56 proceedings and 99 articles) are re- Other databases such as Web of Science and Scopus are used to
viewed. This significant growth in the number of publications is enrich the study by adding those papers not found in the initial
noticeable after 2009. stage. An independent working strategy of authors is carried out to
The publications and distribution of the journals, proceedings adequately investigate publications via a series of search and cross-
and book chapters are presented in Fig. 2, Fig. 3 and Fig. 4, re- check activities.

Fig. 1. Distribution of publications per year.

348 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Fig. 2. Distribution of publications based on various journals (99 papers: 1998–2019).

⎧ N  N
⎪
⎪min wi wj σij ( 1.1 )
⎪
⎪
⎪
⎪ i=1 j=1
⎪
⎪
⎪
⎪Sub ject to :
⎪
⎨N

w i μi = R ∗ ( 1.2 ) (1)
⎪
⎪
⎪
⎪
i=1
⎪
⎪ 
N
⎪
⎪ wi = 1 ( 1.3 )
⎪
⎪
⎪
⎩i=1
0 ≤ wi ≤ 1 , i = 1, . . . , N ( 1.4 )
The objective function given in Eq. (1.1) aims to minimize risk
Fig. 3. Distribution of proceedings (46 proceedings: 1998–2018).
while Eq. (1.2) guarantees to obtain returns at the desired return
level. The investment of total budget is ensured by Eqs. (1.3) and
(1.4) does not allow short sales. The original single objective MV
model can also be rewritten as to maximize the return for a given
level of risk. A portfolio obtained by solving model (1) by taking
into consideration of minimum risk for a given level of return or a
maximum return for a given level of risk is called efficient portfo-
lio. However, to find an efficient portfolio, it is necessary to know
either the level of risk that the investor can endure, or the desired
return defined by the investor. In fact, it may not be quite possible
in real world cases. So, to find the efficient portfolio among various
combinations of assets in the solution space, instead of considering
a single objective, researchers must consider all objectives at once.
Therefore, the researchers transformed the single-objective model
into a multi-objective model.

Fig. 4. Distribution of book chapters (36 book chapters: 1998–2018). 3.2. Multi-objective MVPO model

According to Zitzler (1999), the multi-objective mathematical

3. MVPO model model can be written as follows:
⎧
3.1. Single-objective MVPO model
⎨min f (x ) = ( f1 (x ), f2 (x ) . . . , f p (x ))
subject to: e(x ) = (e1 (x ), e2 (x ) . . . , em (x ) ) ≤ 0 (2)
⎩
The original MV model which aims to minimize risk (variance) where: x = (x1 , x2 , . . . , xn ) ∈ X
for the desired level of return is a single-objective model.
Where the constraints e(x) ≤ 0 determine the set of feasible solu-
Parameters
tions, x = (x1 ,…, xn ) is the vector of decision variables (parameters)
N The number of assets available
and X is the decision space, p is equal to 1 in single-objective mod-
μi The expected return of asset i
els and p is larger than 1 in multi-objective models.
σ ij The covariance between assets i and j
Unlike single-objective optimization, a solution to a multi-
R∗ Desired expected return
objective problem is more of a concept than a definition (Marler
Decision variables & Arora, 2004). Therefore, it is significant to go through basic
wi The proportion held of asset frameworks of this concept. Based on the structure defined by
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 349

Fig. 5. Classification of models and applications on MVPO.

Single-objective MV (SOMV); Multi-objective MV (MOMV); Boundary Constraints (BC); Cardinality Constraints (CC); Transaction Costs (TC); Roundlot Constraints (RL); Sector
capitalization Constraints (SC); Turnover Constraints (TUC); Methodologies to deal with various constraints on MVPO (Constraint handling); Various performance indicators
to test the efficiency of proposed algorithms (Performance Measures);.

Zitzler (1999), the concepts of Feasible Set, Pareto Dominance, 3.2.1. Methods for handling multi-objective MVPO
Pareto Optimality, Nondominated Sets are defined as follows: In multi-objective models, it may not be possible to optimize
Feasible Set all objective functions at once, that’s why, it is needed to give pri-
The feasible set Xf is defined as the set of decision vectors x ority to some objectives over another by using methods such as
that satisfy the constraints e(x). weighted sum method or apply dominance-based approaches to
 tackle this issue.
X f = {x ∈ X |ei (x )  0, i = 1, . . . , m} (3) Weighted sum approach
 In the weighted sum method (model 5), a set of objectives are
The image of Xf denoted as f (X f ) = x∈X f { f (x )} is the feasible
combined into a single objective by assigning an associated weight
region in the objective space.
to prioritize one over another. Thus, it simply relaxes a constraint
Pareto Dominance
into a combined single objective. Because of its simple structure
In model (4), concept of domination may be described as fol-
and ease of implementation, in addition to the fact that it is the
lows (Deb, 2001):
most widely used classical approach for multi objective optimiza-
For any two decision vectors u and v,
tion problems (Deb, 2005), weighted sum approach is the most
popular solution approach for MVPO as well. Nevertheless, in spite
of its simplicity, there is a major difficulty in obtaining Pareto op-
(4) timal solutions by this approach for multi-objective optimization
problems having a non-convex Pareto-optimal front. Therefore, the
main disadvantage of weighted sum approach is that it cannot gen-
Pareto Optimality erate all Pareto optimal solutions with non-convex trade-off sur-
A decision vector x ∈ Xf is said to be nondominated regarding faces (Zitzler, 1999).
a set A ⊆ Xf if and only if there does not exist u ∈ A: u ≺ x. x is ⎧
⎨min  λi fi (x )
p
set to be Pareto optimal if and only if x is nondominated regard-
ing Xf . The entire set of all Pareto-optimal solutions is called the i=1 (5)
⎩
Pareto-optimal set and the corresponding objective vectors form sub ject to: x ∈ X f
the Pareto-optimal front (Zitzler, 1999).
where λi is the weight of objective function fi .
Nondominated Sets
MOMV model can be written as follows using the weighted
Among a set of solutions A, the nondominated set of solutions
sum method (Chang, Meade, Beasley, & Sharaiha, 20 0 0):
are those that are not dominated by any member of the set A. ⎧
Because of the fact that addressing the MVPO by using a sin- ⎪ N  N N
⎪
⎪min λ wi wj σij − (1 − λ ) w i μi ( 6.1 )
gle objective problem-solving structure has several difficulties due ⎪
⎪
⎪
⎪
i=1 j=1 i=1
to real-world conditions, multi-objective approaches have been ap- ⎨Sub ject to :
plied. While solving single objective optimization model, it is pos-
sible to find a global and unique optimal solution, multi objective ⎪
⎪ 
N
⎪
⎪ wi = 1 ( 6.2 )
approach gives a set of optimal solution which is called Pareto op- ⎪
⎪
timal solutions. Single objective model may be defined in two dif- ⎪
⎩
i=1

ferent ways such as aiming risk minimization or return maximiza- 0 ≤ wi ≤ 1 i = 1, . . . , N ( 6.3 )

tion. The multi-objective models on the other hand, are based on As seen in Eq. (6.1), the two conflicting objectives (risk mini-
the idea of simultaneously optimizing conflicting objectives. mization, return maximization) are weighted by a parameter (λ).
350 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Fig. 6. Distribution of studies on single-objective and multi-objective models according to years.

The λ weighting parameter is assigned different values between 0 3.4. Consideration of real-world constraints
and 1. When the λ = 0, the objective function of model seeks the
maximum return, while when λ = 1, the objective function seeks Despite all the advantages of the original MV model, it falls
the minimum risk, and the λ parameter achieves a trade-off be- short in real world applications. The original MV model consid-
tween risk and return for 0 < λ < 1. ers only one hard constraint, setting the sum of asset weights to
Pareto-based approaches one meaning that sum of invested amounts must be equal to the
Pareto-based approaches can handle large search spaces and total budget. Therefore, the MV model needs additional constraints
multiple alternative trade-offs in a single optimization run (Zitzler, to solve realistic PO problems. The real-world constraints that have
1999). However, in contrast to weighted sum approach converting been added to the MV model are summarized in Table 4.
the multi-objective structure into a single objective concept, there Table 5 lists the classification of publications according to con-
is no single criterion to assess the quality of a trade-off front; qual- straint types and Fig. 7 demonstrates the distribution of additional
ity measures are relatively difficult to define. constraints investigated with MV models showing that boundary
In Pareto based approaches, typically, a strategy of solution constraints and cardinality constraints are highly dominant over
ranking based on Pareto optimality concept is used (Horn, Naf- others. Although transaction costs make more sense on multi-
pliotis, & Goldberg, 1994). Many of multi-objective algorithms are period portfolio optimization, researchers widely considered this
based on Pareto ranking, yet several variations are available such as constraint on single-period portfolio optimization for modelling
dominance depth (Deb, Pratap, Agarwal, & Meyarivan, 2002), and purposes. If the selected portfolio is readjusted several times ac-
dominance count (Zitzler et al., 2001). cording to the determined investment horizon, TUC which is only
Based on the concept given by Lwin, Qu, and Kendall (2014), valid for multi period portfolio optimization problems is often
multi-objective MV model can also be written as follows: added into the model.
⎧ While the original MV model is presented with a quadratic ob-
⎪ N  N N
⎪
⎪min wi wj σij and max w i μi ( 7.1 ) jective function and linear constraints, several researchers added
⎪
⎪
⎪
⎪
i=1 j=1 i=1 real life constraints which introduce non-linearity and non-
⎨ Sub ject to : convexity to MVPO. For example, TC may be a nonlinear and non-
convex function of a difference in holdings of new and existing
⎪
⎪ 
N
⎪
⎪ wi = 1 ( 7.2 ) portfolio (Yoshimoto, 1996) which increases the complexity of the
⎪
⎪i=1 MVPO. However, TUC are to be formulated as a linear equation, the
⎪
⎩
0 ≤ wi ≤ 1 i = 1, . . . , N ( 7.3 ) problem can still be solved by QP (Yoshimoto, 1996). On the other
hand, CC together with BC leads to a non-convex search space
As seen in Eq. (7.1), the two conflicting objectives (risk min- (Xidonas & Mavrotas, 2014). Therefore, QP cannot be efficiently uti-
imization, return maximization) are independently evaluated to lized and hence, researchers often head towards to inexact tech-
achieve an optimization for obtaining the Pareto front. niques in realistic cases.

3.3. Single vs. Multi objective MV model 4. Data

It is assumed that the investors are aware of the risk or return Researchers test the performance of their proposed algorithms
levels they desire in single objective models. But in real life, this developed to solve the MVPO problem using several different data
assumption is not always true. So, multi-objective models turn out sets in the literature. Table 6 lists and Fig. 8 demonstrates the dis-
to be more realistic. Multi-objective MVPO models are increasingly tribution of data sets used in MVPO literature. Most of the re-
used in the literature as against single-objective models (Fig. 6). searchers used real data to test their algorithms. Portfolio opti-
Table 3 presents the classification of publications according to mization test problems provided in OR-Library (Beasley, 1990) are
their objective type (multi objective and single objective) and gen- popular benchmark datasets in the MVPO literature used for com-
eration methods (weighted-sum methods and Pareto based meth- paring algorithms. The data refer to the weekly stock prices from
ods). March 1992 to September 1997 for the indexes: Hang Seng with
Table 3 clearly shows that in the literature, multi objective 31 assets, Dax 100 with 85 assets, FTSE 100 with 89 assets, S&P
models are often preferred by the operations research community 100 with 98 assets and Nikkei with 225 assets. In addition, Kalayci,
over single objective models. While weighted sum method is the Ertenlice, Akyer, and Aygoren (2017) introduced new benchmarking
most used, Pareto/dominance-based approaches has gained an in- data which is extracted from daily prices between May 2013 and
creasing attention in recent years. April 2016 from XU030 and XU100 indices to further extend the
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 351

Table 3
Classification of publications based on problem objective types.

Problem objective type Generation method Publications

Multi objective Weighted sum (Ackora-Prah, Gyamerah, & Andam, 2014;Ackora-Prah, Gyamerah, Andam, & Gyamfi , 2014; Anagnostopoulos et al.,
2010; Bacanin & Tuba, 2014; 2015; Bacanin et al., 2014; Baykasoglu et al., 2015; Busetti, 2006; Cesarone et al.,
2013; Chang et al., 20 0 0; Chang et al., 2009; Chen et al., 2013; Chen et al., 2012; Chen et al., 2008; ChiangLin,
2006; Cura, 2009; Deng & Lin, 2010a, b; Deng et al., 2012; Farzi et al., 2013; Fernandez & Gomez, 2007; Gaspero
et al., 2011; Golmakani & Alishah, 2008; Hadi et al., 2016; Hao & Liu, 2009; Kalayci et al., 2017; Kamili & Riffi,
2015, 2016; Kao & Cheng, 2013; Koshino et al., 2007; Li et al., 2010; Lin, Li, & Li, 2005; Lu & Wang, 2013; Lwin &
Qu, 2013; Maringer & Kellerer, 2003; Moral-Escudero et al., 2006; Mozafari et al., 2011; Ni et al., 2017; Niu et al.,
2010; Niu et al., 2009; Pai & Michel, 2009; Peng et al., 2011; Pouya et al., 2016; Qu et al., 2017; Rong et al., 2009;
Sabar & Song, 2014; Sadigh et al., 2012; Schaerf, 2002; Suganya & Vijayalakshmi Pai, 2009; Sun et al., 2011;
Suthiwong & Sodanil, 2016; Tan et al., 2013; Tan et al., 2014; Tuba & Bacanin, 2014a; b; Tuba et al., 2014; Wang
et al., 2015; Wang et al., 2011, 2012; Woodside-Oriakhi et al., 2011; Xia et al., 20 0 0; Xu et al., 2010; Yaakob &
Watada, 2010; Yin, Ni, & Zhai, 2015a; b; Zhu et al., 2010; Zhu et al., 2011)
Pareto-based (Anagnostopoulos & Mamanis, 2010, 2011a; b; Arkeman et al., 2013; Bevilacqua et al., 2011; Branke et al., 2009;
Chen & Zhou, 2018; Chen et al., 2017; Chiam et al., 2007; Chiam et al., 2008; Dreżewski & Doroz, 2017; Duran
et al., 2009; Eftekharian et al., 2017; Ehrgott et al., 2004; Fieldsend et al., 2004; Garcia et al., 2012; Jalota & Thakur,
2018; Kumar & Mishra, 2017; Liagkouras & Metaxiotis, 2014, 2017, 2018; Liang & Qu, 2013; Lwin et al., 2014; Lwin
et al., 2013; Lwin et al., 2017; Macedo et al., 2017; Mishra et al., 2016; Mishra et al., 2014a; b; Mishra et al., 2009;
Ong et al., 2005; Ruiz-Torrubiano & Suárez, 2007; Sen et al., 2015; Skolpadungket et al., 2007; Streichen &
Tanaka-Yamawaki, 2006; Streichert et al., 2004a; b; Zhou & Li, 2014)
Single objective (Abbas & Haider, 2009; Aranha & Iba, 2009; Ban et al., 2018; Bonami & Lejeune, 2009; Cao & Tao, 2010; Cesarone
et al., 2015; Chang & Chen, 2008; Chang & Hsu, 2007; Chen & Cai, 2008; Chen et al., 2006; Chen & Zhang, 2010;
Corazza et al., 2012a; 2013; Coutino-Gomez et al., 2003; Crama & Schyns, 2003; Cui et al., 2014; Cui et al., 2013;
Dehghan Hardoroudi et al., 2017; Fasheng & Wei, 2006; Freitas et al., 2009; Gao & Chu, 2009; Golmakani & Fazel,
2011; He & Qu, 2016; Hoklie & Zuhal, 2010; Hong-mei et al., 2010; Hu & Zhangy, 2010; Huang & Shen, 2010;
Huang, 2012; Jiang et al., 2014; Jiang et al., 2008; Kumar & Bhattacharya, 2012; Lai et al., 2006; Lean et al., 2008; Li
et al., 2006; Lin & Liu, 2008; Loukeris et al., 2009; Mayambala et al., 2015; Reid & Malan, 2015; Ruiz-Torrubiano &
Suarez, 2010; 2015; Sadjadi et al., 2012; Shaikh & Abbas, 2009; Shaw et al., 2008; Shoaf & Foster, 1998; Soleimani
et al., 2009; Strumberger et al., 2017; Talebi et al., 2010; Tang et al., 2009; Thomaidis, 2010; Tian et al., 2016; Tuba
et al., 2013; Wang et al., 2009; Xu & Chen, 2006; Xu et al., 2007; Yu et al., 2009; Zaheer & Pant, 2016; Zhang et al.,
2010)

Table 4
Constraints for realistic portfolio optimization.

Boundary constraints (BC) impose lower and/or upper bounds on the values of each asset weight, also known as buy-in
threshold constraint.
Cardinality constraints (CC) related to the number of assets invested in the portfolio, may be fixed to a certain value, may also
ensure that the number of assets is between the desired range.
Transaction costs (TC) investors pay a fee called transaction costs when they sell or buy stocks influencing the total profit
Roundlot (minimum lots) constraint (RL) ensure that the amount invested in a security is multiples of the minimum transaction lot.
Sector capitalization constraints (SC) impose the assets which belong to the sector with more capitalization value to have more shares
in the final portfolio.
Turnover constraint (TUC) sets the turnover rate of an asset from current period to next period which is especially useful in
multi period portfolio optimization models.

Fig. 7. Distribution of additional constraints used with MV model.

352 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Table 5
Classification of publications according to constraint types.

Constraint type Publications

BC–CC (Ackora-Prah, Gyamerah & Andam 2014; Ackora-Prah, Gyamerah, Andam, & Gyamfi, 2014; Anagnostopoulos & Mamanis, 2010,
2011a; b; Anagnostopoulos et al., 2010; Bacanin & Tuba, 2014; 2015; Bacanin et al., 2014; Baykasoglu et al., 2015; Busetti, 2006;
Cesarone et al., 2013; 2015; Chang et al., 20 0 0; Chang et al., 2009; Chen et al., 2013; Chen et al., 2012; Chen et al., 2017; Chiam
et al., 2007; Corazza et al., 2012a; 2013; Cui et al., 2014; Cui et al., 2013; Cura, 2009; Deng & Lin, 2010a, b; Deng et al., 2012;
Eftekharian et al., 2017; Farzi et al., 2013; Fernandez & Gomez, 2007; Garcia et al., 2012; Gaspero et al., 2011; Golmakani & Alishah,
2008; Golmakani & Fazel, 2011; Hadi et al., 2016; Jalota & Thakur, 2018; Jiang et al., 2014; Jin et al., 2015; Kalayci et al., 2017;
Kamili & Riffi, 2015, 2016; Kao & Cheng, 2013; Kırış & Ustun, 2012; Koshino et al., 2007; Kumar & Mishra, 2017; Kumar &
Bhattacharya, 2012; Liagkouras, 2018; Liagkouras & Metaxiotis, 2014, 2018; Lwin & Qu, 2013; Lwin et al., 2013; Mansour et al.,
2007; Mayambala et al., 2015; Mishra et al., 2016; Mishra et al., 2014a; b; Moral-Escudero et al., 2006; Mozafari et al., 2011; Ni
et al., 2017; Ruiz-Torrubiano & Suarez, 2010; Ruiz-Torrubiano & Suárez, 2007; Sabar & Song, 2014; Sadigh et al., 2012; Sadjadi et al.,
2012; Schaerf, 2002; Streichen & Tanaka-Yamawaki, 2006; Suthiwong & Sodanil, 2016; Tang et al., 2009; Tian et al., 2016; Tuba &
Bacanin, 2014a; b; Tuba et al., 2014; Wang et al., 2015; Wang et al., 2011, 2012; Woodside-Oriakhi et al., 2011; Xu et al., 2010;
Yaakob & Watada, 2010; Yin et al., 2015a; b)
TC (Chen & Zhang, 2010; Huang & Shen, 2010; Li et al., 2010; Lu & Wang, 2013; Paiva et al., 2019; Peng et al., 2011; Tan et al., 2013;
Xia et al., 20 0 0; Zhang et al., 2010)
BC–CC–RL (Chiam et al., 2008; Liagkouras & Metaxiotis, 2017; Lwin et al., 2014; Lwin et al., 2017; Skolpadungket et al., 2007; Streichert et al.,
2004a; Streichert et al., 2004b)
TC–RL (Chen et al., 2008; ChiangLin, 2006; Lin & Liu, 2008; Lin et al., 2005; Niu et al., 2010)
BC–CC–TC (Brito & Vicente, 2014a; Gao & Chu, 2009; Hu & Zhangy, 2010; Ruiz-Torrubiano & Suarez, 2015)
CC (Dehghan Hardoroudi et al., 2017; Maringer & Kellerer, 2003; Xu et al., 2011)
BC (Abbas & Haider, 2009; Chang & Hsu, 2007; Jiang et al., 2008)
BC–TC (Chen & Cai, 2008; Chen et al., 2006; Xu et al., 2007)
BC–CC–SC (Pai & Michel, 2009; Tuba et al., 2013)
BC–RL (Bonami & Lejeune, 2009; Zhou & Li, 2014)
BC–CC–SC–RL (Soleimani et al., 2009)
BC–CC–TC–RL (He & Qu, 2016)
BC–CC–TC–SC (Suganya & Vijayalakshmi Pai, 2009)
BC–TC–TUC (Crama & Schyns, 2003)

BC: Boundary constraints; CC: Cardinality constraints; TC: Transaction costs; RL: Roundlot constraint; SC: Sector capitalization constraints; TUC: Turnover
constraint.

Table 6
Datasets used in MVPO literature.

Datasets Publication

OR-Library (Anagnostopoulos & Mamanis, 2010, 2011a; b; Bacanin & Tuba, 2014; 2015; Baykasoglu et al., 2015; Cesarone et al., 2013; 2015; Chang
et al., 20 0 0; Chang et al., 2009; Chen et al., 2013; Chen et al., 2012; Chen et al., 2017; Chiam et al., 2007; Chiam et al., 2008; Cui et al.,
2014; Cura, 2009; Deng & Lin, 2010a, b; Deng et al., 2012; Fernandez & Gomez, 2007; Gaspero et al., 2011; Golmakani & Alishah, 2008; He
& Qu, 2016; Jin et al., 2015; Kalayci et al., 2017; Kamili & Riffi, 2015, 2016; Kao & Cheng, 2013; Koshino et al., 2007; Kumar & Mishra,
2017; Liagkouras, 2018; Liagkouras & Metaxiotis, 2014, 2017; Lwin & Qu, 2013; Lwin et al., 2014; Lwin et al., 2013; Mayambala et al., 2015;
Mishra et al., 2016; Mishra et al., 2014a; b; Mishra et al., 2009; Moral-Escudero et al., 2006; Mozafari et al., 2011; Ni et al., 2017;
Ruiz-Torrubiano & Suarez, 2010; Ruiz-Torrubiano & Suárez, 2007; Sabar & Song, 2014; Sadigh et al., 2012; Sadjadi et al., 2012; Sen et al.,
2015; Skolpadungket et al., 2007; Streichen & Tanaka-Yamawaki, 2006; Streichert et al., 2004a; b; Tian et al., 2016; Tuba & Bacanin, 2014a;
b; Wang et al., 2015; Wang et al., 2011, 2012; Woodside-Oriakhi et al., 2011; Xu J. et al., 2011; Xu R.T. et al., 2010; Yin et al., 2015a; b;
Zhang et al., 2018a)
Case study (Abbas & Haider, 2009; Ackora-Prah, Gyamerah, Andam, 2014; Ackora-Prah, Gyamerah, Andam, & Gyamfi, 2014; Aouni et al., 2005; Aranha
& Iba, 2009; Arkeman et al., 2013; Bacanin et al., 2014; Bevilacqua et al., 2011; Bonami & Lejeune, 2009; Branke et al., 2009; Brito &
Vicente, 2014a; Busetti, 2006; Cao & Tao, 2010; Chang & Chen, 2008; Chang & Hsu, 2007; Chen & Zhou, 2018; Chen & Cai, 2008; Chen
et al., 2008; Chen et al., 2006; Chen & Zhang, 2010; ChiangLin, 2006; Corazza et al., 2012a; Corazza et al., 2013; Crama & Schyns, 2003;
Cui et al., 2013; Dehghan Hardoroudi et al., 2017; Dreżewski & Doroz, 2017; Duran et al., 2009; Eftekharian et al., 2017; Ehrgott et al.,
2004; Farzi et al., 2013; Fasheng & Wei, 2006; Fieldsend et al., 2004; Freitas et al., 2009; Gao & Chu, 2009; García et al., 2018; Garcia
et al., 2012; Golmakani & Fazel, 2011; Hadi et al., 2016; Hao & Liu, 2009; Hoklie & Zuhal, 2010; Hong-mei et al., 2010; Hu & Zhangy, 2010;
Huang & Shen, 2010; Huang, 2012; Jalota & Thakur, 2018; Jiang et al., 2014; Jiang et al., 2008; Kamali, 2014; Kırış & Ustun, 2012; Kocadağlı
& Keskin, 2015; Kumar & Bhattacharya, 2012; Lai et al., 2006; Lean et al., 2008; Li et al., 2006; Li et al., 2010; Liagkouras, 2018; Liagkouras
& Metaxiotis, 2018; Liang & Qu, 2013; Lin & Liu, 2008; Lin et al., 2005; Loukeris et al., 2009; Lu & Wang, 2013; Lwin et al., 2017; Mansour
et al., 2007; Maringer & Kellerer, 2003; Niu et al., 2010; Niu et al., 2009; Ong et al., 2005; Pai & Michel, 2009; Paiva et al., 2019; Peng
et al., 2011; Pouya et al., 2016; Qu et al., 2017; Reid & Malan, 2015; Rong et al., 2009; Ruiz-Torrubiano & Suarez, 2015; Schaerf, 2002;
Shaikh & Abbas, 2009; Shaw et al., 2008; Shoaf & Foster, 1998; Strumberger et al., 2017; Suganya & Vijayalakshmi Pai, 2009; Sun et al.,
2011; Suthiwong & Sodanil, 2016; Talebi et al., 2010; Tan et al., 2013; Tan et al., 2014; Tang et al., 2009; Thomaidis, 2010; Tuba et al., 2013,
2014; Wang et al., 2009; Xu & Chen, 2006; Xu et al., 2007; Yaakob & Watada, 2010; Yu et al., 2009; Zaheer & Pant, 2016; Zhang et al.,
2010; Zhou & Li, 2014; Zhu et al., 2011)
Hypothetical (Bacanin & Tuba, 2015; Ehrgott et al., 2004; Soleimani et al., 2009; Xia et al., 20 0 0)

publicly available data sets to the attention of researchers, available datasets are not publicly available. Stock indices from China (Chen
on http://www.pau.edu.tr/portfolio/en. In addition to the dataset & Cai, 2008; Chen & Zhou, 2018; Chen, Xu, Yang, & Cai, 2008, Chen,
obtained from OR-Library, many researchers used case studies from Zhang, Cai & Xu et al. 2006, Jiang, Zhang, & Xie, 2008; Lai, Yu,
various stock indices. Instead of testing and comparing the pro- Wang, & Zhou, 2006; Lean, Wang, & Lai, 2008; Liang & Qu, 2013;
posed solution approaches on publicly available data sets, studying Lu & Wang, 2013; Niu, Tan, Xue, Li, & Chai, 2010; Qu, Zhou, Xiao,
on the data belonging to different financial markets is also valu- Liang, & Suganthan, 2017; Rong, Lu, & Deng, 2009; Xu, Lam, & Li,
able because of extended variety of available datasets. However, 2011; Yu, Wang, & Lai, 2009; Zhou & Li, 2014), United Kingdom
in this case, the computational comparison is not possible since (Brito & Vicente, 2014a; Brito & Vicente, 2014b; Corazza, Fasano, &
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 353

Fig. 8. Distribution of data sets.

Gusso, 2012a; Corazza, Fasano, & Gusso, 2012b; Eftekharian, Shoja- – Variance of Return Error: The average value of all the errors
far, & Shamshirband, 2017; Garcia, Quintana, Galvan, & Isasi, 2012; which refers the horizontal distance corresponding a certain re-
Jalota & Thakur, 2018; Kumar & Bhattacharya, 2012; Liagkouras & turn level. (Fernandez & Gomez, 2007).
Metaxiotis, 2018; Loukeris, Donelly, Khuman, & Peng, 2009; Shaw, – Mean return error: The average value of all the errors
Liu, & Kopman, 2008), United States (Aranha & Iba, 2009; Bonami which refers the vertical distance corresponding a certain
& Lejeune, 2009; Corazza, Fasano, & Gusso, 2013; Eftekharian et al., risk(variance) level. (Fernandez & Gomez, 2007).
2017; Ehrgott, Klamroth, & Schwehm, 2004; Fieldsend, et al., 2004; – Mean Percentage Error: shows the average of percentage errors
Lwin, Qu, & MacCarthy, 2017; Sun, Fang, Wu, Lai, & Xu, 2011; (Chang et al., 20 0 0).
Xu, Chen, & Yang, 2007), Japan (Aranha & Iba, 2009; Eftekharian – Median Percentage Error: The median value of all percentage
et al., 2017; Jalota & Thakur, 2018; Kumar & Bhattacharya, 2012; Pai errors (Chang et al., 20 0 0).
& Michel, 2009; Suganya & Vijayalakshmi Pai, 2009), Hong Kong – Generational Distance: measures the average distance between
(Eftekharian et al., 2017; Jalota & Thakur, 2018; Jiang, Li, Gao, & Yu, the non-dominated front obtained via the proposed algorithm
2014; Li, Sun, & Wang, 2006), India (Suganya & Vijayalakshmi Pai, and exact Pareto front (Van Veldhuizen & Lamont, 20 0 0). If the
2009; Zaheer & Pant, 2016), Venezuela (Duran, Cotta, & Fernández, generational distance is equal to zero, then all the solutions
2009), Ghana (Ackora-Prah, Gyamerah, & Andam, 2014; Ackora- given by proposed algorithm are on the exact Pareto front. In
Prah, Gyamerah, Andam, & Gyamfi, 2014), Portuguese (Freitas, De other words, lower the generational distance value means bet-
Souza, & de Almeida, 2009; Paiva, Cardoso, Hanaoka, & Duarte, ter convergence ability and performance.
2019), Turkey (Kırış & Ustun, 2012; Kocadağlı & Keskin, 2015). – Epsilon indicator: is a binary indicator that gives a factor by
Indonesia (Hoklie & Zuhal, 2010), Pakistan (Abbas & Haider, which an approximation set is worse than another considering
2009; Shaikh & Abbas, 2009), Thailand (Suthiwong & Sodanil, all objectives (Zitzler, Thiele, Laumanns, Fonseca, & Da Fonseca,
2016), Taiwan (Chang & Chen, 20 08; ChiangLin, 20 06; Lin & Liu, 2003) giving the idea about how two separate distributions of
2008), Iran (Pouya, Solimanpur, & Rezaee, 2016; Talebi, Molaei, & Pareto-fronts differ from each other according to their degree
Sheikh, 2010), Tunisia (Aouni, Abdelaziz, & Martel, 2005; Mansour, of positive or negative non-domination.
Rebai, & Aouni, 2007) and Poland (Dreżewski & Doroz, 2017) ex- – Convergence metric: is used to compare the quality of two non-
tracted by daily, monthly or yearly stock prices are extensively dominated sets without taking into consideration of the theo-
studied. Lastly, researchers tested performance of proposed so- retical efficient frontier (Khare, Yao, & Deb, 2003).
lution approaches by using hypothetical datasets and benchmark Diversity based indicators: represent the uniformity of the distri-
functions in very limited studies. bution of the obtained solutions along the Pareto front.

5. Performance measures – Spacing metric: defines the distribution characteristic of the

solutions on the Pareto front of non-dominated solutions
Several performance metrics have been used for the comparison (Van Veldhuizen & Lamont, 20 0 0). In other words, it shows
of different algorithms in the MVPO literature. Various indicators how well non-dominated solutions scattered.
of MVPO literature can be categorized in four aspects of a solution – Spread (diversity) metric: measures the extent of the spread i.e.,
set: convergence based, diversity based, hybridization based, and how evenly the points are distributed among the approximation
risk adjusted indicators. sets in objective space (Deb et al., 2002).
Convergence based indicators: represent the closeness of the ob- Hybridization based indicators: represent a combined indicator
tained approximation to the theoretical Pareto optimal front. of convergence and diversity.
– Mean Euclidean Distance: refers to the average straight-line dis- – Hypervolume indicator: measures the volume of the multi-
tance between each portfolio alternatives of the theoretical ef- dimensional region that is dominated by the set of non-
ficient frontier and the efficient frontier obtained (Cura, 2009). dominated solutions provided by a multi-objective algorithm
354 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Fig. 9. Distribution of performance measures.

(Zitzler & Thiele, 1999). A higher hypervolume indicator value while handling the MVPO problem. Therefore, many researchers
means better approximation of solution set. have used inexact techniques to solve more realistic MVPO prob-
– D1R : gives information about both the average distance from lems that incorporate non-convex constraints in the mathematical
the closest solution and convergence to the optimal Pareto front formulation, thus yielding the problem NP-hard (Moral-Escudero,
(Czyzżak & Jaszkiewicz, 1998). Ruiz-Torrubiano, & Suarez, 2006). A broad taxonomy of solution
– Inverted generational distance: as an improvement over the techniques applied to PO based on MV model is given in Fig. 10.
generational distance with a faster calculation advantage, is The following subsections provide a detailed review of solution
based on the idea of reversing the order of the fronts con- techniques developed or adapted to solve MVPO.
sidered as input by the generational distance. In other words,
while generational distance metric calculating the distance 6.1. Exact solution techniques
from the non-dominated solution front to reference front, in-
verted generational distance metric performs the calculation in MVPO can be solved with some exact solution algorithms. The
a reverse order (Van Veldhuizen & Lamont, 20 0 0). most popular of these is QP which is a particular type of nonlinear
programming. Aouni et al. (2005) and Mansour et al. (2007) devel-
Risk adjusted indicators: represent a risk-reward evaluation of
oped goal programming models for MVPO problem using QP. Hu
a returns distribution which incorporates the beneficial impact of
and Zhangy (2010) compared the performance of PO models: MV,
gains as well as the detrimental effect of losses.
Mean Absolute Deviation, Conditional Value-at-Risk and Minimax
– Sharpe ratio: is used for measuring of the risk adjusted return model using linear programming. Peng, Kitagawa, Gan, and Chen
of an investment (Sharpe, 1966). In other words, by using sharp (2011) proposed a MV model with transaction costs in quadratic
ratio performance metric, it can be calculated how well the form, thus, presented a model solved by QP. Brito and Vicente
portfolio return can compensate the risk taken by the investors. (2014b) developed a bi-objective model in which the first objective
Higher Sharpe ratio value means better performance. is minimizing the cardinality and second objective is maximizing
– Omega ratio: captures all the higher moment information in the return and minimizing risk using a derivative-free solver, based on
returns distribution and also incorporates sensitivity to return direct multi-search reporting the solutions obtained by QP for sev-
levels while sharp ratio performance metric requires the as- eral datasets. Mayambala, Rönnberg, and Larsson (2015) presented
sumption of mean-variance structure and normally distributed a new MV model, based on performing eigen decomposition of the
input data (Keating & Shadwick, 2002). covariance matrix and reported the performance of this model us-
Fig. 9 demonstrates distribution of performance measures that ing QP. Qin (2015) introduced MV models for solving a hybrid port-
used in MVPO literature. folio selection problem: simultaneous presence of random and un-
The performance measures calculating the distance between certain returns. The computational results obtained with QP indi-
the constrained efficient frontier obtained by the algorithm and the cate that the proposed models are meaningful and able to be ap-
reference unconstrained efficient frontier, convergence-based per- plied.
formance indicators, are presented in Table 7. Diversity, hybridiza- Although the original MV model is a quadratic model, it is
tion based, and risk adjusted performance indicators utilized in transformed into a nonlinear model when additional integer con-
MVPO literature is listed in Tables 8–10 respectively. straints are added. Hence, QP cannot tackle the nonlinear MV
model but nonlinear programming can.
6. Applications on MVPO model The researchers who solved the MVPO problem using the QP or
nonlinear programming, mostly used CPLEX or MATLAB quadratic
The MVPO problem has a quadratic structure, so exact solution toolbox as the solver. Besides QP and nonlinear programming, La-
algorithms such as quadratic programing can tackle MVPO. How- grange methods (Li et al., 2006; Shaw et al., 2008) and Branch &
ever, when additional constraints are added, or the problem di- Bound algorithm (Bonami & Lejeune, 2009) were also applied to
mension is increased, exact solution algorithms may face troubles MVPO (Table 11).
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 355

Table 7
Convergence based performance indicators in MVPO literature.

Performance Measure Publication

Mean Euclidean Distance (Bacanin & Tuba, 2014; Baykasoglu et al., 2015; Chen et al., 2017; Cura, 2009; Jin et al., 2015; Kalayci et al., 2017; Kamili & Riffi,
2015, 2016; Kao & Cheng, 2013; Kumar & Mishra, 2017; Lwin et al., 2013; Mishra et al., 2016; Mishra et al., 2014a; b; Mozafari
et al., 2011; Ni et al., 2017; Sadigh et al., 2012; Sen et al., 2015; Tuba & Bacanin, 2014a; Wang et al., 2011, 2012; Yin et al., 2015a)
Variance of Return Error (Bacanin & Tuba, 2014; Baykasoglu et al., 2015; Cura, 2009; Fernandez & Gomez, 2007; Golmakani & Alishah, 2008; Kalayci et al.,
and Mean Return Error 2017; Kamili & Riffi, 2015, 2016; Kao & Cheng, 2013; Mishra et al., 2014a; Mozafari et al., 2011; Ni et al., 2017; Sadigh et al., 2012;
Tuba & Bacanin, 2014a; Wang et al., 2011, 2012; Yin et al., 2015a)
Mean Percentage Error (Baykasoglu et al., 2015; Chang et al., 20 0 0; Cui et al., 2014; Deng & Lin, 2010a, b; Deng et al., 2012; Jalota & Thakur, 2018; Jin
et al., 2015; Kalayci et al., 2017; Lwin & Qu, 2013; Pai & Michel, 2009; Sabar & Song, 2014; Tuba & Bacanin, 2014b;
Woodside-Oriakhi et al., 2011; Xu et al., 2010)
Median Percentage Error (Chang et al., 20 0 0)
Generational Distance (Lwin et al., 2017; Skolpadungket et al., 2007)
Epsilon indicator (Liagkouras, 2018; Liagkouras & Metaxiotis, 2017, 2018)
Convergence metric (Mishra et al., 2016; Mishra et al., 2014b; Mishra et al., 2009; Sen et al., 2015)

Table 8 Table 9
Diversity based performance indicators in MVPO literature. Hybridization based performance indicators in MVPO literature.

Performance Measure Publication Performance Measure Publication

Spacing metric (Chen & Zhou, 2018; Eftekharian et al., 2017; D1R (Chen et al., 2013; Chen et al., 2012; Suthiwong &
Kumar & Mishra, 2017; Mishra et al., 2016; Sodanil, 2016)
Mishra et al., 2009) Inverted Generational (Liagkouras, 2018; Liagkouras & Metaxiotis, 2017,
Diversity metric (Chen et al., 2013; Chen et al., 2012; Chen et al., Distance 2018; Lwin et al., 2017)
2017; Liagkouras & Metaxiotis, 2014; Liang & Qu, Hyper Volume (Dreżewski & Doroz, 2017; Liagkouras, 2018;
2013; Lwin et al., 2014; Lwin et al., 2013; Lwin Liagkouras & Metaxiotis, 2017, 2018; Lwin et al.,
et al., 2017; Mishra et al., 2016; Mishra et al., 2017; H. Zhang et al., 2018)
2014b; Mishra et al., 2009; Sen et al., 2015;
Suthiwong & Sodanil, 2016)

6.2. Inexact solution techniques

It is observed in the literature that, the number of publications Approximation techniques have been extensively used in the
that design the exact solution approaches for the MVPO problem literature since the MVPO problem is known to be NP-hard due
is limited. to various real-life constraints. The approximation techniques are

Fig. 10. A taxonomy of solution techniques applied to PO based on MV model.

Artificial Bee Colony (ABC), Ant Colony Optimization (ACO), Artificial Cooperative Search (ACS), Bat Algorithm (BA), Bacterial Foraging Optimization (BFO), Brunch & Bound,
CRO (BnB), Chemical Reaction Optimization (CRO), Cat Swarm Optimization (CSO), Differential Evolutionary (DE), Firefly Algorithm (FA), Fireworks Algorithm (FWO), Genetic
Algorithm (GA), Generalized DE3 (GDE3), Greedy Random Adaptive Search Procedure (GRASP), Guided L S (GL S), Hill Climbing (HC), Hopfield Network (HN), Indicator-Based
Evolutionary Algorithm (IBEA), Invasive Weed Optimization (IWO), Iterated Local Search (ILS), Krill Herd (KH), Local Search (LS), Multi objective co-variance based ABC (M-
CABC), Multi Objective GA (MOGA), Multi Objective PSO (MOPSO), Multi Objective Scatter Search (MOSS), Non-Linear Programing (NLP), Neural Network (NN), Niched Pareto
GA 2 (NPGA-II), Non-dominated Sorting GA 2 (NSGA-II), Pareto Archived Evolution Strategy (PAES), Population Based Incremental Learning (PBIL), Pareto Envelope-based
Selection Algorithm (PESA), Particle Swarm Optimization (PSO), Quadratic Programming (QP), Simulated Annealing (SA), Stochastic L S (SL S), Strength Pareto Evolutionary
Algorithm 2 (SPEA-II), Tabu Search (TS), Vector Evaluated GA (VEGA), Variable Neighborhood Search (VNS).
356 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Table 10 (2006) compared the performance of tabu search and genetic al-
Risk adjusted performance indicators in MVPO literature.
gorithm on MVPO and test results reported that genetic algorithm
Performance Measure Publication was both faster and more efficient. Lin and Liu (2008) proposed
Sharpe ratio (Ban et al., 2018; Ruiz-Torrubiano & Suarez, 2015) models for MVPO problem with minimum transaction lots and
uses genetic algorithm to solve the models. The computational re-
sults showed that the genetic algorithm could obtain near-optimal
examined in two categories: metaheuristics and machine learning solutions within a reasonably short time. Chang, Yang, and Chang
algorithms. (2009) reported that genetic algorithm can provide an efficient and
convenient tool for investors by verifying that investors should not
6.2.1. Metaheuristics consider the number of assets to be held in the portfolio above
Metaheuristics involve iterative high-level procedures guiding one third of total assets since they are obviously dominated by
the sub-level heuristics designed for solving various optimization portfolios with relatively less assets. Loukeris et al. (2009) com-
problems by performing search strategies in order to find near- pared genetic algorithm, differential evolution and particle swarm
optimal solutions in the vast solution space. In each iteration, optimization, reporting that the differential evolution algorithm
metaheuristics may use a single solution or a population. Sub-level obtained the highest and most reliable results for different sce-
heuristics may be composed of a simple local search algorithm or a narios. Xu, Zhang, Liu, and Huang (2010) proposed a population
specific method developed to initialize a solution. In general, meta- based incremental learning algorithm and compared performance
heuristics have two significant search mechanisms: exploration and of the algorithm with genetic algorithm presented by Chang et al.
exploitation. Exploration mechanism guides the main algorithm to (20 0 0) and particle swarm optimization presented by Cura (2009).
discover new promising solution areas providing a less chance of They reported that population based incremental learning algo-
getting trapped in a local optimum while exploitation mechanism rithm obtained similar results with other algorithms. Woodside-
examines the newly discovered solution space carefully to search Oriakhi, Lucas, and Beasley (2011) developed three algorithms (ge-
for a better neighboring solution, hopefully global optimum. A bal- netic algorithm, tabu search and simulated annealing) that out-
anced proportion and timing between these two mechanisms is perform the algorithms designed by Chang et al. (20 0 0) on all
crucial for an efficient and robust metaheuristic algorithm. Meta- performance indicators tested such as mean percentage error and
heuristic algorithms are classified in two categories: population- computation time for all data sets. Sadjadi, Gharakhani, and Safari
based algorithms and single solution-based algorithms. (2012) presented a framework to efficiently solve cardinality con-
strained MVPO considering an uncertainty structure both indepen-
6.2.1.1. Population based algorithms. Population based algorithms dent and correlated by utilizing a genetic algorithm. Liu and Zhang
utilize the power of population during the search in the vast so- (2015) proposed a multi period fuzzy PO model and adapted a ge-
lution space. The population should be adequately diversified so netic algorithm for solving the model.
that a larger area can be scanned. Population-based algorithms Multi objective evolutionary algorithms for MVPO:
generate an initial population randomly or with heuristic support Fieldsend, Matatko, and Peng (2004) applied multi objective
and follow the population improvement strategies in each itera- evolutionary algorithm firstly to MVPO problem. Ong, Huang, and
tion. Population-based algorithms can be examined in two cate- Tzeng (2005) proposed a PO model with three objectives, max-
gories according to the sources of inspiration: evolutionary algo- imizing expected return, minimizing uncertainty risk and mini-
rithms and swarm-based algorithms. mizing relation risk. They NSGA-II in order to solve the three
Evolutionary algorithms: objective model and reported that NSGA-II can provide flexi-
The basic idea of evolutionary algorithms is to generate new so- ble and accurate results. Skolpadungket, Dahal, and Harnpornchai
lutions, represented by chromosomes, from a population of initial (2007) applied various multi-objective evolutionary algorithms to
solutions utilizing the principles of natural selection and crossover solve MVPO problem with cardinality constraints and boundary
or mutation strategies to create new offspring. The most popular constraints such as vector evaluated genetic algorithm, multi ob-
evolutionary algorithm is genetic algorithm that is introduced by jective genetic algorithm, SPEA2 and NSGA-II. According to the test
Holland (1975). Evolutionary algorithms applied to MVPO problem results, even though multi-objective genetic algorithms and SPEA2
are summarized in Table 12. required more complex structures, they outperformed the others.
Single objective evolutionary algorithms for MVPO: Anagnostopoulos and Mamanis (2010) proposed a tri-objective PO
Shoaf and Foster (1998) firstly applied genetic algorithm and model: maximizing return, minimizing risk and minimizing num-
Chang et al. (20 0 0) adapted three algorithms including genetic al- ber of assets held in the portfolio. They experimented with NSGA-
gorithm to MVPO problem. Ehrgott et al. (2004) proposed a model II, Pareto envelope based selection algorithm and SPEA2 for finding
for MVPO problem and designed three algorithms genetic algo- approximation solutions and test results showed that SPEA-II ob-
rithm, tabu search, simulated annealing for solving this model. tained the best results. Anagnostopoulos and Mamanis (2011a) pre-
Evaluation of algorithms in terms of utility value and standard de- sented five multi-objective evolutionary algorithms a single ob-
viation, genetic algorithm produced better results and turned out jective evolutionary algorithm for solving MVPO problem. The
to be the most stable algorithm compared to the others. Busetti

Table 11
Publications that applied exact solution techniques to solve MVPO problem.

Technique Publications

QP (Abbas & Haider, 2009; Aouni et al., 2005; Baykasoglu et al., 2015; Brito &
Vicente, 2014a; Cesarone et al., 2013; 2015; Cui et al., 2014; Cui et al., 2013;
Gaspero et al., 2011; Hu & Zhangy, 2010; Jiang et al., 2014; Jin et al., 2015;
Liagkouras & Metaxiotis, 2014; Mansour et al., 2007; Mayambala et al., 2015;
Moral-Escudero et al., 2006; Peng et al., 2011; Ruiz-Torrubiano & Suarez, 2015)
Nonlinear Programming (Kırış & Ustun, 2012; Kocadağlı & Keskin, 2015)
Branch & Bound (Bonami & Lejeune, 2009; He & Qu, 2016)
Lagrangian (Li et al., 2006; Shaw et al., 2008)
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 357

Table 12
Evolutionary algorithms applied to MVPO.

Evolutionary algorithm Proposed by Objective type

Genetic algorithm (GA) Holland (1975) Single objective

Differential evolution (DE) Storn and Price (1997) Single objective
Population based incremental learning (PBIL) Baluja (1994) Single objective
Non-dominated sorting genetic algorithm 2 (NSGA-II) Deb et al. (2002) Multi objective
Strength Pareto evolutionary algorithm 2 (SPEA2) Zitzler et al. (2001) Multi objective
Pareto envelope-based selection algorithm (PESA) Corne et al. (20 0 0) Multi objective
Pareto archived evolution strategy (PAES) Knowles et al. (2006) Multi objective
Indicator based evolutionary algorithm (IBEA) Zitzler and Künzli (2004) Multi objective
Multi objective genetic algorithm (MOGA) Fonseca and Fleming (1993) Multi objective
Niched Pareto genetic algorithm 2 (NPGA-II) Erickson et al. (2001) Multi objective
Generalized differential evolution 3 (GDE3) Kukkonen and Lampinen (2005) Multi objective
Multi objective scatter search (MOSS) Glover et al. (20 0 0) Multi objective
Vector evaluated genetic algorithm (VEGA) Schaffer (1985) Multi objective
e-multi objective evolutionary algorithm (e-MOEA) Hanne (2007) Multi objective

computational results report that, firstly, all multi-objective evo- with particle swarm optimization showing the robust and effec-
lutionary algorithms outperform the single-objective ones in all tive performance of the algorithm, especially for low-risk invest-
problem instances on epsilon performance indicator and per- ments. Hong-mei, Zhuo-fu, and Hui-min (2010) was first to apply
formed faster, secondly, NSGA-II and SPEA2 are best performing artificial bee colony algorithm to MVPO problem and noted the al-
algorithms among the algorithms in terms of convergence capa- gorithm as a useful approach for MVPO problem. Zhu, Chen, and
bility. Liagkouras and Metaxiotis (2014) proposed a probe guided Wang (2010) proposed two swarm-based algorithms: ant colony
mutation operator and compared their proposed operator with a optimization and particle swarm optimization. Computational re-
classical mutation operator embedded into NSGA-II and SPEA2 al- sults showed that ant colony optimization performed better than
gorithms in terms of performance metrics such as hypervolume, particle swarm optimization in small-scale and large-scale portfo-
epsilon indicator and the spread of solutions Results confirmed lios, however in the medium-scale portfolio turned out to be the
the superiority of the probe guided mutation operator. Lwin et al. opposite utilizing Sharpe ratio performance indicator. Golmakani
(2014) proposed a learning guided multi-objective evolutionary al- and Fazel (2011) compared particle swarm optimization with ge-
gorithm for MVPO problem and compared their proposed algo- netic algorithms reporting the achievement of finding better so-
rithm performed well on diversity and hypervolume performance lutions within a less amount of time. Xu et al. (2011) proposed
indicators, especially on large sized data sets. Macedo, Godinho, chemical reaction optimization by adding a super molecule to the
and Alves (2017) compared two multi-objective evolutionary algo- evolutionary process on performance indicators such as Sharpe ra-
rithms on hypervolume performance indicator and technical anal- tio and execution time showing the improvement on the origi-
ysis based strategies reporting that, NSGA II outperformed SPEA2 nal algorithm. Deng, Lin, and Lo (2012) designed a particle swarm
and SPEA2 frontiers presenting a lesser extent than those of the optimization algorithm with a mutation operator and compared
NSGA II. Evolutionary algorithms applied for MVPO problem are against genetic algorithms, tabu search and simulated annealing al-
shown in Table 13. gorithms (Chang et al., 20 0 0) reporting the superiority of the al-
While GA is the most preferred evolutionary algorithms in the gorithm on mean percentage error performance indicator. Chen,
MVPO literature so far, in recent years the attention has been given Liang, and Liu (2013) proposed an artificial bee colony algorithm
to MOEAs. with two phase encoding in which selection of stocks is deter-
Swarm based algorithms: mined in the first part of the encoding while investment percent-
Swarm based algorithms are based on the study of computa- age is determined in the second part of encoding. Kao and Cheng
tional systems inspired by behaviors of animals living in herds in (2013) applied bacterial foraging optimization to MVPO problem
their natural lives. Swarms such as schools of fish, flocks of birds, and compared with genetic algorithms, tabu search and simulated
colonies of ants and bees liaise for their natural activities (mostly annealing algorithms (Chang et al., 20 0 0) and particle swarm opti-
hunting). Swarm based algorithms are born as a reflection of this mization (Cura, 2009) claiming the superior performance of their
cooperation to computational systems. The most popular swarm algorithm against others in terms of solution quality and time.
based algorithm is particle swarm optimization that introduced by Tuba, Bacanin, and Pelevic (2013) was first to apply firefly al-
Kennedy and Eberhart (1995). The swarm-based algorithms are ex- gorithm to MVPO problem and reported that the algorithm has
amined in two categories: single objective swarm-based algorithm a potential for solving this problem. Bacanin, Tuba, and Pelevic
and multi objective swarm-based algorithm. Swarm based algo- (2014) compared the artificial bee colony algorithm with genetic
rithms applied to MVPO problem are showed in Table 14. algorithm and firefly algorithm and reported that their algorithm
Single objective swarm-based algorithms for MVPO: performed best on variance and return error performance indica-
Chen et al. (2006) was first to apply particle swarm optimiza- tors. Tan, Niu, Wang, Huang, and Duan (2014) proposed a bacterial
tion to MVPO problem. Cura (2009) developed a particle swarm foraging optimization algorithm with neighborhood learning for
optimization algorithm for MVPO with boundary constraints and dynamic MVPO problem and showed its performance against the
cardinality constraints and compared to genetic algorithm, tabu original algorithm. Test results showed that proposed algorithm
search and simulated annealing (Chang et al., 20 0 0) reporting that was superior than original BFO. Tuba and Bacanin (2014b) pro-
none of the four algorithms had clearly outperformed the oth- posed a firefly algorithm based solution approach by introduc-
ers, however, the particle swarm optimization obtained best solu- ing a more intensive exploitation mechanism in the late iterations
tions on low risk levels on performance indicators such as Euclid- of algorithm’s run. On mean and median percentage error per-
ian distance, contribution percentage, variance of return error and formance indicators, they reported that proposed algorithm out-
mean return error criteria. Deng and Lin (2010a) was first to apply performed the original version of the algorithm. Chen (2015) de-
ant colony optimization to MVPO problem and compared results veloped an artificial bee colony algorithm to solve MVPO by
358 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Table 13
Publications that applied evolutionary algorithms to solve MVPO problem.

Category Evolutionary algorithm References

SOEA GA (Ackora-Prah et al., 2014a; Aranha & Iba, 2009; Busetti, 2006; Chang et al., 20 0 0; Chang et al.,
2009; Chen et al., 2008; ChiangLin, 2006; Corazza et al., 2012a; Coutino-Gomez et al., 2003;
Dreżewski & Doroz, 2017; Ehrgott et al., 2004; García et al., 2018; Hadi et al., 2016; Hao & Liu,
2009; Hoklie & Zuhal, 2010; Huang & Shen, 2010; Huang, 2012; Kamali, 2014; Lai et al., 2006;
Lean et al., 2008; Lin & Liu, 2008; Lin et al., 2005; Loukeris et al., 2009; Lu & Wang, 2013;
Moral-Escudero et al., 2006; Pai & Michel, 2009; Rong et al., 2009; Ruiz-Torrubiano & Suarez,
2010; 2015; 2007; Sabar & Song, 2014; Sadjadi et al., 2012; Shaikh & Abbas, 2009; Shoaf & Foster,
1998; Soleimani et al., 2009; Talebi et al., 2010; Thomaidis, 2010; Woodside-Oriakhi et al., 2011;
Xia et al., 20 0 0; Yu et al., 2009)
DE (Loukeris et al., 2009; Lwin & Qu, 2013; Zaheer & Pant, 2016)
PBIL (Jin et al., 2015; Lwin & Qu, 2013; Xu et al., 2010)
MOEA NSGA & NSGA-II (Anagnostopoulos & Mamanis, 2010, 2011a; Anagnostopoulos & Mamanis, 2011b; Arkeman et al.,
2013; Branke et al., 2009; Chen & Zhou, 2018; Chiam et al., 2007; Chiam et al., 2008; Duran et al.,
2009; Fieldsend et al., 2004; Garcia et al., 2012; Liagkouras, 2018; Liagkouras & Metaxiotis, 2014;
Lwin et al., 2014; Macedo et al., 2017; Mishra et al., 2016; Mishra et al., 2014a; Mishra et al., 2009;
Ong et al., 2005; Qu et al., 2017; Sen et al., 2015; Skolpadungket et al., 2007; Streichen &
Tanaka-Yamawaki, 2006; Streichert et al., 2004a; Streichert et al., 2004b)
SPEA2 (Anagnostopoulos & Mamanis, 2010, 2011a; Anagnostopoulos & Mamanis, 2011b; Chen & Zhou,
2018; Duran et al., 2009; Garcia et al., 2012; Liagkouras, 2018; Liagkouras & Metaxiotis, 2014; Lwin
et al., 2014; Macedo et al., 2017; Mishra et al., 2016; Mishra et al., 2014a; Mishra et al., 2009;
Skolpadungket et al., 2007)
PESA & PESA2 (Anagnostopoulos & Mamanis, 2010, 2011a; Anagnostopoulos & Mamanis, 2011b; Lwin et al., 2014;
Mishra et al., 2014a)
PAES (Lwin et al., 2014)
MOGA (Bevilacqua et al., 2011; Skolpadungket et al., 2007)
IBEA (Duran et al., 2009)
GDE3 (Garcia et al., 2012)
MOSS (Lwin et al., 2013)
NPGA-II (Anagnostopoulos & Mamanis, 2011a)
VEGA (Skolpadungket et al., 2007)
e-MOEA (Anagnostopoulos & Mamanis, 2011a)
Co-EMOA (Dreżewski & Doroz, 2017)
MODE-GL (Lwin et al., 2017)
MOEA/D-C (Zhang et al., 2018a)

DE: Differential Evolutionary, GA: Genetic Algorithm, GDE3: Generalized DE3, MOGA: Multi Objective GA, MOPSO: Multi Objective PSO, MOSS:
Multi Objective Scatter Search, NPGA-II: Niched Pareto GA 2, NSGA-II: Non-dominated Sorting Genetic Algorithm II, PAES: Pareto Archived
Evolution Strategy, PBIL: Population Based Incremental Learning, PESA: Pareto Envelope-based Selection Algorithm, SPEA-II: Strength Pareto
Evolutionary Algorithm 2, VEGA: Vector Evaluated GA.

Table 14
Swarm based algorithms applied to MVPO problem.

Swarm based algorithm Proposed by Objective type

Particle swarm optimization (PSO) Kennedy and Eberhart (1995) Single objective
Ant colony optimization (ACO) Dorigo et al. (1996) Single objective
Artificial bee colony (ABC) Karaboga (2005) Single objective
Firefly algorithm (FA) Yang (2010b) Single objective
Bacterial foraging optimization (BFO) Passino (2002) Single objective
Bat algorithm (BA) Yang (2010a) Single objective
Fireworks algorithm (FWA) Tan and Zhu (2010) Single objective
Cat swarm optimization (CSO) Chu and Tsai (2007) Single objective
Krill herd (KH) Gandomi and Alavi (2012) Single objective
Invasive weed optimization (IWO) Mehrabian and Lucas (2006) Single objective
Chemical reaction optimization (CRO) Lam and Li (2010) Single objective
Artificial cooperative search (ACS) Civicioglu (2013) Single objective
Multi objective particle swarm optimization (MOPSO) Coello et al. (2004) Multi objective
Multi objective co-variance based artificial bee colony (M-CABC) Kumar and Mishra (2017) Multi objective
Multi objective bacterial foraging optimization (MOBFO) Panigrahi et al. (2011) Multi objective

introducing a chaotic initialization approach and a hybridization cial bee colony algorithm with feasibly enforcement and infeasibil-
with particle swarm optimization. Kamili and Riffi (2015) applied ity toleration procedures for MVPO with boundary and cardinality
cat swarm optimization to MVPO with boundary and cardinal- constraints achieving much stronger performance than the origi-
ity constraints. Kamili and Riffi (2016) published a comparative nal version of the algorithm as well as demonstrating the supe-
study on MVPO problem by using particle swarm optimization, riority against other algorithms such as genetic algorithms, tabu
cat swarm optimization and bat algorithms. The results claim that search and simulated annealing algorithms (Chang et al., 20 0 0),
the three algorithms have obtained similar results. Ni, Yin, Tian, (Deng et al., 2012), population based incremental learning ap-
and Zhai (2017) proposed a particle swarm optimization with dy- proach (Lwin & Qu, 2013), GRASP-QP (Baykasoğlu, Avci, & Burcin
namic random population topology strategies and reported that Özsoydan, 2016) on various convergence metrics in the literature.
the computational results demonstrate that the proposed dynamic Multi objective swarm-based algorithms for MVPO
random population topology could improve the performance of Mishra, Panda, and Meher (2009) applied multi objective par-
the original algorithm. Kalayci et al. (2017) proposed an artifi- ticle swarm optimization to MVPO problem and compared with
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 359

Table 15
Publications that applied swarm-based algorithm to solve MVPO problem.

Category Swarm based algorithms References

SOSBA PSO (Abbas & Haider, 2009; Cao & Tao, 2010; Chang & Chen, 2008; Chang & Hsu, 2007; Chen & Cai,
2008; Chen et al., 2006; Chen & Zhang, 2010; Corazza et al., 2012a; 2013; Cui et al., 2014; Cura,
2009; Deng & Lin, 2010b; Deng et al., 2012; Farzi et al., 2013; Fasheng & Wei, 2006; Gao & Chu,
2009; Golmakani & Fazel, 2011; Kamali, 2014; Kamili & Riffi, 2016; Koshino et al., 2007; Li et al.,
2010; Loukeris et al., 2009; Mozafari et al., 2011; Ni et al., 2017; Niu et al., 2010; Niu et al., 2009;
Pouya et al., 2016; Reid & Malan, 2015; Sadigh et al., 2012; Sun et al., 2011; Talebi et al., 2010;
Tang et al., 2009; Thomaidis, 2010; Wang et al., 2015; Wang et al., 2009; Xu & Chen, 2006; Xu
et al., 2007; Yaakob & Watada, 2010; Yin et al., 2015a; Yin et al., 2015b; Zhang et al., 2010; Zhu
et al., 2010; Zhu et al., 2011)
ABC (Bacanin et al., 2014; Chen et al., 2013; Chen et al., 2012; Hong-mei et al., 2010; Kalayci et al.,
2017; Suthiwong & Sodanil, 2016; Tuba & Bacanin, 2014a; Wang et al., 2011, 2012)
ACO (Deng & Lin, 2010a; Zhu et al., 2010)
FA (Bacanin & Tuba, 2014; Tuba & Bacanin, 2014a; b; Tuba et al., 2013)
BFO (Kao & Cheng, 2013; Tan et al., 2013; Tan et al., 2014)
BA (Kamili & Riffi, 2016; Strumberger et al., 2017)
FWA (Bacanin & Tuba, 2015)
CSO (Kamili & Riffi, 2015, 2016)
KH (Tuba et al., 2014)
IWO (Pouya et al., 2016)
CRO (Xu et al., 2011)
ACS (Kumar & Bhattacharya, 2012)
MOSBA MOPSO (Chen & Zhou, 2018; Garcia et al., 2012; Lwin & Qu, 2013; Mishra et al., 2016; Mishra et al., 2014a;
Mishra et al., 2009; Qu et al., 2017; Sen et al., 2015; Zhou & Li, 2014)
MOABC (Kumar & Mishra, 2017)
MOBFO (Mishra et al., 2014b)

NSGA-II, SPEA2 and parallel single front genetic algorithm and Table 16
Single-solution-based algorithms applied to MVPO problem.
experimental results showed that their proposed algorithm sig-
nificantly outperforms its counterparts in all experiments based Single-solution-based
on diversity performance indicators. Garcia et al. (2012) proposed algorithm Proposed by Objective type
a robust approach for MVPO and tested through experimenta- Simulated annealing (SA) Kirkpatrick et al. (1983) Single objective
tion using four multi objective algorithms (NSGA-II, SPEA-II, multi- Tabu search (TS) Glover (1986) Single objective
objective particle swarm optimization and generalized differen- Greedy randomize Feo and Resende (1995) Single objective
adaptive search procedure
tial evolution algorithms). According to comparative test results,
(GRASP)
multi-objective particle swarm optimization outperformed others. Hill climbing (HC) Forrest and Mitchell (1993) Single objective
Mishra, Panda, and Majhi (2014b) proposed multi objective bacte- Archived multi objective Bandyopadhyay et al. (2008) Multi objective
rial foraging optimization to MVPO problem with boundary and simulated annealing
(AMOSA)
cardinality constrains and compared with three multi-objective
evolutionary algorithms performing superior against other algo-
rithms. Mishra, Panda, and Majhi (2016) proposed a prediction
based MV model and implemented three multi objective algo- quickly since they improve a single solution instead of a population
rithms (multi-objective particle swarm optimization, NSGA-II and in each iteration. Single solution-based algorithms that applied to
SPEA2) reporting that multi-objective particle swarm optimization MVPO problem are summarized in Table 16.
provided the best Pareto optimal solutions among all. Kumar and Single-solution-based algorithms for MVPO
Mishra (2017) proposed a multi objective co-variance based artifi- Chang et al. (20 0 0) and Ehrgott et al. (2004) applied genetic
cial bee colony to tackle two conflicting objectives simultaneously algorithm, tabu search and simulated annealing and compared all
validating the performance of their proposed algorithm for MVPO by leaving a comment that no algorithm outperformed others.
problem. Swarm based algorithms applied for MVPO problem are Anagnostopoulos, Šević, Chatzoglou, and Katsavounis (2010) de-
summarized in Table 15. signed a greedy randomized adaptive search procedure algorithm
As shown in Table 15, particle swarm optimization is the most to solve MVPO problem confirming its potential on the prob-
popular swarm-based algorithm for MVPO problem since it is the lem. Woodside-Oriakhi et al. (2011) proposed three metaheuris-
oldest algorithm among other swarm-based algorithms and there tics (genetic algorithm, tabu search and simulated annealing) to
is no need for a transformation for the MVPO problem, i.e., the MVPO problem and compared their experimental results with re-
original structure of the algorithm is continuous and has a direct sults obtained by Chang et al. (20 0 0) and obtained better solu-
fit into the problem domain. Despite the popularity of the parti- tions faster. Fastrich and Winker (2012) designed a hybrid heuristic
cle swarm optimization algorithm, it appears that the artificial bee algorithm integrating simulated annealing algorithm as a modifi-
colony algorithm has attracted the researchers’ interest with its su- cation mechanism into the main solution approach. Ackora-Prah,
perior performance and therefore, artificial bee colony algorithm Gyamerah, Andam and Gyamfi et al.(2014) applied pattern search
may become more popular in the near future for the MVPO prob- originally proposed by Hooke and Jeeves (1961) to MVPO prob-
lem. lem reporting the efficacy of pattern search. Sen, Saha, Ekbal,
and Laha (2015) applied archive multi objective simulated anneal-
6.2.1.2. Single-solution-based algorithms. Algorithms working on ing and compared with NSGA-II on diversity performance indica-
single solution are called trajectory methods (Blum & Roli, 2003). tors and multi-objective particle swarm optimization algorithm re-
Single solution-based algorithms focus on modifying and improv- porting their algorithm’s superiority against the others. Chen, Lin,
ing a single solution throughout the iterations. Although, they Zeng, Xu, and Zhang (2017) introduced a local search based multi-
underutilize the power of population, they can get results very objective optimization, a combination of a local search schema and
360 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Table 17
Publications that applied single-solution-based algorithm to solve MVPO problem.

Single-solution-based algorithm References

SA (Busetti, 2006); Chang et al. (20 0 0); Coutino-Gomez et al. (2003); Ehrgott et al. (2004); Fastrich
and Winker (2012); (García et al., 2018); Li et al. (2009); Schaerf (2002); Thomaidis (2010);
(Woodside-Oriakhi et al., 2011)
TS (Chang et al., 20 0 0; Coutino-Gomez et al., 2003; Crama & Schyns, 2003; Ehrgott et al., 2004;
Maringer & Kellerer, 2003; Schaerf, 2002; Thomaidis, 2010; Woodside-Oriakhi et al., 2011)
GRASP (Anagnostopoulos et al., 2010; Baykasoglu et al., 2015)
HC (Coutino-Gomez et al., 2003; Gaspero et al., 2011; Schaerf, 2002)
AMOSA (Sen et al., 2015)
LS Based MOEA Chen et al. (2017)
PS Ackora-Prah, Gyamerah, Andam, & Gyamfi. (2014)

Simulated Annealing (SA); Tabu Search (TS); Greedy Randomized Adaptive Search Procedures (GRASP); Hill Climbing (HC); Archive Multi
Objective Simulated Annealing (AMOSA); Pattern Search (PS).

Table 18 Table 19
Machine learning algorithms applied to MVPO problem. Publications that applied single-solution-based algorithm to solve MVPO problem.

Machine learning algorithm Proposed by Machine learning algorithm References

Hopfield network (HN) Hopfield (1984) Neural networks (Fernandez & Gomez, 2007; Freitas et al.,
Artificial immune systems (AIS) De Castro and Timmis (2002) 2009; Suganya & Vijayalakshmi Pai, 2009)
K-means clustering Lloyd (1982) Artificial immune systems (Abbas & Haider, 2009; Golmakani & Alishah,
2008)
K-means clustering (Jiang et al., 2014; Pai & Michel, 2009;
Suganya & Vijayalakshmi Pai, 2009)
a farthest-candidate approach, to solve MVPO problem. Publica-
tions that applied single-solution-based algorithms for MVPO prob-
lem are summarized in Table 17.
As shown in Table 17, compared to population-based solution
approaches, the single-solution-based algorithms have not drawn cial variants of MVPO when integer constraints are introduced to
much interest to solve MVPO problem. In the MVPO problem, the problem. The distribution of solution techniques applied to the
single-solution-based algorithms have a handicap of efficient ex- MVPO problem in the literature is given in Fig. 11.
ploration capability since it may be more difficult to efficiently ex- One may easily conclude that well-known metaheuristics such
plore the solution space with a single solution since the MVPO as genetic algorithms and particle swarm optimization have drawn
problem lies in a continuous domain instead of discrete domain. a great attention in MVPO literature. However, because of the fact
that studies on GA and PSO have reached an adequate level of ma-
6.2.2. Machine learning algorithms turity and perhaps very limited improvement area has left, these
Machine learning field is evolved from the study of pattern two algorithms have lost researcher’s interest in the past few years.
recognition and computational learning theory in artificial intel- Contrary to this, the artificial bee colony algorithm which is rel-
ligence. Machine learning algorithms explore the problem data, atively a newer algorithm, may get a bigger slice from the cake
make data-driven predictions operating a model from a training due to higher performance against other algorithms reported by
set of input observations in order to make decisions expressed as (Kalayci et al., 2017).
outputs, rather than strictly following static program instructions.
Machine learning algorithms applied to MVPO problem are 6.4. Hybridization techniques
showed in Table 18.
Fernandez and Gomez (2007) applied Hopfield network (HN) Hybridization strategy may be defined as designing a better al-
to solve MVPO problem and compared against GA, SA and TS gorithm by combining favorable characteristics of different algo-
(Chang et al., 20 0 0) reporting HN’s superior performance on big- rithms. Researchers, in the quest of designing more efficient algo-
ger sized instances. Golmakani and Alishah (2008) proposed an ar- rithms for MVPO, used this strategy to benefit from the advantages
tificial immune system algorithm and compared with genetic al- of pure algorithms or to dismiss their drawbacks. Publications that
gorithm, tabu search and simulated annealing (Chang et al., 20 0 0) used hybridization strategy to solve MVPO problem can classified
and Hopfield network (Fernandez & Gomez, 2007) indicating that as in Table 20.
artificial immune system approach could achieve better results, yet Moral-Escudero et al. (2006) designed a hybrid strategy that
performed not as fast as others. Suganya and Vijayalakshmi Pai combines evolutionary techniques with quadratic programming.
(2009) designed an algorithm based on Hopfield network and em- Gaspero, Tollo, Roli, and Schaerf (2011) presented a hybrid tech-
ployed k-means clustering analysis to tackle additional constraints nique that combines a local search metaheuristic, as a master
in this algorithm. Publications that applied machine learning algo- solver, with a quadratic programming procedure, as a sub solver.
rithms for MVPO problem are summarized in the Table 19. Lwin and Qu (2013) proposed a hybrid algorithm by combining
population-based incremental learning and differential evolution
6.3. A peek into the solution approaches for MVPO algorithms adapting a mutation and an elitist strategy to enhance
the evolution over the search space. Tuba and Bacanin (2014a) pre-
The exact and the approximation techniques applied to the sented a hybridization of the artificial bee colony with the fire-
MVPO problem were described in Section 3.1 and Section 3.2, re- fly algorithm for MVPO problem in order to enhance the search
spectively. Among those solution techniques, the metaheuristics capability of the artificial bee colony algorithm for a more effec-
are commonly used to solve the MVPO problem (about 82% of all). tive trade-off between exploitation and exploration. They compare
The primary reason would be the NP-hard property of the spe- the proposed algorithm against genetic algorithm, tabu search and
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 361

Fig. 11. Scatter chart of solution techniques applied MVPO problem.

Table 20
Hybrid algorithms developed to solve MVPO problem.

simulated annealing (Chang et al., 20 0 0) and particle swarm opti- 6.5. Constraint handling methodology
mization (Cura, 2009) to confirm its performance. Ruiz-Torrubiano
and Suarez (2015) proposed a hybrid approach that combines In the MVPO literature, there are several methods to satisfy the
genetic algorithm and QP. Baykasoglu, Yunusoglu, and Ozsoydan constraints of the problem. Literature shows that constraint han-
(2015) developed a hybrid algorithm that selects a predetermined dling is done via a repair procedure, a penalty function or infea-
number of stocks with GRASP algorithm and finds the proportions sibility tolerance. Fig. 12 demonstrates distribution of constraint
of the selected stocks with QP. handling techniques employed in MVPO literature.
362 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Fig. 12. Distribution of constraint handling techniques.

Table 21
Constraint handling techniques for MVPO problem.

Constraint handling technique Publication

Repair procedure (Anagnostopoulos & Mamanis, 2010, 2011a; b; Anagnostopoulos et al., 2010; Bacanin & Tuba, 2014; 2015; Bacanin
et al., 2014; Branke et al., 2009; Chang et al., 20 0 0; Chang et al., 2009; Chen et al., 2013; Chen et al., 2012; Chen
et al., 2017; Chiam et al., 2007; Chiam et al., 2008; ChiangLin, 2006; Coutino-Gomez et al., 2003; Crama & Schyns,
2003; Cui et al., 2014; Cura, 2009; Deng & Lin, 2010a, b; Deng et al., 2012; Eftekharian et al., 2017; Fernandez &
Gomez, 2007; Garcia et al., 2012; Golmakani & Alishah, 2008; Hao & Liu, 2009; Huang & Shen, 2010; Jiang et al.,
2014; Jiang et al., 2008; Kalayci et al., 2017; Kamili & Riffi, 2015; Kao & Cheng, 2013; Koshino et al., 2007; Kumar &
Mishra, 2017; Kumar & Bhattacharya, 2012; Lai et al., 2006; Lean et al., 2008; Liagkouras & Metaxiotis, 2014; Lin
et al., 2005; Lwin & Qu, 2013; Lwin et al., 2014; Lwin et al., 2013; Lwin et al., 2017; Maringer & Kellerer, 2003;
Mishra et al., 2014a; Moral-Escudero et al., 2006; Mozafari et al., 2011; Ong et al., 2005; Pai & Michel, 2009; Reid &
Malan, 2015; Ruiz-Torrubiano & Suarez, 2010; Sabar & Song, 2014; Sadigh et al., 2012; Schaerf, 2002; Shoaf &
Foster, 1998; Skolpadungket et al., 20 07; Streichen & Tanaka-Yamawaki, 20 06; Streichert et al., 20 04a; b; Talebi
et al., 2010; Tang et al., 2009; Thomaidis, 2010; Tuba & Bacanin, 2014a; b; Tuba et al., 2013; Wang et al., 2012;
Woodside-Oriakhi et al., 2011; Xia et al., 20 0 0; Xu J. et al., 2011; Xu R.T. et al., 2010; Yin et al., 2015a; Yu et al.,
2009; Zhou & Li, 2014; Zhu et al., 2011)
Penalty function (Chen & Cai, 2008; Chen et al., 2006; Chen & Zhang, 2010; Corazza et al., 2012a; Corazza et al., 2013; Crama &
Schyns, 2003; Golmakani & Fazel, 2011; Lean et al., 2008; Lu & Wang, 2013; Moral-Escudero et al., 2006; Pouya
et al., 2016; Reid & Malan, 2015; Rong et al., 2009; Xu et al., 2007)
Infeasibility tolerance (Anagnostopoulos & Mamanis, 2010; Fasheng & Wei, 2006; Kalayci et al., 2017; Mishra et al., 2016; Mishra et al.,
2014b; Reid & Malan, 2015; Soleimani et al., 2009; Wang et al., 2011; Xu & Chen, 2006)

Repairing an infeasible solution to get into feasible region is the (ii) if the both solutions are infeasible, select the solution with less
most popular methodology for constraint handling when MVPO violation, (iii) if both solutions are feasible, select the solution that
problem is solved. Many researchers (Babaei, Sepehri, & Babaei, has a better objective function value. Kalayci et al. (2017) used
2015; Chen et al., 2013; Chen et al., 2017; Deng & Lin, 2010a, this infeasibility toleration technique to satisfy binary constraints
2010b; Lwin & Qu, 2013; Lwin, Qu, & Zheng, 2013; Mishra, Panda, and repair procedure to satisfy cardinality constraints in their pro-
& Majhi, 2014a; Pai & Michel, 2009; Schaerf, 2002; Vijayalakshmi posed artificial bee colony algorithm reporting the superior per-
Pai & Michel, 2013; Woodside-Oriakhi et al., 2011) used the repair formance against the repair procedure suggested by Chang et al.
procedure suggested by Chang et al. (20 0 0) to satisfy cardinality (20 0 0). Table 21 shows the classifications of studies according to
and boundary constraints at once. Crama and Schyns (2003) pro- utilized constraint handling methodologies.
posed specific approaches to deal with each specific class of con-
straint, either by explicitly restricting the portfolios to remain in 7. Conclusion and discussion
the feasible region or by penalizing infeasible portfolios. Golmakani
and Fazel (2011) and Corazza et al. (2013) used penalty functions This paper provided a comprehensive survey of deterministic
for constraint handling. Some researchers have applied the selec- models and applications on MVPO. Analyzing the literature con-
tion rules proposed by Deb (20 0 0) to deal with infeasibility. These firmed the growing interest in search of the new trends, devel-
selection rules can be summarized as follows: (i) If a solution is opments and practical factors on portfolio optimization as the re-
feasible and the other solution is infeasible, select the feasible one, search in MVPO continues to be very active.
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 363

A taxonomy of applications based on models, constraints, con- Despite all the efforts made to present a complete review on
straint handling techniques, various solution approaches along MVPO based on the applied research methodology, assuming that
with the performance indicators are presented. The vast majority the risky security returns obey normal distribution, it has some
of the presented work focused on employed exact and heuristic limitations. The review of powerful approaches for optimization
attempts for solving MVPO problem. The main research lines for under uncertainty, robust, fuzzy and dynamic optimization was out
this research area was devoted to the deterministic models with of the scope of this study.
classical formulations including realistic features and the adapta- The analysis of the literature revealed a number of open re-
tion of various algorithms with a computational analysis utilizing search fields requiring an in-depth discussion to guide researchers
difference performance indicators. for future research directions in the light of the findings based on
Although the original mean-variance model had single objec- the solution approaches developed for MVPO. It is evident that the
tive, multi-objective variants that simultaneously or independently field presents significant research opportunities both for academics
consider minimized risk and maximized return turned out to be and practitioners. The identified gaps in literature for the future re-
more popular as years pass since in practice, investors may not search opportunities are as follows:
be aware of the risk and returns levels they desire. In order to
tackle the multi-objective nature of the problem, the majority of – It is also worthy to note that the lack of multiple constraints
researchers had either focused on weighted sum approach that in the problem formulation that the real-life portfolio managers
systematically prioritizes objectives or dealt with efficiency and deal with limits the methodologies’ usefulness to academic pur-
non-dominance rules to obtain Pareto optimal solutions forming poses. Therefore, to reduce the gap between the academia and
the efficient frontier. the real-world, the incorporation of additional real-life con-
In the pursuit of a realistic portfolio management, various stud- straints is still a challenge to be answered.
ies increasingly incorporated additional constraints such as cardi- – Further research is always required towards the design of new
nality, threshold, transaction costs, round lots, sector capitaliza- algorithmic enhancements on solving portfolio optimization
tion and turnover constraints to the original model. Because of variants.
the integer constraints, the classical convex quadratic programming – Because of the fact that different financial markets have differ-
problem with a polynomial worst-case complexity bound had been ent dynamics and features, extensions of the common datasets
transformed into a mixed integer quadratic programming problem would undoubtfully be an important future direction in order to
which introduced the problem to the class of NP-Complete prob- provide an opportunity for testing new solution methodologies
lems. Therefore, the majority of researchers had employed approx- and algorithms in various market conditions. New performance
imate algorithms rather than exact algorithms to overcome com- measures would also serve a better race platform for future al-
putational complexity of the problem where solution time greatly gorithmic studies on testing the performance
matters in the financial business. Metaheuristics are by far the – It appears that most of the researchers focus on the com-
most preferred approximation algorithms adapted to solve MVPO. putational aspects of the problem and overlook its profitabil-
Although evolutionary and swarm intelligence algorithms utilizing ity in practice. A potential path of future research would be
the power of a population of solutions had drawn a great attention to analyze the performance of securities on investment hori-
so far, other successful metaheuristic algorithms such as single- zons in dynamic real-world environment. Therefore, solution
solution-based algorithms may get a bigger slice from the cake in approaches should also be of interest in incorporating multi-
the near future. period portfolio optimization since decision support systems in
In the quest of precise optimization, researchers have employed financial optimization rely on algorithmic decisions in consecu-
various hybridization strategies to benefit from the advantages of tive phases.
different algorithms or to dismiss some drawbacks. Although an – The decision-making process involved in portfolio optimization
exact algorithm may be inadequate to reach optimal solutions is extremely complex and difficult to automate. Therefore, to
alone in a reasonable time, literature analysis revealed that hy- extend the capabilities of efficient algorithms to trade execu-
bridizations of exact and metaheuristic algorithms had shown su- tion, the design of interactive and automated algorithms for the
perior performance against other competing algorithms. This may portfolio optimization problem promises a potential area for fu-
trigger the design and implementation of mat-heuristics for MVPO ture research.
variants. – Today the race is on financial trading. The investigation of per-
Constraint handling, a requirement for constrained portfolio forming short-term forecasting of stock prices with heuristics is
optimization, was done via repair procedures, penalty functions a topic worth exploring that can provide trading decision mod-
or temporarily tolerating infeasibility in the literature. Although els. Furthermore, combing forecasting theory with portfolio op-
a complete repair procedure was widely preferred in the liter- timization is another promising research direction.
ature, comparative results showed that a partial repair mecha- – In spite of the fact that uncertainty in MVPO was not within the
nism might enhance the algorithms with a dramatic effect on the scope of this study, it is evident from the uncertain characteris-
performance. tics of the financial data, making the model dynamic and robust
Considering the intense competition in financial markets to pro- may ensure a more consistent estimation. That’s why, a detailed
vide a better portfolio optimization, developing new methodolo- analysis on MVPO including uncertainty and robustness would
gies to efficiently solve portfolio optimization is a matter of great be very helpful for the researchers and practitioners.
importance. Publicly available benchmark instances for portfolio – As financial systems can be affected by various conditions such
optimization enable the computational comparison of the pro- as social-fact events, politics and investor behaviors, data min-
posed algorithms in the literature. Solution approaches racing on ing and big data analysis come into prominence for enhanced
the benchmarking platform would serve to the future of financial decision support systems.
decision making with the help of algorithmic enhancements. Most
of the studies are either based on some case studies or publicly Acknowledgement
available benchmarking datasets of limited variety and quantity.
The future of financial decisions relies on algorithmic performances This research is funded by the Scientific and Technological
which can only be validated on various large-scale datasets gath- Research Council of Turkey (TUBITAK) with the grant number
ered from complex environments and global markets. 214M224.
364 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

References Cao, J., & Tao, L. (2010). Improved particle swarm algorithm for portfolio optimiza-
tion problem. In ICIMA 2010 - 2010 2nd International Conference on Industrial
Abbas, A., & Haider, S. (2009). Comparison of AIS and PSO for constrained portfolio Mechatronics and Automation: 2 (pp. 561–564).
optimization. In Proceedings - 2009 International Conference on Information and Cesarone, F., Scozzari, A., & Tardella, F. (2013). A new method for mean-variance
Financial Engineering, ICIFE 2009 (pp. 50–54). portfolio optimization with cardinality constraints. Annals of Operations Re-
Ackora-Prah, J., Gyamerah, S. A., & Andam, P. S. (2014a). A heuristic crossover for search, 205, 213–234.
portfolio selection. Applied Mathematical Sciences, 8, 3215–3227. Cesarone, F., Scozzari, A., & Tardella, F. (2015). Linear vs. quadratic portfolio selection
Ackora-Prah, J., Gyamerah, S. A., Andam, P. S., & Gyamfi, D. (2014b). Pattern search models with hard real-world constraints. Computational Management Science, 12,
for portfolio selection. Applied Mathematical Sciences, 8, 7137–7147. 345–370.
Anagnostopoulos, K. P., & Mamanis, G. (2010). A portfolio optimization model with Chang, J. F., & Chen, K. L. (2008). Applying new investment satisfied capability index
three objectives and discrete variables. Computers & Operations Research, 37, and particle swarm optimization to stock portfolio selection. ICIC Express Letters,
1285–1297. 3, 349–354.
Anagnostopoulos, K. P., & Mamanis, G. (2011a). The mean–variance cardinality con- Chang, J. F., & Hsu, S. W. (2007). The construction of stock’s portfolios by using
strained portfolio optimization problem: An experimental evaluation of five particle swarm optimization. Second International Conference on Innovative Com-
multiobjective evolutionary algorithms. Expert Systems with Applications, 38, puting, Information and Control, ICICIC 2007.
14208–14217. Chang, T. J., Meade, N., Beasley, J. E., & Sharaiha, Y. M. (20 0 0). Heuristics for cardi-
Anagnostopoulos, K. P., & Mamanis, G. (2011b). Multiobjective evolutionary algo- nality constrained portfolio optimisation. Computers & Operations Research, 27,
rithms for complex portfolio optimization problems. Computational Management 1271–1302.
Science, 8, 259–279. Chang, T. J., Yang, S. C., & Chang, K. J. (2009). Portfolio optimization problems in
Anagnostopoulos, K. P., Šević, Ž., Chatzoglou, P. D., & Katsavounis, S. (2010). A reac- different risk measures using genetic algorithm. Expert Systems with Applications,
tive greedy randomized adaptive search procedure for a mixed integer portfolio 36, 10529–10537.
optimization problem. Managerial Finance, 36, 1057–1065. Chen, A. H. L., Liang, Y. C., & Liu, C. C. (2013). Portfolio optimization using improved
Aouni, B., Colapinto, C., & La Torre, D. (2014). Financial portfolio management artificial bee colony approach. In Proceedings of the 2013 IEEE Conference on Com-
through the goal programming model: Current state-of-the-art. European Jour- putational Intelligence for Financial Engineering and Economics, CIFEr 2013 - 2013
nal of Operational Research, 234, 536–545. IEEE Symposium Series on Computational Intelligence, SSCI 2013 (pp. 60–67).
Aouni, B., Doumpos, M., Pérez-Gladish, B., & Steuer, R. E. (2018). On the increas- Chen, A. H. L., Yun-Chia, L., & Chia-Chien, L. (2012). An artificial bee colony algo-
ing importance of multiple criteria decision aid methods for portfolio selection. rithm for the cardinality-constrained portfolio optimization problems. In Evolu-
Journal of the Operational Research Society, 1–18. tionary Computation (CEC), 2012 IEEE Congress on (pp. 1–8).
Aouni, B. F., Abdelaziz, Ben, & Martel, J. M. (2005). Decision-maker’s preferences Chen, B. L., Lin, Y. B., Zeng, W. H., Xu, H., & Zhang, D. F. (2017). The mean–
modeling in the stochastic goal programming. European Journal of Operational variance cardinality constrained portfolio optimization problem using a lo-
Research, 162, 610–618. cal search-based multi-objective evolutionary algorithm. Applied Intelligence, 47,
Aranha, C., & Iba, H. (2009). The Memetic Tree-based Genetic Algorithm and its ap- 505–525.
plication to Portfolio Optimization. Memetic Computing, 1, 139–151. Chen, C., & Zhou, Y. S. (2018). Robust multiobjective portfolio with higher moments.
Arkeman, Y., Yusuf, A., Mushthofa, Laxmi, G. F., & Seminar, K. B. (2013). The forma- Expert Systems with Applications, 100, 165–181.
tion of optimal portfolio of mutual shares funds using multi-objective genetic Chen, W. (2015). Artificial bee colony algorithm for constrained possibilistic port-
algorithm. Telkomnika, 11, 625–636. folio optimization problem. Physica a-Statistical Mechanics and Its Applications,
Azmi, R., & Tamiz, M. (2010). A review of goal programming for portfolio selection. 429, 125–139.
Lecture Notes in Economics and Mathematical Systems, 638, 15–33. Chen, W., & Cai, Y. M. (2008). Study on the efficient frontier in portfolio selection by
Babaei, S., Sepehri, M. M., & Babaei, E. (2015). Multi-objective portfolio optimiza- using particle swarm optimizationChinese Control and Decision Conference, 2008,
tion considering the dependence structure of asset returns. European Journal of CCDC 2008 (pp. 269–272).
Operational Research, 244, 525–539. Chen, W., Xu, W. J., Yang, L., & Cai, Y. M. (2008). Genetic algorithm with an application
Bacanin, N., & Tuba, M. (2014). Firefly algorithm for cardinality constrained mean– to complex portfolio selectionProceedings - 4th International Conference on Natural
variance portfolio optimization problem with entropy diversity constraint. Sci- Computation, ICNC 2008: 5 (pp. 333–337). IEEE.
entific World Journal, 2014, 1–16. Chen, W., Zhang, R. T., Cai, Y. M., & Xu, F. S. (2006). Particle swarm optimization for
Bacanin, N., & Tuba, M. (2015). Fireworks algorithm applied to constrained portfolio constrained portfolio selection problems. In Proceedings of the 2006 International
optimization problem. In 2015 IEEE Congress on Evolutionary Computation, CEC Conference on Machine Learning and Cybernetics: 2006 (pp. 2425–2429). IEEE.
2015 - Proceedings (pp. 1242–1249). Chen, W., & Zhang, W. G. (2010). The admissible portfolio selection problem with
Bacanin, N., Tuba, M., & Pelevic, B. (2014). Constrained portfolio selection using ar- transaction costs and an improved PSO algorithm. Physica a-Statistical Mechanics
tificial bee colony (ABC) algorithm. International Journal of Mathematical Models and Its Applications, 389, 2070–2076.
and Methods in Applied Sciences, 8, 190–198. Chiam, S. C., Al Mamun, A., & Low, Y. L. (2007). A realistic approach to evolution-
Baluja, S. (1994). Population-based incremental learning. a method for integrating ge- ary multiobjective portfolio optimization. In 2007 IEEE Congress on Evolutionary
netic search based function optimization and competitive learning. DTIC Document. Computation, CEC 2007 (pp. 204–211).
Ban, G. Y., El Karoui, N., & Lim, A. E. B. (2018). Machine learning and portfolio opti- Chiam, S. C., Tan, K. C., & Mamum, A. (2008). Evolutionary multi-objective portfolio
mization. Management Science, 64, 1136–1154. optimization in practical context. International Journal of Automation and Com-
Bandyopadhyay, S., Saha, S., Maulik, U., & Deb, K. (2008). A simulated anneal- puting, 5, 67–80.
ing-based multiobjective optimization algorithm: AMOSA. IEEE Transactions on ChiangLin, C. Y. (2006). Applications of genetic algorithm to portfolio optimization
Evolutionary Computation, 12, 269–283. with practical transaction constraints. In Proceedings of the 9th Joint Conference
Baykasoğlu, A., Avci, M. G., & Burcin Özsoydan, F. (2016). Erratum to “A GRASP based on Information Sciences, JCIS 2006: 2006.
solution approach to solve cardinality constrained portfolio optimization prob- Chu, S. C., & Tsai, P. W. (2007). Computational intelligence based on the behavior
lems” [Comput. Indus. Eng. 90 (2015) 339–351]. Computers & Industrial Engineer- of cats. International Journal of Innovative Computing Information and Control, 3,
ing, 96, 249–250. 163–173.
Baykasoglu, A., Yunusoglu, M. G., & Ozsoydan, F. B. (2015). A GRASP based solu- Civicioglu, P. (2013). Artificial cooperative search algorithm for numerical optimiza-
tion approach to solve cardinality constrained portfolio optimization problems. tion problems. Information Sciences, 229, 58–76.
Computers & Industrial Engineering, 90, 339–351. Coello, C. A. C., Pulido, G. T., & Lechuga, M. S. (2004). Handling multiple objectives
Beasley, J. E. (1990). Or-Library - distributing test problems by electronic mail. Jour- with particle swarm optimization. IEEE Transactions on Evolutionary Computa-
nal of the Operational Research Society, 41, 1069–1072. tion, 8, 256–279.
Bevilacqua, V., Pacelli, V., & Saladino, S. (2011). A novel multi objective genetic al- Corazza, M., Fasano, G., & Gusso, R. (2012a). Portfolio selection with an alternative
gorithm for the portfolio optimization. In D.-S. Huang, Y. Gan, V. Bevilacqua, & measure of risk: Computational performances of particle swarm optimization
J. C. Figueroa (Eds.), Advanced intelligent computing (pp. 186–193). BerlinHeidel- and genetic algorithms. In C. Perna, & M. Sibillo (Eds.), Mathematical and statis-
berg: Springer Berlin Heidelberg. tical methods for actuarial sciences and finance (pp. 123–130). Milano: Springer
Blum, C., & Roli, A. (2003). Metaheuristics in combinatorial optimization: Overview Milan.
and conceptual comparison. Acm Computing Surveys, 35, 268–308. Corazza, M., Fasano, G., & Gusso, R. (2012b). Portfolio selection with an alternative
Bonami, P., & Lejeune, M. A. (2009). An exact solution approach for portfolio opti- measure of risk: Computational performances of Particle Swarm Optimization
mization problems under stochastic and integer constraints. Operations Research, and Genetic Algorithms. Mathematical and Statistical Methods for Actuarial Sci-
57, 650–670. ences and Finance, 123–130.
Branke, J., Scheckenbach, B., Stein, M., Deb, K., & Schmeck, H. (2009). Portfolio opti- Corazza, M., Fasano, G., & Gusso, R. (2013). Particle Swarm Optimization with non-s-
mization with an envelope-based multi-objective evolutionary algorithm. Euro- mooth penalty reformulation, for a complex portfolio selection problem. Applied
pean Journal of Operational Research, 199, 684–693. Mathematics and Computation, 224, 611–624.
Brito, R. P., & Vicente, L. N. (2014a). Efficient cardinality/mean-variance portfolios Corne, D., Knowles, J., & Oates, M. (20 0 0). The Pareto envelope-based selection al-
(pp. 52–73). BerlinHeidelberg: Springer Berlin Heidelberg. gorithm for multiobjective optimization. In Parallel problem solving from nature
Brito, R. P., & Vicente, L. N. (2014b). Efficient cardinality/mean-variance portfolios. PPSN VI (pp. 839–848). Springer.
IFIP Advances in Information and Communication Technology, 443, 52–73. Coutino-Gomez, C. A., Torres-Jimenez, J., & Villarreal-Antelo, B. M. (2003). Heuris-
Busetti, F. (2006). Heuristic approaches to realistic portfolio optimisation. WIT Trans- tic Methods for Portfolio Selection at the Mexican Stock Exchange. In J. Liu,
actions on Modelling and Simulation, 43, 361–370. Y.-m. Cheung, & H. Yin (Eds.), Intelligent Data Engineering and Automated Learn-
ing: 4th International Conference, IDEAL 2003, Hong Kong, China, March 21-23,
2003. Revised Papers (pp. 919–923). Springer Berlin Heidelberg.
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 365

Crama, Y., & Schyns, M. (2003). Simulated annealing for complex portfolio selection Gandomi, A. H., & Alavi, A. H. (2012). Krill herd: A new bio-inspired optimization
problems. European Journal of Operational Research, 150, 546–571. algorithm. Communications in Nonlinear Science and Numerical Simulation, 17,
Cui, T., Cheng, S., & Bai, R. (2014). A combinatorial algorithm for the cardinality con- 4831–4845.
strained portfolio optimization problem. In Proceedings of the 2014 IEEE Congress Gao, J., & Chu, Z. (2009). An improved particle swarm optimization for the con-
on Evolutionary Computation, CEC 2014 (pp. 491–498). strained portfolio selection problem. In Computational Intelligence and Natural
Cui, X. T., Zheng, X. J., Zhu, S. S., & Sun, X. L. (2013). Convex relaxations and MIQCQP Computing, 2009. CINC’09. International Conference on: 1 (pp. 518–522). IEEE.
reformulations for a class of cardinality-constrained portfolio selection prob- García, F., Guijarro, F., & Oliver, J. (2018). Index tracking optimization with cardinal-
lems. Journal of Global Optimization, 56, 1409–1423. ity constraint: A performance comparison of genetic algorithms and tabu search
Cura, T. (2009). Particle swarm optimization approach to portfolio optimization. heuristics. Neural Computing and Applications, 30, 2625–2641.
Nonlinear Analysis-Real World Applications, 10, 2396–2406. Garcia, S., Quintana, D., Galvan, I. M., & Isasi, P. (2012). Time-stamped resampling
Czyzżak, P., & Jaszkiewicz, A. (1998). Pareto simulated annealing—a metaheuristic for robust evolutionary portfolio optimization. Expert Systems with Applications,
technique for multiple-objective combinatorial optimization. Journal of Multi- 39, 10722–10730.
-Criteria Decision Analysis, 7, 34–47. Gaspero, L. D., Tollo, G. D., Roli, A., & Schaerf, A. (2011). Hybrid metaheuristics for
De Castro, L. N., & Timmis, J. (2002). Artificial immune systems: a new computational constrained portfolio selection problems. Quantitative Finance, 11, 1473–1487.
intelligence approach. Springer Science & Business Media. Glover, F. (1986). Future paths for integer programming and links to artificial-intel-
Deb, K. (20 0 0). An efficient constraint handling method for genetic algorithms. Com- ligence. Computers & Operations Research, 13, 533–549.
puter Methods in Applied Mechanics and Engineering, 186, 311–338. Glover, F., Lagunai, M., & Marti, R. (20 0 0). Fundamentals of scatter search and path
Deb, K. (2001). Multi-objective optimization using evolutionary algorithms: 16. John relinking. Control and cybernetics, 29, 653–684.
Wiley & Sons. Golmakani, H. R., & Alishah, E. J. (2008). Portfolio selection using an artificial im-
Deb, K. (2005). Multi-objective optimization. In E. K. Burke, & G. Kendall (Eds.), mune system. In 2008 IEEE International Conference on Information Reuse and
Search Methodologies: introductory tutorials in optimization and decision support Integration, IEEE IRI-2008 (pp. 28–33).
techniques (pp. 273–316). Boston, MA: Springer US. Golmakani, H. R., & Fazel, M. (2011). Constrained portfolio selection using particle
Deb, K., Pratap, A., Agarwal, S., & Meyarivan, T. (2002). A fast and elitist multiobjec- swarm optimization. Expert Systems with Applications, 38, 8327–8335.
tive genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation, Govindan, K., Soleimani, H., & Kannan, D. (2015). Reverse logistics and closed-loop
6, 182–197. supply chain: A comprehensive review to explore the future. European Journal
Dehghan Hardoroudi, N., Keshvari, A., Kallio, M., & Korhonen, P. (2017). Solving car- of Operational Research, 240, 603–626.
dinality constrained mean-variance portfolio problems via MILP. Annals of Oper- Hadi, A. S., El Naggar, A. A., & Bary, M. N. A. (2016). New model and method for
ations Research, 254, 47–59. portfolios selection. Applied Mathematical Sciences, 10, 263–288.
Deng, G.-F., & Lin, W.-T. (2010a). Ant Colony optimization for markowitz mean-vari- Hanne, T. (2007). A multiobjective evolutionary algorithm for approximating the ef-
ance portfolio model. In B. K. Panigrahi, S. Das, P. N. Suganthan, & S. S. Dash ficient set. European Journal of Operational Research, 176, 1723–1734.
(Eds.), Swarm, Evolutionary, and Memetic Computing: First International Confer- Hao, F. F., & Liu, Y. K. (2009). Mean-variance models for portfolio selection with
ence on Swarm, Evolutionary, and Memetic Computing, SEMCCO 2010, Chennai, In- fuzzy random returns. Journal of Applied Mathematics and Computing, 30, 9–38.
dia, December 16-18, 2010. Proceedings (pp. 238–245). Springer Berlin Heidelberg. He, F., & Qu, R. (2016). Hybridising local search with branch-and-bound for con-
Deng, G.-F., & Lin, W.-T. (2010b). Swarm intelligence for cardinality-constrained strained portfolio selection problems. In Proceedings - 30th European Conference
portfolio problems. In Lecture Notes in Computer Science (including subseries on Modelling and Simulation Ecms 2016 (pp. 446–452).
Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics): 6423 Hoklie, & Zuhal, L. R. (2010). Resolving multi objective stock portfolio optimization
(pp. 406–415). LNAI. problem using genetic algorithm. In 2010 The 2nd International Conference on
Deng, G. F., Lin, W. T., & Lo, C. C. (2012). Markowitz-based portfolio selection with Computer and Automation Engineering (ICCAE): 2 (pp. 40–44). IEEE.
cardinality constraints using improved particle swarm optimization. Expert Sys- Holland, J. H. (1975). Adaptation in natural and artificial systems. Ann Arbor, MI: The
tems with Applications, 39, 4558–4566. University of Michigan Press.
Doering, J., Juan, A. A., Kizys, R., Fito, A., & Calvet, L. (2016). Solving realistic portfolio Hong-mei, L., Zhuo-fu, W., & Hui-min, L. (2010). Artificial bee colony algorithm
optimization problems via metaheuristics: A survey and an example. In Model- for real estate portfolio optimization based on risk preference coefficient. In
ing and simulation in Engineering, economics and management: 254 (pp. 22–30). Management Science and Engineering (ICMSE), 2010 International Conference on
Springer. (pp. 1682–1687). IEEE.
Dorigo, M., Maniezzo, V., & Colorni, A. (1996). Ant system: Optimization by a colony Hooke, R., & Jeeves, T. A. (1961). “Direct Search”Solution of Numerical and Statistical
of cooperating agents. IEEE Trans Syst Man Cybern B Cybern, 26, 29–41. Problems. Journal of the ACM (JACM), 8, 212–229.
Dreżewski, R., & Doroz, K. (2017). An agent-based co-evolutionary multi-objective Hopfield, J. J. (1984). Neurons with graded response have collective computational
algorithm for portfolio optimization. Symmetry, 9, 168–198. properties like those of two-state neurons. Proceedings of the National Academy
Duran, F. C., Cotta, C., & Fernández, A. J. (2009). Evolutionary optimization for mul- of Sciences, 81, 3088–3092.
tiobjective portfolio selection under markowitz’s model with application to the Horn, J., Nafpliotis, N., & Goldberg, D. E. (1994). A niched Pareto genetic algorithm
caracas stock exchange. In R. Chiong (Ed.). In Nature-Inspired algorithms for op- for multiobjective optimization. In Evolutionary Computation, 1994. IEEE World
timisation: 193 (pp. 489–509). BerlinHeidelberg: Springer Berlin Heidelberg. Congress on Computational Intelligence., Proceedings of the First IEEE Conference
Eftekharian, S., Shojafar, M., & Shamshirband, S. (2017). 2-Phase NSGA II: An opti- on (pp. 82–87). Ieee.
mized reward and risk measurements algorithm in portfolio optimization. Algo- Hu, J., & Zhangy, G. (2010). Comparison of portfolio optimization models with real
rithms, 10, 130–143. features: An empirical study based on Chinese stock market. Dynamics of Con-
Ehrgott, M., Klamroth, K., & Schwehm, C. (2004). An MCDM approach to portfolio tinuous, Discrete and Impulsive Systems Series B: Applications and Algorithms, 17,
optimization. European Journal of Operational Research, 155, 752–770. 83–100.
Erickson, M., Mayer, A., & Horn, J. (2001). The niched pareto genetic algorithm 2 Huang, X., & Shen, W. (2010). Multi-period mean-variance model with transaction
applied to the design of groundwater remediation systems. International Confer- cost for fuzzy portfolio selection. In Proceedings - 2010 7th International Con-
ence on Evolutionary Multi-Criterion Optimization, 1993, 681–695. ference on Fuzzy Systems and Knowledge Discovery, FSKD 2010: 2 (pp. 894–898).
Farzi, S., Shavazi, A. R., & Pandari, A. (2013). Using quantum-behaved particle swarm IEEE.
optimization for portfolio selection problem. International Arab Journal of Infor- Huang, X. X. (2012). Mean-variance models for portfolio selection subject to experts’
mation Technology, 10, 111–119. estimations. Expert Systems with Applications, 39, 5887–5893.
Fasheng, X., & Wei, C. (2006). Stochastic portfolio selection based on velocity limited Jalota, H., & Thakur, M. (2018). Genetic algorithm designed for solving portfolio op-
particle swarm optimization. In Proceedings of the World Congress on Intelligent timization problems subjected to cardinality constraint. International Journal of
Control and Automation (WCICA): 1 (pp. 3599–3603). IEEE. System Assurance Engineering and Management, 9, 294–305.
Fastrich, B., & Winker, P. (2012). Robust portfolio optimization with a hybrid heuris- Jiang, K., Li, D., Gao, J., & Yu, J. X. (2014). Factor model based clustering approach for
tic algorithm. Computational Management Science, 9, 63–88. cardinality constrained portfolio selection. In IFAC Proceedings Volumes (IFAC-Pa-
Feo, T. A., & Resende, M. G. C. (1995). Greedy randomized adaptive search proce- persOnline): 19 (pp. 10713–10718). Elsevier.
dures. Journal of Global Optimization, 6, 109–133. Jiang, W., Zhang, Y., & Xie, J. (2008). A particle swarm optimization algorithm
Fernandez, A., & Gomez, S. (2007). Portfolio selection using neural networks. Com- based on diffusion-repulsion and application to portfolio selection. In 2008
puters & Operations Research, 34, 1177–1191. International Symposium on Information Science and Engineering, ISISE 2008: 2
Fieldsend, J. E., Matatko, J., & Peng, M. (2004). Cardinality constrained portfolio opti- (pp. 498–501). IEEE.
misation. In Z. R. Yang, H. Yin, & R. M. Everson (Eds.), Intelligent Data Engineer- Jin, Y., Qu, R., & Atkin, J. (2015). A population-based incremental learning method
ing and Automated Learning – IDEAL 2004: 5th International Conference, Exeter, for constrained portfolio optimisation. In Proceedings - 16th International Sympo-
UK. August 25-27, 2004. Proceedings (pp. 788–793). Berlin, Heidelberg: Springer sium on Symbolic and Numeric Algorithms for Scientific Computing, SYNASC 2014
Berlin Heidelberg. (pp. 212–219). IEEE.
Fonseca, C. M., & Fleming, P. J. (1993). Genetic algorithms for multiobjective opti- Kalayci, C. B., Ertenlice, O., Akyer, H., & Aygoren, H. (2017). An artificial bee colony
mization: formulationdiscussion and generalization. In Icga: 93 (pp. 416–423). algorithm with feasibility enforcement and infeasibility toleration procedures
Citeseer. for cardinality constrained portfolio optimization. Expert Systems with Applica-
Forrest, S., & Mitchell, M. (1993). Relative building-block fitness and the build- tions, 85, 61–75.
ing-block hypothesis. Foundations of Genetic Algorithms 2, 2, 109–126. Kamali, S. (2014). Portfolio optimization using particle swarm optimization and ge-
Freitas, F. D., De Souza, A. F., & de Almeida, A. R. (2009). Prediction-based portfolio netic algorithm. Journal of mathematics and computer science, 10, 85–90.
optimization model using neural networks. Neurocomputing, 72, 2155–2170. Kamili, H., & Riffi, M. E. (2015). Portfolio selection using the cat swarm optimization.
Journal of Theoretical and Applied Information Technology, 74, 374–380.
366 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Kamili, H., & Riffi, M. E. (2016). A comparative study on portfolio optimization Liu, Y. J., & Zhang, W. G. (2015). A multi-period fuzzy portfolio optimization model
problem. In Proceedings - 2016 International Conference on Engineering and MIS, with minimum transaction lots. European Journal of Operational Research, 242,
ICEMIS 2016. IEEE. 933–941.
Kao, Y. C., & Cheng, H. T. (2013). Bacterial foraging optimization approach to portfo- Lloyd, S. P. (1982). Least-Squares Quantization in Pcm. IEEE Transactions on Informa-
lio optimization. Computational Economics, 42, 453–470. tion Theory, 28, 129–137.
Karaboga, D. (2005). An idea based on honey bee swarm for numerical optimization: Loukeris, N., Donelly, D., Khuman, A., & Peng, Y. (2009). A numerical evaluation
200. Kayseri: Erciyes University, Engineering Faculty, Computer Engineering De- of meta-heuristic techniques in portfolio optimisation. Operational Research, 9,
partment. 81–103.
Keating, C., & Shadwick, W. F. (2002). A universal performance measure. Journal of Lu, Z., & Wang, X. (2013). Improved portfolio optimization with non-convex and
performance measurement, 6, 59–84. non-concave cost using genetic algorithms. In Proceedings - 2013 International
Kennedy, J., & Eberhart, R. (1995). Particle swarm optimization. In Neural Conference on Mechatronic Sciences, Electric Engineering and Computer, MEC 2013
Networks, 1995. Proceedings., IEEE International Conference on: 4 (pp. 1942–1948). (pp. 2567–2570). IEEE.
vol. 1944. Lwin, K., & Qu, R. (2013). A hybrid algorithm for constrained portfolio selection
Khare, V., Yao, X., & Deb, K. (2003). Performance scaling of Multi-objective problems. Applied Intelligence, 39, 251–266.
evolutionary algorithms (pp. 376–390). BerlinHeidelberg: Springer Berlin Heidel- Lwin, K., Qu, R., & Kendall, G. (2014). A learning-guided multi-objective evolution-
berg. ary algorithm for constrained portfolio optimization. Applied Soft Computing, 24,
Kırış, Ş., & Ustun, O. (2012). An integrated approach for stock evaluation and port- 757–772.
folio optimization. Optimization, 61, 423–441. Lwin, K., Qu, R., & Zheng, J. (2013). Multi-objective scatter search with external
Kirkpatrick, S., Gelatt, C. D., Jr., & Vecchi, M. P. (1983). Optimization by simulated archive for portfolio optimization. In IJCCI 2013 - Proceedings of the 5th Inter-
annealing. Science, 220, 671–680. national Joint Conference on Computational Intelligence (pp. 111–119).
Knowles, J., Thiele, L., & Zitzler, E. (2006). A tutorial on the performance assessment Lwin, K. T., Qu, R., & MacCarthy, B. L. (2017). Mean-VaR portfolio optimization:
of stochastic multiobjective optimizers. Tik report, 214, 327–332. A nonparametric approach. European Journal of Operational Research, 260, 751–
Kocadağlı, O., & Keskin, R. (2015). A novel portfolio selection model based on fuzzy 766.
goal programming with different importance and priorities. Expert Systems with Macedo, L. L., Godinho, P., & Alves, M. J. (2017). Mean-semivariance portfolio op-
Applications, 42, 6898–6912. timization with multiobjective evolutionary algorithms and technical analysis
Kolm, P. N., Tutuncu, R., & Fabozzi, F. J. (2014). 60 Years of portfolio optimization: rules. Expert Systems with Applications, 79, 33–43.
Practical challenges and current trends. European Journal of Operational Research, Mansini, R., Ogryczak, W., & Speranza, M. G. (2014). Twenty years of linear program-
234, 356–371. ming based portfolio optimization. European Journal of Operational Research, 234,
Konno, H., & Yamazaki, H. (1991). Mean-Absolute Deviation Portfolio Optimiza- 518–535.
tion Model and Its Applications to Tokyo Stock-Market. Management Science, 37, Mansour, N., Rebai, A., & Aouni, B. (2007). Portfolio selection through the impre-
519–531. cise goal programming model: Integration of manager’s preferences. Journal of
Koshino, M., Murata, H., & Kimura, H. (2007). Improved particle swarm optimization Industrial Engineering International, 3, 1–8.
and application to portfolio selection. Electronics and Communications in Japan Maringer, D., & Kellerer, H. (2003). Optimization of cardinality constrained portfolios
Part III-Fundamental Electronic Science, 90, 13–25. with a hybrid local search algorithm. OR Spectrum, 25, 481–495.
Kukkonen, S., & Lampinen, J. (2005). GDE3: The third evolution step of generalized Markowitz, H. (1952). Portfolio Selection. The Journal of Finance, 7, 77–91.
differential evolution. In Evolutionary Computation, 2005. The 2005 IEEE Congress Marler, R. T., & Arora, J. S. (2004). Survey of multi-objective optimization methods
on: 1 (pp. 443–450). IEEE. for engineering. Structural and Multidisciplinary Optimization, 26, 369–395.
Kumar, D., & Mishra, K. K. (2017). Portfolio optimization using novel co-variance Masmoudi, M., & Abdelaziz, F. B. (2018). Portfolio selection problem: A review of
guided Artificial Bee Colony algorithm. Swarm and Evolutionary Computation, 33, deterministic and stochastic multiple objective programming models. Annals of
119–130. Operations Research, 267, 335–352.
Kumar, R., & Bhattacharya, S. (2012). Cooperative search using agents for cardinality Mayambala, F., Rönnberg, E., & Larsson, T. (2015). Eigendecomposition of the mean–
constrained portfolio selection problem. IEEE Transactions on Systems Man and variance portfolio optimization model. Springer Proceedings in Mathematics and
Cybernetics Part C-Applications and Reviews, 42, 1510–1518. Statistics, 130, 209–232.
Lai, K. K., Yu, L., Wang, S., & Zhou, C. (2006). A double-stage genetic optimization al- Mehrabian, A. R., & Lucas, C. (2006). A novel numerical optimization algorithm in-
gorithm for portfolio selection. In I. King, J. Wang, L.-W. Chan, & D. Wang (Eds.), spired from weed colonization. Ecological informatics, 1, 355–366.
Neural Information Processing: 13th International Conference, ICONIP 2006, Hong Metaxiotis, K., & Liagkouras, K. (2012). Multiobjective Evolutionary Algorithms for
Kong, China, October 3-6, 2006. Proceedings, Part III (pp. 928–937). Springer Berlin Portfolio Management: A comprehensive literature review. Expert Systems with
Heidelberg. Applications, 39, 11685–11698.
Lam, A. Y. S., & Li, V. O. K. (2010). Chemical-reaction-inspired metaheuristic for op- Mishra, S. K., Panda, G., & Majhi, B. (2016). Prediction based mean-variance model
timization. IEEE Transactions on Evolutionary Computation, 14, 381–399. for constrained portfolio assets selection using multiobjective evolutionary al-
Lean, Y., Wang, S., & Lai, K. K. (2008). Portfolio optimization using evolutionary al- gorithms. Swarm and Evolutionary Computation, 28, 117–130.
gorithms. Reflexing Interfaces: The Complex Coevolution of Information Technology Mishra, S. K., Panda, G., & Majhi, R. (2014a). A comparative performance assessment
Ecosystems, 235–245. of a set of multiobjective algorithms for constrained portfolio assets selection.
Li, D., Sun, X. L., & Wang, J. (2006). Optimal lot solution to cardinality constrained Swarm and Evolutionary Computation, 16, 38–51.
mean-variance formulation for portfolio selection. Mathematical Finance, 16, Mishra, S. K., Panda, G., & Majhi, R. (2014b). Constrained portfolio asset selection
83–101. using multiobjective bacteria foraging optimization. Operational Research, 14,
Li, L., Xue, B., Tan, L., & Niu, B. (2010). Improved particle swarm optimizers with 113–145.
application on constrained portfolio selection. In Lecture notes in computer sci- Mishra, S. K., Panda, G., & Meher, S. (2009). Multi-objective particle swarm opti-
ence (including subseries lecture notes in artificial intelligence and lecture notes in mization approach to portfolio optimization. In 2009 World Congress on Nature
Bioinformatics): 6215 (pp. 579–586). LNCS. and Biologically Inspired Computing, NABIC 2009 - Proceedings (pp. 1612–1615).
Li, X., Zhang, Y., Wong, H. S., & Qin, Z. F. (2009). A hybrid intelligent algorithm for IEEE.
portfolio selection problem with fuzzy returns. Journal of Computational and Ap- Moral-Escudero, R., Ruiz-Torrubiano, R., & Suarez, A. (2006). Selection of optimal
plied Mathematics, 233, 264–278. investment portfolios with cardinality constraints. In Evolutionary Computation,
Liagkouras, K. (2019). A new three-dimensional encoding multiobjective evolu- 2006. CEC 2006. IEEE Congress on (pp. 2382–2388). IEEE.
tionary algorithm with application to the portfolio optimization problem. Mozafari, M., Jolai, F., & Tafazzoli, S. (2011). A new IPSO-SA approach for cardinality
Knowledge-Based Systems, 163, 186–203. https://doi.org/10.1016/j.knosys.2018.08. constrained portfolio optimization. International Journal of Industrial Engineering
025. Computations, 2, 249–262.
Liagkouras, K., & Metaxiotis, K. (2014). A new Probe Guided Mutation operator Ni, Q. J., Yin, X. S., Tian, K. W., & Zhai, Y. Q. (2017). Particle swarm optimization
and its application for solving the cardinality constrained portfolio optimization with dynamic random population topology strategies for a generalized portfolio
problem. Expert Systems with Applications, 41, 6274–6290. selection problem. Natural Computing, 16, 31–44.
Liagkouras, K., & Metaxiotis, K. (2017). Handling the complexities of the multi– Niu, B., Tan, L., Xue, B., Li, L., & Chai, Y. (2010). Constrained portfolio selection using
constrained portfolio optimization problem with the support of a novel MOEA. multiple swarms. In IEEE Congress on Evolutionary Computation (pp. 1–7). IEEE.
Journal of the Operational Research Society, 1–19. Niu, B., Xue, B., Li, L., & Chai, Y. (2009). Symbiotic multi-swarm PSO for portfolio op-
Liagkouras, K., & Metaxiotis, K. (2018). A new efficiently encoded multiobjective al- timization. In D.-S. Huang, K.-H. Jo, H.-H. Lee, H.-J. Kang, & V. Bevilacqua (Eds.),
gorithm for the solution of the cardinality constrained portfolio optimization Emerging Intelligent Computing Technology and Applications. With Aspects of Arti-
problem. Annals of Operations Research, 267, 281–319. ficial Intelligence: 5th International Conference on Intelligent Computing, ICIC 2009
Liang, J. J., & Qu, B. Y. (2013). Large-scale portfolio optimization using multiobjec- Ulsan, South Korea, September 16-19, 2009 Proceedings (pp. 776–784). Springer
tive dynamic mutli-swarm particle swarm optimizer. In Proceedings of the 2013 Berlin Heidelberg.
IEEE Symposium on Swarm Intelligence, SIS 2013 - 2013 IEEE Symposium Series on Ong, C. S., Huang, H. J., & Tzeng, G. H. (2005). A novel hybrid model for portfolio
Computational Intelligence, SSCI 2013 (pp. 1–6). selection. Applied Mathematics and Computation, 169, 1195–1210.
Lin, C. C., & Liu, Y. T. (2008). Genetic algorithms for portfolio selection problems Özceylan, E., Kalayci, C. B., Güngör, A., & Gupta, S. M. (2018). Disassembly line bal-
with minimum transaction lots. European Journal of Operational Research, 185, ancing problem: A review of the state of the art and future directions. Interna-
393–404. tional Journal of Production Research, 1–23.
Lin, D., Li, X., & Li, M. (2005). In A genetic algorithm for solving portfolio opti- Pai, G. A. V., & Michel, T. (2009). Evolutionary optimization of constrained k-means
mization problems with transaction costs and minimum transaction lots: 3612 clustered assets for diversification in small portfolios. IEEE Transactions on Evo-
(pp. 808–811). BerlinHeidelberg: Springer Berlin Heidelberg. lutionary Computation, 13, 1030–1053.
 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368 367

Paiva, F. D., Cardoso, R. T. N., Hanaoka, G. P., & Duarte, W. M. (2019). Decision-mak- Streichert, F., Ulmer, H., & Zell, A. (2004b). Evolutionary algorithms and the cardi-
ing for financial trading: A fusion approach of machine learning and portfolio nality constrained portfolio optimization problem. In Operations Research Pro-
selection. Expert Systems with Applications, 115, 635–655. ceedings 2003: 2003 (pp. 253–260). Springer.
Panigrahi, B. K., Pandi, V. R., Sharma, R., Das, S., & Das, S. (2011). Multiobjective bac- Strumberger, I., Bacanin, N., & Tuba, M. (2017). Constrained portfolio optimization by
teria foraging algorithm for electrical load dispatch problem. Energy Conversion hybridized bat algorithm. In 7th International Conference on Intelligent Systems,
and Management, 52, 1334–1342. Modelling and Simulation (pp. 83–88). IEEE.
Passino, K. M. (2002). Biomimicry of bacterial foraging for distributed optimization Suganya, N. C., & Vijayalakshmi Pai, G. A. (2009). Constrained portfolio rebalancing
and control. IEEE Control Systems Magazine, 22, 52–67. with transaction costs using evolutionary wavelet hopfield network strategy. In
Peng, H., Kitagawa, G., Gan, M., & Chen, X. H. (2011). A new optimal portfolio selec- 2009 World Congress on Nature and Biologically Inspired Computing, NABIC 2009
tion strategy based on a quadratic form mean-variance model with transaction - Proceedings (pp. 451–456).
costs. Optimal Control Applications & Methods, 32, 127–138. Sun, J., Fang, W., Wu, X. J., Lai, C. H., & Xu, W. B. (2011). Solving the multi-stage
Ponsich, A., Jaimes, A. L., & Coello, C. A. C. (2013). A survey on multiobjective portfolio optimization problem with a novel particle swarm optimization. Expert
evolutionary algorithms for the solution of the portfolio optimization problem Systems with Applications, 38, 6727–6735.
and other finance and economics applications. IEEE Transactions on Evolutionary Suthiwong, D., & Sodanil, M. (2016). Cardinality-constrained Portfolio optimization
Computation, 17, 321–344. using an improved quick Artificial Bee Colony Algorithm. In Computer Science
Pouya, A. R., Solimanpur, M., & Rezaee, M. J. (2016). Solving multi-objective portfolio and Engineering Conference (ICSEC), 2016 International (pp. 1–4). IEEE.
optimization problem using invasive weed optimization. Swarm and Evolution- Talebi, A., Molaei, M. A., & Sheikh, M. J. (2010). Performance investigation and com-
ary Computation, 28, 42–57. parison of two evolutionary algorithms in portfolio optimization: Genetic and
Qin, Z. F. (2015). Mean-variance model for portfolio optimization problem in the si- particle swarm optimization. In Proceedings - 2010 2nd IEEE International Confer-
multaneous presence of random and uncertain returns. European Journal of Op- ence on Information and Financial Engineering, ICIFE 2010 (pp. 430–437).
erational Research, 245, 480–488. Tan, L., Niu, B., Lin, F., Duan, Q., & Li, L. (2013). Modified bacterial foraging opti-
Qu, B. Y., Zhou, Q., Xiao, J. M., Liang, J. J., & Suganthan, P. N. (2017). Large-scale mization for constrained portfolio optimization. Information Technology Journal,
portfolio optimization using multiobjective evolutionary algorithms and prese- 12, 7918–7921.
lection methods. Mathematical Problems in Engineering, 2017, 1–14. Tan, L., Niu, B., Wang, H., Huang, H., & Duan, Q. (2014). In Bacterial foraging
Reid, S. G., & Malan, K. M. (2015). Constraint handling methods for portfolio opti- optimization with neighborhood learning for dynamic portfolio selection: 8590
mization using particle swarm optimization. In Proceedings - 2015 IEEE Sympo- (pp. 413–423). Cham: Springer International Publishing.
sium Series on Computational Intelligence, SSCI 2015 (pp. 1766–1773). IEEE. Tan, Y., & Zhu, Y. (2010). Fireworks algorithm for optimization. In Y. Tan, Y. Shi, &
Rockafellar, R. T., & Uryasev, S. (20 0 0). Optimization of conditional value-at-risk. K. C. Tan (Eds.), Advances in Swarm Intelligence: First International Conference, ICSI
Journal of risk, 2, 21–42. 2010, Beijing, China, June 12-15, 2010, Proceedings, Part I (pp. 355–364). Springer
Rong, X., Lu, M., & Deng, L. (2009). Multi-period model of portfolio investment and Berlin Heidelberg.
adjustment based on hybrid genetic algorithm. Transactions of Tianjin University, Tang, J., Wang, K., & Zhao, X. (2009). A novel constraint handling technique for
15, 415–422. complex portfolio selection. In 6th International Conference on Fuzzy Systems and
Rubinstein, M. (2002). Markowitz’s “Portfolio Selection”: A Fifty-Year Retrospective. Knowledge Discovery, FSKD 2009: 2 (pp. 156–160). IEEE.
The Journal of Finance, 57, 1041–1045. Thomaidis, N. S., et al. (2010). Active portfolio management from a fuzzy multi-ob-
Ruiz-Torrubiano, R., & Suarez, A. (2010). Hybrid approaches and dimensionality re- jective programming perspective. In C. Chio, A. Brabazon, G. A. Caro, M. Ebner,
duction for portfolio selection with cardinality constraints. IEEE Computational M. Farooq, & A. Fink, et al. (Eds.), Applications of Evolutionary Computation:
Intelligence Magazine, 5, 92–107. EvoApplications 2010: EvoCOMNET, EvoENVIRONMENT, EvoFIN, EvoMUSART, and
Ruiz-Torrubiano, R., & Suarez, A. (2015). A memetic algorithm for cardinality-con- EvoTRANSLOG, Istanbul, Turkey, April 7-9, 2010, Proceedings, Part II (pp. 222–231).
strained portfolio optimization with transaction costs. Applied Soft Computing, Springer Berlin Heidelberg.
36, 125–142. Tian, Y., Fang, S., Deng, Z. B., & Jin, Q. W. (2016). Cardinality constrained portfo-
Ruiz-Torrubiano, R., & Suárez, A. (2007). Use of heuristic rules in evolutionary meth- lio selection problem: a completely positive programming approach. Journal of
ods for the selection of optimal investment portfolios. In 2007 IEEE Congress on Industrial and Management Optimization, 12, 1041–1056.
Evolutionary Computation, CEC 2007 (pp. 212–219). IEEE. Tuba, M., & Bacanin, N. (2014a). Artificial bee colony algorithm hybridized with fire-
Sabar, N. R., & Song, A. (2014). Dual population genetic algorithm for the cardinality fly algorithm for cardinality constrained mean-variance portfolio selection prob-
constrained portfolio selection problem (pp. 703–712). Cham: Springer Interna- lem. Applied Mathematics & Information Sciences, 8, 2831–2844.
tional Publishing. Tuba, M., & Bacanin, N. (2014b). Upgraded firefly algorithm for portfolio optimiza-
Sadigh, A. N., Mokhtari, H., Iranpoor, M., & Ghomi, S. M. T. F. (2012). Cardinality tion problem. In Proceedings - UKSim-AMSS 16th International Conference on
constrained portfolio optimization using a hybrid approach based on particle Computer Modelling and Simulation, UKSim 2014 (pp. 113–118). IEEE.
swarm optimization and hopfield neural network. Advanced Science Letters, 17, Tuba, M., Bacanin, N., & Pelevic, B. (2013). Framework for constrained portfolio se-
11–20. lection by the firefly algorithm. International Journal of Mathematical Models and
Sadjadi, S. J., Gharakhani, M., & Safari, E. (2012). Robust optimization framework for Methods in Applied Sciences, 7, 888–896.
cardinality constrained portfolio problem. Applied Soft Computing, 12, 91–99. Tuba, M., Bacanin, N., & Pelevic, B. (2014). Krill Herd (KH) algorithm applied to the
Schaerf, A. (2002). Local search techniques for constrained portfolio selection prob- constrained portfolio selection problem. International Journal of Mathematics and
lems. Computational Economics, 20, 177–190. Computers in Simulation, 8, 94–102.
Schaffer, J. D. (1985). Multiple objective optimization with vector evaluated genetic Van Veldhuizen, D. A., & Lamont, G. B. (20 0 0). Multiobjective evolutionary al-
algorithms. In Proceedings of the 1st international Conference on Genetic Algo- gorithms: Analyzing the state-of-the-art. Evolutionary computation, 8, 125–
rithms (pp. 93–100). L. Erlbaum Associates Inc. 147.
Sen, T., Saha, S., Ekbal, A., & Laha, A. K. (2015). Bi-objective portfolio optimization Vijayalakshmi Pai, G. A., & Michel, T. (2013). Fuzzy decision theory based optimiza-
using archive multi-objective simulated annealing. 2014 International Conference tion of constrained portfolios using metaheuristics. IEEE International Conference
on High Performance Computing and Applications, ICHPCA 2014. IEEE. on Fuzzy Systems. IEEE.
Shaikh, R. A., & Abbas, A. (2009). Genetic algorithm and MS solver for portfolio op- Wang, J.-b., Chen, W.-N., Zhang, J., & Lin, Y. (2015). A dimension-decreasing parti-
timization under exogenous influence. In 2009 International Conference on Com- cle swarm optimization method for portfolio optimization. In Proceedings of the
puter and Electrical Engineering, ICCEE 2009: 1 (pp. 555–558). Companion Publication of the 2015 on Genetic and Evolutionary Computation Con-
Sharpe, W. F. (1966). Mutual fund performance. The Journal of Business, 39, 119– ference - GECCO Companion ’15 (pp. 1515–1516). ACM.
138. Wang, W., Wang, H., Wu, Z., & Dai, H. (2009). A simple and fast particle swarm
Shaw, D. X., Liu, S. C., & Kopman, L. (2008). Lagrangian relaxation procedure for optimization and its application on portfolio selection. In Intelligent Systems and
cardinality-constrained portfolio optimization. Optimization Methods & Software, Applications, 2009. ISA 2009. International Workshop on (pp. 1–4). IEEE.
23, 411–420. Wang, Z., Liu, S., & Kong, X. (2011). An Improved artificial bee colony algorithm for
Shoaf, J., & Foster, J. A. (1998). Efficient set GA for stock portfolios. In Proceedings portfolio optimization problem. International Journal of Advancements in Com-
of the IEEE Conference on Evolutionary Computation, ICEC (pp. 354–359). IEEE. puting Technology, 3, 67–74.
Skolpadungket, P., Dahal, K., & Harnpornchai, N. (2007). Portfolio optimization using Wang, Z., Liu, S., & Kong, X. (2012). Artificial bee colony algorithm for portfolio opti-
multi-objective genetic algorithms. In 2007 IEEE Congress on Evolutionary Com- mization problems. International Journal of Advancements in Computing Technol-
putation, CEC 2007 (pp. 516–523). IEEE. ogy, 4, 8–16.
Soleimani, H., Golmakani, H. R., & Salimi, M. H. (2009). Markowitz-based portfolio Woodside-Oriakhi, M., Lucas, C., & Beasley, J. E. (2011). Heuristic algorithms for
selection with minimum transaction lots, cardinality constraints and regarding the cardinality constrained efficient frontier. European Journal of Operational Re-
sector capitalization using genetic algorithm. Expert Systems with Applications, search, 213, 538–550.
36, 5058–5063. Xia, Y. S., Liu, B. D., Wang, S. Y., & Lai, K. K. (20 0 0). A model for portfolio selec-
Storn, R., & Price, K. (1997). Differential evolution–a simple and efficient heuristic tion with order of expected returns. Computers & Operations Research, 27, 409–
for global optimization over continuous spaces. Journal of Global Optimization, 422.
11, 341–359. Xidonas, P., & Mavrotas, G. (2014). Multiobjective portfolio optimization with non–
Streichen, F., & Tanaka-Yamawaki, M. (2006). The effect of local search on the con- convex policy constraints: Evidence from the Eurostoxx 50. European Journal of
strained portfolio selection problem. In 2006 IEEE Congress on Evolutionary Com- Finance, 20, 957–977.
putation, CEC 2006 (pp. 2368–2374). IEEE. Xu, F., & Chen, W. (2006). Stochastic portfolio selection based on velocity limited
Streichert, F., Ulmer, H., & Zell, A. (2004a). Evaluating a hybrid encoding and three particle swarm optimization. In Intelligent Control and Automation, 2006. WCICA
crossover operators on the constrained portfolio selection problem. In Evolution- 2006. The Sixth World Congress on: 1 (pp. 3599–3603). IEEE.
ary Computation, 2004. CEC2004. Congress on: 1 (pp. 932–939). Vol. 931.
368 C.B. Kalayci, O. Ertenlice and M.A. Akbay / Expert Systems With Applications 125 (2019) 345–368

Xu, F., Chen, W., & Yang, L. (2007). Improved particle swarm optimization for re- Zhang, H., Zhao, Y., Wang, F., Zhang, A., Yang, P., & Shen, X. (2018a). A new
alistic portfolio selection. In Proceedings - SNPD 2007: Eighth ACIS International evolutionary algorithm based on MOEA/D for portfolio optimization. In Ad-
Conference on Software Engineering, Artificial Intelligence, Networking, and Paral- vanced Computational Intelligence (ICACI), 2018 Tenth International Conference on
lel/Distributed Computing: 1 (pp. 185–190). IEEE. (pp. 831–836). IEEE.
Xu, J., Lam, A. Y. S., & Li, V. O. K. (2011). Stock portfolio selection using chemical Zhang, X. L., Zhang, W. G., Xu, W. J., & Xiao, W. L. (2010). Possibilistic approaches
reaction optimization. World Academy of Science, Engineering and Technology, 5, to portfolio selection problem with general transaction costs and a CLPSO Algo-
458–463. rithm. Computational Economics, 36, 191–200.
Xu, R. T., Zhang, J., Liu, O., & Huang, R. Z. (2010). An estimation of distribu- Zhang, Y., Li, X., & Guo, S. (2018b). Portfolio selection problems with Markowitz’s
tion algorithm based portfolio selection approach. In Proceedings - International mean–variance framework: A review of literature. Fuzzy Optimization and Deci-
Conference on Technologies and Applications of Artificial Intelligence, TAAI 2010 sion Making, 17, 125–158.
(pp. 305–313). IEEE. Zhou, J., & Li, J. (2014). An improved multi-objective particle swarm optimization for
Yaakob, S. B., & Watada, J. (2010). A hybrid particle swarm optimization approach constrained portfolio selection model. 11th International Conference on Service
to mixed integer quadratic programming for portfolio selection problems. Inter- Systems and Service Management, ICSSSM 2014 - Proceeding. IEEE.
national Journal of Simulation: Systems, Science and Technology, 11, 68–74. Zhu, H., Chen, Y., & Wang, K. (2010). Swarm intelligence algorithms for portfolio op-
Yang, X.-S. (2010a). A new metaheuristic bat-inspired algorithm. In J. R. González, timization. In Y. Tan, Y. Shi, & K. C. Tan (Eds.), Advances in Swarm Intelligence:
D. A. Pelta, C. Cruz, G. Terrazas, & N. Krasnogor (Eds.). In Nature inspired cooper- First International Conference, ICSI 2010, Beijing, China, June 12-15, 2010, Proceed-
ative strategies for optimization (NICSO 2010): 284 (pp. 65–74). BerlinHeidelberg: ings, Part I: 6145 (pp. 306–313). Springer Berlin Heidelberg.
Springer Berlin Heidelberg. Zhu, H. H., Wang, Y., Wang, K. S., & Chen, Y. (2011). Particle Swarm Optimization
Yang, X. S. (2010b). Engineering optimization: an introduction with metaheuristic ap- (PSO) for the constrained portfolio optimization problem. Expert Systems with
plications. John Wiley & Sons. Applications, 38, 10161–10169.
Yin, X., Ni, Q., & Zhai, Y. (2015a). A novel particle swarm optimization for portfolio op- Zitzler, E. (1999). Evolutionary algorithms for multiobjective optimization: methods and
timization based on random population topology strategies (pp. 164–175). Cham: applications: 63. Citeseer.
Springer International Publishing. Zitzler, E., & Künzli, S. (2004). Indicator-based selection in multiobjective search.
Yin, X., Ni, Q., & Zhai, Y. (2015b). A novel PSO for portfolio optimization based on In International Conference on Parallel Problem Solving from Nature: 3242
heterogeneous multiple population strategy. In 2015 IEEE Congress on Evolution- (pp. 832–842). Springer.
ary Computation, CEC 2015 - Proceedings (pp. 1196–1203). IEEE. Zitzler, E., Laumanns, M., Thiele, L., Zitzler, E., Zitzler, E., Thiele, L., et al. (2001).
Yoshimoto, A. (1996). The mean-variance approach to portfolio optimization sub- SPEA2: Improving the strength Pareto evolutionary algorithm. Eidgenössische
ject to transaction costs. Journal of the Operations Research Society of Japan, 39, technische hochschule zürich (ETH). Institut für Technische Informatik und Kom-
99–117. munikationsnetze (TIK).
Young, M. R. (1998). A minimax portfolio selection rule with linear programming Zitzler, E., & Thiele, L. (1999). Multiobjective evolutionary algorithms: A comparative
solution. Management Science, 44, 673–683. case study and the Strength Pareto approach. IEEE Transactions on Evolutionary
Yu, L., Wang, S. Y., & Lai, K. K. (2009). Multi-attribute portfolio selection with genetic Computation, 3, 257–271.
optimization algorithms. INFOR, 47, 23–30. Zitzler, E., Thiele, L., Laumanns, M., Fonseca, C. M., & Da Fonseca, V. G. (2003). Per-
Zaheer, H., & Pant, M. (2016). Solving portfolio optimization problem through dif- formance assessment of multiobjective optimizers: An analysis and review. IEEE
ferential evolution. In International Conference on Electrical, Electronics, and Opti- Transactions on Evolutionary Computation, 7, 117–132.
mization Techniques, ICEEOT 2016 (pp. 3982–3987). IEEE.