Accepted Manuscript

Multiobjective credibilistic portfolio selection model with fuzzy chance-con‐

straints

Pankaj Gupta, Masahiro Inuiguchi, Mukesh Kumar Mehlawat, Garima Mittal

PII: S0020-0255(12)00794-3
DOI: http://dx.doi.org/10.1016/j.ins.2012.12.011
Reference: INS 9857

To appear in: Information Sciences

Received Date: 24 November 2011

Revised Date: 8 October 2012
Accepted Date: 9 December 2012

Please cite this article as: P. Gupta, M. Inuiguchi, M.K. Mehlawat, G. Mittal, Multiobjective credibilistic portfolio
selection model with fuzzy chance-constraints, Information Sciences (2012), doi: http://dx.doi.org/10.1016/j.ins.
2012.12.011

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
 Multiobjective credibilistic portfolio selection model with fuzzy
chance-constraints
Pankaj Gupta , Masahiro Inuiguchi b , Mukesh Kumar Mehlawat c
a, ∗

and Garima Mittal a

a
Department of Operational Research, University of Delhi, Delhi, India
b
Department of Systems Innovation, Graduate School of Engineering Science,
Osaka University, Japan
c
Department of Decision Sciences, Apeejay School of Management, Dwarka,
New Delhi, India
Abstract
In this paper, we propose a multiobjective credibilistic model with fuzzy
chance constraints of the portfolio selection problem. The key financial crite-
ria used are short-term return, long-term return, risk and liquidity. The model
generates portfolios which are optimal to the extent of achieving the highest cred-
ibility values for the objective functions. The problem is solved using a hybrid
intelligent algorithm that integrates fuzzy simulation with a real-coded genetic
algorithm. The approach adopted here has advantage of handling the multi-
objective portfolio selection problem where fuzzy parameters are characterized
by general functional forms. Numerical examples are provided to demonstrate
effectiveness of the solution approach and efficiency of the model.
Keywords: Fuzzy portfolio selection; Credibility measure; Chance-constrained
programming; Multiobjective programming; Real-coded genetic algorithm
1. Introduction
Portfolio selection has attracted much attention since the introduction of the
mean-variance model [42], in which return is quantified using the mean and risk
using the variance. In most subsequent studies, return and risk are used as
the two fundamental criteria that govern investors’ choices. However, it is often
found that not all the relevant information for portfolio selection can be captured
by these two criteria alone. Other criteria might be of equal, if not greater,
importance to the investor. By considering other criteria in a portfolio selection
model, it may be possible to obtain portfolios in which a less favorable return or
risk is compensated by the portfolio’s performance quantified on other criteria,
which can result in greater overall satisfaction for the investor. For these reasons,
multiple-criteria portfolio selection models [1, 3, 5, 12, 13, 14, 15, 16, 17, 31, 44]
have been the subject of great interest from researchers in the recent past.
Traditionally, portfolio selection models also make the restrictive assumption
that the investor has all of the information that is necessary for decision making.
Corresponding Author. Tel/FAX: 91-11-27666672
∗

Address for correspondence: Flat No.-01, Kamayani Kunj, Plot No. 69, Indraprastha Extension,
Delhi - 110092, India, E-mail : pgupta@or.du.ac.in
The first author acknowledges the research grant received under a scheme for strengthening
R & D Doctoral Research Programme of University of Delhi, Delhi, India.
However, the information available in financial markets is often incomplete, and
thus, decisions are made under uncertainty. Additionally, markets are affected by
vagueness and ambiguity caused by the use of expressions such as ‘high risk’, ‘low
profit’ and ‘low liquidity’ by the investors and the investment experts. Conse-
quently, fuzzy set theory [51, 52] has been used in portfolio selection models to in-
tegrate quantitative and qualitative information about the subjective preferences
of investors and expert knowledge. There exists a large body of literature on port-
folio selection using possibility measure [3, 4, 9, 18, 28, 29, 30, 40, 41, 45, 46, 48]
in which the returns are assumed to be fuzzy. Though the possibility measure
is widely used in the literature, it has certain limitations. For example, a fuzzy
event with maximum possibility value 1 may still not occur. Additionally, it is
possible that two fuzzy events with different chances of occurrence may have the
same possibility value; thus, the possibility value is of little information to the
investor. The primary reason for these limitations is that the possibility measure
is not self-dual, which is an important property both in theory and practice. To
alleviate this limitation, Liu and Liu [39] proposed a credibility measure that is
self-dual and thus overcomes the limitations inherent in the possibility measure.
Many authors have developed fuzzy portfolio selection models using credibility
measures [11, 21, 22, 23, 24, 33, 34, 35, 36, 49, 50].
In this paper, we propose a new multicriteria credibilistic portfolio selection
model. The fuzzy portfolio selection model developed here is a credibility max-
imization model in which we desire to maximize the credibility such that the
short-term return, long-term return and liquidity of the portfolio are not less
than some given target levels and that the portfolio meets credibility-based fuzzy
chance constraints which represent portfolio risk. For the portfolio return, we
consider the short-term return (the average performance of the asset during a
12-month period) and the long-term return (the average performance of the asset
during a 36-month period). We use both measures of return to capture the sub-
jective preferences of the investor regarding portfolio return. We also consider
several realistic constraints, namely, a capital budget constraint, a cardinality
constraint and diversification constraints that apply lower and upper bounds for
investments in individual assets. To capture the uncertain behavior of financial
markets more realistically, we assume the fuzzy parameters used in the model
are characterized by general functional forms. Under this assumption, the con-
version of the fuzzy portfolio selection model into the crisp equivalent becomes
computationally difficult. Therefore, we develop a hybrid intelligent algorithm
that integrates fuzzy simulation with a real-coded genetic algorithm to solve the
fuzzy multiobjective portfolio selection problem.
The paper is organized as follows: in Section 2, we present some basic def-
initions and notation. In Section 3, we present the multiobjective credibilistic
portfolio selection model with fuzzy parameters and also present details of the
hybrid intelligent algorithm. The proposed model is tested in Section 4 using
a 36-month data series regarding assets listed on the National Stock Exchange
(NSE) of Mumbai, India. This section also includes a discussion of the results

2
obtained. We conclude the paper in Section 5.
2. Preliminaries
2.1. Credibility fundamentals
Credibility theory [32, 38, 39] is a branch of mathematics for studying the be-
havior of fuzzy events using a credibility measure. Credibility theory is based on
five axioms from which a credibility measure is defined. Let Θ be a nonempty
set and let P (Θ) be the power set of Θ (i.e., the collection of all subsets of Θ ).
Each element in P (Θ) is called an event. To present an axiomatic definition of
credibility, it is necessary to assign to each event A , a number Cr{A} , which
indicates the credibility that A will occur. Furthermore, to ensure that the num-
ber Cr{A} has certain mathematical properties, the following five axioms should
hold:
1. Cr{Θ} = 1 .
2. Cr is increasing, i.e., Cr{A} ≤ Cr{B} whenever A ⊂ B .
3. Cr is self-dual, i.e., Cr{A} + Cr{Ac } = 1 for any A ∈ P (Θ) .
4. Cr{∪i Ai } ∧ 0.5 = sup Cr{Ai } for any Ai with Cr{Ai } ≤ 0.5 .
i
5. Let Θk be nonempty sets for which Crk satisfies the first four axioms for
k = 1, 2, . . . , n and let Θ = Θ1 ×Θ2 ×. . .×Θn . Then, Cr{(θ1 , θ2 , . . . , θn )} =
Cr1 {θ1 } ∧ Cr2 {θ2 } ∧ ... ∧ Crn {θn } for each
′ ′
(θ1 , θ2 , . . . , θn ) ∈ Θ , where ∧ represents the minimal operator.
It may be noted that Cr{∅} = 0 and that the credibility measure takes values
between 0 and 1. Traditionally, a fuzzy variable is defined by a membership
function [51]. Liu [38] defined it as a function on a credibility space in a manner
analogous to that by which a random variable is defined as a measurable function
on a probability space.
Definition 1 [38]. A fuzzy variable is a function from a credibility space
(Θ, P (Θ), Cr) to the set of real numbers.
An n -dimensional fuzzy vector is defined as a function from a credibility
space to the set of n -dimensional real vectors. Moreover, (ξ1 , ξ2, . . . , ξn ) is a
fuzzy vector if and only if ξ1 , ξ2, . . . , ξn are fuzzy variables.
Definition 2 [38]. Let f : ℜn → ℜ and let ξ1 , ξ2 , . . . , ξn be fuzzy variables
on the credibility space (Θ, P (Θ), Cr) . Then, ξ = f (ξ1, ξ2 , . . . , ξn ) is a fuzzy
variable defined as ξ(θ) = f (ξ1 (θ), ξ2(θ), . . . , ξn (θ)) for any θ ∈ Θ .
If the fuzzy variables are defined on different credibility spaces, then
ξ = f (ξ1, ξ2 , . . . , ξn ) is a fuzzy variable defined on the product credibility space
(Θ, P (Θ), Cr) as ξ(θ1 , θ2 , . . . , θn ) = f (ξ1(θ1 ), ξ2 (θ2 ), . . . , ξn (θn )) for any
(θ1 , θ2 , . . . , θn ) ∈ Θ .
Definition 3 [38]. Let ξ be a fuzzy variable with membership function µ and
u and r be real numbers. The credibility of the fuzzy event characterized by
ξ ≥ r , is defined as

3
 1
Cr{ξ ≥ r} = (sup µ(u) + 1 − sup µ(u)).
2 u≥r u<r
It may be noted that the credibility of a fuzzy event can also be defined as
an average of possibility and necessity, i.e.,
Cr{ξ ≥ r} = 21 (P os{ξ ≥ r} + Nes{ξ ≥ r}),
where P os{ξ ≥ r} = (sup µ(u)) and Nes{ξ ≥ r} = 1 − sup µ(u) . The use of
u≥r u<r
such a credibility measure can also be justified from the perspective of relative
possibility, as was investigated earlier by Inuiguchi and Ichihashi [25].
Definition 4 [38]. Let ξi be fuzzy variables with membership functions µi ,
and let ui be real numbers, where i = 1, 2, . . . , n . Define f : ℜn → ℜ . The
credibility of the fuzzy event characterized by f (ξ1 , ξ2 , . . . , ξn ) ≥ 0 is given as
1
Cr{f (ξ1 , ξ2, . . . , ξn ) ≥ 0} = ( sup { min µξ (ui )|f (u1, u2 , . . . , un ) ≥ 0}
2 u1 ,u2 ,...,un∈R 1≤i≤n i
+1 − sup { min µξi (ui )|f (u1, u2, . . . , un ) < 0}).
u1 ,u2 ,...,un ∈R 1≤i≤n

2.2. Risk curve and confidence curve

In portfolio selection, investors are generally concerned about each likely loss of
the portfolio return and the corresponding chance of its occurrence rather than
a single pre-set level of the loss. In such situations, the portfolio risk, instead of
being set at a specific threshold level, is represented using a curve called the risk
curve.
Definition 5 [22]. Let ξ denote the fuzzy return of the portfolio and b the
target return. The risk curve is defined as
f (r) = Cr{(b − ξ) ≥ r}, ∀r ≥ 0.
Here, (b − ξ) denotes the potential loss of the portfolio return and r is the
severity level of the loss. Larger values of the indicator r correspond to greater
losses (b−ξ) . The risk curve f (r) gives the credibility measure corresponding to
all possible losses of the portfolio return. It may be noted that when the investor
is only concerned with one fixed loss amount, i.e., (b − ξ) ≥ r0 , where r0 is the
fixed severity level of the loss, the chance of occurrence of such a risk becomes
Cr{(b − ξ) ≥ r0 } .
Given any severity loss indicator r , the investor can express his maximal
tolerance for the chance of occurrence of potential losses equal to or greater than
r , using a curve called the confidence curve.
Definition 6 [22]. The confidence curve α(r) is the curve that gives the credi-
bility levels for all possible losses for any given r that would be tolerated by the
investor.
Usually, when r is low, the investor can tolerate a relatively high occurrence
credibility for losses equal to or greater than r . However, when r is high, the
investor can only tolerate a low occurrence credibility for the loss. Thus, the area
below the confidence curve α(r) is regarded as the low-risk area and the area
above the confidence curve α(r) as the high-risk area, see Fig. 1 [22]. It may be

4
noted that different investors may have different confidence curves; however, the
common trend for any confidence curve is that severer losses are assigned lower
tolerances for the chance of occurrence of the loss.
1
.0

y
Highrisk area
it
bil
di
e
Cr
α(r)

Lowrisk area

00 S
.
everity of risk ′࢘ ′

Figure 1. The confidence curve and investment risk

Let ξ be the fuzzy return of the portfolio and α(r) be the investor’s confi-
dence curve. We say the portfolio is safe if
Cr{(b − ξ) ≥ r} ≤ α(r), ∀r ≥ 0,
i.e., the risk curve of the portfolio is completely below the investor’s confidence
curve. The above expression of risk represents a fuzzy chance constraint for
every randomly generated pair ( r, α(r) ). In chance-constrained programming
[7], one requires that the objectives must be achieved with the stochastic con-
straints having a finite probability of violation. Following the idea of stochastic
chance-constrained programming, Liu [37] developed fuzzy chance-constrained
programming. It may be noted that the introduction of chance-constraint con-
cepts to fuzzy programming was proposed earlier in Inuiguchi et al. [26, 27] but
was referred to as modality-constrained programming.
3. Research methodology
3.1. Portfolio selection model
In this section, we formulate the multiobjective credibilistic model of portfolio
selection problem. We assume that investors allocate their wealth among n assets
that offer fuzzy returns. The parameters and variables used to formulate the
mathematical model are the following:
λi : the fuzzy short-term return of the i -th asset ,
γi : the fuzzy long-term return of the i -th asset ,
βi : the fuzzy turnover rate of the i -th asset ,
r1 : the lower limit on the expected short-term return of the portfolio ,
r2 : the lower limit on the expected long-term return of the portfolio ,
r3 : the lower limit on the expected liquidity of the portfolio ,
b : the target return of the portfolio ,
r : the severity indicator for a possible loss ,
α(r) : the value of the investor’s confidence curve that corresponds to a given r ,
ui : the maximal fraction of the capital budget allocated to the i -th asset ,
li : the minimal fraction of the capital budget allocated to the i -th asset ,

5
 xi : the proportion of the total funds invested in the i -th asset , and
yi : a binary variable indicating whether the i -th asset is contained in the port-
folio, where (
1, if the i-th asset is contained in the portfolio,
yi =
0, otherwise.

3.1.1 Objectives
• Short-term return
The form that maximizes the credibility that the short-term return of the portfolio
is not less than r1 is expressed as
Max f1 (x) = Cr{λ1x1 + λ2 x2 + . . . + λn xn ≥ r1 }.
• Long-term return
The form that maximizes the credibility that the long-term return of the portfolio
is not less than r2 is expressed as
Max f2 (x) = Cr{γ1x1 + γ2 x2 + . . . + γn xn ≥ r2 }.
• Liquidity
For any asset, liquidity may be measured using the turnover rate as characterized
by the ratio of the average trading volume of the stocks traded in the market
and the trading volume of the tradable stock (i.e., shares held by the public)
corresponding to the asset. Because of incomplete information, the turnover rates
are only vague estimates. To treat uncertainty, liquidity has been considered as
a fuzzy variable [12, 14, 15, 16]. The form that maximizes the credibility that
the portfolio liquidity is not less than r3 is expressed as
Max f3 (x) = Cr{β1x1 + β2 x2 + . . . + βn xn ≥ r3 }.

3.1.2 Constraints
• Risk
Portfolio risk is represented by credibility-based fuzzy chance constraints. These
constraints are defined using the risk curve of the portfolio constructed using
simulated long-term returns and the investor’s confidence curve as follows:
Cr{b − (γ1 x1 + γ2 x2 + . . . + γn xn ) ≥ r} ≤ α(r), ∀r ≥ 0.
• Capital budget constraint
X n
xi = 1.
i=1
• Maximal fraction of the capital that can be invested in a single asset
xi ≤ ui yi i = 1, 2, . . . , n.
• Minimal fraction of the capital that can be invested in a single asset
xi ≥ li yi i = 1, 2, . . . , n.
The maximal and minimal fractions of the capital budget allocated to the assets
in the portfolio depend upon a number of factors. For example, one may consider
the price or value of the asset relative to the average price or value of all the assets

6
in the chosen portfolio, the minimal lot size that can be traded in the market, the
past behavior of the price or volume of the asset, the information available about
the issuer of the asset and trends in the industry of which it is a part. Thus, the
investor considers many fundamental and technical analysis factors that affect
the company and the industry. Because investors differ in their interpretation of
the available information, they may allocate the same capital budget differently.
The constraints corresponding to lower bounds li and upper bounds ui on the
investment in individual assets (0 ≤ li , ui ≤ 1, li ≤ ui , ∀i) are included to avoid
a large number of very small investments (this is ensured by the lower bounds)
and to ensure sufficient diversification of the investments (this is ensured by the
upper bounds).
• Number of assets held in the portfolio (the cardinality constraint)
Xn
yi = h.
i=1
Here, h is the number of assets that the investor includes in the portfolio. Of all
the available assets, the investor would select those that are most likely to satisfy
his preferences. Investors may also differ in their opinion about the number of
assets they can effectively manage in the portfolio.
• No short selling of assets
xi ≥ 0, i = 1, 2, . . . , n.

3.1.3 The decision problem

The multiobjective credibilistic portfolio selection model with fuzzy chance con-
straints is formulated as

(P1) Max f1 (x) = Cr{λ1x1 + λ2 x2 + . . . + λn xn ≥ r1 }

Max f2 (x) = Cr{γ1x1 + γ2 x2 + . . . + γn xn ≥ r2 }
Max f3 (x) = Cr{β1 x1 + β2 x2 + . . . + βn xn ≥ r3 }
subject to
Cr{b − (γ1 x1 + γ2 x2 + . . . + γn xn ) ≥ r} ≤ α(r), ∀r ≥ 0 , (1)
Xn
xi = 1 , (2)
i=1
Xn
yi = h , (3)
i=1

xi ≤ ui yi , i = 1, 2, . . . , n , (4)
xi ≥ li yi , i = 1, 2, . . . , n , (5)
xi ≥ 0 , i = 1, 2, . . . , n , (6)
yi ∈ {0, 1} , i = 1, 2, . . . , n . (7)

7
It may be noted that when solving the model, the risk constraint (1) is replaced
by fuzzy chance constraints corresponding to randomly generated pairs ( r, α(r) ).
3.2. Hybrid intelligent algorithm
The credibility-based fuzzy multiobjective model (P1) of the portfolio selection
problem involves fuzzy parameters. Several researchers have used heuristic and
meta-heuristic methods [2, 6, 10, 19, 47] to solve fuzzy portfolio selection prob-
lem. When the membership functions of fuzzy parameters are characterized by
general functional forms, the conversion of a fuzzy model into the crisp equivalent
becomes computationally difficult. Therefore, we develop a hybrid intelligent al-
gorithm (HIA) as a solution method for multiobjective portfolio selection problem
involving general fuzzy functional forms.

3.2.1 Fuzzy simulation

The technique of fuzzy simulation was discussed in detail by Liu [37]. In the
proposed HIA, fuzzy simulation is applied to compute the credibility values for
the uncertain objective functions.
Assuming that x is a decision vector and ξ is a fuzzy vector, let G(x, ξ)
represents a fuzzy quantity. Let ξ = (ξ1 , ξ2, . . . , ξn ) and x = (x1 , x2 , . . . , xn ) .
Moreover, let µ = (µ1 , µ2 , . . . , µn ) denote the membership function vector of
ξ . To solve the decision model (P 1) , we must handle the following uncertain
function:
U : x → Cr{G(x, ξ) ≥ r},
where r ≥ 0 is given by the investor. It may be noted that for an expression of
the type Cr{(b − (ξ1 x1 + ξ2 x2 + . . . + ξn xn )) ≥ r} ,
G(x, ξ) = b − (ξ1 x1 + ξ2 x2 + . . . + ξn xn ).
The following fuzzy simulation algorithm is used in this paper:
Step 1 Let j = 1 .
Step 2 Randomly generate uij for the j− th simulation run using the ε− level
sets of fuzzy variables ξi such that µij (uij ) ≥ ε, where i = 1, 2, . . . , n, j =
1, 2, . . . , N , where ε is a sufficiently small positive number and N is a
sufficiently large number.
Step 3 Set uj = (u1j , u2j , . . . , unj ) and µ(uj ) = µ1j (u1j ) ∧ µ2j (u2j ) ∧ . . . ∧
µnj (unj ) .
Step 4 j ← j + 1 . Go to Step 2 if j ≤ N ; otherwise, go to Step 5.
Step 5 Return the credibility Cr{G(x, ξ) ≥ r} which is calculated as follows:
1
Cr{G(x, ξ) ≥ r} = ( max {µ(uj )|G(x, uj ) ≥ r}
2 1≤j≤N
+1 − max {µ(uj )|G(x, uj ) < r}).
1≤j≤N

8
3.2.2 Real-coded genetic algorithm
Genetic algorithms (GA) were proposed by Holland [20] and have since been
well developed and documented in the literature. Here, we use RCGA in the
hybrid intelligent algorithm. The basic difference between the conventional GA
and RCGA is how the encoding of chromosomes is performed. Coding of the
variables is essential for an efficient GA. The RCGA, which uses real numbers for
encoding, convergences more quickly toward optima than GA. It also overcomes
the difficulty of the Hamming cliff, which may arise in the conventional GA when
the Hamming distance (which is defined in binary coding in conventional GA)
between two adjacent integers (in decimal code) is very large. In such cases, a
large number of bits must be altered to change an integer to the adjacent one,
which causes a reduction in the efficiency of conventional GA. An overview of
conventional GA is given below.
To solve a problem with GA, an encoding mechanism must first be designed to
represent each solution as a chromosome. A fitness function is then defined to
measure the goodness of a chromosome. The GA searches the solution space
using a population, which is a set of chromosomes at each generation. During
each generation, the three genetic operators selection, crossover and mutation are
applied to the population several times to form a new population. The selection
operation forms a parent population that is used for creating the next genera-
tion. Given a crossover rate, the crossover operation recombines the two selected
chromosomes to form offspring. Given a mutation rate, the mutation operation
randomly alters selected positions in a selected chromosome. The new population
is then generated by replacing some chromosomes in the parent population with
the offspring. This process is repeated until some termination condition, e.g., a
maximum number of generations, is reached.
We now present the details of RCGA used in this paper.
• Chromosome encoding
In the proposed encoding method, the length of the chromosome is set to n ,
the number of available assets. Let the solution vector x = (x1 , x2 , . . . , xn ) be
represented by the chromosome Chk , which is encoded as an array as follows:
Chk = Xk [i] = xi , i = 1, 2, . . . , n, k = 1, 2, . . . , popsize.
Here, popsize defines the number of chromosomes initialized to constitute the
population of one generation. In this encoding method, the position of the gene,
xi for i = 1, 2, . . . , n , is used to represent the ID number of the asset and
the gene’s value is used to represent the corresponding proportion of the capital
budget that is allocated to the i -th asset. The initialization algorithm to create
first generation of feasible chromosomes of size popsize is as follows:
Step 1 For k = 1 to popsize, repeat Step 2 to Step 6.
Step 2 Randomly select h assets out of the n available assets for initialization
to satisfy the cardinality constraint (3).
Step 3 For i = 1 to n , repeat Step 4.

9
Step 4 If the i -th asset has been selected in Step 2, then assign yi = 1 and
randomly generate xi ∈ [li yi , ui yi ], i.e., xi ∈ [li , ui ]; otherwise, assign
yi = 0 and thus xi = 0 . Thus, Xk [i] = xi . This step ensures the
constraints (4)-(7) are satisfied.
Step 5 To ensure that constraint (2) holds, set
xi
Xk [i] = .
x1 + x2 + . . . + xn

Step 6 Check the feasibility of the chromosome corresponding to constraints

(4)-(5) which may be violated as a result of performing normalization in
Step 5. If either of the constraints is violated, go to Step 3.

• Fitness evaluation The fitness evaluation function must consider all the
desired objective functions and make rational trade-offs among them. The only
constraint of the model (P 1) that is not incorporated in the chromosome design
is the portfolio risk constraint (1). It may be noted that the constraint (1)
corresponds to the N ′ constraints that result from the randomly generated pairs
(rj , α(rj )) , where j = 1, 2, . . . , N ′ . To design the fitness function, these N ′
constraints are incorporated in the RCGA process by assigning a penalty P to
the infeasible chromosomes. The penalty parameter is used to apply sufficient
selective pressure on the fitness function to avoid infeasible chromosomes. In
the case of no violation of the constraints corresponding to portfolio risk, the
penalty parameter P will be zero; it is positive otherwise. It may be noted
that if the penalty is too high or too low, then the problem might become very
difficult for the GA to solve. At the beginning of the search process, a large
penalty discourages the exploration of the infeasible region. Conversely, if the
penalty is too low, significant search time (generations) will be spent exploring
the infeasible region because the penalty will be negligible compared with the
objective function(s). It has been observed in the literature that penalties that
are functions of the distance from feasibility (the completion cost) perform better.
The selection of appropriate penalty is vital for faster convergence and more
precision. However, it is difficult to determine an appropriate penalty parameter
that is problem specific. In this paper, we use a static penalty in which the
penalty parameter remains constant during the entire RCGA process.
For j = 1, 2, . . . , N ′ , let
f4 (x, rj ) = Cr{b − (γ1 x1 + γ2 x2 + . . . + γn xn ) ≥ rj } − α(rj ) . (8)
The penalty levied ( on the infeasible chromosomes is
102 ∗ f4 (x, rj ), if f4 (x, rj ) > 0; j = 1, 2, . . . , N ′ ,
Pj =
0, otherwise.
Therefore, the net penalty is
XN′
P = Pj .
j=1

10
Because the model (P 1) involves multiple objective functions, the simplest
method to combine multiple objective functions into a scalar fitness solution
is the weighted-sum approach. It may be noted that if we consider the weighted
sum in the fitness function using constant weights, the search direction in the GA
is also constant. To explore various search directions to find good solutions, we
adopt the random weight approach proposed by Murata and Ishibuchi [43]. In
this approach, the weights are randomly generated for the fitness evaluation of
each chromosome. This approach explores the entire solution space to avoid local
optima. Thus, the resulting fitness function f itk corresponding to chromosome
Chk , where k = 1, 2, . . . , popsize , is defined as weighted sum of the objective
functions of model (P1) with a penalty parameter for infeasible chromosomes as
follows:
f itk = w1 f1 (x) + w2 f2 (x) + w3 f3 (x) − P
where wj for j = 1, 2, 3 is the weight of the j -th objective function, which en-
codes the relative importance of the objective. We use weights w k = (w1 , w2 , w3 )Tk
to be assigned for the three objective functions of model (P 1) in k -th solution
of a population. These weights are determined randomly and later normalized
X 3
such that wj = 1 as follows:
j=1
For j = 1,2,3 generate randomj ∼ U(0, 1). Then,
randomj
wj = P3 , j = 1, 2, 3
j=1 randomj
It may be noted that the uncertain objective functions fi (x) , where i = 1, 2, 3, of
the model (P1) used in the fitness evaluation function are approximated using the
fuzzy simulation process. The objective is now to find the solution chromosome
Chk corresponding to the optimum (maximum value) of the fitness function f itk .
• Elitism
To preserve and use the previously determined best solution in subsequent genera-
tions, an elite-preserving operator is often used. In addition to an overall increase
in performance, there is another advantage of using elitism. In an elitist GA, the
statistics of the population of best solutions cannot degrade with generations.
The elite count ( t ) indicates the number of individuals that are guaranteed to
be included in the next generation without the selection, crossover and mutation
operations being performed. In this paper, we use t = 4 to retain the four fittest
individuals of the current population when constructing the population for the
next generation.
• Selection
The selection method determines how chromosomes are selected from the cur-
rent population to be considered parents for the crossover operation. The goal
of the selection(reproduction) operator is to choose individuals that, on average,
are more fit than others to pass their genes to the next generation. We em-
ploy 4-player tournament selection as a selection mechanism in this study. Four
individuals are randomly selected and the individual with the highest fitness is

11
selected for the parent population. Recall that we already have four members
of the next generation as a result of performing the elitism operation. Let Ch′k ,
where k = 5, 6, . . . , popsize , constitute the parent population that will yield the
remaining popsize − 4 members of the next generation after the crossover and
mutation operations are performed. The remaining popsize − 4 chromosomes
for the parent population are generated using the 4-player tournament selection
as follows:
Step 1 For k = 5 to popsize, repeat Step 2 to Step 5.
Step 2 Randomly generate four integers sel1 , sel2 , sel3 , sel4 ∈ [1, popsize] . These
represent the selection of chromosomes Chsel1 , Chsel2 , Chsel3 , Chsel4 for
4-player tournament selection.
Step 3 If f itsel1 ≥ f itsel2 , let m = sel1 ; otherwise, m = sel2 .
Step 4 If f itsel3 ≥ f itsel4 , let t = sel3 ; otherwise, t = sel4 .
Step 5 If f itm ≥ f itt , let Ch′k = Chm ; otherwise, Ch′k = Cht .

• Crossover operator
Crossover is the primary genetic operator. The two parent chromosomes, if se-
lected for mating pool, reproduce two child chromosomes (offspring) using the
crossover operation. The crossover probability pc ∈ (0, 1) represents the chance
that the two selected chromosomes will crossover. For each potential crossover, a
random number between 0 and 1 is generated. If then umber of selected chro-
mosomes is odd, then the above procedure is repeated until one more chromo-
some is selected or the number of selected chromosomes becomes even. Standard
crossover operators have a high probability of violating the cardinality constraint
(3) of the model (P 1) . Thus, we use the shrinking crossover (SX) operator [8].
SX revises the two-point crossover by moving the second crossover point leftward
until there are an equal number of selected assets between and including the two
crossover points for both the selected parents and then exchanging the gene val-
ues of the parent chromosomes to produce offspring. The algorithm of the SX
operation is given below:
Step 1 For k = 5 to popsize, repeat Step 2.
Step 2 Randomly generate a real number r from the interval (0, 1) . The chro-
mosome Ch′k is selected as a parent for crossover if r < pc .
Step 3 Denote the selected parents as S1 , S2 , . . . and divide them into the fol-
lowing pairs: (S1 , S2 ), (S3 , S4 ), . . . .
Step 4 For each pair of selected parents, e.g., (S1 , S2 ) , randomly select two
positions a, b ∈ [1, n] . If a < b then pos1 = a and pos2 = b ; otherwise,
pos1 = b and pos2 = a .
Step 5 Until S1 &S2 have an equal number of selected assets between and in-
cluding pos1 &pos2 , repeat Step 6.
Step 6 pos2 = pos2 − 1 .

12
Step 7 If pos1 = pos2 go to Step 4; otherwise, go to Step 8.
Step 8 For i = pos1 to pos2 , repeat Step 9 to Step 11.
Step 9 temp = S1 [i] .
Step 10 S1 [i] = S2 [i] .
Step 11 S2 [i] = temp .
Fig. 2 depicts the SX operation between two randomly generated positions,
pos1 =14 and pos2 =18, for the selected parents. Here, pos1 =14 corresponds to
asset A14 having fractions of the capital budget x14 =0.1463 and x14 =0.1231
for the first and the second parents, respectively. pos2 =18 corresponds to asset
A18 having fractions of the capital budget x18 =0.2749 and x18 =0 for the first
and the second parents, respectively. Because these two positions do not have the
same number of selected assets for both the parents, pos2 is shifted left. Finally,
two-point crossover occurs between pos1 =14 and pos2 =17 when the number of
selected assets between and including the two positions stated above matches for
both the parents.
A1 A 14 A 15 A 16 A 17 A 18 A 19 A2 0
872 FFF 1463 0 0915 1187 749 0 0
࢞૚ ൌ.0 ࢞૚૝ ൌ. ࢞૚૞ ൌ ࢞૚૟ ൌ. ࢞૚ૠ ൌ. ࢞૚ૡ ൌ.2 ࢞૚ૢ ൌ ࢞૛૙ ൌ
Parents
0 ^^ 1231 2314 0 0821 0 0 9374
࢞૚ ൌ ࢞૚૝ ൌ. ࢞૚૞ ൌ. ࢞૚૟ ൌ ࢞૚ૠ ൌ. ࢞૚ૡ ൌ ࢞૚ૢ ൌ ࢞૛૙ ൌ.0
_

Crossover
872 ||| 1231 2314 0 0821 749 0 0
࢞૚ ൌ.0 ࢞૚૝ ൌ. ࢞૚૞ ൌ. ࢞૚૟ ൌ ࢞૚ૠ ൌ. ࢞૚ૡ ൌ.2 ࢞૚ૢ ൌ ࢞૛૙ ൌ
Offspring
0 ””” 1463 0 0915 1187 0 0 9374
࢞૚ ൌ ࢞૚૝ ൌ. ࢞૚૞ ൌ ࢞૚૟ ൌ. ࢞૚ૠ ൌ. ࢞૚ૡ ൌ ࢞૚ૢ ൌ ࢞૛૙ ൌ.0

Figure 2. The shrinking crossover operation

It may be noted that after each crossover operation, we must perform the nor-
malization again to ensure that constraint (2) holds, i.e.,
xi
Xk [i] = , k = 5, 6, . . . , popsize.
x1 + x2 + . . . + xn
The constraint (3) is treated by the SX operation. The constraints (6) and
(7) will always hold because of the design of the chromosomes. The offspring
generated after the SX operation can still violate constraints (4) and/or (5) when
the normalization is performed; thus, such offspring are rejected. It may be
noted that only the feasible offspring will replace the corresponding parents to
constitute the population for the next generation; otherwise, both the selected
parents will be retained in the population for the next generation. Note that all
the parents that are not selected for the crossover operation will be retained in
the population for the next generation.
• Mutation operator
Mutation is another operator that produces spontaneous random changes in var-
ious chromosomes by altering one or more genes. In this paper, out of several
available mutation operators, we use swap mutation to ensure that the capital
budget constraint (2) and the cardinality constraint (3) of model (P 1) are not
violated.
Given some probability of mutation pm ∈ (0, 1) , a chromosome is selected for
the process of mutation. The mutation process is summarized as follows:

13
Step 1 Set k = 5 .
Step 2 If k <= popsize , go to Step 3; otherwise, stop.
Step 3 Randomly generate a real number r from the interval (0, 1) . If r < pm ,
select the chromosome Ch′k for mutation and go to Step 4; otherwise,
k = k + 1 and go to Step 2.
Step 4 Randomly select two positions, pos1 , pos2 ∈ [1, n] and swap the gene
values at these two selected positions.
Fig 3. depicts the process of swap mutation in a selected chromosome with
pos1 =14 and pos2 =16. Here, in the selected chromosome, pos1 =14 corresponds
to asset A14 having a fraction of the capital budget x14 =0.1231 and pos2 =16
corresponds to asset A16 having a fraction of the capital budget x16 =0. After
the swap mutation is performed, we get x14 =0 and x16 =0.1231.
A1 A14 A15 A16 A17 A18 A19 A20
23 2314 0 821 2749 0 0
࢞૚ ൌǤ ૙ૡૠ૛ ËË ࢞૚૝ ൌ.1 1 ࢞૚૞ ൌ. ࢞૚૟ ൌ ࢞૚ૠ ൌ.0 ࢞૚ૡ ൌ. ࢞૚ૢ ൌ ࢞૛૙ ൌ
Ì
Mut ati on
872 èèè 2314 1231 821 2749 0 0
࢞૚ ൌ.0 ࢞૚૝ ൌ ૙ ࢞૚૞ ൌ. ࢞૚૟ ൌ. ࢞૚ૠ ൌ.0 ࢞૚ૡ ൌ. ࢞૚ૢ ൌ ࢞૛૙ ൌ

Figure 3. The swap mutation operation

It may be noted that swap mutation does not disturb the feasibility of the chro-
mosomes obtained after the crossover operation.

3.2.3 Step-wise description of HIA

We summarize the HIA algorithm as follows:

Step 1 Initialize appropriate chromosomes with length equal to the desired pop-
ulation size.
Step 2 Use fuzzy simulation to calculate the credibility values for the uncertain
objective functions of the model ( P 1 ).
Step 3 Compute the fitness of each chromosome using the credibility values
obtained in Step 2 and penalize the infeasible chromosomes that violate
the fuzzy chance constraints given by (8).
Step 4 Perform the elitism operation.
Step 5 Select the chromosomes for the parent population at any subsequent
generation using the 4-player tournament selection.
Step 6 Update the parent population using the crossover and mutation opera-
tions.
Step 7 Repeat Step 2 to Step 6 for a given number of generations.
Step 8 Select the best chromosome from all the generations as the solution of
the portfolio selection problem.

14
4. Numerical experiments: results and analysis
Now, we present computational results to illustrate portfolio selection using the
model (P 1) . We also test the effectiveness of the designed HIA applied to fuzzy
input data expressed using both special and general functional forms. The HIA
is coded in C++ on a personal computer with a 2.8 GHz Intel Core2Duo CPU
and 4GB of RAM. Table 1 gives the fuzzy data regarding short-term return,
long-term return and liquidity for 20 assets listed on the NSE of Mumbai, India.
We used historical data to create these fuzzy estimates. Note that the input data
regarding assets A1 to A17 are represented in fuzzy form using a trapezoidal
possibility distribution expressed as (a, b, c, d) . Here, b and c are the center
values and a and d are the left endpoint and right endpoint, respectively. The
input data regarding assets A18 to A20 are represented using general functional
forms.
Table 1: Input data
Asset Short-term return Long-term return Liquidity
A1 (-0.86, -0.165, 0.465, 0.9) (-0.38, -0.175, 0.425, 0.78) (.0065, .0115, .0205, .032)
A2 (-0.51, -0.23, 0.43, 0.75) (-0.5, -0.203, 0.547, 0.75) (.0003, .0009, .0027, .0033)
A3 (-0.85, -0.18, 0.54, 0.64) (-0.383, -0.003, 0.51, 0.65) (.0003, .00125, .00515, .01145)
A4 (-0.49, -0.125, 0.325, 0.78) (-0.455, -0.095, 0.625, 0.78) (.0014, .0017, .0047, .008)
A5 (-0.38, -0.035, 0.235, 0.35) (-0.265, -0.145, 0.305, 0.482) (.0002, .000375, .000825, .001)
A6 (-0.44, -0.065, 0.265, 0.56) (-0.4, -0.03, 0.51, 0.92) (.000276, .000465, .000855, .0022)
A7 (-0.22, 0.025, 0.535, 0.69) (-0.38, -0.15, 0.45, 0.69) (.0008, .00115, .00205, .003)
A8 (-0.66, -0.325, 0.245, 0.57) (-0.66, -0.275, 0.475, 0.66) (.000045, .0001625, .0004175, .000546)
A9 (-0.87, 0, 0.72, 0.88) (-0.75, -0.45, 0.52, 0.88) (.0011875, .0013875, .0079125, .013)
A10 (-0.68, -0.085, 0.605, 0.94) (-0.68, -0.22, 0.62, 1.04) (.0013, .00285, .00735, .013)
A11 (-0.15, -0.05, 0.37, 0.58) (-0.72, -0.3, 0.3, 0.7) (.001, .001575, .002625, .003185)
A12 (-0.85, -0.14, 0.4, 0.79) (-0.74, -0.25, 0.41, 0.79) (.000845, .0027, .0081, .015)
A13 (-0.55, -0.395, 0.115, 0.44) (-0.52, -0.275, 0.475, 0.77) (.0008, .0022, .0082, .010773)
A14 (-0.42, -0.175, 0.215, 0.4) (-0.76, -0.31, 0.29, 0.86) (.000256, .0014, .008, .015493)
A15 (-0.5, -0.07, 0.47, 0.8) (-0.79, -0.18, 0.51, 0.92) (.000252, .0008, .0038, .00576)
A16 (-0.66, -0.145, 0.425, 0.85) (-0.72, -0.27, 0.63, 0.85) (.000691, .00145, .00415, .005058)
A17 (-0.22, -0.025, 0.425, 0.52) (-0.58, -0.18, 0.66, 0.88) (.00039, .001065, .003855, .00472)
1 1 1
A18 exp(0.29r)
, r = [4.5, 5.5] exp(0.32r)
, r = [2.2, 3] exp(2r)
, r = [2.4, 3]
1 1 1
A19 exp(0.32r)
, r = [4, 5] exp(0.4r)
, r = [2.5, 3.5] (350−r)
, r = [4, 6]
1 1 1
A20 (r+27)
, r = [0.9, 1.2] (r+0.39)
, r = [3, 4] exp(8−r)
, r = [0.7, 0.9]

We solve three portfolio selection problems using three different confidence

curves. These curves are distinguished using the shape parameter K . Using
different values of K , we can explore different confidence curves of the investor
that reflect his tolerance of portfolio risk. The primary attributes of the problems
solved are summarized in Table 2.
1
Table 2: Primary attributes of the problems solved using confidence curve α(r) = exp(Kr)
∀r ≥ 0

Cases
Case-I, K=3 Case-II, K=2 Case-III, K=1.3
1 1 1
α(r) exp(3r) exp(2r) exp(1.3r)
r1 0.10 0.12 0.14
r2 0.14 0.16 0.18
r3 0.0035 0.0045 0.0055
Target return of portfolio (b) 0.18 0.21 0.25
Number of pairs (r, α(r)) generated (N ′ ) 35 35 35
Length of a chromosome (n) 20 20 20
h 7 7 7
ui , ∀ i 0.3 0.3 0.3
li , ∀ i 0.08 0.08 0.08

15
4.1. Computational results and sensitivity analysis for case-I
• Stability analysis of the HIA
To test the effectiveness of the HIA, we solve the portfolio selection problem
using different values for the HIA parameters. We performed 20 HIA runs for
each set of the parameter settings and report the best fitness value determined.
The parameters used and the corresponding computational results are shown in
Table 3. To compare the results, we use the relative error (RE) index, which is
defined as follows:
RE = (Maximal fitness- Actual fitness)/Maximal fitness × 100%
where the maximal fitness is the maximum of the fitness values for all the com-
putational results obtained.
Table 3: Results corresponding to different settings of the HIA parameters for case-I
pc pm popsize Simulation runs (N) Generations Fitness Relative error (%)
0.5 0.3 50 3000 3000 0.757136 1.068
0.6 0.4 50 4000 3000 0.754406 1.425
0.7 0.3 50 3000 3000 0.758344 0.91
0.7 0.5 40 4000 2000 0.764366 0.123
0.8 0.4 50 3000 2000 0.761957 0.438
0.8 0.6 40 4000 3000 0.765308 0
0.9 0.5 50 3000 2000 0.764527 0.102
0.9 0.8 40 4000 3000 0.759791 0.721
From Table 3,we see that the relative error corresponding to each parame-
ter setting does not exceed 2% , which demonstrates that the proposed HIA is
effective at setting the parameters. We use the parameter settings from Table
3 that correspond to the maximum fitness, i.e., popsize = 40, pc = 0.8, pm =
0.6, generations = 3000, and simulation runs (N) = 4000 , to run the HIA to
obtain the optimal portfolio. To further demonstrate the efficiency of the al-
gorithm, we performed 20 runs of HIA to check the stability of the solutions
obtained. Solution statistics for the 20 runs are reported in Table 4.
Table 4: Solution statistics for 20 HIA runs for case-I
Best fitness Average fitness Standard deviation Coefficient of variation (%)
0.765308 0.7364408 0.015702387 2.132199467
In Fig. 4, we show the sensitivity of the maximum fitness attained in each
HIA run.
1

0 5
.9
0
.9
0 5
.8
s
nes
0
Fi t .8
0 5
um .7
im 0
.7
M ax
0 5
.6
0
.6
0
.55
0
.5
0 2 4 6 8 10 12 14 16 18 20

HIA Run

Figure 4. Maximum fitness vs. HIA run for case-I

16
 Recall that each HIA run includes 3000 generations. In Fig. 5, we show the
variation of the maximum fitness (out of 40 chromosomes) attained at various
generations for the HIA run that yields the best fitness value of those reported
in Table 4.
0.8

0 5
.7
s
es
0.7
it n
F

um 0 5
.6
im

Max
0.6

0
.55

0.5

0 500 1000 1500 2000 2500 3000

Generation

Figure 5. Maximum fitness vs. generation for case-I

• Portfolio selection
The credibility values of the objective functions of the portfolio selection model
(P 1) that correspond to the best fitness found in the HIA runs reported in Table
4 are presented in Table 5. The fraction of the capital budget invested in each
asset is reported in Table 6. A graphical representation of the asset allocation is
shown in Fig. 6.
Table 5: The attainment values of the various objective functions for case-I
Maximum credibility level of objectives
Short term return Long term return Liquidity
0.507766 0.571469 0.782493

Table 6: The asset allocation for case-I

Asset
A1 A2 A3 A4 A5 A6 A7 A8 A9 A10
0.197785 0 0.089247 0 0 0 0 0 0.156369 0.088035
Asset
A11 A12 A13 A14 A15 A16 A17 A18 A19 A20
0 0.181678 0 0.089344 0 0 0 0.197542 0 0
0
.2
0 18
.
0 16
.
0 14
.
0 12
.
ti ona
0
.1
lAl oc 0 08
.
0 06
.
0 04
.
0 02
.
0
A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11 A12 A13 A14 A15 A16 A17 A18 A19 A20
Asset ID Number

Figure 6. The asset allocation for case-I

Recall that we desire to invest in seven assets. Thus, the optimal portfolio

17
should be a combination of the seven assets for which the obtained risk curve f (r)
is below the given confidence curve α(r) of the investor, i.e., the low-risk area.
A combination of assets with a risk curve that lies above the given confidence
curve, i.e., the high-risk area, is considered infeasible. Fig. 7 shows the risk curve
of the portfolio generated for case-I that correspond to the investible assets listed
in Table 6. Further, Fig. 7 demonstrates that the risk curve of the generated
portfolio is completely below the confidence curve of the investor. Thus, the
combination of these seven assets satisfies the investor’s preferences.
1

0.9

0.8

0.7
y
0.6
it
bil
di
0.5
e α (r)

C r 0.4

0.3
f(r)

0.2

0.1

0
0 0 1 0 2 0 3 0 4 0 6 0.7 0.8 0.9 1 1 1
. . . . .5 0. .

Severity of risk 'r '

Figure 7. The confidence curve and risk curve of the portfolio obtained for
case-I

• Sensitivity analysis of the risk curve w.r.t. the parameter b

We perform a sensitivity analysis w.r.t. the parameter b. Fig. 8 shows a graphical
representation of the sensitivity analysis results.
1
alpha( r)=1/exp(3r)

f (r ) for b=0.18
0.9
f (r ) for b=0.3
0.8
f (r ) for b=0.5

0.7 f (r ) for b=0.6

ity 0.6
il
dib
0.5

C re
0.4

0.3

0.2

0.1

0 0.1 0.2 03 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1

.

Severity of risk 'r'

Figure 8. Effect of changes in parameter b on risk curve for case-I

Fig. 8 demonstrates that as the target value of the portfolio return increases,
the corresponding risk curve of the generated portfolio approaches the investor’s
confidence curve. After a threshold value of b is reached, the portfolio selection
model (P 1) becomes infeasible because the risk curve of the generated portfolio
moves beyond the confidence curve of the investor and enters the high-risk area,
which is undesirable. As stated earlier, the portfolio risk constraint (1) of model
(P 1) is replaced by fuzzy chance constraints that correspond to randomly gen-
erated pairs (rj , α(rj )) , where j = 1, 2, . . . , 35 . The infeasibility of the portfolio
selection model is due to the violation of such fuzzy chance constraints.

18
 4.2. Computational results and sensitivity analysis for case-II and
case-III
• Stability analysis of the HIA
More numerical experiments are performed to demonstrate the adaptability of
the portfolio selection model (P 1) to different investor-preferences regarding
the short-term return, long-term return, risk and liquidity characteristics of
the desired portfolio. We use the following parameter settings to run the HIA
for case-II and case-III: popsize = 40, pc = 0.8, pm = 0.6, generations =
3000, and simulation runs (N) = 4000 . Solution statistics for the 20 runs of
HIA are reported in Table 7 for case-II and case-III. Fig. 9 and Fig. 10 show the
sensitivity of the maximum fitness attained in each HIA run.
Table 7: Solution statistics for 20 HIA runs for case-II and case-III
Best fitness Average fitness Standard deviation Coefficient of variation (%)
Case-II 0.70474 0.6764113 0.016100095 2.380222707
Case-III 0.674705 0.65462205 0.013686557 2.090757043
1
1
0.95
0.95
0.9
0.9
snes 0.85
snes 0.85
Fit 0.8 Fi t 0.8
um um 0.75
im
0.75

M ax 0.7 im 0.7
M ax
0.65
0.65
0.6
0.6
0.55
0.55
0.5 0.5
0 2 4 6 8 10 12 14 16 18 20
0 2 4 6 8 10 12 14 16 18 20
HIA Run HIA Run
Figure 9. Maximum fitness vs. HIA run for case-II Figure 10. Maximum fitness vs. HIA run for case-III

• Portfolio selection
In Table 8, we give the values attained by maximizing credibilistic objective
functions for case-II and case-III. Table 9 gives the fraction of the capital budget
invested in each asset for the portfolios corresponding to the best fitness reported
in Table 7 for both the cases. Graphical representations of the asset allocation
for both the cases are presented in Fig. 11 and Fig. 12.
Table 8: Values of the various objective functions for case-II and case-III
Maximum credibility level of objectives
Short term return Long term return Liquidity
Case-II 0.530641 0.761438 0.520641
Case-III 0.541246 0.712166 0.560898

Table 9: Asset allocation for case-II and case-III

Asset
A1 A2 A3 A4 A5 A6 A7 A8 A9 A10
Case-II 0.190104 0 0 0 0 0.096162 0.118304 0 0 0.081834
Case-III 0.243503 0 0.141348 0 0 0.086343 0 0 0 0
Asset
A11 A12 A13 A14 A15 A16 A17 A18 A19 A20
0 0 0 0 0 0 0 0.140712 0.257111 0.115773
0 0 0 0 0 0.082819 0 0.244763 0.116417 0.084807

19
 0.26 0.26
0.24 0.24
0.22 0.22
02 02
0.18. 0.18.
0.16 0.16
it ona 0.14 n 0.14
lA oc 0.0121
ti o 0.12
ca
01
0.08. 0.08.
lAl o

0.06 0.06
0.04 0.04
0.02 0.02
0 0
A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11 A12 A13 A14 A15 A16 A17 A18 A19 A20 A1 A2 A3 A 4 A 5 A 6 A 7 A 8 A 9 A 10 A 11 A 12 A 13 A 14 A 15 A 16 A 17 A 18 A 19 A 20
Asset ID Number Asset ID Number

Figure 11. Asset allocation for case-II Figure 12. Asset allocation for case-III
Again, because we wish to invest in seven assets, the optimal portfolio should
be a combination of the seven assets for which the risk curve f (r) is below the
investor’s confidence curve α(r) . Fig. 13 and Fig. 14 show the risk curve of the
portfolio generated for both the cases.
1 1

0.9 0
.9
0.8 0
.8
y
0.7 0
it .7
bil y
0.6 0 α (r)
di .6
e it
0.5 il 0
Cr dib .5
0.4 0
α (r) C re .4
0.3 0
.3 f(r)
0.2 f(r) 0
.2
0.1 0
.1
0 0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0 0.1 0 5 06 07 08 09 1 11 12
.2 0.3 0.4 0. . . . . . .
Severity of risk 'r '
Severity of risk 'r '

Figure 13. The confidence curve and risk curve Figure 14. the confidence curve and risk curve
for the obtained portfolio for case-II for the obtained portfolio for case-III
Fig. 13 and Fig. 14 demonstrate that the risk curve of the portfolio generated
in both cases is below the corresponding confidence curve of the investor.
• Sensitivity analysis of the risk curve w.r.t. the parameter b
We perform a sensitivity analysis w.r.t. the parameter b for both cases. Fig. 15
and Fig. 16 show graphical representations of the sensitivity analysis results.
1 1
alpha(r )=1/exp(2r) alpha( r)=1/exp(1.3r)
09 f(r ) for b=0.21 0.9 f(r ) for b=0 25
. .
f( r) for b=0.4 f(r ) for b=0 5
08 0.8 .
. f(r ) for b=0.6 f(r ) for b=0 7
.
07 f(r ) for b=0.7 0.7 f(r ) for b=0 9
. .
y y 0.6
06
it . it
ilb il 0.5
05 dib
di .
e 0.4
Cr 04 C re
.
03 0.3
.
02
0.2
.
01
0.1
.
0
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 2
0 01 02 0.3 0.4 0 5 0.6 0.7 0 9 1 1.1 12 .8 0.9 1 1.1 1.
. . . .8 0. .
Severity of risk 'r'
Severity of risk 'r'

Figure 15. Effect of changes in the parameter b Figure 16. Effect of changes in the parameter b
on the risk curve for case-II on the risk curve for case-III
Fig. 15 and Fig. 16 also demonstrate that after a threshold value of b is
reached, the portfolio selection problem becomes infeasible because the risk curve

20
of the generated portfolio moves into the high-risk area, which is undesirable.
5. Concluding remarks
In this paper, we extended the credibility-based portfolio selection to a fuzzy
multiple-criteria framework where we consider the short-term return, long-term
return, risk and liquidity characteristics of the portfolio. The fuzzy multiobjective
portfolio selection model developed here is a credibility maximization model in
which short-term return, long-term return and liquidity are considered as objec-
tive functions. The portfolio risk was characterized by a risk curve that represents
each likely loss of the portfolio return and the corresponding chance of its occur-
rence rather than a single pre-set level of the loss. The risk curve must be below
the confidence curve of the investor to build a safe portfolio. The portfolio se-
lection problem was solved using a HIA under the assumption that the fuzzy
parameters have general functional forms. We conducted numerical experiments
corresponding to three different confidence curves for portfolio selection. These
curves represented the tolerances of the investor regarding the chances of losses
equal to or greater than a given severity indicator. Corresponding to each confi-
dence curve, the best combination of seven assets was reported; this combination
gives not only the safe portfolio but also one that is optimal in the sense of achiev-
ing the highest credibility values for the objective functions. We also performed
a sensitivity analysis by perturbing the target value of the portfolio return and
analyzing the variation in the solutions obtained. The results of the numerical
analysis demonstrated the efficiency of the algorithm and the adaptability of the
model to different situations. The primary advantage of the proposed model is
that it is not only a multiple-criteria model but it also treats the uncertainty of
the financial markets more realistically because the fuzzy parameters short-term
return, long-term return, risk and liquidity can have general functional forms.
One can, however, identify and add more criteria to capture investor preferences.
Furthermore, the model can be tested in other environments using random, fuzzy
random or random fuzzy parameters.
Acknowledgements
We are thankful to the Editor-in-chief Professor W. Pedrycz and the anonymous
referees for their valuable comments and suggestions to improve presentation of
the paper.
References
[1] K. P. Anagnostopoulos, G. Mamanis, A portfolio optimization model with
three objectives and discrete variables, Computers & Operations Research
37 (2010) 1285-1297.
[2] J. D. Bermúdez, J. V. Segura, E. Vercher, A multi-objective genetic algo-
rithm for cardinality constrained fuzzy portfolio selection, Fuzzy Sets and
Systems 188 (2012) 16–26.
[3] R. Bhattacharyya, S. Kar, D. D. Majumder, Fuzzy mean-variance-skewness
portfolio selection models by interval analysis, Computers and Mathematics
with Applications 61 (2011) 126-137.
[4] C. Carlsson, R. Fullér, P. Majlender, A possibilistic approach to selecting
portfolios with highest utility score, Fuzzy Sets and Systems 131 (2002) 13-
21.

21
 [5] I. G. Carrasco, R. C. Palacios, J. L. L. Cuadrado, Á. G. Crespo, B. R.
Mezcua, PB-ADVISOR: A private banking multi-investment portfolio advi-
sor, Information Sciences 206 (2012) 63–82.
[6] J. F. Chang, P. Shi, Using investment satisfaction capability index based
particle swarm optimization to construct a stock portfolio, Information Sci-
ences 181 (2011) 2989–2999.
[7] A. Charnes, W. W. Cooper, Chance-constrained programming, Management
Science 6 (1959) 73-79.
[8] J. S. Chen, J. L. Hou, S. M. Wu, Y. W. Chang-Chien, Constructing invest-
ment strategy portfolios by combination genetic algorithms, Expert Systems
with Applications 36 (2009) 3824-3828.
[9] W. Chen, Weighted portfolio selection models based on possibility theory,
Fuzzy Information and Engineering 2 (2009) 115-127.
[10] H. Dastkhan, N. S. Gharneh, H. Golmakani, A linguistic-based portfolio se-
lection model using weighted maxmin operator and hybrid genetic algorithm,
Expert Systems with Applications 38 (2011) 11735–11743.
[11] W. Dong, J. Peng, Credibilistic risk optimization models and algorithms. In
Proceedings of the International Conference on Computational Intelligence
and Natural Computing IEEE (2009) 378-381.
[12] Y. Fang, K. K. Lai, S. Y. Wang, Portfolio rebalancing model with transac-
tion costs based on fuzzy decision theory, European Journal of Operational
Research 175 (2006) 879-893.
[13] Y. Fang, K. K. Lai, S. Y. Wang, Fuzzy portfolio optimization: theory and
methods, Lecture Notes in Economics and Mathematical Systems 609 (2008)
Springer.
[14] P. Gupta, M. K. Mehlawat, A. Saxena, Asset portfolio optimization using
fuzzy mathematical programming, Information Sciences 178 (2008) 1734-
1755.
[15] P. Gupta, M. K. Mehlawat, A. Saxena, A hybrid approach to asset allocation
with simultaneous consideration of suitability and optimality, Information
Sciences 180 (2010) 2264-2285.
[16] P. Gupta, M. Inuiguchi, M. K. Mehlawat, A hybrid approach for constructing
suitable and optimal portfolios, Expert Systems with Applications 38 (2011)
5620-5632.
[17] P. Gupta, M. K. Mehlawat, G. Mittal, Asset portfolio optimization using
support vector machines and real-coded genetic algorithm, Journal of Global
Optimization 53 (2012) 297-315.
[18] T. Hasuike, H. Katagiri, H. Ishii, Portfolio selection problems with random
fuzzy variable returns, Fuzzy Sets and Systems 160 (2009) 2579-2596.
[19] N. Hachicha, B. Jarboui, P. Siarry, A fuzzy logic control using a differen-
tial evolution algorithm aimed at modelling the financial market dynamics,
Information Sciences 181 (2011) 79–91.
[20] J. H. Holland, Adaptation in Natural and Artificial Systems, University of
Michigan Press, Ann Arbor, 1975.
[21] X. Huang, Portfolio selection with fuzzy returns, Journal of Intelligent and
Fuzzy Systems 18 (2007) 383-390.
[22] X. Huang, Risk curve and fuzzy portfolio selection, Computers and Mathe-
matics with Applications 55 (2008) 1102-1112.

22
[23] X. Huang, A review of credibilistic portfolio selection, Fuzzy Optimization
and Decision Making 8 (2009) 263-281.
[24] X. Huang, An entropy method for diversified fuzzy portfolio selection, Inter-
national Journal of Fuzzy Systems 14 (2012) 160-165.
[25] M. Inuiguchi, H. Ichihashi, Relative modalities and their use in possibilistic
linear programming, Fuzzy Sets and Systems 35 (1990) 303-323.
[26] M. Inuiguchi, H. Ichihashi, Y. Kume, Relationships between modality con-
strained programming problems and various fuzzy mathematical program-
ming problems, Fuzzy Sets and Systems 49 (1992) 243-259.
[27] M. Inuiguchi, H. Ichihashi, Y. Kume, Modality constrained programming
problems: a unified approach to fuzzy mathematical programming problems
in the setting of possibility theory, Information Sciences 67 (1993) 93-126.
[28] M. Inuiguchi, J. Ramik, Possibilistic linear programming: a brief review
of fuzzy mathematical programming and a comparison with stochastic pro-
gramming in portfolio selection problem, Fuzzy Sets and Systems 111 (2000)
3-28.
[29] M. Inuiguchi, T. Tanino, Portfolio selection under independent possibilistic
information, Fuzzy Sets and Systems 115 (2000) 83-92.
[30] P. Jana, T.K. Roy, S.K. Mazumder, Multi-objective possibilistic model for
portfolio selection with transaction cost, Journal of Computational and Ap-
plied Mathematics 228 (2009) 188-196.
[31] Z. Jianguo, Z. Liang, Multi-objective model for uncertain portfolio opti-
mization problems, International Journal of Advancements in Computing
Technology 3 (2011) 138-145.
[32] X. Li, B. Liu, A sufficient and necessary condition for credibility measures,
International Journal of Uncertainity, Fuzziness & Knowledge-Based Sys-
tems 14 (2006) 527-535.
[33] X. Li, B. Shou, Z. Qin, An expected regret minimization portfolio selection
model, European Journal of Operational Research 218 (2012) 484-492.
[34] X. Li, Y. Zhang, H. S. Wong, Z. Qin, A hybrid intelligent algorithm for
portfolio selection problem with fuzzy returns, Journal of Computational
and Applied Mathematics 233 (2009) 264-278.
[35] X. Li, Z. Qin, S. Kar, Mean-Variance-Skewness model for portfolio selection
with fuzzy returns, European Journal of Operational Research 202 (2010)
239-247.
[36] Y. Li, B. Wang, J. Watada, Building a fuzzy multi-objective portfolio se-
lection model with distinct risk measurements. In Proceedings of the IEEE
International Conference on Fuzzy Systems (FUZZ-IEEE 2011) (2011) 1096-
1102.
[37] B. Liu, Theory and Practice of Uncertain Programming, Physica-Verlag,
Heidelberg, 2002.
[38] B. Liu, A survey of credibility theory, Fuzzy Optimization and Decision
Making 5 (2006) 387-408.
[39] B. Liu, Y. K. Liu, Expected value of fuzzy variable and fuzzy expected value
models, IEEE Transactions on Fuzzy Systems 10 (2002) 445-450.
[40] S. T. Liu, A fuzzy modeling for fuzzy portfolio optimization, Expert Systems
with Applications 38 (2011) 13803–13809.

23
[41] E. A. Luengo, Fuzzy mean-variance portfolio selection problems, Advanced
Modelling and Optimization 12 (2010) 399-410.
[42] H. M. Markowitz, Portfolio selection, Journal of Finance 7 (1952) 77-91.
[43] T. Murata, H. Ischibuchi, MOGA: Multi-objective genetic algorithms. In
Proceedings of the second IEEE International Conference on Evolutionary
Computation (1995) 289-294.
[44] R. Rodrı́guez, M. Luque, M. González, Portfolio selection in the Spanish
stock market by interactive multiobjective programming TOP 19 (2011) 213-
231.
[45] S. Y. Wang, S. S. Zhu, On fuzzy portfolio selection problem, Fuzzy Opti-
mization and Decision Making 1 (2002) 361-377.
[46] Y. Yoshida, An estimation model of value-at-risk portfolio under uncertain-
ity, Fuzzy Sets and Systems 160 (2009) 3250-3262.
[47] W. G. Zhang, Q. Mei, Q. Lu, W. L. Xiao, Evaluating methods of investment
project and optimizing models of portfolio selection in fuzzy uncertainty,
Computers & Industrial Engineering 61 (2011) 721–728.
[48] W. G. Zhang, Y. L. Wang, Z. P. Chen, Z. K. Nie, Possibilistic mean-variance
models and efficient frontiers for portfolio selection problem, Information
Sciences 177 (2007) 2787-2801.
[49] Y. Zhang, X. Li, H. -S. Wong, L. Tan, Fuzzy multi-objective portfolio selec-
tion model with transaction costs. In Proceedings of the IEEE Internation
Conference on Fuzzy Systems (FUZZ-IEEE 2009) (2009) 273-278.
[50] X. Zhang, W. G. Zhang, R. Cai, Portfolio adjusting optimization under
credibility measures, Journal of Computational and Applied Mathematics
234 (2010) 1458-1465.
[51] L. A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338-353.
[52] L. A. Zadeh, Towards a generalized theory of uncertainity (GTU)-an outline,
Information Sciences 172 (2005) 1-40.

24
 We propose a new multiobjective portfolio selection model in fuzzy environment
 We model fuzziness using credibility measure of fuzzy events
 We use fuzzy-chance constraints for measuring portfolio risk
 A hybrid intelligent algorithm is designed to solve the mathematical model
 The approach is useful when the fuzzy parameters take any general functional
form