MCDM in A Fuzzy Setting: Investment Projects Assessment Application

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

ARTICLE IN PRESS

Int. J. Production Economics 100 (2006) 10–29


www.elsevier.com/locate/ijpe

MCDM in a fuzzy setting: Investment projects


assessment application
L. Dimovaa, P. Sevastianova, D. Sevastianovb,,1
a
Institute of Computers and Information Science, Technical University of Czestochowa, Dabrowskiego 73, 42-200 Czestochowa, Poland
b
Reuters Research, Inc., 3 Times square, location 17-W15, New York, NY 10036, USA
Received 1 February 2004; accepted 30 September 2004
Available online 7 January 2005

Abstract

The problem of investment project estimation is considered as the multiple criteria hierarchical task of choosing the
optimal alternative (project). The mathematical tools of fuzzy sets theory are used for representation of uncertainty and
building of local criteria, based on the quantitative and qualitative parameters characterizing the considered projects.
The problems of ranked local criteria aggregation are analyzed and some new theoretical results, which can be useful
for proper choice of aggregation method, are presented. The new method for generalization of aggregation schemes
based on level-2 fuzzy sets mathematical tools is proposed and illustrated. The method for dealing with hierarchical
structures of local criteria is presented as well. The theoretical consideration is illustrated with simple examples and their
realization by authors’ software.
r 2004 Elsevier B.V. All rights reserved.

Keywords: Multiple criteria optimization; AHP; Decision making; Investment project estimation; Fuzzy sets

1. Introduction of considered projects, such as net present value


(NPV), internal rate of return (IRR) and so on.
Traditional approaches to the investment pro- The estimation of the investment efficiency, as well
ject estimation are usually based on the budgeting, as any forecasting, is rather an uncertain problem
i.e. analysis of the discounted financial parameters and so the proper methods for operating in
uncertain setting must be used. Since the applic-
Corresponding author. Tel.: +1 646 223 8182; fax: ability of traditional probability methods is
+1 646 223 4123. often restricted by absence of objective probabil-
E-mail addresses: sevast@icis.pcz.czest.pl (P. Sevastianov), istic information about future events, during
Dmitry.Sevastyanov@reuters.com (D. Sevastianov).
1
The statements made and the views expressed are solely last two decades growing interest in the applica-
those of the author and should not be attributed to his tion of interval and fuzzy methods in budgeting
employer. has been observed; see Ward (1985), Kuchta

0925-5273/$ - see front matter r 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijpe.2004.09.014
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 11

(2000), Dimova et al. (2000) and Kahraman et al. practical applications of multiple criteria analysis,
(2002). the key problems we discuss further in this paper
On the other hand, when analyzing the invest- may be formulated as follows:
ment project we consider (perhaps, implicitly) 1. The local criteria may be constructed on the
some local criteria based on calculated financial basis of quantitative parameters such as financial
parameters. So, the project estimation is in essence ones, as well as using expert subjective estimations
a multiple criteria problem and as examples of (verbal assessments of project’s scientific, techno-
successful systematization, the local criteria sets logical level, etc). It is known that experts prefer to
proposed by Lopes and Flavel (1998), and by give rather ‘‘fuzzy’’ advice on linguistic level of
Wang (2004) may be considered. presentation to avoid possible mistakes connected
Even skin-deep analysis of these criteria system- with qualitative prediction of future events. At the
atization allows us to conclude that investment same time, human experience and intuition play an
and project quality estimation is the complicated important role in project estimation and they
multiple criteria problem frequently having certain cannot be ignored, although the specific uncer-
hierarchical structure. tainty is their inherent property. This uncertainty
There are a lot of multiple criteria methods is of subjective, fuzzy nature and cannot be
proposed in literature for solving economical and described in usual probabilistic way.
financial problems. Steuer and Na (2003) pre- So a proper methodology is needed to take into
sented the widest review of this problem based on account the uncertainty factors, which will allow
more than 250 literature indices. Nevertheless, we to build a set of comparable local criteria based on
can cite the only few papers devoted to the directly measurable quantitative parameters, as
multiple criteria financial project estimation; see well as on linguistically formulated assessments.
Wang (2004), Mohamed and McCowan (2001) The mathematical tools of fuzzy sets theory
and Li and Sterali (2003). elaborated for dealing with subjective kind of
It is worth noting that in all these works the uncertainty (Zadeh, 1965) may be successfully
fuzzy sets theory concepts were used. The method used for this purpose.
of Mohamed and McCowan (2001) is based on a 2. Some criteria are competitive ones, i.e. an
representation of the local criteria by membership improvement of one criterion inevitably leads to a
functions and their aggregation using simple fuzzy deterioration of another. This feature must be
summation. The ranks of local criteria and taken into account when creating any multiple
possible hierarchical structure of the problem were criteria method since the compromise among
not taken into account. competitive, sometimes antagonistic, local criteria
The hierarchical structure of the problem is is the major problem of decision making.
considered in Weck et al. (1997) and the well- 3. Usually criteria are not equivalent, i.e. they
known AHP method is used for its building, but make different contribution to the integral estima-
simple normalization of financial parameters tion of project quality. So, the problem of local
(dividing them by their maximal values) is applied criteria ranking arises and suitable well-defined
instead of natural local criteria. An interesting method should be chosen to resolve it.
example of practical application of the multiple 4. Real world decision problems may involve a
criteria hierarchical analysis is presented by Li and lot of local criteria to be analyzed simultaneously.
Sterali (2003). The generalized AHP method has Regrettably, the human ability to do this is
been used for estimation of 103 mutually depen- strongly restricted by the known empirical 772
dent investment projects proposed for the Tumen law of psychology, according to which a person
river region (China) industrial development. can normally distinguish no more than 7 plus
We do not intend to make here the detailed minus 2 classes or grades on some feature scale. If
review of these works, but as a result of the the number of grades is greater, the adjacent
analysis we have done in the field of investment grades start to merge and cannot be clustered
project estimation as well as in some other confidently; see Miller (1956) and Milner (1970).
ARTICLE IN PRESS

12 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

To solve this problem the relevant aggregation of mathematical technique is needed to represent the
local criteria, taking into account their ranking, local criteria in some general form allowing their
can be used to create some generalized criteria. direct comparison. It is suitable to use mathema-
5. Criteria may compose a multilevel hierarch- tical tools of fuzzy sets theory for this purpose.
ical structure when a given set of local criteria Let us consider such important quantitative
consists of certain subgroups connected logically. financial parameters as IRR, and NPV. Since IRR
The list of problems can be continued. For is measured in percentages whereas NPV ac-
instance, some problems appear when fuzzy valued counted in currency units, it seems impossible to
financial parameters are used as arguments of compare them while estimating the project. How-
functions representing the local criteria (Dimova ever, it can be done using functions presenting the
et al., 2000) or when such functions are fuzzy local criteria based on these financial parameters.
as well (Dymova et al., 2002; Dymova, 2003), but It is worth noting that introducing of such
detailed analysis of these issues is beyond the scope function does not lead to the loss of the original
of this paper. information, quite the contrary. Really, if x is
We can say that almost all components for some parameter of analyzed system representing
building the efficient method that may be used in its quality to certain extent, then some scale of
the majority cases of real-life project estimations preference—presented on numerical or verbal
are already elaborated and described in literature. level—for this parameter inevitably exists at least
What is needed is their critical analysis from the in the decision maker’s mind. Indeed, such
viewpoint of the considered problem, and proper preference type information is the key to deci-
synthesis of them into an integrated method for sion-making and local criteria are the mathema-
hierarchical multiple criteria estimation of invest- tical tools for its formalization.
ment projects. For these reasons the rest of the Often they can be built using the following
paper is organized as follows. In Section 2, the simple and natural procedure.
fuzzy sets theory method for building local criteria, For example, when considering IRR, it is easy
based on the quantitative and qualitative para- to see that there always exists some lower bound
meters, is presented and illustrated. Section 3 is for permissible values of IRR, usually being equal
devoted to the ranking of local criteria on the base to the bank rate, r. Further, there is some interval
of their pair comparison. The problems of local rpIRRpIRRm ; where project’s quality rises
criteria aggregation are analyzed in Section 4. gradually with IRR increase. Finally, it is expected
Some new theoretical results, which can be useful that if IRRXIRRm ; the project’s efficiency with
for proper choice of the aggregation method, as respect to IRR is so high that it is difficult to make
well as the new method for generalizing the a reasonable choice among such excellent projects.
aggregation schemes, based on the level-2 fuzzy To transform this description into a mathema-
sets mathematical tools, are presented as well. In tical form the membership function, which is the
Section 5, the method for dealing with hierarchical pivotal concept of Fuzzy Sets Theory, may be
structures of criteria is described. Everywhere we used. The membership function m(IRR) represent-
will try to illustrate our mathematical considera- ing the local criterion based on IRR for some
tions with the use of simple illustrative examples analyzed example is shown in Fig. 1.
realized through the author’s software. Finally, the The values of membership functions change
concluding section summarizes the paper and from zero (for the worst values of quality
discusses future research issues. parameters) to maximum value equal to 1 in the
area of best values of analyzed quality parameters.
So, in a context of the considered problem the
2. Local criteria presentation values of membership functions may be treated as
degrees of quality parameter’s preference.
In practice, local criteria are usually based on The linear form of membership function is not a
parameters of different nature. Hence, the special dogma. However, in the practice what we usually
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 13

(IRR) know is only that some value is more preferable


1 than other without any certain quantitative
estimation of this preference. In other words, in
most cases we deal with the so-called ‘‘linear
ordering’’ and for such situations the linear form
0.5 of criterion function is proved to be the best one
(Yager, 1980).
Since the membership function must be convex
IRR and normalized to 1, the only few forms of such
0 function, presented in Fig. 2, may normally be used.
r IRRm
Of course, if the probability distribution of quality
Fig. 1. The membership function representing the local parameter is known, the corresponding membership
criterion based on IRR. function may have a more complex form.

Fig. 2. The typical forms of membership functions.


ARTICLE IN PRESS

14 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

Consider a qualitative parameter presented in a for formalization of local criteria in the majority of
verbal form, e.g. such as ‘‘ecological impact on real-life situations (of course, in some cases the
region’’. It can be described by a set of statements use of more complex descriptions, e.g. on the base
that linguistically represents the degrees to which of type 2 fussy sets, may be preferable (Dymova
an analyzed project may affect the ecology of et al., 2002; Dymova, 2003).
region: ‘‘not significant’’, ‘‘slightly significant’’, Finally, we can see that all the qualitative and
‘‘noticeably significant’’ and so on. As it was noted quantitative local criteria can be naturally pre-
above, no more than nine such linguistic degrees sented within the universal scale of membership
may be used in practice. The linguistic variables function.
may be translated into a mathematical form by To make our further analysis more transparent,
presenting them in a form of triangular trapezoidal consider a simple example which is not the
fuzzy numbers. This approach is in accordance description of any real project estimation situation
with a spirit of fuzzy sets theory and undoubtedly but reflects the main advantages of proposed
is very fruitful in a great number of applications, approach.
especially in fuzzy logic. Nevertheless, when deal- Assume we have four projects to be compared
ing with the decision-making problems, we usually taking into account five quality parameters
do not have enough reliable information to build (Fig. 4). Obviously, the number of the parameters
such fuzzy numbers. Frequently, it is hard to can be much greater in a real situation, but four
choose the base for such triangular trapezoidal following main financial parameters are almost
fuzzy numbers, due to the absence of any evident obligatory for consideration: IRR, NPV, profit-
reasons to prefer some base as the best one. In fact, ability index (PI), payback period (PB). The
the set of linguistic terms such as ‘‘not significant’’, project risk (R) is, generally speaking, a complex
‘‘slightly significant’’, ‘‘noticeably significant’’ and aggregated characteristic estimated on the basis
so on in practice often represents only some labels of the quantitative data as well as expert qualita-
signed on the levels of the decision maker’s tive estimations. Details can be found in Doumpos
preference scale. et al. (2002).
So the membership function shown in Fig. 3 Suppose for the sake of simplicity that in our
seems to be a quite sufficient level of abstraction example the risk is estimated in such a way that it

1


0.5

0
equivalence slight weak noticeable strong
preference preference preference preference
the grade of preference

Fig. 3. The membership functions of qualitative local criterion.


ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 15

Fig. 4. The values of parameters used in illustrative example.

ranges in interval from 0 to 1. All numerical values 3. The ranking of local criteria
characterizing considered projects are presented
in Fig. 4. Since in real decision problems the local criteria
The next step is building of the membership are usually expected to bring different contribu-
functions. In fact, the decision maker has to select tion to the final aggregated estimation of alter-
an appropriate type of the function (see Fig. 2) and natives, the appropriate method for local criteria
choose no more than four reference points on a ranking is needed. It must be emphasized that
scale of the analyzing parameter to define com- simple qualitative ordering of local criteria is not
pletely the corresponding membership function. enough for practical purposes: usually some
One may assign the reference points on the basis of quantitative indices, representing the local criteria
expert evaluations, statistical analysis of similar contributions to the overall alternative estimation,
projects or strong (e.g. banking) standards and so are necessary. Experience shows that quantitative
on. We will try to use the simpler approach, ranking of the criteria is more difficult task for a
because in our case we just need to choose the best decision maker than building of the membership
of four projects. In our example (see Fig. 4) the function. Though it is normally hard for decision
worst value of NPV among all the projects is equal makers to rank the set of local criteria as whole,
to 2500 and the best one is 4000. Therefore, it they usually can confidently state preference, at
seems reasonable to assume for the first reference least verbally, when comparing only a pair of
point x1 ¼ 2000; i.e. less than 2500 because we do criteria. Therefore, a proper criteria ranking
not want to reject the project No 1 when using technique should use this pair comparison in a
some types of local criteria aggregations (see verbal form. Such technique is based on the so-
details in the Section 4). Obviously, for the next called matrix of linguistic pair comparisons,
point it is quite natural to assume x2 ¼ 4000; and suggested by Saaty (1997). The procedure of
to obtain a complete description of the function we building this matrix for our example is illustrated
introduce an auxiliary point x3 ¼ 6000 (see Fig. 5). in Fig. 6.
Other membership functions for considered Of course, only nine basic verbal estimates are in
example are built in a similar way. use. Linguistic scales used in such estimations may
ARTICLE IN PRESS

16 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

Fig. 5. The building of membership function.

Fig. 6. The matrix of pair comparison.

have different sense, but a number of the scale feature of human thinking (Borisov and Korneeva,
levels (linguistic granules) cannot be more than 1980). Nevertheless, to be able to make calcula-
nine in any natural languages: this is an inherent tions some natural numbers are assigned to the
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 17

verbal estimations. Nevertheless, the numbers in  the geometric row means method (GRM),
Fig. 6 are shown only to illustrate our theoretical  the weighted least squares method (WLSM)
considerations. In practice, it is not advised to
and a category of methods that involve only
show any numbers to the experts. More reliable
arithmetic operations:
result can be obtained when only the linguistic
opinions of an expert are taken into consideration.  the row means of normalized columns approach
The fact is that if you propose to a group of (Saaty, 1977),
experts to estimate some known objects, their  the normalized row sum and the inverted
verbally expressed opinions usually are quite column sum methods (Lootsma, 1981).
similar. We cannot expect any other result: these
people learned using the same manuals, read the The relative advantages and drawbacks of these
same articles, worked in the same field. However, methods are reviewed and discussed in literature
if we force them to use some numbers for the not only for the case of crisp aij (Chu et al., 1979),
estimations (usually it is not easy to do, since but when the entries of pair comparison matrix A
nobody loves numbers) we do not obtain any themselves are represented by fuzzy numbers
consensus (Zollo et al., 1999). The matter is that (Chang and Lee, 1995; Mikhailov, 2003).
‘‘In the beginning was the Word’’ (The Holy Nevertheless, in general, since in practice we
Gospel of Jesus Christ). Numbers had appeared deal with an approximate equality (1), the natural
much later and during the last few millennia, criterion of pair comparison efficiency can be
which are a minute from historical viewpoint, presented as
people have not learned to use numbers properly n  
X n X
ai 2
yet. So far we have been thinking using words, not S¼ aij  ; (2)
numbers, and even trying to teach our computers i¼1 j1
aj
this trick.
More strictly, let C i ; i ¼ 1; . . . ; n; be the local where ai are ranks obtained using considered
criteria to be ranked. Then pair comparison matrix method.
A can be created, such that any entry aij 2 A For this reason, it is quite natural to propose a
represents relative preference of the criterion C i method based on the minimization of S, i.e.,
when it is compared with the criterion C j : The pair Xn X n  
ai 2
comparison matrix is reciprocal, meaning that S¼ aij  ! min (3)
aj
aij ¼ 1=aji and aii ¼ 1: If ai and aj represent values i¼1 j1
of ranks, then in a perfect case with a consistent
matrix, aij ¼ ai =aj ; aij ¼ aik akj and ranks ai ; i ¼ s:t:
1; . . . ; n; can be calculated easily. Unfortunately, in X n
real-world situations, we usually only have esti- ai ¼ 1
mates of aij ; and actual value for aij may be i¼1
Xn
unknown. Then the question arises: how to find ai ; ðsometimes restriction ai ¼ n is usedÞ: ð4Þ
i¼1
i ¼ 1; . . . ; n; such that
Using rich experimental data, Wagenknecht and
ai
aij  : (1) Hartmann (1983) have earnestly shown that the
aj method (3)–(4) (LSM) is the best one as it gives the
Various approaches have been proposed to least final values of S. On the other hand, in some
obtain these crisp ranks. In Chang and Lee cases the simplest row means of normalized
(1995), they were roughly classified as follows: columns method (Saaty, 1977)
pffi Qn
½n j¼1 aij
 the eigenvector method, ai ¼ n (5)
P p ffi Qn
 the least squares method (LSM), ½n j¼1 aij
 the logarithmic least squares method (LLSM), i¼1
ARTICLE IN PRESS

18 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

may produce results good from the practical More strictly, let A and B be local criteria and
viewpoint (Wagenknecht and Hartmann, 1983). mA ; mB be their membership functions. Then for
Of course, criteria weights in some special cases each x 2 X ; where X is a set of alternatives,
may be interpreted differently. Some not inex- the artificial functions mA ðxÞ; mB ðxÞ can be for-
haustive classification of such possible interpreta- mally introduced. Of course, x is not a variable
tions and an overview of corresponding ranking in a common sense, factually, it is only a label
methods are presented in Choo et al. (1999). (of number) assigned to the corresponding
However, in our case the method based on alternative.
expressions (3), (4) seems to be completely corres- Hence, we can say that if for some x1 2 X ; we
ponding to the nature of the problem and suffi- have mA ðx1 Þ ¼ mB ðx1 Þ; then the alternative x1
ciently justified mathematically. satisfies the local criteria A and B in equal
extent, and if for some x2 2 X ; we have
mA ðx2 Þ4mB ðx2 Þ; then the alternative x2 satisfies
the local criterion A in a greater degree than
4. Aggregation of local criteria criterion B. In this way the initially multidimen-
sional problem can be formally transformed into
When all the membership functions of the local one-dimensional with the alternative number
criteria are built and corresponding ranks are (label) as the only variable.
calculated, the next step is their aggregation for the
generalized project quality assessment. To do 4.1. Equally ranked local criteria
this, various aggregation methods may be used.
The most
P popular is the simplest additive method To make our consideration more transferable,
D ¼ ni¼1 ai mi ; where ai are ranks of the local let us at first consider the situation when we have
criteria presented by corresponding membership only two local criteria A and B which are equally
functions mi : It is used in many well-known important for a decision maker and therefore have
decision-making models such as the analytic equal ranks, e.g. aA ¼ aB ¼ 1:
hierarchy process (AHP) (Saaty, 1977), Multi-
attribute utility analysis (Pardalos et al., 1995) and (a) So if X is a set of the alternatives and
in specific applications (Büyüközkan and Feyzio- mA ðxÞ; mB ðxÞ; x 2 X are membership func-
ǧlu, 2004; Erol and Ferrell, 2003), but often tions representing formally—as it is described
without any critical analysis. Nevertheless, the above—the local criteria A and B respectively,
problem of choice of appropriate aggregation then the best (optimal) alternative xo will be
method is of perennial interest, because of its such that:mA ðxo Þ ¼ mB ðxo Þ (since criteria A and
direct relevance to practical decision-making B are of equal importance),
(Yager, 1979; Zimmerman and Zysno, 1980; (b) the value mA ðxo Þ is maximal in comparison
Peneva and Popchev, 2003). with all alternatives for which condition (a) is
Although Zimmerman and Zysno (1980) verified.
stated that choice of aggregation scheme is a
context-dependent problem, we want to con- In this spirit, the theorem useful for our further
tribute to its consideration from some other point analysis had been proved by Sevastianov and
of view. Tumanov (1990). It can be formulated as follows:
Firstly, it is possible to represent the member- Theorem 1. If A n B are equally ranked local
ship function of the local criteria as some criteria represented on a set of the alternatives X
artificial functions of alternatives (comparing by corresponding membership functions mA ðxÞ;
investment projects). Such a transformation mB ðxÞ; x 2 X ; such that they have unique maximal
can be carried out formally after the calculation points xA ; xB 2 X respectively, and
of the membership functions values for all
alternatives. mA ðxA Þ4mB ðxA Þ; mB ðxB Þ4mA ðxB Þ; (6)
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 19

then optimal alternative xo can be found as considered). Since subset C (crosshatched region
in Fig. 7) is an intersection of subsets A and B
xo ¼ arg maxx2X ðmC ðxÞÞ; (7)
representing the local criteria, the optimal alter-
native must be a maximum point of function
mC ðxÞ ¼ minðmA ðxÞ; mB ðxÞÞ: (8)
mC ðxÞ; which represents a degree of compromise
between the local criteria. If in a framework of set-
It is easy to see that function mC ðxÞ can be type representation of A, B and C we consider a
naturally treated as an aggregation of local criteria case when A ¼ B; then it is quite natural to expect
A and B. C ¼ A \ B ¼ A \ A ¼ A (this is a simple idempo-
What can be achieved using some other popular tency rule). On the other hand, it is possible only
aggregation methods is shown in Fig. 7. Ob- when mC ðxÞ ¼ mA ðxÞ: It is easy to see that only the
viously, only the min-type aggregation (8) derives min-type aggregation gives us proper result in this
the optimal alternative x2 (see Fig. 7) fulfilling the case. It is important that only min-type aggrega-
natural restriction (a). All other aggregation tion guarantees the idempotency, i.e. A \ A ¼ A:
methods get the optimal alternatives x1 or x3 All other aggregation methods based on arithme-
which is in Pareto region, but far from the actual tical operations, such as addition or multiplication
optimum (see Fig. 7). ones, give us the result A \ AaA that cannot be
Consider an important feature of the min-type explained reasonably. The only exception is the
aggregation (8). Function case mA ðxÞ ¼ 0:5mA ðxÞ þ 0:5mA ðxÞ: In practice we
mC ðxÞ ¼ minðmA ðxÞ; mB ðxÞÞ can meet more complicated situations than con-
sidered above. For example, the membership
may be interpreted as a membership function of
functions can have several points of extreme. This
the set C ¼ A \ B and value mA ðxÞ can be treated
problem can be resolved by clustering a subset of
as a degree to which an alternative x belongs to a
the alternatives into some Pareto regions as it is
subset of alternatives that satisfy the local criterion
shown in Fig. 8. Obviously, within such regions
A (in a similar way the value mB ðxÞ can be
(I, II, III in Fig. 8) all the conditions of Theorem 1
are verified.
Consider a situation when condition (6) of
µA (x) Theorem 1 is violated. In this case (see Fig. 9)
1.0 2 the maximum of the function mC ðxÞ does not
coincide with any of the intersection points of the
0.8 3 µB (x) functions mA ðxÞ; mB ðxÞ:
Observe that in this case condition (a) cannot
0.6 1 be satisfied for optimal alternative but the min-
type aggregation gives us the best decision, as
µ

usual.
0.4
Summarizing, we can state that the min-type
aggregation is the best choice in a case of two
0.2 equally ranked local criteria.

0.0 4.2. The local criteria of different importance


x1 x2 x3
x
Assume that local criteria A and B are of
Fig. 7. The methods of equally ranked local criteria aggrega- different importance for a decision maker, i.e. for
tion: 1mC ðxÞ ¼ mA ðxÞmB ðxÞ; 2  mC ðxÞ ¼ 0:5 mA ðxÞ þ 0:5mB ðxÞ;
their ranks we have aA aaB : Since the additive and
3  mC ðxÞ ¼ maxð0; mA ðxÞ þ mB ðxÞ  1Þ; x1 is the optimal alter-
native for 2 and 3 types of aggregation; x3 is the optimum of 1 multiplicative aggregations mC ðxÞ ¼ aA mA ðxÞ þ
type of aggregation; x2 is the optimal alternative for the aB mB ðxÞ and mC ðxÞ ¼ ðaA mA ðxÞÞðaB mB ðxÞÞ in a case
aggregation mC ðxÞ ¼ minðmA ðxÞ; mB ðxÞÞ: of aA  aB result in inappropriate decisions
ARTICLE IN PRESS

20 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

1
µB (x)

µA (x)

0.5

0
I II III
x

Fig. 8. The case of multi-extreme membership functions.

Theorem 2. Let the local criteria A and B satisfy all


1.0 µA (x)
the conditions of Theorem 1 except the equal
0.8 µB (x) importance condition, and criterion A is more
important than criterion B, i.e. aA 4aB :
0.6
Let x0 be an optimal alternative for a case of the
µ

0.4 equally ranked criteria A and B, i.e. x0 ¼


arg maxx2X ðminðmA ðxÞ; mB ðxÞÞÞ; and x10 be an opti-
0.2 mal alternative for a case of the weighted criteria,
i.e. x10 ¼ arg maxx2X ðminðaA mA ðxÞ; aB mB ðxÞÞÞ:
0.0
x* xj Then alternative x10 satisfies criterion B to a
x greater extent than criterion A, which is in contra-
Fig. 9. The case of condition (6) violation.
diction with an initial assumption for A to be more
important than B.
Proof. Observe that x10 is one of the roots of the
equation
(see Fig. 7), we looked for more correct aggrega-
tion rules. aA mA ðxÞ ¼ aB mB ðxÞ; (10)
Zimmerman (1987) proposed the following which can be rewritten as
aggregation mode:
bmA ðxÞ ¼ mB ðxÞ; (11)
m0C ðxÞ ¼ minðaA mA ðxÞ; aB mB ðxÞÞ: (9)
b ¼ aA =aB 41:
It is easy to see that in asymptotic case aA ¼ If xA ; xB 2 X are unique maximal points of the
aB ¼ 1 expression (9) reduces to the optimal min- membership functions mA ðxÞ; mB ðxÞ respectively,
type aggregation (8) and weighting in (9) appears then without loss of generality we can assume
to be logically justified. that xA oxB : In this case the function mA ðxÞ always
Nevertheless, in practice such aggregation can decreases in interval ½xA ; xB  and mB ðxÞ always
produce completely absurd results. increases in it.
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 21

µA (x) µB (x) the natural inequality mA ðx10 Þ4mB ðx10 Þ always takes
1.0 place.
In a general form expression (12) can be
0.8 µA′ (x) presented as
m0C ðxÞ ¼ ma11 ðxÞ ^ ma22 ðxÞ ^    ^ mann ðxÞ;
0.6
1X n
µ

a1 ; a2 ; :::; an 40; ai ¼ 1; ð13Þ


0.4 n i¼1
µB′ (x)
where 4 is a min-operator and n is a number of
0.2 the local criteria.
Thus, it can be stated that aggregation (12) is the
0.0 best one, but it is true only when the conditions of
x0 x0′
x Theorem 1 are verified. In practice, the situations
presented in Fig. 11 may occur, when the
Fig. 10. The min-type aggregation of ranked local criteria: conditions of Theorem 1 are violated.
m0A ðxÞ ¼ 0:8mA ðxÞ; m0B ðxÞ ¼ 0:2mB ðxÞ:
Obviously, in such a case the use of aggregation
(12) gives us an optimal alternative, which cannot
be considered as satisfactory decision, especially if
Then, taking into account that in Eq. (11), b41; the local criterion A is more important than
we can state that x10 4x0 and so mA ðx10 ÞomB ðx10 Þ: criterion B.
This means that less important criterion B is
satisfied in x10 better, than more important
criterion A. &
The theorem is illustrated in Fig. 10. 1.0
Obviously, in order to obtain qualitatively µA (x)
proper result the most important criterion should
be multiplied by a minimal rank, but such an
approach looks quite artificial and can cause 0.8
certain difficulties when the number of the local
criteria is greater.
An additional drawback of the considered
aggregation is that the general criterion m1C ðxÞ ¼ 0.6
minðaA mA ðxÞ; aB mB ðxÞÞ is not normalized to 1.
µ

Hence, it is difficult to assess the closeness of the


optimal decision to the global optimum.
Yager (1979) proposed the aggregation method, 0.4
which properly reflects the sense of ranking:
µB (x)
m0C ðxÞ ¼ minðmaAA ðxÞ; maBB ðxÞÞ; (12)
0.2
ðaA þ aB Þ=2 ¼ 1:
Observe that for aA ¼ aB ¼ 1 expression (12)
reduces to (8).
It is shown in Yager (1979) that if aA 4aB ; then 0.0
in optimal point x
x10 ¼ arg maxx2X minðmaAA ðxÞ; maBB ðxÞÞ Fig. 11. The violation of Theorem 1’s conditions.
ARTICLE IN PRESS

22 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

1.0 µA (x) µB (x)

µC (x)
0.8 (µA (x)+µB (x)+µC (x))/3
min (µA (x), µB (x), µC (x))
µ (x)

0.6 µA (x)µB (x)µC (x)

0.4

0.2

0.0
x0 x1 x2 x3
x

Fig. 12. The case of three non-ranked local criteria. Fig. 13. Different types of aggregating operator.

Factually, the min-type aggregation, which does of only two local criteria such aggregations
not compensate in considered situation small frequently cannot reveal any preferences in Par-
values of one local criterion by greater values of eto-region (see Fig. 13).
another, does not properly represent contributions The results of theoretical analysis and our
of the local criteria to the overall estimation. practical experience make it possible to state that
Indeed, Theorem 1 is proved only for a case of the most reliable aggregation approach lies in a use
two local criteria. of the min-type operator (13).
The problem already arises in a case of three The multiplicative approach
local criteria. The case of three non-ranked local
criteria mA ðxÞ; mB ðxÞ; mC ðxÞ is shown in Fig. 12. It is Y
n
mC ðxÞ ¼ mai i ðxÞ (14)
hard to say that the alternative x2 is the optimal i¼1
one in a sense of Theorem 1, since there is no point
x 2 X in the Pareto-region such that mA ðxÞ ¼ appears to be somewhat less reliable and, finally,
mB ðxÞ ¼ mC ðxÞ: the additive (weighted sum) method
Moreover, the min-type aggregation sometimes
does not comply with intuitive concepts of decision X
n
mC ðxÞ ¼ ai mi ðxÞ (15)
makers about optimality; see Dubois and Koenig i¼1
(1991).
Therefore, when dealing with a complex task may be considered as unreliable and insensi-
characterized by a great number of local criteria, it tive when choosing an alternative in Pareto-
seems reasonable to use all possible types of region.
aggregations. If the results obtained using different The results of projects estimations for the
aggregation modes are similar, this fact may be example considered in Section 3 using all three
considered as a good confirmation of their aggregating methods are presented in Fig. 14.
optimality. In the opposite case an additional Observe that the min-type and multiplicative
analysis of local criteria and their ranking should aggregations give us similar resulting estimations,
be advised. which are far from those obtained using additive
It is worth noting that addition-type aggrega- aggregations.
tion and the so-called bounded difference aggrega- The proposed approach allows to estimate
tion, maxð0; mA ðxÞ þ mB ðxÞ  1Þ; might be used directly the contribution of each local criterion to
only with a great prudence, since even in a case the generalized project assessment (see Fig. 15).
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 23

Fig. 14. Result estimations of projects presented in Fig. 4 (min-type, multiplicative and additive aggregations are denoted by
D1 ; D2 ; D3 respectively).

Fig. 15. Contributions of local criterion to the final project’s assessment.


ARTICLE IN PRESS

24 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

4.3. The aggregation of aggregation methods with whereas their ranking seems to be a more
the help of level-2 fuzzy sets important issue than choosing g: Finally, the
generalization modes (16)–(18) do not involve all
The natural consequence of problems men- possible approaches to the aggregation.
tioned above is a growing interest to the methods Here we propose a simple but intuitively
of generalization of the aggregating operators obvious and mathematically strong approach to
(aggregation of aggregation modes); see Roubens the aggregation of aggregating methods based on
(1997). the level-2 fuzzy sets.
For this purpose it is proposed to apply the Originally, level-2 fuzzy sets were presented by
possibility theory (Dubois and Koenig, 1991) as Zadeh (1971) and were more elaborately studied
well as the weighted averaging aggregation; see by Gottwald (1979) and in the work of Tre and
Yager (1988). Caluwe ( 2003). As proposed by Zadeh (1974), the
Also, Yager’s t-norms are used in Hauke (1999), level-2 fuzzy set is such a fuzzy set, of which
and the hierarchical aggregation approach is membership grades assigned to the elements of the
developed; see Dyckhoff (1985), Migdalas and universal set are ordinary fuzzy sets.
Pardalos (1996). Nowadays the most popular is Since we deal with restricted number of com-
the so-called g-operator (Zimmerman and Zysno, pared projects and aggregating methods, it is quite
1983). sufficient to consider only discrete representation
!1g !g of level-2 fuzzy set.
Y Y More strictly, if fuzzy set A is defined on a
Z¼ mi 1  ð1  mi Þ ;
i i
discrete set xi, i ¼ 1; . . . ; N; and xi are presented by
ordinary fuzzy sets defined on discrete universe set
i ¼ 1; 2; :::; n; 0pgp1; ð16Þ
zj ; j ¼ 1; . . . ; M; then A is a level-2 fuzzy subset
where mi are membership functions corresponding defined by following expressions:
to local criteria.    
mA ðxi Þ hi ðzj Þ
Since expression (16) is based only on the A¼ ; xi ¼ ; (19)
multiplicative aggregation, more general approach xi zj
has been proposed by Mitra (1988): ( )
!, max½mA ðxi Þhi ðzj Þ
i
X A¼ ;
Zor ¼ g maxðmi Þ þ ð1  gÞ mi n; (17) zj
i
i
i ¼ 1; . . . ; N; j ¼ 1; . . . ; M: ð20Þ
!,
X It can be seen when analyzing expression (20)
Zand ¼ g minðmi Þ þ ð1  gÞ mi n: (18) that final grade of membership of zj to the set A
i
i
may be presented as
Expressions (17), (18) have been used in Shih
mA ðzj Þ ¼ maxi ½mA ðxi Þhi ðzj Þ; j ¼ 1; . . . ; M: (21)
and Lee (2000) to solve the multiple-level decision-
making problem. As a key issue, the lack of strong Suppose there are sets of projects zj ; j ¼
rules for choosing values of g is mentioned. 1; . . . ; M; and N types of aggregating operators
The work of Choi and Oh (2000) is specifically Di ; i ¼ 1; . . . ; N; under consideration.
devoted to this problem, but the method proposed Since it is possible to estimate the relative
by the authors for choosing g demands too much reliability of aggregation modes on a verbal level,
additional information to be presented by the as it has been done above, it is natural to introduce
decision maker in a quantitative form. In practice some membership function, mðDi Þ; i ¼ 1; . . . ; N;
it is hard to obtain such information since it is not representing expert’s opinions about closeness of
directly related to the decision maker’s real considering aggregating operators Di to some
problems. It is easy to see that local criteria in perfect type of aggregation, which can be treated
(17), (18) are considered to be non-ranked, as the best one or ‘‘ideal’’ method of aggregation.
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 25

Then such ‘‘ideal’’ method, Dopt ; can be Using the method presented in Section 3 from
represented through its membership function and this matrix we obtain
a set of compared aggregation modes as in the
mðD1 Þ ¼ 0:7; mðD2 Þ ¼ 0:25; mðD3 Þ ¼ 0:05:
following fuzzy set:
  Thus, ‘‘ideal’’ method of aggregation in our case
mðDi Þ
Dopt ¼ ; i ¼ 1; . . . ; N: (22) can be presented as
Di    
mðD1 Þ mðD2 Þ mðD3 Þ 0:7 0:25 0:05
In turn, each Di can be described formally by a Dopt ¼ ; ; ¼ ; ; ;
D1 D2 D3 D1 D2 D3
fuzzy set of considered projects, zj ; j ¼ 1; . . . ; M;
on which it is, factually, defined and for which the where
aggregated estimations Di ðzj Þ have been calcu- 
Di ðProject_1Þ Di ðProject_2Þ Di ðProject_3Þ
lated. It is clear that the values Di ðzj Þ may be Di ¼ ; ; ;
treated as a degree to which project zj satisfies the Project_1 Project_2 Project_3

aggregated criterion Di or as an extent to which Di ðProject_4Þ
; i ¼ 1; 2; 3:
project zj belongs to a set of projects satisfying Di : Project_4
Then
Hence, Dopt can be presented as
  
Di ðzj Þ mopt ðProject_1Þ mopt ðProject_2Þ
Di ¼ ; i ¼ 1; . . . ; N; j ¼ 1; . . . ; M: Dopt ¼ ; ;
zj Project_1 Project_1
(23) 
mopt ðProject_3Þ mopt ðProject_4Þ
; ;
Substituting (22) into (23) we have Project_3 Project_4
 
mopt ðzj Þ where
Dopt ¼ ; j ¼ 1; . . . ; M; (24)
zj mopt ðProject_jÞ ¼ maxi ðmðDi ÞDi ðProject_jÞÞ;
where i ¼ 1; 2; 3:
mopt ðzj Þ ¼ max½mðDi ÞDi ðzj Þ: (25) Finally, after all calculations using the results
i
from Fig. 14 we obtain
It is clear that the best project can be found as  
0:18 0:05 0:42 0:31
zopt ¼ arg max mopt ðzj Þ: (26) Dopt ¼ ; ; ; :
j Project_1 Project_2 Project_3 Project_4
To illustrate, let us continue the consideration of So, the third project is the best one. Factually, we
the example. can infer the same result analyzing the data
At first, we have to calculate the values mðDi Þ; presented in Fig. 14, but it would be hard to do
i ¼ 1; 2; 3: As it has been stated above, the min- for a greater number of projects or/and aggregat-
type aggregation, D1 is more reliable than the ing operators under consideration.
multiplicative aggregation, D2 ; and both are
noticeably more reliable than the additive aggre-
gation D3 : 5. Hierarchical system of criteria
Such linguistic assessments, in accordance with
the method described in Section 3, can be As it has been noticed above, many of the real-
represented by following pair comparison matrix: life problems of investment (and other) project
estimations are not only multi-criteria but also
D1 D2 D3
multi-level (hierarchical) ones.
D1 1 3 9 The method presented in previous sections
D2 1/3 1 9 allows to build, in a natural way the branched
D3 1/9 1/9 1 hierarchical structures as in Fig. 16. It can be seen
that each criterion of upper k level is built on the
ARTICLE IN PRESS

26 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

Dn

Dn-1,1 Dn-1,2 Dn-1,in-1 Dn-1,mn-1

Dn-2,in-1,1 Dn-2,in-1,2 Dn-2,in-1,in-2

Dn-3,in-1,in-2,1

Fig. 16. The hierarchical structure of local criteria.

basis of local criteria of underlying (k1) levels functions representing the local criteria based on
using one of the aggregation methods or their origin parameters of project’s quality are used, i.e.
generalization as proposed in Section 4. The
general expression for calculation of criteria on D1;in1 ;in2 ;...;i1
intermediate levels of hierarchy is as follows: ¼ f 1;in1 ;in2 ;...;i1 ðm0;in1 ;in2 ;...;i1 ;1 ; a0;in1 ;in2 ;...;i1 ;1 ; . . . ;
Dk;in1 ;in2 ;...;ik m0;in1 ;in2 ;...;i1; m1;i ; a0;in1 ;in2 ;...;ik; m1;in1 ;in2 ;...;i1 Þ;
n1 ;in2 ;...;i1

¼ f k;in1 ;in2 ;...;ik ðDk1;in2 ;...;ik ;1 ; ak1;in2 ;...;ik ;1 ; . . . ;


where m1;in1 ;in2 ;...;i1 —the number of initial local
Dk1;in1 ;in2 ;...;ik ;mk;in1 ;in2 ;...;ik ; criteria on the lowest level.
ak1;in1 ;in2 ;...;ik ;mk;in1 ;in2 ;...;ik Þ; The method for building of hierarchical local
criteria systems is based on the multiple criteria
where f k;in1 ;in2 ;...;ik is an operator of criteria approach described in previous sections and
aggregation, mk;in1 ;in2 ;...;ik —the number of local generalizes it. The method is developed in a form
criteria on (k1) level, aggregating to intermediate of user-friendly software (Fig. 17).
local criterion Dk;in1 ;in2 ;...;ik : It is important that the presented method not
It is worthy to note that values Dn1;in1 always only gives us some generalized quantitative esti-
are in interval [0,1] and may be interpreted as some mation of projects as a whole but makes it possible
intermediate local criteria assessments. On the to assess the contribution of each local criterion in
lowest level of hierarchy the initial membership this final estimation.
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 27

Fig. 17. Building of hierarchical local criteria system.

6. Conclusion of project’s local criteria. The theoretical consid-


eration is illustrated with simple examples and
The paper presents the generalized method for their realization by author’s software.
multiple criteria hierarchical estimation of invest- The main direction of future research will be
ment projects in fuzzy setting. The key issue is the generalization of proposed approach using the
analysis of the familiar approaches to aggregation concept of hyperfuzzy sets (Dymova et al., 2002),
of local criteria. The problems of ranked local which being the useful particle case of type-2 fuzzy
criteria aggregation are analyzed and some new sets for representation of verbally formulated local
theoretical results, which can be useful for proper criteria and parameters of project’s quality.
choice of aggregation method, are presented. It is
proved that most popular weighted sum method is
the most unreliable one and can provide wrong
results. All the known aggregation modes have References
their own advantages and drawbacks and it seems
Borisov, A.N., Korneeva, G.V., 1980. Linguistic approach to
impossible to choose the best one, especially when decision making model building under uncertainty: Meth-
dealing with complicated hierarchical problem. ods of decision making under uncertainty. Riga 4–6
Therefore, the new approach which makes it (in Russian).
possible to generalize the aggregating schemes Büyüközkan, G., Feyzioǧlu, O., 2004. A fuzzy-logic-based
into some ‘‘ideal’’ criterion is elaborated using the decision-making approach for new product development.
International Journal of Production Economics 90, 27–45.
mathematical tools of level-2 fuzzy sets. It is Chang, P.-T., Lee, E.S., 1995. The estimation of normalized
shown that proposed method allows to build in a fuzzy weights. Computers and Mathematics with Applica-
natural way the branched hierarchical structures tions 29, 21–42.
ARTICLE IN PRESS

28 L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29

Choi, D.Y., Oh, K.W., 2000. Asa and its application to multi- Migdalas, A., Pardalos, P.M. (Eds.), 1996. Editorial: Hier-
criteria decision making. Fuzzy Sets and Systems 114, archical and bilevel programming, Journal of Global
89–102. Optimization 8, 209–215.
Choo, E.U., Schoner, B., Wedley, W.C., 1999. Interpretation of Mikhailov, L., 2003. Deriving priorities from fuzzy pairwise
criteria weights in multicriteria decision making. Computers comparison judgments. Fuzzy Sets and Systems 134,
and Industrial Engineering 37, 527–541. 365–385.
Chu, A., Kalaba, R., Springarn, R., 1979. A comparison of two Miller, G.A., 1956. The magical number seven plus or minus
methods for determining the weights of belonging to fuzzy two: Some limits on our capacity for processing informa-
sets. Journal of Optimization Theory and Applications 27, tion. Psychological Review 63, 81–97.
531–538. Milner, P.M., 1970. Physiological Psychology. Holt, New York.
Dimova, L., Sevastianov, D., Sevastianov, P., 2000. Applica- Mitra, G., 1988. Mathematical Models for Decision Support.
tion of fuzzy sets theory, methods for the evaluation of Springer, Berlin.
investment efficiency parameters. Fuzzy Economic Review Mohamed, S., McCowan, A.K., 2001. Modelling project
5, 34–48. investment decisions under uncertainty using possibility
Doumpos, M., Kosmidou, K., Baourakis, G., Zopounidis, C., theory. International Journal of Project Management 19,
2002. Credit risk assessment using a multicriteria hierarch- 231–241.
ical discrimination approach: A comparative analysis. Pardalos, P.M., Siskos, Y., Zopounidis, C., 1995. Advances in
European Journal of Operational Research 138 (2), Multicriteria Analysis. Kluwer Academic Publishers, Dor-
392–412. drecht.
Dubois, D., Koenig, J.L., 1991. Social choice axioms for fuzzy Peneva, V., Popchev, I., 2003. Properties of the aggregation
set aggregation. Fuzzy Sets and Systems 43, 257–274. operators related with fuzzy relations. Fuzzy Sets and
Dyckhoff, H., 1985. Basic concepts for theory of evaluation: Systems 139, 615–633.
Hierarchical aggregation via autodistributive connectives in Roubens, M., 1997. Fuzzy sets and decision analysis. Fuzzy
fuzzy set theory. European Journal of Operational Research Sets and Systems 90, 199–206.
20, 221–233. Saaty, T., 1977. A scaling method for priorities in hierarchical
Dymova, L., 2003. A constructive approach to managing fuzzy structures. Journal of Mathematical Psychology 15,
subsets of type 2 in decision making. TASK Quarterly 7, 234–281.
157–164. Sevastianov, P., Tumanov, N., 1990. Multi-criteria identifica-
Dymova, L., Róg, P., Sevastianov, P., 2002. Hyperfuzzy tion and optimization of technological processes. Science
estimations of financial parameters. Proceedings of the and Engineering, Minsk (in Russian).
Second International Conference on Mathematical Meth- Shih, H.S., Lee, E.S., 2000. Compensatory fuzzy multiple level
ods in Finance and Econometrics, pp. 78–84. decision making. Fuzzy Sets and Systems 114, 71–87.
Erol, I., Ferrell Jr., W.G., 2003. A methodology for selection Steuer, R.E., Na, P., 2003. Multiple criteria decision making
problems with multiple conflicting objectives and both combined with finance. A categorical bibliographic study.
qualitative and quantitative criteria. International Journal European Journal of Operational Research 150, 496–515.
of Production Economics 86, 187–199. The Holy Gospel of Jesus Christ, According to St. John 1:1.
Gottwald, S., 1979. Set theory for fuzzy sets of higher level. Tre, G., Caluwe, R., 2003. Level-2 fuzzy sets and their
Fuzzy Sets and Systems 2, 125–151. usefulness in object-oriented database modeling. Fuzzy Sets
Hauke, W., 1999. Using Yager’s t-norms for aggregation of and Systems 140, 29–49.
fuzzy intervals. Fuzzy Sets and Systems 101, 59–65. Wagenknecht, M., Hartmann, K., 1983. On fuzzy rank
Kahraman, C., Ruan, D., Tolga, E., 2002. Capital budgeting ordering in polyoptimisation. Fuzzy Sets and Systems 11,
techniques using discounted fuzzy versus probabilistic cash 253–264.
flows. Information Sciences 142, 57–76. Wang, W.-C., 2004. Supporting project cost threshold decisions
Kuchta, D., 2000. Fuzzy capital budgeting. Fuzzy Sets and via a mathematical cost model. International Journal of
Systems 111, 367–385. Project Management 22, 99–108.
Li, Q., Sterali, H.D., 2003. An approach for analyzing foreign Ward, T.L., 1985. Discounted fuzzy cash flow analysis.
direct investment projects with application to China’s Proceeding of the 1985 Fall Industrial Engineering Con-
Tumen River Area development. Computers and Opera- ference, pp. 476–481.
tions Research 3, 1467–1485. Weck, M., Klocke, F., Schell, H., Rüenauver, E., 1997.
Lootsma, F.A., 1981. Performance evaluation of non-linear Evaluating alternative production cycles using the extended
optimization methods via multi-criteria decision analysis fuzzy AHP method. European Journal of Operational
and via linear model analysis. In: Powell, M.J.D. (Ed.), Research 100, 351–366.
Nonlinear Optimization, pp. 419–453. Yager, R., 1979. Multiple objective decision-making using fuzzy
Lopes, M.D.S., Flavel, R., 1998. Project appraisal-a framework sets. International Journal of Man–Machine Studies 9,
to assess non-financial aspects of projects during the project 375–382.
life cycle. International Journal of Project Management 16, Yager, R., 1980. A foundation for a theory of possibility.
223–233. Journal of Cybernetics 10, 177–209.
ARTICLE IN PRESS

L. Dimova et al. / Int. J. Production Economics 100 (2006) 10–29 29

Yager, R., 1988. On ordered weighted averaging aggregation Zimmerman, H.J., 1987. Fuzzy Sets, Decision-Making and
operators in multicriteria decision making. IEEE Transac- Expert Systems. Kluwer Academic Publishers, Dordrecht.
tions on Systems Man and Cybernetics 18, 183–190. Zimmerman, H.J., Zysno, P., 1980. Latent connectives in
Zadeh, L.A., 1965. Fuzzy sets. Information and Control 8, human decision making. Fuzzy Sets and Systems 4, 37–51.
338–358. Zimmerman, H.J., Zysno, P., 1983. Decision and evaluations
Zadeh, L.A., 1971. Quantitative fuzzy semantics. Information by hierarchical aggregation of information. Fuzzy Sets and
Sciences 3, 177–200. Systems 104, 243–260.
Zadeh, L.A., 1974. Fuzzy logic and its application to Zollo, G., Iandoli, L., Cannavacciuolo, A., 1999. The
approximate reasoning. Information Processing 74, performance requirements analysis with fuzzy logic. Fuzzy
591–594. Economic Review 4, 35–69.

You might also like