E½n¼
Z
1
0
Crfn P rgdr
Z
0
1
Crfn 6 rgdr ð2Þ
provided that at least one of the two integrals is finite.
If n
i
, i =1,2,...,n, are fuzzy variables, then we call them mutu-
ally independent fuzzy variables if and only if (Liu & Gao, 2007)
Crfn
i
2 B
i
; i ¼ 1; 2; ...; ng¼min
16i6n
Crfn
i
2 B
i
gð3Þ
for any subsets B
1
,B
2
,...,B
n
of R.
In this article, we adopt the concept of fuzzy random variable
defined in Liu and Liu (2003) for portfolio selection problems. For
convenience, we give its definition as follows.
Let (
X
,
R
,Pr) be a probability space and F
n
v
a collection of n-
dimensional fuzzy vectors. A map n ¼ðn
1
; ...; n
n
Þ
T
:
X
!F
n
v
is
called an n-dimensional fuzzy random vector if for any Borel sub-
set B of R
n
, the function Cr{n(
x
) 2 B} is measurable with respect to
x
.Ifn = 1, then n is called a fuzzy random variable.
The various measurability criteria for fuzzy random variables
were discussed in Feng and Liu (2006), which provided various
methods to check if a map from
X
to F
v
is a fuzzy random variable.
From the definition above, we can see that if n is a fuzzy random
variable, then for any r 2 R; Crfnð
x
Þ P rg is measurable with re-
spect to
x
, which leads to the following definition.
Let n be an n-dimensional fuzzy random vector. For any r 2 R,
the chance distribution of a fuzzy random event {n P r} was de-
fined as (Liu & Liu, 2005)
G
n
ðrÞ¼Chfn P rg¼
Z
X
Crfnð
x
Þ P rgPrðd
x
Þ: ð4Þ
On the basis of chance distribution (4), the expected value of a
fuzzy random variable n can be represented as (Liu & Liu, 2005)
E½n¼
Z
1
0
Chfn P r gdr
Z
0
1
Chfn 6 rgdr ð5Þ
provided that at least one of the two integrals is finite.
Let n be a fuzzy random variable with a finite expected value
E[n]. Then its variance was defined as (Liu & Liu, 2003)
V½n ¼E½ðn E½nÞ
2
: ð6Þ
In this article, we adopt the following convergence mode for a
sequence of fuzzy random variables (Liu, Liu, & Gao, 2009).
Let G
n
n
ðtÞ and G
n
(t) be chance distributions of fuzzy random
variables n
n
and n, respectively. The sequence {n
n
} of fuzzy random
variables is said to converge in distribution to n, denoted by n
n
!
d:
n,
if G
n
n
ðtÞ!G
n
ðtÞ for all continuity points t of distribution function
G
n
(t).
3. Analytical expressions of variance and chance distribution
In this section, we consider the case when the returns are char-
acterized by trapezoidal fuzzy random variables, and derive some
useful computational formulas of variance and chance distribution.
In the next section, these formulas will be used to derive the equiv-
alent stochastic programming problems of C–V models.
Let n be a trapezoidal fuzzy random variable such that for each
x
, n(
x
)=(X(
x
) d, X(
x
)
a
, X(
x
)+
a
, X(
x
)+b) is a trapezoidal
fuzzy variable with the following possibility distribution
l
nð
x
Þ
ðxÞ¼
xXð
x
Þþd
d
a
; if Xð
x
Þd 6 x < Xð
x
Þ
a
1; if Xð
x
Þ
a
6 x < Xð
x
Þþ
a
xþXð
x
Þþb
b
a
; if Xð
x
Þþ
a
6 x < Xð
x
Þþb
0; otherwise;
8
>
>
>
>
>
>
<
>
>
>
>
>
>
:
where d >
a
>0,b >
a
> 0, and X is a random variable. Then for each
x
, according to (2), we have E[n(
x
)] = (4X(
x
) d + b)/4. Moreover,
by (5), we have E[n]=(4E[X] d + b)/4.
Given a random event
x
, we now compute the quadratic devi-
ation of fuzzy variable (n(
x
) E[n])
2
, and the computational re-
sults are summarized in the following lemma.
Lemma 3.1. Let n be a trapezoidal fuzzy random variable such that
for each
x
, n(
x
) = (X(
x
) b,X(
x
)
a
,X(
x
)+
a
,X(
x
)+b), and
denote m = E[n].
(i) If X(
x
) 6 m b, then
E½ðnð
x
ÞmÞ
2
¼X
2
ð
x
Þ2mXð
x
Þþm
2
þ
a
2
þ b
2
þ
a
b
3
:
(ii) If m b < X(
x
) 6 m
a
, then
E½ðnð
x
ÞmÞ
2
¼
1
6ðb
a
Þ
½X
3
ð
x
Þ3ðb 2
a
þ mÞX
2
ð
x
Þþð3ðb þ mÞ
2
12
a
mÞXð
x
Þðb þ mÞ
3
þ 2
a
3
þ 6
a
m
2
:
(iii) If m
a
< X(
x
) 6 m, then
E½ðnð
x
ÞmÞ
2
¼X
2
ð
x
Þ
b
a
2
þ 2m
Xð
x
Þþm
2
þ
b
a
2
m þ
4
a
2
þ
a
b þ b
2
6
:
(iv) If m < X(
x
) 6 m+
a
, then
E½ðnð
x
ÞmÞ
2
¼X
2
ð
x
Þþ
b
a
2
2m
Xð
x
Þþm
2
b
a
2
m þ
4
a
2
þ
a
b þ b
2
6
:
(v) If m +
a
< X(
x
) 6 m+b, then
E½ðnð
x
ÞmÞ
2
¼
1
6ðb
a
Þ
½X
3
ð
x
Þþ3ðb 2
a
mÞX
2
ð
x
Þ
þð3ðb mÞ
2
þ 12
a
mÞXð
x
Þþðb mÞ
3
2
a
3
6
a
m
2
:
(vi) If X(
x
) > m+b, then
E½ðnð
x
ÞmÞ
2
¼X
2
ð
x
Þ2mXð
x
Þþm
2
þ
a
2
þ b
2
þ
a
b
3
:
Proof. According to (1) and (2), the assertions of the lemma can be
checked easily. h
In this article, we use X Nða;
r
2
Þ to represent X is a normal
random variable whose probability density function f
X
(t)is
Nða;
r
2
Þ, where a and
r
2
are the expected value and variance of
X, respectively. Particularly, if a = 0 and
r
= 1, then X Nð0; 1Þ
means that X a standard normal random variable whose probabil-
ity density function and probability distribution function are usu-
ally denoted by /(t) and
U
(t), respectively.
As a consequence of Lemma 3.1, when X is a normal random
variable, the following theorem gives the variance formula of a
trapezoidal fuzzy random variable.
Theorem 3.1. Let n be a trapezoidal fuzzy random variable such that
for each
x
, n(
x
) = (X(
x
) b,X(
x
)
a
,X(
x
)+
a
,X(
x
)+b). If
X Nða;
r
2
Þ, then
6516 Y. Liu et al. / Expert Systems with Applications 39 (2012) 6514–6526