http://ijmms.hindawi.com
© Hindawi Publishing Corp.
NULL DISTRIBUTION OF MULTIPLE CORRELATION COEFFICIENT UNDER MIXTURE NORMAL MODEL
HYDAR ALI and DAYA K. NAGAR Received 14 April 2001
The multiple correlation coefficient is used in a large variety of statistical tests and regres- sion problems. In this article, we derive the null distribution of the square of the sample multiple correlation coefficient,R2, when a sample is drawn from a mixture of two multi- variate Gaussian populations. The moments of 1−R2and inverse Mellin transform have been used to derive the density ofR2.
2000 Mathematics Subject Classification: 62H10, 62H15.
1. Introduction. Suppose thatx(p×1),µ(p×1), andΣ(p×p) >0 are partitioned as x=x
1 x(2)
,µ=µ
µ(2)1
, andΣ=σ
11σ21 σ21Σ22
, wherex(2)=(x2, . . . , xp)andµ(2)=(µ2, . . . , µp) are(p−1)×1 andΣ22is(p−1)×(p−1), so that Var(x1)=σ11, Cov(x(2))=Σ22, and σ12 is the (p−1)×1 vector of covariances betweenx1 andx2, . . . , xp. The multiple correlation coefficient betweenx1andx(2), denoted by ¯R1·2···p, is defined as
R¯1·2···p=
σ21Σ−221σ21
σ11
1/2
. (1.1)
LetAbe the sample sum of squares and products matrix formed fromNindepen- dent observations onx. PartitionAasA=a
11 a21 a21 A22
, whereA22is(p−1)×(p−1). The sample multiple correlation coefficient betweenx1andx(2)is defined by
R=a21A−122a21
a11
1/2
. (1.2)
It is well known that, when the underlying population is normal, the random matrixA has Wishart distribution withn=N−1 degrees of freedom and parameter matrixΣ.
Further, ¯R1·2···p=0 if and only ifx1is independent ofx(2)=(x2, . . . , xp). Furthermore, when the population multiple correlation coefficient ¯R1·2···pis zero, the distribution ofR2is beta with parameters(1/2)(p−1)and(1/2)(N−p).
In practice, it is often the case that the random variables are not normally dis- tributed. When such is the case, how would the departure from the normality affect the conventional inference procedure? Specifically, one may wonder what would be the sampling distributions of some commonly used statistics? For providing some answers to the above questions, Srivastava and Awan [9] and Tan [11] derived the distribution of the sample sum of squares and products matrix when sampling from a mixture of two multivariate normal distributions. The normal mixture is defined as follows:
f (x)=Np
µ1,Σ;x
+(1−)Np
µ2,Σ;x
, x∈Rp, (1.3)
where Np(µ,Σ;x)
=(2π )−(1/2)pdet(Σ)−1/2exp
−1
2(x−µ)Σ−1(x−µ)
, x∈Rp,µ∈Rp,Σ>0, (1.4) and 1− is known as the degree of contamination. This model is very common in medical, biological, and agricultural experiments (Titterington et al. [12]). For results on the distribution theory and robustness studies of certain test statistics when sam- pling from a mixture normal model, see Srivastava [8], Srivastava and Awan [9,10], Kabe and Gupta [5], Amey and Gupta [2], and Nagar and Castañeda [7].
Srivastava [8], using certain transformations, derived the null distribution of multi- ple correlation coefficient when sampling from a mixture of two multivariate normal distributions (see also Gupta and Kabe [3]). Amey [1] integrated the joint density of a11,a21, andA22suitably to derive the density ofR2and studied its robustness.
In this article, we derive the null distribution ofR2when sampling from a mixture of two multivariate normal distributions. First, we derive thehth null moment of 1−R2. Then, by using the inverse Mellin transform, the density of 1−R2is obtained from which the density ofR2is deduced.
Note thatR2is a function of the elements of sample sum of squares and products matrixA. Therefore, in our derivation, we use the distribution of Awhen sampling from the above model. Srivastava and Awan [9] and Tan [11] have shown that the density ofA, when sampling from (1.3), is a binomial sum of linear noncentral Wishart densities:
f (A)=
N γ=0
N γ
γ(1−)N−γWp
n,Σ, c2γΣ−1νν;A
, (1.5)
wheren=N−1,cγ2=γ(N−γ)/N, andν=(µ1−µ2). HereWp(n,Σ, cγ2Σ−1νν;A)rep- resents the noncentral Wishart density withndegrees of freedom and noncentrality parameter matrixc2γΣ−1ννdefined by
Kp(n,Σ,ν)etr
−1 2Σ−1A
det(A)(1/2)(n−p−1) 0F1(p)
1 2n;1
4cγ2Σ−1AΣ−1νν
, (1.6) where
Kp(n,Σ,ν)=
2(1/2)pnΓp
1 2n
det(Σ)(1/2)n −1
etr
−1
2cγ2Σ−1νν
(1.7) andΓm(a)=π(1/4)m(m−1)m
j=1Γ(a−(1/2)(j−1)).
2. Null moments of 1−R2. In this section, we derive moments of 1−R2 when R¯1·2···p=0 (or equivalently σ21=0). LetΣ0=σ11 0
0 Σ22
andU=1−R2. Sincea11 is scalar, then
U=1−R2=1−a21A−221a21
a11 = det(A) a11det
A22. (2.1)
Thehth null moment ofUis given by E
Uh
=
N γ=0
N γ
γ(1−)N−γEγ
Uh
, (2.2)
Eγ Uh
=Kp
n,Σ0,ν
A>0etr
−1 2Σ−01A
a−h11det A22−h
×det(A)(1/2)(n−p−1)+h 0F1(p)
1 2n;1
4cγ2Σ−10 AΣ−10 νν
dA.
(2.3)
Replacinga−11hand det(A22)−hby their integral representations, namely a−11h= 1
2hΓ(h) ∞
0 exp
−1 2a11y1
y1h−1dy1, Re(h) >0, det
A22
−h
= 1
2(p−1)hΓp−1(h)
Y22>0etr
−1 2A22Y22
×det Y22
h−(1/2)(p−1+1)
dY22, Re(h) >1 2(p−2),
(2.4)
respectively, in (2.3) and integratingA, the moment expression is rewritten as Eγ
Uh
=2(1/2)npKp
n,Σ0,ν Γp
(1/2)n+h Γ(h)Γp−1(h)
× ∞
0 y1h−1
Y22>0det Y22
h−(1/2)p
det
Σ−01+Y−(1/2)n−h
×1F1(p) 1
2n+h;1 2n;1
2cγ2Σ−10
Σ−10 +Y−1
Σ−10 νν
dy1dY22,
(2.5)
whereY=y1 0 0 Y22
and1F1(p)is the confluent hypergeometric function of matrix argu- ment (Gupta and Nagar [4]). Since rank(Σ−10 (Σ−10 +Y )−1Σ−10 νν)=1, the only nonzero characteristic root of the matrixΣ−01(Σ−01+Y )−1Σ−01ννis tr((Σ−01+Y )−1Σ−01ννΣ−01) and therefore,
1F1(p) 1
2n+h;1 2n;1
2cγ2Σ−01
Σ−01+Y−1
Σ−01νν
=1F1
1 2n+h;1
2n;1 2cγ2tr
Σ−01+Y−1
Σ−01ννΣ−01
,
(2.6)
where1F1is the confluent hypergeometric function of scalar argument (see [6]). Substi- tuting (2.6) in (2.5) and expanding1F1in series form, the moment expression simplifies to
Eγ Uh
=2(1/2)npKp
n,Σ0,ν Γp
(1/2)n+h Γ(h)Γp−1(h)
∞ t=0
cγ2 2
t
(1/2)n+h t
(1/2)n
tt!
× ∞
0
y1h−1
Y22>0
det
Y22h−(1/2)pdet
Σ−10 +Y−(1/2)n−h
× νΣ−10
Σ−10 +Y−1Σ−10 νt
dy1dY22,
(2.7)
where(a)r=a(a+1)···(a+r−1)and(a)0=1. Noting thatΣ0is a block diagonal matrix, we obtain
νΣ−01
Σ−01+Y−1Σ−01νt
= ν12σ11−1
1+σ11y1−1+ν2Σ−122
Σ−122+Y2−1Σ−122ν2t
=
k+=t
t!
k!!
ν12σ11−1
1+σ11y1−1k ν2Σ−122
Σ−122+Y22−1Σ−122ν2
,
det
Σ−10 +Y
= σ11−1
1+σ11y1 det
Σ22−1det
Ip−1+Σ22Y22 .
(2.8)
Now substituting (2.8) in (2.7), we have Eγ
Uh
=2(1/2)npKp
n,Σ0,ν
Γp((1/2)n+h) det(Σ0)(1/2)n+hΓ(h)Γp−1(h)
×
∞ t=0
cγ2 2
t
(1/2)n+h t
(1/2)n
t k+=t
1 k!!
ν12 σ11
k
× ∞
0 y1h−1
1+σ11y1−((1/2)n+h+k)
dy1
×
Y22>0det Y22
h−(1/2)p
det
Ip−1+Σ22Y22
−(1/2)n−h
× ν2Σ−221
Σ−221+Y22
−1
Σ−221ν2
dY22.
(2.9)
SubstitutingZ=(Ip−1+Σ1/222Y22Σ1/222 )−1, the integral involvingY22is evaluated as
Y22>0det Y22
h−(1/2)p
det
Ip−1+Σ22Y22
−(1/2)n−h ν2Σ−221
Σ−221+Y22
−1
Σ−221ν2
dY22
=det Σ22−h
0<Z<Ip−1det(Z)(1/2)(n−p)
×det
Ip−1−Zh−(1/2)p
ν2Σ−221/2ZΣ−221/2ν2
dZ
=det
Σ22−h ∂
∂η
η=0
0<Z<Ip−1
det(Z)(1/2)(n−p)det
Ip−1−Zh−(1/2)p
×etr
ην2Σ−221/2ZΣ−221/2ν2 dZ
=det
Σ22−hΓp−1
(1/2)n Γp−1(h) Γp−1
(1/2)n+h ∂
∂η
η=01F1(p−1) 1
2n;1
2n+h;ηΣ−122ν2ν2
=det Σ22
−hΓp−1
(1/2)n Γp−1(h) Γp−1
(1/2)n+h
(1/2)n
(1/2)n+h
ν2Σ−221ν2
,
(2.10) where1F1(p−1)is the confluent hypergeometric function of matrix argument (see [4]).
Collecting terms containingy1and integrating, we obtain ∞
0 y1h−1
1+σ11y1
−((1/2)n+h+k)
dy1=σ11−hΓ (1/2)n
Γ(h) Γ
(1/2)n+h
(1/2)n k
(1/2)n+h
k
. (2.11) Substituting (2.10), (2.11), and (1.7) in (2.9) and simplifying the resulting expression using results on gamma function, we get
Eγ
Uh
=exp
−1
2cγ2νΣ−01ν
Γ (1/2)n Γ
(1/2)(n−p+1) ∞
t=0+k=t
cγ2 2
t (1/2)n
k
(1/2)n
(1/2)n
t
×
ν12/σ11
k
ν2Σ−221ν2
k!!
Γ
(1/2)(n−p+1)+h Γ
(1/2)n+t+h Γ
(1/2)n+k+h Γ
(1/2)n++h
=exp
−1
2cγ2νΣ−10 νΓ
(1/2)n Γ
(1/2)(n−p+1)+h Γ
(1/2)n+h Γ
(1/2)(n−p+1) ∞
k=0
cγ2 2
k
×
ν12/σ11
k
k! 2F2
1
2n+h+k,1 2n;1
2n+k,1 2n+h;1
2cγ2ν2Σ−122ν2
,
(2.12) where2F2is the generalized hypergeometric function of scalar argument (see [6]).
3. Distribution ofR2under mixture normal model. The density functionf (u)of U=1−R2is obtained by taking the inverse Mellin transform ofE(Uh)as
f (u)=
N γ=0
N γ
γ(1−)N−γfγ(u) (3.1) with
fγ(u)=(2π ι)−1
CEγ
Uh
u−h−1dh, 0< u <1, (3.2) whereι=√
−1 andCis a suitable contour. Substituting (2.12) in (3.2), we obtain fγ(u)=exp
−1
2cγ2νΣ−01ν
Γ
(1/2)n Γ
(1/2)(p−1) Γ
(1/2)(n−p+1) ∞
t=0k+=t
cγ2 2
t
×
(1/2)n
k
(1/2)n
(1/2)n
t
ν12/σ11
ν2Σ−221ν2
k
k!! u(1/2)n+k+−1(1−u)(1/2)(p−3)
×2F1
1
2(p−1)+k,1
2(p−1)+;1
2(p−1); 1−u
, 0< u <1,
(3.3) where2F1is the Gauss hypergeometric function (see [6]). To obtain (3.3) we have used the result
1 0
u(1/2)n+h+k+−1(1−u)(1/2)(p−3)2F1
1
2(p−1)+k,1
2(p−1)+;1
2(p−1); 1−u
du
=Γ
(1/2)(p−1) Γ
(1/2)(n−p+1)+h Γ
(1/2)n+t+h Γ
(1/2)n+k+h Γ
(1/2)n++h .
(3.4)
The density ofR2=1−Uis now derived from the density ofUas g
R2
=
N γ=0
N γ
γ(1−)N−γgγ
R2
, (3.5)
where gγ
R2
=exp
−1
2c2γνΣ−01ν
Γ (1/2)n Γ
(1/2)(p−1) Γ
(1/2)(n−p+1) ∞
t=0k+=t
cγ2 2
t
×
(1/2)n
k
(1/2)n
(1/2)n
t
ν12/σ11
ν2Σ−122ν2
k
k!!
×
1−R2(1/2)n+k+−1
R2(1/2)(p−3)
×2F1
1
2(p−1)+k,1
2(p−1)+;1
2(p−1);R2
, 0< R2<1.
(3.6) By using the result2F1(a, b;c;z)=(1−z)c−a−b2F1(c−a, c−b;c;z), the above density can be rewritten as
gγ R2
=exp
−1
2cγ2νΣ−10 ν Γ (1/2)n Γ
(1/2)(p−1) Γ
(1/2)(n−p+1)
×
R2(1/2)(p−3)
1−R2(1/2)(n−p−1) ∞ t=0k+=t
cγ2 2
t (1/2)n
k
(1/2)n
(1/2)n
t
×
ν12/σ11
ν2Σ−221ν2
k
k!! 2F1
−k,−;1
2(p−1);R2
, 0< R2<1.
(3.7) It is interesting to note that ifν=0, then the densityg(R2)reduces to
g R2
= Γ
(1/2)n Γ
(1/2)(p−1) Γ
(1/2)(n−p+1)
×
R2(1/2)(p−3)
1−R2(1/2)(n−p−1)
, 0< R2<1.
(3.8)
References
[1] A. K. A. Amey,Robustness of the multiple correlation coefficient when sampling from a mixture of two multivariate normal populations, Comm. Statist. Simulation Com- put.19(1990), no. 4, 1443–1457.
[2] A. K. A. Amey and A. K. Gupta,Testing sphericity under a mixture model, Austral. J. Statist.
34(1992), 451–460.
[3] A. K. Gupta and D. G. Kabe,On some noncentral distribution problems for the mixture of two normal populations, Metrika38(1991), 1–10.
[4] A. K. Gupta and D. K. Nagar,Matrix Variate Distributions, Chapman & Hall/CRC, Florida, 2000.
[5] D. G. Kabe and A. K. Gupta,Hotelling’sT2-distribution for a mixture of two normal popu- lations, South African Statist. J.24(1990), 87–92.
[6] Y. L. Luke,The Special Functions and Their Approximations, Vol. I, Academic Press, New York, 1969.
[7] D. K. Nagar and M. E. Castañeda,Distribution of correlation coefficient under mixture normal model, to appear in Metrika, 2002.
[8] M. S. Srivastava,On the distribution of Hotelling’sT2and multiple correlationR2when sampling from a mixture of two normals, Comm. Statist. A—Theory Methods12 (1983), no. 13, 1481–1497.
[9] M. S. Srivastava and H. M. Awan,On the robustness of Hotelling’sT2-test and distribution of linear and quadratic forms in sampling from a mixture of two multivariate normal populations, Comm. Statist. A—Theory Methods11(1982), no. 1, 81–107.
[10] ,On the robustness of the correlation coefficient in sampling from a mixture of two bivariate normals, Comm. Statist. A—Theory Methods13(1984), 371–382.
[11] W. Y. Tan,On the distribution of the sample covariance matrix from a mixture of normal densities, South African Statist. J.12(1978), 47–55.
[12] D. M. Titterington, A. F. M. Smith, and U. E. Makov,Statistical Analysis of Finite Mixture Distributions, John Wiley, Chichester, 1985.
Hydar Ali: Department of Mathematics and Computer Science, the University of the West Indies, St. Augustine, Trinidad and Tobago
Daya K. Nagar: Departamento de Matemáticas, Universidad de Antioquia, Medellín, A. A.1226, Colombia
Special Issue on
Intelligent Computational Methods for Financial Engineering
Call for Papers
As a multidisciplinary field, financial engineering is becom- ing increasingly important in today’s economic and financial world, especially in areas such as portfolio management, as- set valuation and prediction, fraud detection, and credit risk management. For example, in a credit risk context, the re- cently approved Basel II guidelines advise financial institu- tions to build comprehensible credit risk models in order to optimize their capital allocation policy. Computational methods are being intensively studied and applied to im- prove the quality of the financial decisions that need to be made. Until now, computational methods and models are central to the analysis of economic and financial decisions.
However, more and more researchers have found that the financial environment is not ruled by mathematical distribu- tions or statistical models. In such situations, some attempts have also been made to develop financial engineering mod- els using intelligent computing approaches. For example, an artificial neural network (ANN) is a nonparametric estima- tion technique which does not make any distributional as- sumptions regarding the underlying asset. Instead, ANN ap- proach develops a model using sets of unknown parameters and lets the optimization routine seek the best fitting pa- rameters to obtain the desired results. The main aim of this special issue is not to merely illustrate the superior perfor- mance of a new intelligent computational method, but also to demonstrate how it can be used e
ffectively in a financial engineering environment to improve and facilitate financial decision making. In this sense, the submissions should es- pecially address how the results of estimated computational models (e.g., ANN, support vector machines, evolutionary algorithm, and fuzzy models) can be used to develop intelli- gent, easy-to-use, and/or comprehensible computational sys- tems (e.g., decision support systems, agent-based system, and web-based systems)
This special issue will include (but not be limited to) the following topics:
• Computational methods
: artificial intelligence, neu- ral networks, evolutionary algorithms, fuzzy inference, hybrid learning, ensemble learning, cooperative learn- ing, multiagent learning
• Application fields
: asset valuation and prediction, as- set allocation and portfolio selection, bankruptcy pre- diction, fraud detection, credit risk management
• Implementation aspects
: decision support systems, expert systems, information systems, intelligent agents, web service, monitoring, deployment, imple- mentation
Authors should follow the Journal of Applied Mathemat- ics and Decision Sciences manuscript format described at the journal site
http://www.hindawi.com/journals/jamds/.Prospective authors should submit an electronic copy of their complete manuscript through the journal Manuscript Track- ing System at
http://mts.hindawi.com/, according to the fol-lowing timetable:
Manuscript Due December 1, 2008 First Round of Reviews March 1, 2009 Publication Date June 1, 2009
Guest Editors
Lean Yu,
Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China;
Department of Management Sciences, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong;
[email protected]
Shouyang Wang,
Academy of Mathematics and Systems Science, Chinese Academy of Sciences, Beijing 100190, China; [email protected]
K. K. Lai,
Department of Management Sciences, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong; [email protected]
Hindawi Publishing Corporation http://www.hindawi.com
