Vous êtes sur la page 1sur 7

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 4, ISSUE 03, MARCH 2015

ISSN 2277-8616

Mixed Estimation In Seemingly Unrelated


Regression Equation Model Some Finite Sample
Properties Results
Ghazal. A. Ghazal, Salwa.A.Hegazy
Abstract: in (1964) Nagar and Kakwani analysis the results of the first and second moment that proposed by Theil and Goldberger (1961) ,derive the
bias estimators and second moment matrix of mixed model estimators consider with the order of magnitude criteria ,derived biased estimators to order

OP (T 1 )

that refer to big (O) in probability where

(T )

begin the number of observations. In this paper we derive mixed seemingly unrelated

regression equations (SURE) by combining the prior information and sample information in a single model, derive the bias estimator to order
and the moment matrix to order OP (T

OP (T 1 )

by using the methodology of Nagar's expansion for the moment of estimator. the bias has been derived

conceder the normality distribution assumption of the random disturbances.


Keywords: mixed seemingly unrelated regression model, moment matrix, Nagar's expansion, prior information

1. Introduction

(2)THE MODEL & ASSUMPTION

There are two popular types of the prior information. Exact


linear prior information on the coefficient according to
economic theorem, and the stochastic linear prior information
it is involved exact part and stochastic part.Thus we have two
types of estimators. First type is pure estimator that depends
upon the data of original sample, second type that the mixed
estimator obtained from unifying data of exact information and
stochastic information. The problem of incorporating the two
types of prior information discuss by Theil and Goldberger
(1961) assumes additional information of the coefficient
comes from previous statistical investigation, economic
theorem and previous experience about the phenomena. ,
combining stochastic prior information with the available
sample information when estimating the parameters of linear
regression model to obtain the estimator of mixed regression
model. In Section (2) presents the description of the model
with their assumption and properties, estimating (sure) model
under prior information. Section (3) drive mixed (sur)
estimator. we are going to derive the feasible mixed (SUR)
model and estimator, this paper ended with section (4) that
1
we are going to derive the bias to order (T ) & variance
covariance matrix of the feasible mixed (SUR) regression

Let us consider the following model of (GLS)

estimator to order
Nagar's procedure.

(T ) . This approximation in view of

__________________________
Ghazal. A. Ghazal. Professor of Statistics &
Econometrics, Department of Applied Statistics and
Econometrics, Institute of Statistical Studies & Research
Cairo University Giza Egypt. More than 45 years
teaching experience, University of North Carolina and
Cairo University Number of published papers
specialized Applies Statistics and Econometrics
Salwa .A. Hegazy. Applied Statistic And Econometrics
Department, Institute of Statistical Studies & Research Cairo University Giza Egypt. 14 years teaching
experience, of 6 October University, Faculty of
Economics and Management Giza Egypt
Salwahegazy2@yahoo.com

Yi

(T .1)

X i Bi U i

(T .ki ) ( K i .1)

i 1,2

(T .1)

(2-1)

K k1 k 2
That is the sample information

Where y i is a (T.1) vector of observations on the i th


dependent variable (the variable to be "explained" by the

i th

regression equation). ,

matrix of observation on

is a (T .k i ) block diagonal

( K i ) nonstochastic

variable, each column of which consists of

independent

T observation on
i th equation of

a regressors (explanatory variable ) in the


B
the model, with rank ( K i T ) . , i is a (k i .1) vector of
regression coefficients that unknown parameters in the
equation of the model.

ui

is the corresponding

of random disturbances term in the


equation. We can write (2-1) as

Y1 X 1

Y2 0

i th

(T.1)

i th

vector

regression

0 1 U 1

X 2 2 U 2

We have the following assumption that the disturbance terms


have a normally distribution with zero mean vector and
variance

ii I , U T .1 ~ N T (0, ii I )

173
IJSTR2015
www.ijstr.org

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 4, ISSUE 03, MARCH 2015

(u i ) 0 (T .1)
Var (u i ) E (u i2 ) ii I

, i 1,2
,i

Cov(u i , u j ) E (u i , u j ) ij I

u
I
E 1 u1 u2 11
u2
21I
11 12

I I (T .T )
21 22

(2-2)

by unifying the sample information model (2-1) and the


extraneous prior information (3-1) in a single model. We get

Y
TMx1
r
qx1

12 I

22 I

Where

Y
*

i th equation for each observation in the

denotes the usual expectation operation. ,


the covariance between the disturbances of

,and E (.)

i j

i th

j th

equations

x1 x2 x2 x1 0
3. ESTIMATING MIXED (SURE) MODEL
Suppose the following stochastic linear extraneous prior
information " it called stochastic linear restriction on the
regression coefficients of (2-1) is available and it can be
written as.

is a

, R is a

(3-1)

(qx1) column random vector of "Known" elements


(qxk ) matrix of prior information of "known"

q( k ) , V
is a
(qx1) disturbance random vector with (q) component which
are "Independently" distributed of the element of u random
elements

with

rank

being

vector.

(TM q) x1

We have assumption

E ( w) 0

u 0
E (W ) E 0
V 0 (TM q ) x1

E (WW )
u u
I 0
COV (W ) COV E u V

V V
0


In view of above assumption. Applying (GLS) we get

b1M
( Z 1 Z ) 1 ( Z 1 y * )
2M

E (V ) 0

I 0 y
x R

1

Where

11 12

21
22

Applying Aitken's (GLS) procedure to obtain the Mixed (SUR)

estimator

Assume that.
( qX 1)

u
, W TMx1
V
qx1

(TM q) xk

I 0 x
x R

1

Where:

X
TMx
R
qxk

,Z

(TM q) x1

r R V

( qx1)

Y
TMx1
r
qx1

( qxk )

Represents

and
equations for each observation in the sample
Assume that the regressors in two equations are orthogonal

( qxk ) ( kx1)

(TM q) xk (TM q) x1

Y * ZB W

is an identity matrix of order

( qx1)

( kx1)

u
TMx1
V
qx1
(3-2)

is scalar and Represents the variance of the random

sample,

rewrite model as

Nonsingular matrix
Where:

disturbance in the

X
TMx
R
qxk

(TM q) x1

ii

ISSN 2277-8616

b
1M
b2 M

we get

X ( 1 I ) X R 1 R

. X ( 1 I ) y R 1 r

COV (V ) E (VV )

( qxq)

is a

(qxq )

matrix of "Known" elements and nonsingular


174
IJSTR2015
www.ijstr.org

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 4, ISSUE 03, MARCH 2015

b
b

1M

2M

x1 0 11 12 x1 0

21

22
0 x 2
0 x 2

0 R1 0
R1 0 111


0 R 2 0 221 0 R2

2
122 122 22 122 22

11 11

3
2
22
22
22

122 123 124



2 3

22 22 22
2
2
2 2
12 12 12 212 22 12 312 22
22
22
22

x1 0 11 12 y1

0 x 2 21 22 y 2
.

1
0 r1
R1 0 11
0 R 0 1 r
2
22 2

With simply calculation and Using orthogonal condition we get

11 x x R 1 R 1
1 1
1 11 1
b
11
1M
. x1 y1 12 x1 y 2 R1111 r1

(3-3)

1M

is the Mixed (SUR) estimator.

Thus consider the alternative estimator


replacing

ij

by

s ij

s ij

derived by

122 122 22 2 12 12

22
22 222

122 222 122 2 12 12 22

11 11 3

22
22

22
123 2 12 12 222 124
3
2
3

22
22
22
Then we have

then we get

11
2 1
s s22 (s11s22 s12 )

2 1
s 22 s11 (s11s22 s12
)

2 1
s 21 s12 s12 (s11s22 s12
)

2 1
let (s11s 22 s12
)

b1 x1 x1
R1111 R1
s 22
FM

x1 y1 12 x1 y 2
R1111 r1
s 22
s 22

Now we are going to find

where

(3-4)

2
1
11 11 12 12 22 22
s22
1 22
1

22 22

(3-5)

223
1 22 22

......

22 22 222 223

(3-7)

s 22

122
122 22

11 2 1
11
22 O (T 0 )
22 O (T 2 )

s 22
122 222


......
11 223 1

O (T )

From (3-4) we have

This is called the Feasible Mixed (SUR) Model (FMSUR),


is the Feasible Mixed (SUR) estimator.

22 22 1

ISSN 2277-8616

1
FM

b1 B x1 x1
R1111 R1
s 22
FM

s
s

. x1u1
R1111V 12 x1 x2 B2 12 x1u 2
s 22
s 22
s 22

Using orthogonal condition

x1 x2 x2 x1 0

1
x1 x1

R111 R1
s 22

b B

FM
x u R 1V s12 x u
1 2
1 1 s 22 1 11
s 22

Using (3-7) we get


(3-6)

Substitute (3-6) in (3-5) and rearrange we get

175
IJSTR2015
www.ijstr.org

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 4, ISSUE 03, MARCH 2015

2
x1 x1 11 12 R1111 R1

22

b B
1

FM

122


11 2 22 R111 R1
22

2
x1u1 11 12 R1111V

22

2
12 R 1V 12 xu
1 2
11 2 22 1 11
22
22

x1u1 11 12 R1111V

22

2
12 R 1V s12 xu
1 2
11 2 22 1 11
s22
22

4.

estimator to order

1
The bias to order OP (T ) is zero under the assumptions of
the disturbance term that follow normal distribution

PROOF
From (3-8) we have

2
let ( A1 ) 122 22R1111 R1 x1 x1 2 11 12 R1111 R1
22

22

By using
Then

let
let

let

11 122 22 R1111V 12 x1u 2

22
22

( AB) 1 B 1 A1

2
. x1 x1 2 11 12 R1111 R1

22

2
B x1x1 2 11 12 R1111 R1
1
22

12
1

u1
11

C
x

R
V
1
11
1

1
22

R1111V 12 x1u 2
D0 11 12

22
2

22
22

b B I A B C D
1

1
1 1
0
FM

2
2

x1x1 2 11 12 R1111 R1

22

122

I 2 22R1111 R1

22
b B
1

1
FM

122
R1111 R1
x1 x1 2 11


22

2
2
x1 x1 2 11 12 R1111 R1 I 122 22R1111 R1 x1 x1 2 11 12 R1111 R1

22
22
22

2
. x1u1 11 12 R1111V

22

(T 2 )

THEOREM (1)

2
1
x1u1 11 12 R111

22

12
12

11 2 22 R1 11 V x1u 2
22
22

Take a common factor

BIAS AND MOMENT MATRIX OF (FMSUR) ESTIMATOR

(T 1 )

2
2

x1u1 11 12 R1111V 11 122 22 R1111V

22
22

12 xu

1 2

22

R111
R
1

(3-8)

we are going to derive the bias to order


& variance
covariance matrix of the feasible mixed (SUR) regression

2
2
x1x1 2 11R1111R1 12 R1111R1 122 22R1111R1
22
22

x x 2 12
1
1
11

22


2
12
1
2 22R111 R1

22

ISSN 2277-8616

Expanding the right hand side of this equation by using


binomial expansion we get

I A1 A1 A1 A1 A1 A1
2
2
2
2
2
2

B1C 1 B1 D0

B1C 1

B1 D0 A1 B1C 1

2 O (T 2 )
2
2 O ( T 1 )

A1 A1 B1C 1 A1 B1 D0 3

2 2

2
2
O (T 2 )

A1 A1 B1 D0 A1 A1 A1 B1C 1

2
2
2
2 O (T 2 )
2 2

A A A B D

21 21 21 1 0 25

O (T )

176
IJSTR2015
www.ijstr.org

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 4, ISSUE 03, MARCH 2015

1
FM

1
To derive the bias of estimator
to order O(T ) we are
going to calculate the expectation of the first and second
terms


R1111 R1
E B1C 1 x1 x1 2 11

22

2
12


R1111V
E x1u1 11

22

E (u ) 0
E (V ) 0
( qX 1)

E B1C 1 0
2

2
E B1 D0 x1 x1 2 11 12 R1111 R1
22

I A1 A1 A1 B1C 1 B1 D0

2
2
2
2

I A1 A1 A1 B1C 1 B1 D0

2
2
2
2

E B1 D0 0

( qX 1)

B1C 1 B1 D0 A1 B1C 1 A1 B1 D0

2
2
2
2

A1 A1 B1C 1 A1 A1 B1 D0

2
2
2
2
2

From above expectation we get


1

E b1 B E I A1 B1 C 1 D0 0

FM

2
2

b
1
Then FM is unbiased estimator for B

THEOREM (2)
Under the assumptions of normal distribution of the
disturbance term .the moment of feasible mixed (SUR)
b
1
estimator FM to order

(T 2 )

is given by

I 2 A1 2

Proof
Note the order of magnitude of
2
12

22

is O(T )

3
3
12
2 2
2 12 312 22 is O(T 2 )
22
22

Expanding the right hand side of this equation by using


binomial expansion we get

11 12 22 R1111V 12 x1u 2

22
222

E (u ) 0
E (V ) 0
(TX 1)

2
122 22
2 2
3 12 12 212 22 is O(T 1 )
22
22
22

b
1
The moment matrix of FM estimator can be derive as
1


I A1 B1 C 1 D0

2
2

E b1 B b1 B E

1
FM
FM



I A1 B1 C 1 D0
2
2

2
12

(TX 1)

ISSN 2277-8616

1
122 22 2 12 12
2

is
O
(
T
)
2

22
22

4
12

3 is O(T 2 )
22
,

B1C 1 B1 D0 A1 B1C 1 A1 B1 D0

2
2
2
2
.

A1 A1 B1C 1 A1 A1 B1 D0

2
2
2
2
2

Expanding the right hand side and Rearrange these values


according to the order of magnitude we get 12 items required
to expected values of it,

B1C1 D0 B1 B1C1 C1 B1 A1

2 2
2
B1C1 C1 B1

2 2 O (T 1 ) B1D0C1 B1 A1 B1C1 C1 B1 3

2
2
2 2

O (T 2 )

E B1C1 D0 B1 A1 B1C1 C1 B1 A1 A1

2
2
2 2
2 2
5

B1D0 D0 B1 B1D0C1 B1 A1
O(T 2 )

2
2

A1 B1C1 D0 B1 A1 B1C1 C1 B1 A1 A1 B1D0C1 B1

2
2 2
2
2
2
O (T 2 )
2 2

Note that.

177
IJSTR2015
www.ijstr.org

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 4, ISSUE 03, MARCH 2015

The

Second

term

&

the

fourth

term

symmetric,

E B1 D0 C 1 B1 3 E B1C 1 D0 B1 3
2
2

O (T 2 )

O (T 2 )
Third term & fifth term are symmetric, then

Six term & 12-th are symmetric, then

E A1 B1C 1 D0 B1
E B1 D0 C 1 B1 A1
2
2
2 O (T 2 )
2
O (T 2 )

122

x1 x1 11 12 2 22

22
22

2 A1 B1
B1
122
2


1

11 R111 R1

22

Ninth term & tenth are symmetric, then

E A1 B1 D0 C 1 B1
2
2

=
Then by using the expectation we get

122

R1111 R1 B1 A1
A1 B1 11 x1 x1 11

22

2
2

122

122 22

x1 x1 11
2

22

22

2 B1
B1 A1

122

2
1

11 R111 R1

22

E B1C 1 C 1 B1 A1 3 E A1 B1C 1 C 1 B1 3
2 2
2 O (T 2 )
2 2

2
O (T 2 )

E B1C 1 D0 B1 A1
2
2

ISSN 2277-8616

E b B b B E B1C 1 C 1 B1
FM
FM
2 2

O (T 1 )

(i)

2 E B1C 1 D0 B1 2 E B1C 1 C 1 B1 A1

3
2
2
2
2 O (T 2 )

Let

2 2

1
12

11x1 x1 11
R111 R1

22

Combine the first, third, fourth & sixth terms we get

B1C1 C1 B1 A1 A1 B1D0 D0 B1

2 2
2
2
E

A1 B1C1 C1 B1 A1

2 2
2
2
O ( T 2 )

B1 2 B1 1 B1 A1

2
2

B1 B1 A1 A1 A1 B1 B1 A1

- 2E B1 D0 C 1 B1 A1
2 E B1C 1 D0 B1 A1
2
2 O (T 2 )
2
2 O (T 2 )

Then we get

122

R1111 R1 B1
E b1 B b1 B B1 11x1 x1 11

22
FM FM

122

122

x x
22 1 1 11 22
2 B1
B1
122 22


1
11 2 R111 R1
22

B1 B1 I 2 A1 A1 A1 A1 A1
2
2
2
2
2

B1 B1 I 2A1
2

(ii)

Let

2
2
12 x1 x1 11 12 11 12 2 22 R1111 R1
22
22

22

Combine the second, seventh & eighth terms we get

122
1

B A

2 B1 11 x1 x1 11
R

R
1
11
1

22

2
2


B1 11 x1 x1 11 12 R1111 R1 B1 A1 A1

22

2
2

122
122 22
1

R111 R1
x1 x B
B1 11
2

22
22

2B1 B1 - 2 B1 B1 A1 2 B1 B1 A1
2
2

2B1 B1 I - A1 A1
2
2

2B1 B1 I - 2 A1
2

(iii)

Let

178
IJSTR2015
www.ijstr.org

INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 4, ISSUE 03, MARCH 2015

ISSN 2277-8616

122
122 22
1

11
R

2
1 11 1 x1 x

22

22

Then the fifth term

B1B1

From combine these terms we get

B1 B1 I 2 A1 2B1 B1 I - 2 A1 B1B1
2
2


B1 B1
B B
E b1 B b1 B I 2 A1
2 B B 1 1
FM FM

1
2

let B1 B1 , B1 B1 ,

B1 B1

From above we get the variance covariance of feasible


b
1

(T 2 )

mixed (SUR) estimator FM to order


E b1 B b1 B I 2 A1 2
FM
FM

REFRENCES
[1] Theil,H ., and A.S.Goldberger,"On Pure And Mixed
Statistical Estimation In Economics" , international
economics review ,vol.2,pp.65-78 , (1961).
[2] Theil,H,"On The Use Of Incomplete Prior Information
In Regression Analysis",Journal of the American
Statistical Association, Vol. 58, 401-414, .(1963).
[3] Nagar,A.L,and N.C.Kakwani, "The Bias And Moment
Matrix Of A Mixed Regression Estimator ",
Econometrica , vol. 32,pp.174-182, (1964).
[4] Nagar,A.L,and N.C.Kakwani,"Note On The Bias Of A
Mixed
Simultaneous
Equation",
International
Economic Review, vol.7,No.1,pp 65-71, (1966).
[5] Zellner , A, "An Efficient Method Of Estimating
Seemingly Unrelated Regression Equations And
Tests For Aggregation Bias" , Journal of the American
Statistical Association , vol.57,pp. 348-368., .(1962).
[6] _______, Estimators For Seemingly Unrelated
Regression Equations:,"some exact finite sample
results", Journal of the American Statistical
Association ,vol. 58, pp.977-992, (1963)

179
IJSTR2015
www.ijstr.org