# Chapter 6 Regression Analysis Under Linear...

date post

03-Apr-2018Category

## Documents

view

266download

6

Embed Size (px)

### Transcript of Chapter 6 Regression Analysis Under Linear...

Econometrics | Chapter 6 | Linear Restrictions and Preliminary Test Estimation | Shalabh, IIT Kanpur 1 1

Chapter 6

Regression Analysis Under Linear Restrictions and Preliminary Test Estimation

One of the basic objectives in any statistical modeling is to find good estimators of the parameters. In the

context of multiple linear regression model y X = + , the ordinary least squares estimator

( ) 1' 'b X X X y= is the best linear unbiased estimator of . Several approaches have been attempted in the

literature to improve further the OLSE. One approach to improve the estimators is the use of extraneous

information or prior information. In applied work, such prior information may be available about the

regression coefficients. For example, in economics, the constant returns to scale imply that the exponents

in a Cobb-Douglas production function should sum to unity. In another example, absence of money illusion

on the part of consumers implies that the sum of money income and price elasticities in a demand function

should be zero. These types of constraints or the prior information may be available from

(i) some theoretical considerations.

(ii) past experience of the experimenter.

(iii) empirical investigations.

(iv) some extraneous sources etc.

To utilize such information in improving the estimation of regression coefficients, it can be expressed in the

form of

(i) exact linear restrictions

(ii) stochastic linear restrictions

(iii) inequality restrictions.

We consider the use of prior information in the form of exact and stochastic linear restrictions in the model

y X = + where y is a ( 1)n vector of observations on study variable, X is a ( )n k matrix of

observations on explanatory variables 1 2, ,..., , ( 1)kX X X is a k vector of regression coefficients and is

a ( 1)n vector of disturbance terms.

Econometrics | Chapter 6 | Linear Restrictions and Preliminary Test Estimation | Shalabh, IIT Kanpur 2 2

Exact linear restrictions: Suppose the prior information binding the regression coefficients is available from some extraneous sources

which can be expressed in the form of exact linear restrictions as

r R=

where r is a ( 1)q vector and R is a ( )q k matrix with ( ) ( ).rank R q q k= < The elements in

and are known.r R

Some examples of exact linear restriction r R= are as follows:

(i) If there are two restrictions with 6k = like

2 43 4 52 1

=+ + =

then

0 0 1 0 1 0 0 0,

1 0 0 1 2 1 0 0r R

= =

.

(ii) If 3k = and suppose 2 3, = then

[ ] [ ]3 , 0 1 0r R= =

(iii) If 3k = and suppose 1 2 3: : :: : :1ab b

then 0 1 00 , 0 1 .0 1 0

ar R b

ab

= =

The ordinary least squares estimator 1( ' ) 'b X X X y= does not uses the prior information. It does not obey

the restrictions in the sense that .r Rb So the issue is how to use the sample information and prior

information together in finding an improved estimator of .

Econometrics | Chapter 6 | Linear Restrictions and Preliminary Test Estimation | Shalabh, IIT Kanpur 3 3

Restricted least squares estimation The restricted least squares estimation method enables the use of sample information and prior information

simultaneously. In this method, choose such that the error sum of squares is minimized subject to linear

restrictions r R= . This can be achieved using the Lagrangian multiplier technique. Define the Lagrangian

function

( , ) ( ) '( ) 2 '( )S y X y X R r =

where is a ( 1)k vector of Lagrangian multiplier.

Using the result that if a and b are vectors and A is a suitably defined matrix, then

' ( ')

' ,

a Aa A A aa

a b ba

= +

=

we have

( , ) 2 ' 2 ' 2 ' ' 0 (*)

( , ) 0.

S X X X y R

S R r

= =

= =

Pre-multiplying equation (*) by 1( ' ) ,R X X we have

1 12 2 ( ' ) ' 2 ( ' ) ' ' 0R R X X X y R X X R = 1or ( ' ) ' ' 0R Rb R X X R =

11' ( ' ) ' ( )R X X R Rb r =

using 1( ' ) ' 0.R X X R >

Substituting in equation (*), we get 112 ' 2 ' 2 ' ( ' ) ' ( ) 0X X X y R R X X R Rb r + =

or ( ) 11' ' ' ( ' ) ' ( )X X X y R R X X R Rb r = .

Pre-multiplying by ( ) 1'X X yields

( ) ( ) ( )

( ) ( ) ( )

11 1 1

11 1

' ' ' ' ( ' ) '

' ' ' ' .

R X X X y X X R R X X R r Rb

b X X R R X X R Rb r

= +

=

This estimation is termed as restricted regression estimator of .

Econometrics | Chapter 6 | Linear Restrictions and Preliminary Test Estimation | Shalabh, IIT Kanpur 4 4

Properties of restricted regression estimator

1. The restricted regression estimator R obeys the exact restrictions, i.e., .Rr R= To verify this,

consider

( ) { } ( )11 1 ' ' ( ' )

.

RR R b X X R R X X r Rb

Rb r Rbr

= +

= + =

2. Unbiasedness

The estimation error of R is

( ) ( ) ( )

( ) { } ( )( )

11

11 1

( ' ) ' ' '

' ' ( ' ) '

R b X X R R X X R R Rb

I X X R R X X R R b

D b

= + =

=

where

( )111( ' ) ' ' .D I X X R R X X R R =

Thus

( ) ( )0

RE DE b =

=

implying that R is an unbiased estimator of .

3. Covariance matrix

The covariance matrix of R is

( ) ( )( )( )( )

( )

( ) ( ) ( ) ( )

12

11 1 1 12 2

'

' '( ) '

' '

' ' ' ' ' ' '

R R RV E

DE b b DDV b D

D X X D

X X X X R R X X R R X X

=

=

=

=

=

which can be obtained as follows:

Econometrics | Chapter 6 | Linear Restrictions and Preliminary Test Estimation | Shalabh, IIT Kanpur 5 5

Consider

( ) ( ) ( ) ( ) ( )

( ) ( ) ( ) ( ){ } ( ) ( ) ( ){ }( ) ( ) ( ) ( ) ( ) ( ) ( )

( ) ( )

11 1 1 1 1

'1 11 1 1 1 1 1 1

11 1 1 1 1 1 1

11 1

' ' ' ' ' ' '

' ' ' ' ' ' ' ' ' ' ' '

' ' ' ' ' ' ' ' ' ' ' '

' ' ' '

D X X X X X X R R X X R R X X

D X X D X X X X R R X X R R X X I X X R R X X R R

X X X X R R X X R R X X X X R R X X R R X X

X X R R X X R R X

=

=

=

+ ( ) ( ) ( )

( ) ( ) ( ) ( )

11 1 1

11 1 1 1

' ' ' ' '

' ' ' ' ' ' .

X R R X X R R X X

X X X X R R X X R R X X

=

Maximum likelihood estimation under exact restrictions:

Assuming 2~ (0, )N I , the maximum likelihood estimator of and 2 can also be derived such that it

follows r R= . The Lagrangian function as per the maximum likelihood procedure can be written as

( ) ( )22 2 21 1 ( ) '( ), , exp '2 2n

y X y XL R r

=

where is a ( )1q vector of Lagrangian multipliers. The normal equations are obtained by partially

differentiating the log likelihood function with respect to 2, and and equated to zero as

( ) ( )

( ) ( )( ) ( ) ( )

2

2

2

2

2 2 4

ln , , 1 ' ' 2 ' 0 (1)

ln , ,2 0 (2)

ln , , 2 '2 0. (3)

LX X X y R

LR r

L y X y Xn

= + =

= =

= + =

Let 2, andR R denote the maximum likelihood estimators of 2, and respectively which are

obtained by solving equations (1), (2) and (3) as follows:

From equation (1), we get optimal as

( ) ( )

11

2

' '.

R X X R r R

=

Substituting in equation (1) gives

( ) ( ) ( )11 1' ' ' 'R X X R R X X R r R = +

Econometrics | Chapter 6 | Linear Restrictions and Preliminary Test Estimation | Shalabh, IIT Kanpur 6 6

where ( ) 1' 'X X X y = is the maximum likelihood estimator of without restrictions. From equation (3),

we get

( ) ( )2 ' .R

y X y X

n

=

The Hessian matrix of second order partial derivatives of 2and is positive definite at

2 2and .R R = =

The restricted least squares and restricted maximum likelihood estimators of are same whereas they are

different for 2 .

Test of hypothesis It is important to test the hypothesis

01

::

H r RH r R

=

before using it in the estimation procedure.

The construction of the test statistic for this hypothesis is detailed in the module on multiple linear regression

model. The resulting test statistic is

11( ) ' ( ' ) ' ( )

( ) '(

r Rb R X X R r Rbq

Fy Xb y Xb

n k

=

which follows a F -distribution with q and ( )n k degrees of freedom under 0.H The decision rule is to

reject 0 atH level of significance whenever

1 ( , ).F F q n k

Econometrics | Chapter 6 | Linear Restrictions and Preliminary Test Estimation | Shalabh, IIT Kanpur 7 7

Stochastic linear restrictions: The exact linear restrictions assume that there is no randomness involved in the auxiliary or prior

information. This assumption may not hold true in many practical situations and some randomness may be

present. The prior information in such cases can be formulated as

r R V= +

where r is a ( )1q vector, R is a ( )q k matrix and V is a ( )1q vector

*View more*