MONASH UNIVERSITY

DEPARTMENT OF ECONOMETRICS AND BUSINESS STATISTICS

ETC3400 PRINCIPLES OF ECONOMETRICS

SEMESTER 2, 2022

Tutorial Exercise No. 6

Learning Objectives

1. Students to produce analytical and numerical results relevant to vector MLEs

Background Reading:

1. Essential: Sections 3.1 and 3.2 in the Lecture Notes, plus the relevant EVIEWS
simulation programs.

2. Recommended: Johnston and Dinardo (1997), Section 5.1.

3. Recommended: Casella and Berger (2002), Section 5.5.4 and Section 10.1, up to
end of Section 10.1.3.

4. Recommended: Green (2008), Sections 16.1 to 16.4 (in particular Section 16.4.6).

1
The following questions are to be attempted in the tutorial of Week 7. Those questions
not covered in the tutorial should be completed as homework.

PART I: Analytical

2
1. Assume n i:i:d: draws from Y N( ; ):

(a) Derive the vector MLE, b, of the true parameter vector, 0 = ( 0; 0

0) : (Note
the de…nition of 0 )
(b) Theorem 14 states that:
p d
n(b 0) ! N (0; i( 0 ) 1 ): (1)
1
Derive i( 0 ) in this case.
(c) Why does (1) indicate that b is asymptotically e¢ cient as an estimator of
0?

(d) From (1) it follows that:

b asy N ( 0 ; I( 0 ) 1 );

where
varcov(b) = I( 0 ) 1

is the asymptotic variance-covariance matrix of the non-normalized estima-

tor b: From the form of I( 0 ) 1 , specify the asymptotic variance of b and b
respectively.

2. Assume n i:i:d: draws from Y N ( ; 2 ): The delta method gives the form of the
limiting distribution of the vector MLE of any set of k continuous functions of
= ( ; )0 ; ( ), namely:
p d
n( (b) ( 0 )) ! N (0; ( 0 )i( 0 ) 1 ( 0 )0 ); (2)
| {z }| {z }| {z }
k p p p p k

where ( 0 ) is as de…ned in lectures. De…ne:

1( ; 2) =
( )= 2 2
2( ; )=

Specify both (b) and the variance-covariance matrix of the limiting distribution of
(b). How do these results compare with the results produced in lectures, in which
the parameterization: = ( ; 2 )0 is used and the distributional result
p d
n(b 1
0 ) ! N (0; i( 0 ) )

is applied directly to the MLE of this parameter vector?

2
 3. Repeat the exercise above (i.e. specify both (b) and the variance-covariance matrix
of the limiting distribution of (b)) for the following speci…cations of ( ) (where,
again, = ( ; )0 ):

(a)
2
) = 2=
1( ;
( )= 2
2( ; ) = ln( 2 ) +
(b) 2 3
2
1( ; )= =
( )=4 2( ;
2
)= 2+ 4 5
2
3( ; )= 3
(c)
2 2
( )= 1( ; )= (= 1 +0 )
(d) p
2 2)
( )= 1( ; )= (= 0 +1

PART II: Numerical

1. Now use mc_simulation.prg (setting reps = 1) to generate a single set of n = 50

i:i:d: draws from Y N ( ; 2 ), where = 0 = 10 and = 0 = 2.

(a) Produce the MLE of 0 = ( 0 ; 20 )0 ; using the logl object. Note: you have
two parameters to estimate now and two starting values to set.
(b) Compare the estimate of I( 0 ) 1 ; as produced by EVIEWS, with the analyti-
cal form produced in lectures. Note, there are two main options for estimating
I( 0 ) when I( 0 ) is a matrix:
i. The (negative of the) Hessian evaluated at b :
h i 1 @ 2l
1
Pn @ 2 ln l 1
d b) = H(b)
i
varcov( = 0 = 0 ;
@ @ =b i=1 @ @ =b

where li = log L( jyi ) = log f (yi j ):

ii. The outer product of the gradiant evaluated at b :
1
Pn @ ln l @ ln li
d b) = OPG(b) =
i
varcov( :
i=1 @ =b @ 0 =b

EVIEWS logl object uses (ii). (Note: this second expression is exploiting
the likelihood equality in equation (6), Section 2.3.4 in the lecture notes,
as it applies in the vector case).

3
2. Now use mc_simulation.prg (setting reps = 1) to generate a single set of n = 50
i:i:d: draws from Y Gamma( ; ), with pdf:
1 y
f (yj ; ) = y( 1)
exp( ) for y > 0; ; >0
( )
2
E(Y j ; ) = ; var(Y j ; ) = ;

where = 0 = 2 and = 0 = 1:5 (Note: I am now reserving the symbol for

the vector of two unknown parameters, and ).

(a) Produce the MLE of 0 = ( 0 ; 0 )0 ; using the logl object. Save both b and
d b), produced in the EVIEWS logl output (see the
the (2 2) matrix, varcov(
relevant code in the EVIEWS Command Reference for doing the latter).
(b) State the vector MLE, (b), where

1( ; )=
( )= 2
2 ( ; )=

The two parameter functions in ( ) de…ne, respectively, the mean and vari-
ance of Y:
(c) Produce
@ ( )
( )=
@ 0
analytically.
(d) Using the estimate of I( 0 ) 1 ; as produced by EVIEWS, produce an estimate
of the asymptotic variance matrix for (b) using the formula:
0
@ ( ) @ ( )
d
varcov[ (b)] = d b)
asyvar( (3)
@ 0 =b @ 0 =b

Note: this will involve two steps:

i. Use b from a. to produce

@ ( )
(b) =
@ 0 =b

ii. Use the appropriate EVIEWS matrix commands, and varcov(d b) saved
from a., to numerically evaluate the matrix product in (3) that de…nes
d
varcov[ (b)].

4
3. Now perform the same sorts of steps as in Part II, Q2 but assuming n = 50 i:i:d:
draws from Y W eibull( ; ), with pdf:

y 1 h y i
f (yj ; ) = exp for y > 0; ; >0

where = 0 = 3 and = 0 = 2 and the parameter function of interest,

( )= 1( ; )= ( 1);

de…nes the mode of Y: