Solution of Exercise Session 14, Dec 8th ; 2006

Mathematics for Economics and Finance

Prof: Norman Schürho¤
TAs: Zhihua (Cissy) Chen, Natalia Guseva

Problem 1 Let X1 ; :::; Xn be i.i.d N ( ; 2 ).

(a) Show the statistics sample variance S 2 is unbiased estimator for 2 :(b)
2
Compute MSE of the estimator. (hints: (n 1)S 2
2 2
n 1 and Var n 1 = 2(n
1):) (c) An alternative estimator for 2
is the maximum likelihood estimator ^ 2 ,
2
show ^ = n S . (d) Is ^ a biased estimator? What is the variance of ^ 2 ?
n 1 2 2

(e) Show that ^ 2 has smaller MSE than S 2 :Explain why.

Solution:
(a)
P
n
E X = E( n1 Xi ) = 1
n nEX1 = ;
i=1
P
n P
n 2
V ar(X) = V ar( n1 Xi ) = 1
n2 V ar( Xi ) = 1
n2 nV ar(X1 ) = n ;
i=1 i=1
2
E(X 2 ) = V ar(X) + (E X)2 = n + 2
;

n
X
2 1
E(S ) = Ef (Xi X)2 g
n 1 i=1
n
X
1
= Ef (Xi2 2Xi X + X 2 )g
n 1 i=1
n
X n
X n
X
1
= f E(Xi2 ) 2E(X Xi ) + E( X 2 )g
n 1 i=1 i=1 i=1
| {z }
=nX
1
= nE(X12 ) 2nE(X 2 ) + nE(X 2 )
n 1
2
1 2 2 2
= fn( + ) n( + )g
n 1 n
2
= :
2
So, Sample variance is unbiased estimator for :
(b) We get

(n 1)S 2
V ar( 2
) = 2(n 1) )
(n 1)2
4
V ar(S 2 ) = 2(n 1) )
4
2
V ar(S 2 ) = :
n 1

1
 (c) Compute MLE ^ 2 ;
2 1 1
P
n
(xi )2
L( ; j X) = n
2) 2
expf 2 2 g;
(2 i=1
n n 2 1
Pn
(xi )2
lnL = 2 ln 2 2 ln 2 2 ;
i=1
FOC:

n
@ ln L 1X
= 2
(xi )=0 (1)
@ i=1
n
@ ln L n 1 X
= + (xi )2 = 0 (2)
@ 2 2 2 2 4 i=1

P
n
From (2) ) ^ 2 = 1
n (xi X)2 ) ^ 2 = n 1 2
n S :
i=1
(d)
E ^ 2 = E( nn 1 S 2 ) = nn 1 2
(clearly, it is biased. )
Bias: E(^ 2 2
) = nn 1 2 2
= n1 2 (Notes: as n ! 1; bias ! 0 )
4
2(n 1)
V ar(^ 2 ) = V ar( nn 1 S 2 ) = ( nn 1 )2 V ar(S 2 ) = n2 :
(e) MSE of ^ 2 is given as

E(^ 2 2 2
) = V ar(^ 2 ) + (Bias(^ 2 ))2
2(n 1) 4 1
= + 2 4
n2 n
2n 1 4
= :
n2
Compare M SE(^ 2 ) with M SE(S 2 ); we …nd
2n 1 4 2 4
< )
n2 n 1
M SE(^ 2 ) < M SE(S 2 )

This shows there is trade-o¤ between bias and variance.

Problem 2 Take the regression model y = X + " and assume that

n 1 n KK 1 n 1
it ful…lls the main assumptions of the linear regression model. Furthermore,
assume that y=X N (X ; 2 In ):
(a) Write the log likelihood function. (b) Find the MLE estimators for
and 2 :(c) Compute the Cramer-Rao lower bound and the Fisher Information
matrix. (d) Find the asymptotic distribution of the estimators.

Solution:

2
 (a) The likelihood function is given by

2 n=2 1
L= 2 exp 2
(Y X )0 (Y X )
2
The log likelihood function is then
n n 2 1
lnL = ln 2 ln 2|
(Y X )0 (Y X )
2 2 2 {z }
Y 0Y 2 0X0Y + 0X0X

2
(b) The score vector is de…ned by the FOC. Let =

@ ln L 1
= ( 2X 0 Y + 2X 0 X )
@ 2
1
= (X 0 Y X 0X ) = 0 )

^ 1
M LE = (X 0 X) X 0 Y:

Replace this result in the FOC( )

@ ln L n 1
= + 2 (Y X )0 (Y X )=0
@ 2 2
n 1 0
) + 2 (Y X ^ M LE ) (Y X ^ M LE ) = 0
2 2 | {z }
e
1
) ^ M LE = e0 e
n
Note that the MLE estimator for the is the OLS estimator, however, the
^ M LE is di¤erent from the OLS estimator (check its small sample properties,
recall ^ OLS = n 1 k e0 e ).
(c) The SOC are:
@ ln L 1
= X 0X
@ @ 0
@ 2 ln L n 1
= e0 e
@ 2 2 2 3

@ ln L 1
= 2
X 0e
@ @
Thus, the Fisher Information Matrix is given by
1 1
X 0X 2 X 0e
I( ; ) = E 1 0 n 1 0
2e X 2 2 3e e

1
X 0X 0
= n
0 2 2

3
The Cramér-Rao lower bound is then
" #
1
1 (X 0 X) 0
I( ; ) = 2 2
0 n
2 1
(X 0 X) 0
= 2 4
0 n

(d) According to the properties of MLE, the asymptotic distribution is given

by
^ 2 1
M LE d (X 0 X) 0
! N (0; ):
^ 2M LE 2 0 2 4
n

Problem 3 (One-Sector Growth Model)

Consider the intertemporal optimization problem
1
X
t
max1 u(ct ) s:t: kt+1 + ct = f (kt ); k0 0 given.
fct ;kt gt=0
t=0

Assume u(c) = ln c; f (k) = Ak ; with A > 0; 2 (0; 1); 2 (0; 1):

Solution: ( it is in your lecture notes)

Problem 4 Minimizing quadratic costs. The agent solves

1
X
t
min x2t + vt2
fvt g
t=0
st: xt+1 = 2xt + vt
x0 is given, 0 < <1

Assume the transversality condition is satis…ed.

Solution:
We approach this problem by using the Bellman equation and the guess and
verify method.
The Bellman equation is given by

V (xt ) = min x2t + vt2 + V (xt+1 ) :

fvt g

We use the following guess for the value function,

V (xt ) = Ax2t

Replacing the guess in the Bellman, and using the transition equation, we get
n o
2
Ax2t = min x2t + vt2 + A (2xt + vt )
fvt g

4
The FOC gives the optimality condition that links xt and vt ;

2vt + 2 A (2xt + vt ) = 0
2 Axt
) vt =
1+ A
This result is used to …nd the constant A: Putting it back in the Bellman, we
get,
2 2
2 Axt 2 Axt
Ax2t = x2t + + A 2xt ;
1+ A 1+ A
!
2 2
2 A 2 A
Ax2t = 1+ x2t + A 2 x2t
1+ A 1+ A

Note that we can drop the term x2t , since the Bellman is valid for all values of xt
and we obtain a quadratic equation for A; which gives the value of the constant
given ; !
2 2
2 A 2 A
A= 1+ + A 2 :
1+ A 1+ A

(You can stop here in this question, since it is a polynomial of order 3) .