CHAPTER 11

Exercise Solutions

The authors produced these solutions. They will be independently checked and reformatted in the future. We urge
instructors to download them again next term.

1
 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 2

EXERCISE 11.1
(a)

The system of two equations has two unknowns.

y1 = α1 y2 + e1
y2 = α 2 y1 + β1 x1 + β2 x2 + e2

Substitute equation (1) into equation (2) and simplify:

y2 = α 2 y1 + β1 x1 + β2 x2 + e2
= α 2 ( α1 y2 + e1 ) + β1 x1 + β2 x2 + e2
= α1α 2 y2 + α 2 e1 + β1 x1 + β2 x2 + e2

Rearrange terms with y2 on the left,

y2 (1 − α1α 2 ) = β1 x1 + β2 x2 + e2 + α 2 e1

Solve for y2 :

β1 β2 e + α 2 e1
y2 = x1 + x2 + 2
(1 − α1α 2 ) (1 − α1α 2 ) (1 − α1α 2 )
= π1 x1 + π2 x2 + v2

To show the correlation

 β1 β2 e + α 2 e1  
cov ( y2 , e1 | x ) = E ( y2 , e1 | x ) = E  x1 + x2 + 2  e1 | x 
 (1 − α1α 2 ) (1 − α1α 2 ) (1 − α1α 2 )  
 β1   β2   e + α 2 e1  
= E x1e1 | x  + E  x2 e1 | x  + E  2  e1 | x 
 (1 − α1α 2 )   (1 − α1α 2 )   (1 − α1α 2 )  
 e e + α 2 e12  
= 0 + 0 + E  2 1  | x 
 (1 − α1α 2 )  

The first two terms are zero because the x’s are exogenous and uncorrelated with the random
errors. Then, assuming that the two equation errors are uncorrelated (see p. 533 in POE5),

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 3

cov ( y2 , e1 | x ) = E ( y2 , e1 | x )
E ( e2 e1 | x ) + α 2 E ( e12 | x )
=
(1 − α1α 2 )
α2
= σ12
(1 − α1α 2 )
This is not zero unless α 2 = 0 , in which case there is no simultaneity.

(b)

Neither of the structural equations. Both equations (1) and (2) have an endogenous variable on
the right-hand side. OLS is biased and inconsistent. On the other hand the reduced form
equation parameters can be estimated consistently using OLS because only exogenous
variables appear on the right-hand side.

(c) Which parameters are “identified,” in the simultaneous equations sense? Explain your
reasoning.

There are M = 2 equations. Identification requires that M – 1 variables be omitted from each
equation. Equation (2) is not identified. Equation (1) is identified because x1 and x2 are omitted.
It is possible to estimate α1 consistently.

(d)

These moment conditions arise from the assumptions that the x’s are exogenous. It follows that

E ( xi1vi1 | x ) = E ( xi 2 vi 2 | x ) = 0

From part (a), the reduced form equation for y2 is

β1 β2 e + α 2 e1
y2 = x1 + x2 + 2
(1 − α1α 2 ) (1 − α1α 2 ) (1 − α1α 2 )
= π1 x1 + π2 x2 + v2

The reduced form error is uncorrelated with the x’s because

  e + α 2 e1    1   α2 
E  xik  2  | x  = E  xik e2 | x  + E  xik e1 | x  = 0 + 0
  (1 − α1α 2 )    (1 − α1α 2 )   (1 − α1α 2 ) 

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 4

(e)

The sum of squares function, omitting the subscript i for convenience, is

S ( π1 , π2 | y , x ) =  ( y2 − π1 x1 − π2 x2 ) . The first derivatives are

2

∂S ( π1 , π2 | y , x )
= 2 ( y2 − π1 x1 − π2 x2 ) x1 = 0
∂π1
∂S ( π1 , π2 | y , x )
= 2 ( y2 − π1 x1 − π2 x2 ) x2 = 0
∂π2

Divide these equations by 2, and multiply the moment equations by N to see that they are
equivalent.

(f)

The moment conditions are

N −1  xi1 ( y2 − π1 xi1 − π2 xi 2 ) = 0
N −1  xi 2 ( y2 − π1 xi1 − π2 xi 2 ) = 0

Multiplying these out we have

x y − π1  xi21 − π2  xi1 xi 2 = 0
i1 i 2

x i2 yi 2 − π1  xi1 xi 2 − π2  xi22 = 0

Inserting the given values, we have

3 − πˆ 1 = 0  πˆ 1 = 3
4 − πˆ 2 = 0  πˆ 2 = 4

(g)

The first structural equation is y1 = α1 y2 + e1 , so that

E ( π1 x1 + π2 x2 ) e1 | x  = E ( π1 x1 + π2 x2 )( y1 − α1 y2 ) | x  = 0

The empirical analog of this moment condition is

N −1  ( π1 xi1 + π2 xi 2 )( yi1 − α1 yi 2 ) = 0

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 5

If we knew π1 and π2 we could solve this moment condition for an estimator of α1 . While we
do not know these parameters we can consistently estimate them from the reduced form
equations. In large samples the consistent estimators converge to the true parameter values,

plim πˆ 1 = π1 and plim πˆ 2 = π2

In a sense, having consistent estimators of parameters is “just as good as” knowing the
parameter values. Replacing the unknowns by their estimates in the empirical moment
condition we have

 ( πˆ x1 i1 + πˆ 2 xi 2 )( yi1 − α1 yi 2 ) =  yˆi 2 ( yi1 − α1 yi 2 ) = 0

So that

 yˆ yi1
 yˆ yi1 − α1  yˆi 2 yi 2 = 0  αˆ 1, IV =
i2

 yˆ
i2
i2 yi 2

(h)

The least squares estimator of the simple regression model with no intercept is given in Exercise
2.4. Applying that result here, and substituting ŷ2 for x and y1 for y, we have

αˆ 1,2SLS =
 yˆi 2 yi1
 yˆi22
To show that the equations are equivalent, recall that vˆ2 = y2 − yˆ 2  yˆ 2 = y2 − vˆ2 and therefore

 yˆi22 =  yˆi 2 ( y2 − vˆ2 ) =  yˆi 2 y2 −  yˆi 2 vˆ2 =  yˆi 2 y2

The term

 yˆi 2 vˆi 2 =  ( πˆ 1 xi1 + πˆ 2 xi 2 )vˆi 2 = πˆ 1  xi1vˆi 2 + πˆ 2  xi 2 vˆi 2 = 0

Because  xi1vˆi 2 = 0 and  xi 2 vˆi 2 = 0 . This is a fundamental property of OLS that is illustrated
in Exercises 2.3(f) and 2.4(g).

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 6

EXERCISE 11.2
(a)

Yes. 3 ( Γ11 q + Γ 21 p + Β11 + Β 21 x + E1 ) = 0 and 3 ( Γ12 q + Γ 22 p + Β12 + Β 22 x + E 2 ) = 0 are true if

the original equations are true.

(b)

Yes, for the same reason as in part (a).

( −1 Γ11 )( Γ11q + Γ 21 p + Β11 + Β 21 x + E1 ) = 0

 − q − ( Γ 21 Γ11 ) p − ( Β11 Γ11 ) − ( Β 21 Γ11 ) x − ( E1 Γ11 ) = 0

(c)

−q − ( Γ 21 Γ11 ) p − ( Β11 Γ11 ) − ( Β21 Γ11 ) x − ( E1 Γ11 ) = 0


q = α 21 p + β11 + β21 x + e1

(d)

( − 1 Γ 22 ) [ Γ12 q + Γ 22 p + Β12 + Β 22 x + E 2 = 0 ]

( −Γ12 Γ 22 ) q − p − ( Β12 Γ 22 ) − ( Β 22 Γ 22 ) x − ( E 2 Γ 22 ) = 0

p = α 12 q + β12 + β 22 x + e2

(e)

That is true. We can multiply by the negative reciprocal of any Γ or Β and solve.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 7

EXERCISE 11.3

(a)

Multiply demand by Γ12 and supply by −Γ11 and add.

( Γ12 Γ 21 − Γ11Γ 22 ) p + ( Γ12 Β11 − Γ11Β12 ) + ( Γ12Β21 − Γ11Β22 ) x + ( Γ12 E1 − Γ11 E2 ) = 0

Then solve for p,

p=
( Γ12 Β11 − Γ11Β12 ) + ( Γ12 Β21 − Γ11Β22 ) x + ( Γ12 E1 − Γ11 E2 )
( Γ11Γ 22 − Γ12 Γ21 ) ( Γ11Γ 22 − Γ12 Γ 21 ) ( Γ11Γ 22 − Γ12 Γ 21 )
= π1 + π2 x + v

(b)

The reduced form equation will be p = π1* + π*2 x + v* . Express π1* and π*2 in terms of parameters

Γij and Βij .

( Γ12 Β11′ − Γ11′ Β12 ) ( Γ12 Β′21 − Γ11′ Β22 ) ( Γ12 E1 − Γ11′ E2 )
p= + x+
( Γ11′ Γ 22 − Γ12 Γ′21 ) ( Γ11′ Γ 22 − Γ12 Γ′21 ) ( Γ11′ Γ 22 − Γ12 Γ′21 )
Then,

π1* =
( Γ ( 3Β + 2Β ) − ( 3Γ + 2Γ ) Β )
12 11 12 11 12 12

( ( 3Γ + 2Γ ) Γ − Γ ( 3Γ + 2Γ ) )
11 12 22 12 21 22

3Γ12 Β11 + 2Γ12 Β12 − 3Γ11Β12 − 2Γ12 Β12

=
3Γ11Γ 22 + 2Γ12 Γ 22 − 3Γ12 Γ 21 − 2Γ12 Γ 22
3 ( Γ12 Β11 − Γ11Β12 )
=
3 ( Γ11Γ 22 − Γ12 Γ 21 )
= π1

Similarly,

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 8

( Γ12 Β′21 − Γ11′ Β22 )

π*2 =
( Γ11′ Γ 22 − Γ12 Γ′21 )
=
( Γ12 ( 3Β11 + 2Β12 ) − ( 3Γ11 + 2Γ12 ) Β22 )
( ( 3Γ11 + 2Γ12 ) Γ 22 − Γ12 ( 3Γ 21 + 2Γ 22 ) )
3Γ12 Β11 + 2Γ12 Β12 − 3Γ11Β 22 − 2Γ12 Β 22
=
3Γ11Γ 22 + 2Γ12 Γ 22 − 3Γ12 Γ 21 − 2Γ12 Γ 22
3 ( Γ12 Β11 − Γ11Β 22 )
=
3 ( Γ11Γ 22 − Γ12 Γ 21 )
= π2

What is the point of this long, tedious exercise? Based on the reduced form coefficients we
cannot determine the true demand or supply equation from an arbitrary mixture, linear
combination, of the original equations. This is unfortunate because we can consistently estimate
the reduced form coefficients, but without further information about the structural equations
we cannot tell anything about the structural parameters.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 9

EXERCISE 11.4

(a)

q = − p + 3 + 2 x + e1 = − ( q + 1 + x + e2 ) + 3 + 2 x + e1

1 1
q =1+ x + ( e1 − e2 )
2 2

Then

 1 1 
p =  1 + x + ( e1 − e2 )  + 1 + x + e2
 2 2 
3 1
= 3 + x + ( e1 + e2 )
2 2

(b)

Demand: q = − 5 p + 11 + 8 x + e1*

Supply: p = q + 1 + x + e2

q = − 5 p + 11 + 8 x + e1*
= − 5 ( q + 1 + x + e2 ) + 11 + 8 x + e1*
1
=1+ x + ( e1* − 5e2 )
2

Then

 1 
p = 1 + x + ( e1* − 5e2 )  + 1 + x + e2
 2 
3
= 2 + x + ( e1 − 4 e2 )
*

(c)

3 ( − q − p + 3 + 2 x + e1 ) + 2 ( q − p + 1 + x + e2 )
= ( − 3q + 2 q ) + ( − 3 p − 2 p ) + ( 9 + 2 ) + ( 6 x + 2 x ) + ( 3e1 + 2 e2 )
= − q − 5 p + 11 + 8 x + 3e1 + 2 e2 = 0

q = − 5 p + 11 + 8 x + ( 3e1 + 2 e2 )

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 10

(d)

No, the demand equation is not identified. In a system of M = 2 equations there must be at least
M – 1 variables omitted. This demand equation has no variables omitted.

(e)

Yes that is true. If x is exogenous it is uncorrelated with e1 and e2 and thus the reduced form
errors as well.

(f)

No. Both the demand equations lead to the same reduced forms.

(g)

No, the demand equation is not identified. In a system of M = 2 equations there must be at least
M – 1 variables omitted. This demand equation has no variables omitted.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 11

EXERCISE 11.5

(a)

For q:

q = − p + 3 + 2 x + e1
= − ( q + 1 + e2 ) + 3 + 2 x + e1


q =1+ x +
( e1 − e2 )
2

For p:

 (e − e ) 
p = 1 + x + 1 2  + 1 + e2
 2 
( e + e )
=2+ x+ 1 2
2

(b)

Demand: q = −5 p + 11 + 6x + e1*

Supply: p = q + 1 + e2

For q:

q = −5[ q + 1 + e2 ] + 11 + 6x + e1*
( e1 − e2 )
=1+ x +
2

This is the same reduced form equation as in part (a) for q. Therefore the reduced form for p
stays the same as well.

p = 2+ x+
( e1 + e2 )
2

(c)

Demand: q = − p + 3 + 2 x + e1  − q − p + 3 + 2 x + e1 = 0

Supply: p = q + 1 + e2

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 12

q = − p + 3 + 2 x + e1  − q − p + 3 + 2 x + e1 = 0

p = q + 1 + e2  q − p + 1 + e2 = 0

Then

3 ( − q − p + 3 + 2 x + e1 ) + 2 ( q − p + 1 + e2 ) = 0
 − q − 5 p + 11 + 6 x + ( 3e1 + 2e2 ) = 0
 q = −5 p + 11 + 6 x + ( 3e1 + 2e2 )

(d)

No. There is an endogenous variable on the right hand side. The supply equation is

p = q + 1 + e2 and the reduced form equation for q is q = 1 + x +

( e1 − e2 ) , which is a function
2
of e2 . Therefore, assuming E ( e2 | x ) = 0 because x is exogenous, and E ( xe2 | x ) = 0 because

x is exogenous, and assuming cov ( e1 , e2 | x ) = 0 and that the errors are conditionally

homoskedastic, E ( e22 | x ) = σ 22 , we have

 (e − e )  
cov ( q, e2 | x ) = E ( qe2 | x ) = E  1 + x + 1 2  e2 | x 
 2  

= E  e2 + xe2 +
( e1e2 − e22 )  
 | x
 2  
 

σ 22
=0+0+0−
2

Therefore the OLS estimator of the supply equation is biased and inconsistent.

(e)

Yes. The reduced form equations are

( e1 − e2 )
q =1+ x +
2
( e1 + e2 )
p =2+ x+
2

The right-hand side variable x is exogenous and uncorrelated with the random errors e1 and e2
, so that the OLS estimator is consistent.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 13

(f)

No. Presumably x is “income”, which affects demand and not supply. So the given equation
could not be a supply equation based on economic theory.

(g)

There are M = 2 equations, so to be identified at least M − 1 = 1 variable must be omitted from

an equation. The demand equation includes all the variables and is not identified. Supply omits
x and satisfies the necessary condition for identication.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 14

EXERCISE 11.6

(a)

The reduced form for q is

q = α1 ( β1q + β2 + e2 ) + α 2 + α3 x + e1
= α1β1q + α1β2 + α1e2 + α 2 + α3 x + e1

q (1 − α1β1 ) = ( α1β2 + α 2 ) + α3 x + ( α1e2 + e1 )


q=
( α1β2 + α 2 ) + α3 x + ( α1e2 + e1 )
(1 − α1β1 ) (1 − α1β1 ) (1 − α1β1 )
= π11 + π21 x + v1

The reduced form for p is

 (α β + α2 ) α3 (α e + e ) 
p = β1  1 2 + x + 1 2 1  + β2 +
 (1 − α1β1 ) (1 − α1β1 ) (1 − α1β1 ) 
β1 ( α1β 2 + α 2 ) β1α 3 β (α e + e )
= + β2 + x + 1 1 2 1 + e2
(1 − α1β1 ) (1 − α1β1 ) (1 − α1β1 )
β ( α β + α 2 ) + β 2 (1 − α1β1 ) β1α 3 β ( α e + e ) + (1 − α1β1 ) e2
= 1 1 2 + x+ 1 1 2 1
(1 − α1β1 ) (1 − α1β1 ) (1 − α1β1 )
β1α 2 + β 2 β1α 3 βe +e
= + x+ 1 1 2
(1 − α1β1 ) (1 − α1β1 ) (1 − α1β1 )
= π12 + π 22 x + v2

(b)

Demand is not identified so only the supply parameters can be obtained.

α3 β1α 3
π21 = and π22 =
(1 − α1β1 ) (1 − α1β1 )
So that

π22 6 5
β1 = = =2
π21 3 5

To solve for β2 we use

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 15

β1α 2 + β2 ( α β + α2 )
π12 = and π11 = 1 2
(1 − α1β1 ) (1 − α1β1 )
Then

β2 + α 2β1 − α1β1β2 − α 2β1 β2 (1 − α1β1 )

π12 − β1π11 = = = β2
1 − α1β1 1 − α1β1

That is,

7 1
β2 = − 2  = 1
5 5

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 16

EXERCISE 11.7

(a)

First the reduced form for q.

q = α1 p + α 2 x + α 3 w + e1
= α1 ( β1q + e2 ) + α 2 x + α 3 w + e1

So that

α2 α3
q= x+ w + ( α1e2 + e1 )
1 − α1β1 1 − α1β1
= π11 x + π 21 w + v1

Then the reduced form for p is

 α2 α3 
p = β1  x+ w + ( α1e2 + e1 )  + e2
 1 − α1β1 1 − α1β1 
βα βα
= 1 2 x + 1 3 w + β
1 − α1β1 1 − α1β1  1 ( α1e2 + e1 ) + e2 
= π12 x + π 22 w + v2

(b)

π12 π22 2 5
Only supply is identified. β1 = = = = 2 . In this system there are two ways to
π11 π21 1 5

estimate β1 , and if we know the true reduced form parameters each approach gives the same
solution. If we did not know the true reduced form parameters then we would wind up with
two different estimates of β1 . This is an example of an overidentified equation. There are M =
2 equations and we must omit at least M − 1 = 1 to identify it. The supply equation omits 2
variables.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 17

EXERCISE 11.8
(a)

CONSUMPt = β1 + β 2 INCOMEt + et
= β1 + β 2 ( CONSUMPt + INVt ) + et

β1 β 1
CONSUMPt = + 2 INVt + et
1 − β2 1 − β2 1 − β2

(b)

β1 β 1
INCOMEt = CONSUMPt + INVt = + 2 INVt + et + INVt
1 − β2 1 − β2 1 − β2
β1 β 1 − β2 1
= + 2 INVt + INVt + et
1 − β2 1 − β2 1 − β2 1 − β2
β1 1 1
= + INVt + et
1 − β2 1 − β2 1 − β2

Then, assuming the expected value of the random error is zero, and that the variance of the
random error is homoskedastic, we have

 β 1 1  
cov ( INCOMEt , et ) = E ( INCOMEt et ) = E  1 + INVt + et  et 
 1 − β2 1 − β 2 1 − β2  
 β 1 1 2 
= E  1 et + INVt et + et  
 1 − β2 1 − β2 1 − β2  
 β 1 1 2 
= E  1 et + INVt et + et  
 1 − β2 1 − β2 1 − β2  
β 1 1
= 1 E ( et ) + E ( INVt et ) + E ( et2 | INVt )
1 − β2 1 − β2 1 − β2
1 1
=0+ E ( INVt et ) + σe2
1 − β2 1 − β2

Thus INCOME is endogenous in the consumption relationship even if E ( INVt et ) = 0

(c)

cov ( INCOMEt , INVt ) = E ( INVt − E ( INVt ) ) ( INCOMEt − E ( INCOMEt ) ) 

The expected value of INCOME is

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 18

β1 1 1
E ( INCOMEt ) = + E ( INVt ) + E ( et )
1 − β 2 1 − β2 1 − β2
β1 1
= + E ( INVt )
1 − β 2 1 − β2

Therefore
β1 1 1  β 1 
INCOMEt − E ( INCOMEt ) = + INVt + et −  1 + E ( INVt ) 
1 − β2 1 − β 2 1 − β2  1 − β2 1 − β2 
INVt − E ( INVt ) 1
= + et
1 − β2 1 − β2

Then

cov ( INCOMEt , INVt ) = E ( INVt − E ( INVt ) ) ( INCOMEt − E ( INCOMEt ) ) 

  INVt − E ( INVt ) 1 
= E ( INVt − E ( INVt ) )  + et  
  1 − β2 1 − β2  
  INV − E ( INV )  2 1 
= E 
t 


t

1 − β2
+
1 − β2
( t ( t ) ) t 
INV − E INV e
 
var ( INVt ) cov ( INVt , e )
= +
1 − β2 1 − β2

(d)

Yes. If INVt is uncorrelated with the random error then we find that it is positively correlated
with INCOMEt which is one primary property an instrument must have to be useful in
estimating the consumption relation.

var ( INVt )
cov ( INCOMEt , INVt ) = >0
1 − β2

It also simplifies the endogeneity story as the correlation between income and the random error
becomes

1
cov ( INCOMEt , et ) = σe2 > 0
1 − β2

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 19

EXERCISE 11.9

(a)

yi1 = α1 yi 2 + α 2 xi1 + ei1 = α1 ( α 2 yi1 + β1 xi1 + ei 2 ) + α 2 xi1 + ei1

=
( α1β1 + α 2 ) x +
α1ei 2 + ei1
i1
1 − α1β1 1 − α1β1

(b)

cov ( yi1 , ei 2 | x1 ) = E ( yi1ei 2 | x1 )

  ( α β + α2 ) α e +e  
= E 1 1 xi1 + 1 i 2 i1  ei 2 | x1 
 1− α β 1 − α1β1  
 1 1 
 ( α β + α2 )   α e2 + e e 
= E 1 1 xi1ei 2 | x1  + E  1 i 2 i1 i 2 | x1 
 1 − α1β1   1 − α1β1 
2
α σ + σ12
= 1 2
1 − α1β1

(c)

α1σ22 + σ12 0σ22 + 0 0

cov ( yi1 , ei 2 | x1 ) = = = =0
1 − α1β1 1 − 0β1 1

(d)

Under the conditions in (c) the two equations are

yi1 = α 2 xi1 + ei1

yi 2 = α 2 yi1 + β1 xi1 + ei 2

The first equation can be consistently estimated by OLS because it has a single exogenous
variable on the right-hand side.

(e)

Yes, because under the conditions in (c) yi2 is not correlated with ei2 and is thus not endogenous.
OLS can be used to estimate the second equation consistently.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 20

EXERCISE 11.10
(a)

M = 2, so the necessary condition is that M − 1 = 1 variable be omitted from each equation. In

the supply equation PS is omitted. In the demand equation PF is omitted. Thus both equations

satisfy the necessary condition.

(b)

Column (2), the reduced form equation for P is relevant for both supply and demand equations.

For the supply equation the coefficient of PS must be significantly different from zero. The t –

value is t = 2.4522/0.3991 = 6.14. The corresponding F-statistic is F = t2 = 37.75 > 10 satisfying

the rule of thumb threshold.

For the demand equation the coefficient of PF must be significantly different from zero. The t-

value is t = 2.1635/0.3053 = 7.09. Using F = t2 = 50.22 we find that the rule of thumb threshold

is also satisfied.

(c)

Let a1, a2, and a3 be the estimated demand coefficients, and let b1, b2, and b3 be the estimated

supply coefficients. Let PS * and PF * be the given values. Then

Qˆ d = ( a1 + a3 PS * ) + a2 P = a0 + a2 P = 26.439895 − 0.12766334 P
Qˆ s = ( b1 + b3 PF * ) + b2 P = b0 + b2 P = −4.0139699 + 0.35417327 P

Plotting these curves we have

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 21

(d)

Setting the two equations in (c) equal and solving yields an equilibrium price of $63.20 and an
equilibrium quantity of 18.37.

(e)

Increasing the cost of an input reduces supply. The new supply curve equation is
Qˆ s′ = −11.311152 + 0.35417327 P . The graphical representation is

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 22

(f)

The new equilibrium price is $78.348 and the new equilibrium quantity is 16.438. The
percentage change in quantity is −10.524%, and the percentage change in PF is 30.437%. Thus
the “elasticity” is −0.3458%

(g)

Suppose the fitted reduced form is Pˆ = πˆ 1 + πˆ 2 PF + πˆ 3 PS . The estimated elasticity is

PF
εˆ = πˆ 2
πˆ 1 + πˆ 2 PF + πˆ 3 PS

Using PF = 23 and PS = 22, the estimated elasticity is −0.346. In order to compute the standard

error you must use the delta method and the estimated coefficients and their variances and

covariances. Using appropriate software you will find the standard error to be 0.138, with a

95% interval estimate [−0.6296, −0.0620], using the critical value t( 0.975,27 ) = 2.0518 .

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 23

EXERCISE 11.11
(a)

Demand is not identified. There are M = 2 equations so the necessary condition is that at least
M − 1 = 1 variables omitted from an equation for it to be identified. Demand has no variables
omitted. Supply is identified, as it excludes PS and DI.

(b)

Do you judge the omitted exogenous variables (instruments) strong enough to estimate the
identified equation(s)? Explain.

To be adequately strong in the first stage regression P = π1 + π 2 PS + π3 DI + v we must reject

the null hypothesis H 0 : π 2 = 0, π3 = 0 with a large F-value. The rule of thumb threshold to
reject the notion that the instruments are weak is F > 10. Here the overall F = 54.21 which is
large, and we conclude the instruments are not weak. Furthermore, the F-test statistic for
H 0 : π 2 = 0 is F = 10.45, and the F-test statistic for H 0 : π3 = 0 is F = 48.34, so each of the
demand shift variables is strongly significant.

(c)

Qˆ = 8.6455 + 0.1564 P = 8.6455 + 0.1564 ( 62.724 ) = 18.458333 = Q

(d)

εˆ = b2 ( P Q ) = 0.1564 ( 62.724 18.458 ) = 0.532

Using the estimates from Table 11.3b

εˆ = b2 ( P Q ) = 0.338 ( 62.724 18.458 ) = 1.149

With the revised supply equation the estimated price elasticity is only 0.532, compared to 1.149
from the full model.

(e) Comparing the results in parts (b) and (c) to those in Example 11.1, do you think we should
include PF in the supply equation? Explain.

PF should be included. In the full model the reduced form for PRICE is

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 24

P = π12 + π 22 PS + π32 DI + π 42 PF + v2

The F-test of H 0 : π22 = 0, π32 = 0 yields an F = 41.49, so that PS and DI are strong instruments.
The t-value for PF in the estimated supply equation is t = −12.13. We reject the null hypothesis
that β3 = 0 . The 95% interval estimate is [ −1.17, − 0.83] which is reasonably narrow.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 25

EXERCISE 11.12
(a)

Because of the auxiliary supply relationship, if we assume market wage and hours are jointly

determined, then HOURS in the wage equation is endogenous and correlated with the

random error e1. In this case the OLS estimator of the wage equation will be biased and

inconsistent.

(b)

In a system of M = 2 equations at least M − 1 = 1 variables must be omitted from an equation,

and appear elsewhere in the model, to be identified. For the wage equation KIDS and

ln ( HWAGE ) are omitted. Variation in these variables shifts the supply curve relative to the

demand, identifying it. In a figure analogous to Figure 11.4, the supply curve would be

shifting relative to a fixed demand (wage) equation.

(c)

Two stage least squares can be described as follows. In the first stage we estimate

HOURS = π1 + π2 EDUC + π3 EXPER + π4 EXPER 2 + π5 KIDS + π6 ln ( HWAGE ) + v

 from the reduced form equation. Note that the fitted values
Obtain the fitted values HOURS

depend only on exogenous variables.

 in the wage equation. Estimate this

In the second stage we replace HOURS by HOURS

resulting equation by OLS. That is,

 + β EDUC + β EXPER + β EXPER2 + e*

ln(WAGE ) = β1 + β2 HOURS 3 4 5 1

This estimation process yields consistent estimators for the wage equation parameters. The

standard errors from the second stage OLS estimation are not correct, as the error variance

σ12 = var ( e1 ) is not correctly computed.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 26

(d)

In order for 2SLS estimation process to be reliable the null hypothesis H 0 : π5 = 0, π6 = 0

must produce a large F-statistic value, greater than 10. This ensures that the two variables

in the supply equation that are not in the demand equation do substantially shift supply

relative to demand, making demand identified.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 27

EXERCISE 11.13
This exercise is adapted from “Econometric Models, Techniques and Applications, Second
Edition,” by Michael D. Intriligator, Ronald G. Bodkin and Cheng Hsiao (1996) Prentice-Hall,
pp. 468-469. The reported results are from Teigen, R. (1964) “Demand and Supply Functions
for Money in the United States: Some Structural Estimates,” Econometrica, 32, 476-509. Any
errors are ours.

(a)

M t −1 is an exogenous variable in the demand equation with a positive and significant

coefficient. If ΔM t −1 = 1 then the demand curve shifts right (up) indicating an increase in

demand. On the other hand ΔM t −1 > 0 has no effect on the supply curve.

(b)

∂Mˆ t ∂GNPt = 0.0618 − 0.0025R . If the short term interest rate is less than 24.72%, then the
increase in GNP will increase demand, shifting the demand curve to the right (up). A change
in GNP does not affect supply.

(c)

∂Mˆ t ∂Rd ,t = −0.0751M t* . Since M t* > 0 , the increase in the discount rate reduces the money

supply, shifting the supply curve to the left (down).

(d)

Identification requires that at least one variable be omitted from each of the two equations that
appears elsewhere in the system. The demand equation omits Rd ,t , the discount rate. Changs

in Rd ,t shift the supply curve relative to demand.

GNP and M t −1 are omitted from supply. Changes in these variables shift demand relative to
supply.

Both equations satisfy the necessary condition for identification.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 28

EXERCISE 11.14
Note: This question is adapted from “Modelling the Demand and Supply Determinants of
Australian Wine Grapes,” by Edward Oczkowski (March 2014) Economic Papers, (33,1), 1-
12. We have omitted equation intercepts and other variables reported in the paper. The
estimation technique is actually more complex than 2SLS

(a)

INCOME, XRATE and STOCKS.

(b)

Qt − 4 , TIME, TEMP, NON t − 4 and PREM t − 4

(c)

In the demand equation the signs of the coefficients are consistent with expectations. Q is
inversely proportional to P, as predicted by the law of demand. The effect of INCOME is
positive, hence wine is a normal good. The exchange rate is foreign currency per Australian
dollar. When the exchange rate increases the price of Australian wine increases in the U.S., for
example, reducing demand. The sign of XRATE’s coefficient should be negative. Because Qt
is demand for current production, if there are larger stocks from the previous year the effect of
increased stocks should be negative.

(d)

ε p = −0.278 ( 6.06 4.98 ) = −0.338

ε INC = 2.884 (1.66 4.98 ) = 0.961

(e)

To explain the coefficients of the supply equation let Qt* = αPt + et be the desired production

at time t. The partial adjustment model assumes Qt − Qt − 4 = λ ( Qt* − Qt − 4 ) where 0 < λ < 1 .

Substituting and rearranging we obtain

Qt = λαPt + (1 − λ ) Qt − 4 + λet

Therefore 1
− λ = 0.682 , or λˆ = 0.318 and αˆ = βˆ 1 λˆ = 0.824 0.318 = 2.59 .

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 29

The positive coefficient on TIME reflects technological progress. Higher temperatures during
the summer reduce output. The positive sign attached to the price of non-premium grapes
suggests that they are a production substitute and the positive sign on the price of premium
wine grapes suggests that they are a production complement.

(f)

The elasticity at the means is

αˆ ( P Q ) = 2.59 ( 6.06 4.98 ) = 3.15

This suggests a very elastic supply, and that in equilibrium suppliers are very responsive to
price changes.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 30

EXERCISE 11.15
(a)

The signs of β5 and β6 are probably negative. An increase in the number of children will
reduce supply, shifting it down (right). This will reduce the number of hours worked and
increase the market wage rate, as illustrated in the figure below.

(b)

An increase in experience will increase demand, shifting the demand relation upwards (right).
This will result in higher equilibrium wages and hours. This is illustrated in the figure below.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 31

(c)

In a system of M = 2 equations at least M – 1= 1 variable must be omitted from an equation to

identify it. The supply equation omits experience and its square, the necessary condition holds.

The reduced form equation is

HOURS = π1 + π2 EDUC + π3 AGE + π4 KIDSL6 + π5 KIDS 618

+ π6 NWIFEINC + π7 EXPER + π8 EXPER 2 + v1

To test instrument strength we test the joint null hypothesis H 0 : π7 = 0, π8 = 0 .

(d)

The demand equation omits AGE, KIDSL6, KIDS618, and NWIFEINC. The necessary
condition that one variable in the system be omitted is satisfied.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 32

ln(WAGE ) = γ1 + γ 2 EDUC + γ 3 AGE + γ 4 KIDSL6 + γ 5 KIDS 618

+ γ 6 NWIFEINC + γ 7 EXPER + γ 8 EXPER 2 + v2

Test the joint null hypothesis H 0 : γ 3 = 0, γ 4 = 0, γ 5 = 0, γ 6 = 0

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 33

EXERCISE 11.16
(a)

Set Q from demand and supply equal and simplify.

α1 + α2 Pi + edi = β1 + β2 Pi + β3Wi + esi ,

Then

( α 2 − β2 ) P = (β1 − α1 ) + β3W + ( es − ed )


P=
(β1 − α1 ) + β3 W + ( es − ed ) = π + π W + v
( α 2 − β2 ) ( α 2 − β2 ) ( α 2 − β2 ) 1 2 1

Then

 ( β − α1 ) 3 (e − e ) 
Qi = α1 + α2  1 + W + s d  + edi
 ( α2 − β2 ) ( α2 − β2 ) ( α2 − β2 ) 
α2β1 − β2 α1 α 2β
= + W + v2 = θ1 + θ2W + v2
( α 2 − β2 ) ( α 2 − β2 )
(b)

The demand equation is identified because it omits W. The supply equation is not identified
because no variables in the system are omitted. That means we can solve for the demand
parameters from the reduced form parameters.

(c)

αˆ 2 = θˆ 2 πˆ 2 = 0.5 and αˆ 1 = θˆ 1 − αˆ 2 πˆ 1 = 3.8

(d)

The 2SLS estimates are identical to the estimates in part (c). The Stata estimates are

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 34

Instrumental variables (2SLS) regression

Source SS df MS Number of obs = 5

F( 1, 3) = 0.12
Model .7 1 .7 Prob > F = 0.7533
Residual 25.3 3 8.43333333 R-squared = 0.0269
Adj R-squared = -0.2974
Total 26 4 6.5 Root MSE = 2.904

q Coef. Std. Err. t P>|t| [95% Conf. Interval]

p .5 1.45201 0.34 0.753 -4.120944 5.120944

_cons 3.8 6.519509 0.58 0.601 -16.94799 24.54799

Instrumented: p
Instruments: w

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 35

EXERCISE 11.17
(a)

There are M = 8 equations requiring 7 omitted variables in each equation. There are a total of
16 variables in the system.

The consumption equation includes 6 variables and omits 10. The necessary condition is
satisfied. The investment equation includes 5 variables and omits 11. The necessary condition
is satisfied. The private sector wage equation includes 5 variables and omits 11. The necessary
condition is satisfied.

(b)

The consumption equation has 2 RHS endogenous variables and excludes 5 exogenous
variables. The investment and private wage equations have 1 RHS endogenous variable and
omit 5 exogenous variables.

(c)

W1t = π1 + π2Gt + π3W2t + π4TX t + π5TIMEt + π6 Pt −1 + π7 Kt −1 + π8 Et −1 + v

(d)

Obtain fitted values Wˆ1t from the estimated reduced form equation in part (c) and similarly

obtain Pˆt . Create Wt* = Wˆ1t + W2t . Regress CN t on Wt * , Pt and Pt −1 plus a constant by OLS.

(e)

The coefficient estimates will be the same. The t-values will not be because the standard errors
in part (d) are not correct 2SLS standard errors.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 36

EXERCISE 11.18
(a)

Identification requires 1 omitted variable from each equation. Lagged capital stock appears in
the investment equation, but it is not in the consumption equation. Therefore the consumption
equation is identified. The investment equation excludes TIME which does appear in the
consumption equation, therefore the necessary equation is met.

(b)

CN t = α1 + α 2 ( β1 + β2 CN t + β3 K t −1 + e2t ) + α3TIMEt + e1t



CN t =
α1 + α 2β1 α 2β3
+
α ( e + α 2 e2 )
K t −1 + 3 TIMEt + 1
1 − β2 1 − β2 1 − β2 1 − β2
= π1 + π2 K t −1 + π3TIMEt + v1t

(c)

The OLS estimates appear in columns (1) and (2) of Table XR 11.18. The coefficients on
investment and time are positive and significant. In the investment equation the coefficient of
consumption is positive and significant and that for lagged capital stock is negative and
significant. The signs are as one would expect.
Table XR 11.18
(1) (2) (3) (4) (5) (6)
CN OLS I OLS CN IV I IV CN RF I RF
C 52.0478 27.5973 54.0562 25.3488 48.6524 23.0853
(98.82) (2.71) (26.77) (1.82) (1.85) (1.37)
I 1.3552 -0.2341
(9.09) (-0.18)
TIME 1.0054 0.7888 0.7975 -0.0371
(12.58) (3.06) (3.71) (-0.27)
CN 0.3337 -0.0465
(4.10) (-0.27)
KLAG -0.2208 -0.1079 0.0255 -0.1091
(-3.92) (-1.26) (0.19) (-1.30)
N 22 22 22 22 22 22
R2 .9113729 .5430357 .3814895 .0169123 .5270229 .1410895
SSE 100.4832 116.2004 701.2517 249.9871 536.2496 218.4104
t statistics in parentheses

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 37

(d)

These estimates are in columns (3) and (4) of Table XR 11.18. The results for the consumption
equation are quite different. The estimated coefficient of investment is negative but
insignificant. The coefficient on time remains positive and significant. For the investment
equation both coefficients are negative and insignificant.

(e)

These estimates are in columns (5) and (6) of Table XR 11.18. In the reduced form equation
for consumption the lagged capital stock is insignificant. That means that despite the necessary
condition being satisfied in theory for the consumption equation, it is not in practice. We cannot
conclude that the lagged capital stock is not a weak instrument. Thus the 2SLS results are not
reliable.

In the investment reduced form TIME is not significant so we have the same situation as for
the consumption equation. Without TIME being very significant there is no real omitted
variable in the investment equation, making 2SLS unreliable.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 38

EXERCISE 11.19
(a)

Summary statistics for women not in the labor force, LFP = 0

variable N mean Std. dev. min max

AGE 325 43.28308 8.467796 30 60

KIDSL6 325 .3661538 .6368995 0 3

KIDS618 325 1.356923 1.327065 0 7

NWIFEINC 325 21698.05 12728.15 1500 96000

Summary statistics for women in the labor force, LFP = 1

variable N mean Std. dev. min max

AGE 428 41.97196 7.721084 30 60

KIDSL6 428 .1401869 .3919231 0 2

KIDS618 428 1.350467 1.315935 0 8

NWIFEINC 428 18937.48 10591.35 -29.0575 91000

Women not in the labor force are on average slightly older, and have on average more kids
under 6 years of age, and a higher income from their partners.

(b)

β2 > 0 : A higher wage leads to an increased quantity of labor supplied.

β3 : The effect of an increase in education is unclear.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 39

β4 : This sample has been taken for working women between the ages of 30 and 60. It is

not certain whether hours worked increases or decreases over this age group.

β5 < 0, β6 < 0 : The presence of children in the household reduces the number of hours

worked because they demand time from their mother.

β7 < 0 : As income from other sources increases, it becomes less necessary for the woman

to work.

NWIFEINC measures the sum of all family income excluding the wife’s income.

(c)

The estimates are in column (1) of Table XR 11.19. Well we certainly did not expect the
coefficient of ln(WAGE ) to be negative and insignificant. It is meant to be a supply equation
so that coefficient should be positive and significant.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 40

Table XR 11.19
(1) (2) (3)
OLS Red Form 2SLS
C 2114.6973 -0.1620 2478.4349
(6.22) (-0.54) (3.78)
LWAGE -17.4078 1772.3233
(-0.32) (2.98)
EDUC -14.4449 0.1011 -201.1870
(-0.80) (6.69) (-2.88)
AGE -7.7300 -0.0055 -11.2289
(-1.40) (-1.04) (-1.07)
KIDSL6 -342.5048 -0.0697 -191.6588
(-3.42) (-0.79) (-0.98)
KIDS618 -115.0205 -0.0207 -37.7325
(-3.73) (-0.74) (-0.59)
NWIFEINC -0.0042 0.0000 -0.0100
(-1.16) (1.80) (-1.39)
EXPER 0.0175
(3.60)
N 428 428 428
R2 0.0670 0.1572 .
RMSE 755.1606 0.6686 1430.5254
t statistics in parentheses

(d)

The estimates are in column (2) of Table XR 11.19. The coefficient of education implies an
additional year of education will increase wages by approximately 10.11% holding other
factors constant. And the estimate is statistically significant a the 1% level.

(e)

In order for the supply equation to be identified the coefficient on EXPER should be very
significant, with an F – test statistic value of at least 10. The t-value is 3.60 which translates
into F = 12.96, satisfying the rule of thumb threshold for an instrument that is not weak.

(f)

These estimates are in column (3) of Table XR 11.19. We see that ln(WAGE ) now has a
positive and statistically significant coefficient. The coefficient of EDUC is negative and
significant, but the rest of the coefficients are insignificant.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 41

EXERCISE 11.20
(a)

The endogenous variables are the logs of Q = per capita consumption of chicken and P = real
price of chicken. The exogenous variables are logs of Y = real per capita income and PB = real
price of beef, and POPGRO = rate of population growth

(b)

The estimates are in Table XR 11.20 column (1). We see that the coefficient of log price (LP)
is negative but not significant. The coefficient of log income (LY) is positive and significant. It
suggests that a 1% increase in income increases the consumption of chicken by 0.99%. The
coefficient of the log of the price of beef is negative and insignificant. Population growth is
positive and significant. We estimate that a 1% increase in population growth increases per
capita chicken consumption by 0.0017 pounds
Table XR 11.20
(1) (2) (3) (4) (5)
part (b) part (d) part (e) part (g) part (h)
C -6.232*** -5.708*** -5.412 -5.412 -5.708***
(0.701) (0.876) (6.297) (10.18) (0.545)
LP -0.155 -0.255* -0.255***
(0.0864) (0.125) (0.0487)
LY 0.990*** 0.932*** 0.856 0.856 0.932***
(0.0693) (0.0867) (0.630) (1.062) (0.0507)
LPB -0.155 -0.0990 0.219 0.219 -0.0990**
(0.0927) (0.0897) (0.234) (0.345) (0.0342)
POPGRO 0.170*** 0.223*** -0.0231 -0.0231 0.223***
(0.0355) (0.0375) (0.118) (0.185) (0.0544)
TIME -0.0505* -0.0505
(0.0216) (0.0333)
LPF 0.177 0.177
(0.108) (0.127)
[Link] -0.141 -0.141
(0.327) (0.269)
[Link] 2.322** 2.322**
(0.709) (0.844)
N 40 39 39 39 39
R2 0.981 0.982 0.906 0.982
RMSE 0.0428 0.0375 0.0699 0.0375
Standard errors in parentheses. The prefix L. indicates a lagged value. The prefix L indicates the
natural log
*
p < 0.05, ** p < 0.01, *** p < 0.001

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 42

(c)

The correlogram is below. It shows the first two autocorrelations, 0.5663 and 0.4038, to be
significant.

The LM test for AR(1) errors yields a test statistic of 13.362. The test critical value is
χ(20.95,1) = 3.841 . We reject the null hypothesis of no serial correlation.

(d)

These estimates are in column (2) of Table XR 11.20. We find the price elasticity estimated to
be −0.255 and significant at the 5% level. The other results are essentially unchanged. But the
effect of using 2SLS is have a demand curve that is estimated to be downward sloping, as we
would expect.

(e)

These estimates are in column (3) of Table XR 11.20. The coefficient of TIME is statistically
significant at the 5% level, and ln ( EXPTSt −1 ) is significant at the 1% level. The F-test of joint

significance yields F = 3.81, with a p-value of 0.0124 using the F( 4,31) distribution. Thus the

variables are jointly significant at the 5% level, but that is not really the important point. In
order to conclude that the instruments are not weak we require F > 10. This condition does not
hold and we should not conclude that our instruments are adequately strong. Thus we cannot
have too much confidence in the 2SLS estimates.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 43

(f)

The software generated test statistic value for AR(1) errors is 7.047 with a p-value of 0.0079.
We reject the null hypothesis of no serial correlation.

(g)

These estimates are in column (4) of Table XR 11.20. The HAC standard errors are larger and
we no longer find TIME significant, through ln ( EXPTSt −1 ) still is. The F-test of joint

significance yields F = 3.81, with a p-value of 0.0124 using the F( 4,31) distribution. Thus the

variables are jointly significant at the 5% level, but that is not really the important point. In
order to conclude that the instruments are not weak we require F > 10. This condition does not
hold and we should not conclude that our instruments are adequately strong. Thus we cannot
have too much confidence in the 2SLS estimates.

(h)

The correlogram for the 2SLS residuals shows one significant autorcorrelation.

The 2SLS estimates using HAC standard errors are in column (5) of Table XR 11.20. We see
a huge change in the significance levels of the coefficients. They are all now significant at the
0.001 level of significance. The coefficient estimates themselves are unchanged. The sign of
the log of the real price of beef is still a bit of a puzzle. If beef is a substitute for chicken then

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 44

its sign should be positive, indicating that as the price of beef rises the consumption of chicken
increases.

(i)

The Sargan test statistic value is 22.4073 with a p-value < 0.0001, using the χ(23) critical value.

Clearly the validity of the surplus IV is rejected. One or more of the IV are invalid based on
this test, which casts doubt on all the 2SLS results.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 45

EXERCISE 11.21
(a)

QPROD, production, and P, price, are the endogenous variables. The exogenous variables are
PF, feed price, TIME and lagged log production, ln ( QPRODt −1 ) . In this supply relation we

expect β2 > 0, β3 < 0, β4 > 0, β5 > 0 . Higher price should elicit more production, higher feed
price reduces supply, over time technology should improve the production, and desired level
of production may not be achieved in one year, so lagged production is included to capture
“partial adjustment.”

(b)

The estimates are in the first column of Table XR 11.21. All the coefficients have the
anticipated sign, but the coefficient of price is insignificant. The LM test statistic for serial
correlation, AR(1), is 0.039 with a p-value of 0.8444. The correlogram also shows no sign of
serial correlation. The lagged value of production controls for such a problem.
Table XR 11.21
(1) (2) (3)
ols red form 2sls
C 2.154** -5.412 2.214**
(0.782) (6.297) (0.800)
LP 0.0252 0.0446
(0.0671) (0.123)
LPF -0.0999* 0.177 -0.105*
(0.0421) (0.108) (0.0486)
TIME 0.0113* -0.0505* 0.0120*
(0.00503) (0.0216) (0.00589)
[Link] 0.727*** -0.141 0.718***
(0.104) (0.327) (0.110)
LY 0.856
(0.630)
LPB 0.219
(0.234)
[Link] 2.322**
(0.709)
POPGRO -0.0231
(0.118)
N 39 39 39
R2 0.997 0.906 0.997
RMSE 0.0304 0.0699 0.0284
Standard errors in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 46

(c)

The first stage estimates are in column (2) of Table XR 11.21. The only significant external IV
is the log of lagged exports. The joint F-test of the four instruments yields F = 2.71, which is
significant at the 0.05 level, but does not reach the rule of thumb threshold of 10. We cannot
reject the null hypothesis that the IV are weak using the Stock-Yogo test critical values.

(d)

The 2SLS estimates are in column (3) of Table XR 11.21. The results are not qualitatively
different from the OLS estimates.

(e)

The value of the Sargan LM test statistic is 6.5754 with p = 0.0867, using χ(23) . Thus at 5% we

cannot reject the validity of the three surplus IV.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 47

EXERCISE 11.22
(a)

The transformation eliminates the intercept, and the explanatory variables become log-
differences. The parameters of interest are not affected. If ρ = 1, and the AR(1) model is correct,
then we have eliminated serial correlation. The random error vtd is i.i.d. The variables such as

Δ ln ( Qt ) are approximate percentage changes. The accuracy of the approximation declines the

larger the percentage changes. That is, it works best for small percentage changes. The
parameter α 2 is the income elasticity of demand for edible chicken. We expect α 2 < 0
because there is an inverse relationship between price and quantity demanded. We expect
α 3 > 0 because chicken is what economists call a “normal” good. If beef is a substitute for

chicken we expect α 4 > 0 . We expect α 5 > 0 ; as population grows the consumption of

chicken should increase.

(b)

These estimates are in column (1) of Table XR 11.22. We see that among the external IV,
ln ( PFt ) , TIMEt , ln ( QPRODt −1 ) and ln ( EXPTSt −1 ) , only ln ( PFt ) is significant at the 5%

level. The joint F-test statistic is 2.87 which is significant at the 5% level, F( 4,31) , but not near

the rule-of-thumb threshold 10 for IV that are not weak.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 48

Table XR 11.22
(1) (2) (3) (4) (5)
part (b) part (c) part (d) part (e) part (f)
C -3.686 -0.0335 -0.0410* 0.00876
(2.046) (0.0251) (0.0181) (0.0109)
[Link] -0.423** -0.387*
(0.128) (0.153)
[Link] 0.238* 0.242*
(0.0957) (0.0893)
[Link] 2.318** 1.380 1.566* 0.833*** 0.566
(0.655) (0.679) (0.638) (0.157) (0.379)
[Link] 0.446* 0.373 0.357 0.285* 0.295**
(0.194) (0.207) (0.197) (0.116) (0.107)
[Link] 0.0939
(0.843)
[Link]

[Link] -0.0614 -0.0104 0.0632 -0.0152 -0.0214

(0.0996) (0.126) (0.0958) (0.0402) (0.0394)
LPF 0.150
(0.0754)
TIME -0.00961
(0.00959)
[Link] 0.437
(0.260)
[Link] -0.0743
(0.380)
[Link] -0.479
(0.412)
N 39 38 39 39 39
2
R 0.520 0.462 0.458 0.287
RMSE 0.0599 0.0621 0.0607 0.0260 0.0252
Standard errors in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

(c)

These estimates are in column (2) of Table XR 11.22. Among the external IV Δ ln ( PFt ) ,

Δ ln ( QPRODt −1 ) and Δ ln ( EXPTSt −1 ) only Δ ln ( PFt ) is significant at the 5% level. These IV

make more sense logically because we have differenced variables throughout the equation. The
joint F-test statistic is 2.27 which is not significant at the 5% level, F( 3,31) , and not near the

rule-of-thumb threshold 10 for IV that are not weak.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 49

(d)

These estimates are in column 3 of Table XR 11.22. The variable Δ ln ( PFt ) is significant at

the 5% level and the joint F-test statistic is 7.32 which is significant at the 5% level, F(1,34) , but

not near the rule-of-thumb threshold 10 for IV that are not weak. If we are willing to accept a
t-test at 5% on the endogenous variable that actually has a 20% rejection rate, the Stock-Yogo
critical value is 6.66, so with that criteria we can reject the null hypothesis that the IV is weak.

(e)

These estimates are in column 4 of Table XR 11.22. The coefficients on the changes in log
PRICE, log INCOME and log PB, the price of beef have the anticipated signs and are
significant. Only the change in population growth is not significant.

(f)

These estimates are in column 5 of Table XR 11.22. The coefficients on the changes of log
PRICE, and log PB, the price of beef have the anticipated signs and are significant. The change
in population growth is not significant nor is the change in log INCOME.

(g)

An argument for including the constant is that it represents the first difference of TIME. The
significant coefficients are altered. The coefficient of the change in price of chicken is smaller
in magnitude and has a larger standard error. The coefficient on the change in income becomes
smaller, and has a larger standard error, making it statistically insignificant. A type of
collinearity exists when a variable has little variation, thus making it “correlated” with the
constant term. This is perhaps the explanation.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 50

(h)

The autocorrelations are small and insignificant, except for the 3rd lag.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 51

EXERCISE 11.23
(a)

The OLS estimates of the supply equation are in column (1) of Table XR 11.23. We see that
the coefficient of log price is negative and significant. That is quite opposite of what we expect.
The coefficient of STORMY is insignificant. The supply and demand equations in (11.13) and
(11.14) are

ln ( QUAN t ) = α1 + α 2 ln ( PRICEt ) + α3 MON t + α 4TUEt + α5WEDt + α 6THU t + edt

β1 + β 2 ln ( PRICEt ) + β3 STORMYt + est

The reduced form for log price is obtained by setting the log quantity from supply and demand
equal, in equilibrium they are,

β1 + β 2 ln ( PRICEt ) + β3 STORMYt + est = α1 + α 2 ln ( PRICEt ) + α 3 MON t

+ α 4TUEt + α5WEDt + α 6THU t + edt

β 2 ln ( PRICEt ) − α 2 ln ( PRICEt ) = α1 + α 2 ln ( PRICEt ) + α 3 MON t
+ α 4TUEt + α 5WEDt + α 6THU t − β3 STORMYt + edt − est

α1 α3 α4
ln ( PRICEt ) = + MON t + TUEt
β2 − α2 β2 − α2 β2 − α2
α5 α6 β3 e −e
+ WEDt + THU t − STORMYt + dt st
β2 − α2 β2 − α2 β2 − α2 β2 − α2

Then the covariance between ln ( PRICEt ) and est is

 edt − est   −σ 2s
E ln ( PRICEt ) est | X  = E   est | X  =
 β 2 − α 2   β2 − α2

The sign of the covariance is negative, because β2 > 0 and α 2 < 0 so that β 2 − α 2 > 0 . This is
the same result as in Section 11.3.1 and the outcome is the same as well. The OLS estimator of
the supply equation coefficient of log price is negatively biased.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 52

(b)

β1 + β 2 ln ( PRICEt ) + β 3 STORMYt + est = α1 + α 2 ln ( PRICEt ) + α 3 MON t

+ α 4TUEt + α5WEDt + α 6THU t + α 7 RAINY + α8COLD + edt

β 2 ln ( PRICEt ) − α 2 ln ( PRICEt ) = α1 + α 2 ln ( PRICEt ) + α3 MON t
+ α 4TUEt + α 5WEDt + α 6THU t + α 7 RAINY + α8COLD − β3 STORMYt + edt − est

α1 α3 α4
ln ( PRICEt ) = + MON t + TUEt
β2 − α2 β2 − α2 β2 − α2
α5 α6 α7 α8 β3 e −e
+ WEDt + THU t + RAINY + COLD − STORMYt + dt st
β2 − α2 β2 − α2 β2 − α2 β2 − α2 β2 − α2 β2 − α2

(c)

These estimates are in column (2) of Table XR 11.23. We see that the coefficients of RAINY
and COLD are not significant and the joint F-statistic is 0.60. For a F( 2,103) distribution the p =

0.5519. Thus the coefficients of RAINY and COLD are not significant

(d)

The joint F = 0.61. For a F( 6,103) distribution the p = 0.7229. The variables are not jointly

significant.

(e)

These estimates are in column (3) of Table XR 11.23. As you can see nothing is significant and
the wrong sign on log price persists. What the previous problem tells us is that there are no
variables in the demand equation that shift it relative to supply. So the supply equation is not
identified and the estimates are not meaningful.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 53

EXERCISE 11.24
(a)

These estimates are in column (1) of Table XR 11.24. Note that both STORMY and MIXED are
significant at the 1% level. The joint F-test of their significance yields F = 16.14, which is
greater than 10. Thus we can conclude that they are not weak instruments. Using the Stock-
Yogo test critical values, we find that we can reject the null hypothesis that the IV are weak
using the test size criterion, if we are willing to accept Type I error of 15% for a 5% test. The
Stock-Yogo critical value is 11.59.
Table XR 11.24
(1) (2) (3)
red form 2SLS Sargan
C -0.360*** 8.540*** -0.0252
(0.0792) (0.157) (0.159)
LPRICE -0.930**
(0.353)
STORMY 0.446*** -0.0174
(0.0792) (0.159)
MON -0.108 -0.0119 0.0101
(0.103) (0.208) (0.208)
TUE -0.0661 -0.526* -0.00917
(0.101) (0.202) (0.203)
WED -0.0493 -0.563** -0.0195
(0.104) (0.207) (0.209)
THU 0.0393 0.0999 -0.00496
(0.101) (0.203) (0.202)
MIXED 0.237** 0.114
(0.0785) (0.158)
N 111 111 111
R2 0.245 0.185 0.00696
RMSE 0.341 0.685 0.686
Standard errors in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

(b)

These estimation results are in column (2) of Table XR 11.24. Comparing these estimates to
those in Table 11.5, we find that the price elasticity of demand has fallen slightly, to −0.93 but
its standard error is slightly smaller. It is still significant at the 1% level. The daily indicator
variables show the same pattern as in Table 11.5. Friday is the omitted day, and sales on
Tuesday and Wednesday are significantly less, with Monday and Thursday not significantly
different from Friday.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 54

(c)

The Sargan regression of the 2SLS residuals on all the exogenous variables is in column (3) of
Table XR 11.24. We see that none of the variables is significant, which they shouldn’t be if
they are valid. The NR 2 = 0.7722 which is less than the χ(21) critical value of 3.841. Therefore

we fail to reject the validity of the surplus IV.

(d)

The joint F = 0.63, using F( 4,104) the p-value is 0.6437, so we fail to reject the null hypothesis

that the coefficients of these variables are all zero. The supply equation is not identified.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 55

EXERCISE 11.25
(a)

These estimates are in Table XR 11.25a, in column (1).

Table XR 11.25a: Change=1
(1) (2) (3) (4)
red form Hausman OLS 2SLS
C -0.482*** 8.363*** 8.533*** 8.363***
(0.0975) (0.180) (0.167) (0.191)
LPRICE -1.019** -0.435* -1.019**
(0.317) (0.175) (0.336)
STORMY 0.444***
(0.0824)
MON 0.0379 0.295 0.293 0.295
(0.123) (0.209) (0.215) (0.222)
TUE 0.0678 -0.345 -0.350 -0.345
(0.123) (0.209) (0.215) (0.222)
WED 0.143 -0.362 -0.408 -0.362
(0.129) (0.221) (0.225) (0.234)
THU 0.257* 0.394 0.269 0.394
(0.127) (0.226) (0.224) (0.240)
VHAT1 0.821*
(0.376)
N 77 77 77 77
R2 0.320 0.304 0.256 0.140
RMSE 0.340 0.580 0.595 0.615
Standard errors in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

The p-value for testing the null hypothesis that the coefficient of STORMY is zero is 0.0000.
Since this value is less than the level of significance, 0.05, we reject the null hypothesis and
conclude that this coefficient is significantly different from zero. The F-test value is 29.00, well
above the rule of thumb threshold of 10.

It is important to test for the statistical significance of STORMY because it is the supply
equation’s shift variable. It is required to be statistically significant for the demand equation
to be identified. If STORMY is not statistically significant, then the two-stage least squares
regression and the estimation procedure will be unreliable.

The Stock-Yogo test for weak instrument critical value, using the criteria of test size, is 16.38
[Table 10E.1] if we can tolerate a test with Type I error of 10% for a 5% nominal test.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 56

(b)

These estimates are in column (2) of Table XR 11.25a. The null hypothesis of this Hausman
test is H 0 : cov(ln( PRICE ), e) = 0 , which is tested by testing for the significance of the

coefficient of v̂1 in

ln(QUAN ) = α1 + α 2 ln( PRICE ) + α3 MON + α 4TUE + α5WED + α6THU + δvˆ1 + error

In the results below v̂1 is denoted VHAT1.

The t-statistic and p-value for the null hypothesis H 0 : δ = 0 are 2.1847 and 0.0323
respectively. Since this p-value is less than the level of significance, 0.05, we reject the null
hypothesis and conclude that ln(PRICE) is endogenous. The robust version of this test yields
t-statistic of 2.27, and thus our conclusion is unchanged.

(c)

The 2SLS estimates are in column (4) of Table XR 11.25a and the OLS estimates are in column
(3). These estimates have the expected signs. The two-stage least squares and least squares
estimates are very similar in values with the exception of the coefficient of ln(PRICE). Both
estimation procedures conclude that the day indicator variables are not significant at a 5% level
of significance.

Compared to Table 11.5 all estimated coefficients have the same sign except for the coefficient
of MON. Also the intercept estimate and the coefficient estimate of ln(PRICE) are similar but
the coefficient estimates for TUE, WED and THU are quite different. Furthermore, all of the
part (c) two-stage least squares estimates of the weekday indicator variables are insignificant
whereas Table 11.5 shows that TUE and WED are statistically significant.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 57

(d)

These estimates are in column (1) of Table XR 11.25b.

Table XR 11.25b: Change=0
(1) (2) (3) (4)
red form Hausman OLS 2SLS
C -0.0103 8.777*** 8.776*** 8.777***
(0.112) (0.270) (0.264) (0.266)
LPRICE -0.868 -0.975* -0.868
(2.677) (0.439) (2.632)
STORMY 0.149
(0.167)
MON -0.171 -0.901 -0.912 -0.901
(0.219) (0.549) (0.478) (0.540)
TUE -0.0322 -0.843 -0.841 -0.843
(0.186) (0.427) (0.417) (0.420)
WED -0.190 -0.872 -0.890* -0.872
(0.171) (0.607) (0.406) (0.597)
THU -0.244 -0.355 -0.379 -0.355
(0.165) (0.720) (0.397) (0.707)
VHAT0 -0.110
(2.715)
N 34 34 34 34
2
R 0.118 0.304 0.304 0.303
RMSE 0.336 0.806 0.792 0.792
Standard errors in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

These results are very different to those obtained in part (a). All the coefficients of the weekday
indicator variables have opposite signs and the coefficient for STORMY is smaller. In addition,
in part (a) the only variables which were not statistically significant were MON, TUE and WED.
In part (d) all exogenous variables are statistically insignificant.

Comparing these results to Table 11.4(b), all of the estimated coefficients have very different
values, although the only estimated coefficient with the opposite sign is the coefficient of THU.
All weekday indicator variables are statistically insignificant in both estimated regressions.
However, STORMY is statistically significant in Table 11.4b and not statistically significant in
the above regression.

(e)

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 58

These estimates are in column (2) of Table XR 11.25b. As described in part (b), the Hausman
test is a test for the endogeneity of ln(PRICE), which is tested by testing for the significance of
the coefficient of v̂2 in

ln(QUAN ) = α1 + α 2 ln( PRICE ) + α 3 MON + α 4TUE + α 5WED + α 6THU + δvˆ2 + e d

The variable VHAT0 = v̂2 . The t-statistic and p-value for the null hypothesis H 0 : δ = 0 are
−0.0405 and 0.9680, respectively. Since this p-value is greater than the level of significance,
0.05, we do not reject the null hypothesis and conclude ln(PRICE) does not show signs of
endogeneity. This is consistent with Graddy and Kennedy’s expectation that when inventory
changes are small, simultaneity between demand and supply does not exist.

(f)

These estimates are in columns (3) and (4) of Table XR 11.25b. All the estimates have the
expected signs and are almost identical. The major difference between the two sets of estimates
is that, as a consequence of the smaller least squares standard errors, all of the least squares
coefficient estimates are significantly different from zero except those for MON, TUE and THU
whereas none of the two-stage least squares coefficient estimates are significantly different
from zero.

Comparing these values to those in part (c), we find that the coefficient estimates for ln(PRICE)
appear to be quite similar with the exception of the least squares coefficient estimate of
ln(PRICE) in part (c), which is likely to exhibit simultaneous equation bias. Also, the
coefficient of ln(PRICE) is always significantly different from zero in part (c) and only
significant in the least squares part (f) estimation. The estimated values of the coefficients of
the weekday indicator variables are very different.

Part (c) models the demand for fish when there are large changes in inventory, and part (f)
models the demand for fish for small changes in inventory. It has been postulated that when
more fish are sold and bought, causing large changes in inventory, sellers are more responsive
to prices and therefore endogeneity is present and on the days where there is little change in
inventory endogeneity should not be present. This is supported by our estimates which show
that the two stage least squares and least squares coefficient estimates of ln(PRICE) are similar
when CHANGE = 0 but very different when CHANGE = 1. This discrepancy suggests that a

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 59

coefficient bias exists when CHANGE = 1 due to endogeneity. Also note that the least squares
estimate of the price elasticity of demand when CHANGE = 0 is similar in magnitude to the
two-stage least squares estimate of the price elasticity of demand when CHANGE = 1.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 60

EXERCISE 11.26
The estimates are in columns (1) and (2) of Table XR 11.26.

Table XR 11-26
(1) (2) (3) (4) (5)
2SLS Dem 2SLS Sup Dem BP Sup BP 2SLS rob
C -4.2795 20.0328 26.7949 7.8826 20.0328
(5.5439) (1.2231) (50.6363) (3.5084) (1.1462)

P -0.3745 0.3380 0.3380

(0.1648) (0.0249) (0.0198)

PF -1.0009 -2.0350 -0.2728 -1.0009

(0.0825) (1.8931) (0.1312) (0.0777)

PS 1.2960 1.2668 -0.3672

(0.3552) (2.2253) (0.1542)

DI 5.0140 3.5933 2.3901

(2.2836) (10.9358) (0.7577)
N 30 30 30 30 30
R2 . 0.9019 0.0614 0.2988 0.9019
RMSE 4.9300 1.4976 41.8415 2.8991 1.4976
Standard errors in parentheses

(a)

The price elasticities of supply and of demand, at the mean, are calculated as

%ΔQ ΔQ Q P % ΔQ ΔQ Q P
εS = = = β2 εD = = = α2
%ΔP ΔP P Q % ΔP ΔP P Q

Using our estimates

P 62.724
εˆ S = βˆ 1 = 0.3380 × = 1.1485
Q 18.458

P 62.724
εˆ D = αˆ 1 = −0.3745 × = −1.2725
Q 18.458

The signs of the elasticities are as expected; we expect ε S to be positive because quantity

supplied increases as price increases and we expect ε D to be negative because quantity

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 61

demanded decreases as price increases. Both elasticities have a magnitude greater than 1 which
indicates that both supply and demand considered elastic and therefore responsive to prices; a
percentage increase in price leads to a larger than 1% change in supply and demand.

(b)

The BP regression is in column (3) of Table XR 11.26. The value of the N × R 2 test statistic is
1.842, while the 5% critical value for χ(23) is 7.8147, so we fail to find any significant evidence

of heteroskedasticity.

(c)

The BP regression is in column (4) of Table XR 11.26. The value of the N × R 2 test statistic is
8.965, while the 5% critical value for χ(23) is 7.8147, so we reject the null hypothesis of

homoskedasticity.

(d)

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 62

The variation of the 2SLS residuals shows the greatest increase when plotted versus disposable
income. This is not surprising as higher incomes lead to more choices in consumption
expenditures.

(e)

The supply equation regression with robust standard errors is in column (5) of Table XR 11.26.
The robust standard errors are smaller than the conventional 2SLS standard errors. Whether to
report them is something of a judgement call. There is evidence of heteroskedasticity from the
residual plot. The BP test and the properties of 2SLS with and without the robust standard

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 63

errors are justified in large samples. The sample of 30 observations may not convince a reader.
The smaller standard errors make the already large t-values even larger.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 64

EXERCISE 11.27
The estimates are in Table XR 11.27.
Table XR 11.26
(1) (2) (3) (4)
Demand_OLS Demand_2SLS Supply_OLS Supply_2SLS
C 1.0910 -4.2795 20.0328*** 20.0328***
(3.7116) (5.5439) (1.2220) (1.2231)

P 0.0233 -0.3745* 0.3380*** 0.3380***

(0.0768) (0.1648) (0.0217) (0.0249)

PS 0.7100** 1.2960**
(0.2143) (0.3552)

DI 0.0764 5.0140*
(1.1909) (2.2836)

PF -1.0009*** -1.0009***
(0.0764) (0.0825)
N 30 30 30 30
Standard errors in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

Comparing the demand equation estimates in columns (1) and (2) we see dramatic differences.
The OLS estimate of the coefficient of price is insignificant, whereas the 2SLS estimate is
negative and significant as it should be, even though it has a larger standard error. The OLS
and 2SLS estimates of the price of the substitute are positive and significant, although the 2SLS
coefficient and its standard error are larger. The OLS estimate of the coefficient of income is
not significant whereas the 2SLS estimate is positive and significant which is what we would
expect.

The OLS and 2SLS estimates for the supply equation are virtually identical and both satisfy
our economic reasoning.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 65

EXERCISE 11.28
(a)

Rearranging the demand equation,

Q = α1 + α 2 P + α 3 PS + α 4 DI + e d

yields

1
P=
α2
( Q − α1 + α3 PS + α 4 DI + ed )
= δ1 + δ2Q + δ3 PS + δ4 DI + u d

According to economic theory, it is expected that there is an inverse relationship between price
and quantity demanded, so we expect δ2 < 0 . If the price of a substitute increases the demand

for truffles increases, increasing the price of truffles, so we expect δ3 > 0 . If disposable income
increases, and if truffles are a normal good, then demand increases and equilibrium price
increases. We expect δ4 > 0.

Rearranging the supply equation, Q = β1 + β 2 P + β3 PF + e s , yields

1
P=
β2
( Q − β1 + β3 PF + es )
= φ1 + φ2Q + φ3 PF + u s

According to economic theory, there is a positive relationship between quantity supplied and
price. Thus we expect φ2 > 0 . An increase in the price of a factor of production reduces supply

and increases equilibrium price, so we expect φ3 > 0.

(b)

These 2SLS estimates are in columns (1) and (3) of Table XR 11.28.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 66

Table XR 11.28
(1) (2) (3) (4)
Dem 2SLS Dem OLS Sup 2SLS Sup OLS
C -11.4284 -13.6195 -58.7982*** -52.8763***
(13.5916) (9.0872) (5.8592) (5.0238)

Q -2.6705* 0.1512 2.9367*** 2.6613***

(1.1750) (0.4988) (0.2158) (0.1712)

PS 3.4611** 1.3607*
(1.1156) (0.5940)

DI 13.3899*** 12.3582***
(2.7467) (1.8254)

PF 2.9585*** 2.9217***
(0.1560) (0.1482)
N 30 30 30 30
Standard errors in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

The signs are as we expected in part (a) and all coefficients are significantly different from zero
since all p-values are less than the level of significance of 0.05.

(c)
The price elasticity of demand at the mean is calculated as

%ΔQ ΔQ Q 1 P
εD = = = ×
%ΔP ΔP P δ2 Q

Using our estimates

1 P 1 62.724
εˆ D = × = × = −1.2725
δˆ 2 Q −2.6705 18.458

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 67

(d)

(e)

The estimated equilibrium values from part (d) are given by equating the supply and demand
equations after substituting in the given exogenous variable values. Therefore equating these
equations yields

111.5801 − 2.6705QEQM = 9.2470 + 2.9367QEQM

QEQM =18.2503

When QEQM is substituted into the demand equation (substituting into the supply equation will

yield the same result) we find the equilibrium value of P, thus

PEQM = 111.5801 − 2.6705 × 18.2503 = 62.8427

Using the reduced form estimates in Tables 11.2a and 11.2b, the predicted equilibrium values
are

QEQM _ RF = 7.8951 + 0.6564 × 22 + 2.1672 × 3.5 − 0.5070 × 23 = 18.2604

PEQM _ RF = −32.5124 + 1.7081 × 22 + 7.6025 × 3.5 + 1.3539 × 23 = 62.8154.

Comparing the equilibrium values calculated using the results from part (d) to those calculated
using the reduced form estimates, we find them to be almost equal.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 68

(f) Estimate the supply and demand equations that you have formulated in (a) using ordinary least
squares. Are the signs correct? Are the estimated coefficients significantly different from zero?
Compare the results to those in part (b).

The OLS estimates for the demand equation are in column (2) of Table XR 11.28. The OLS
estimates for the supply equation are in column (2) of Table XR 11.28. All estimated
coefficients are significantly different from zero except for the intercept term and the
coefficient of Q. The sign for the coefficient of Q is incorrect because it suggests that there is
a positive relationship between price and quantity demanded. Compared to the results from
part (b), the coefficient of Q has the opposite sign and the estimated intercept and the coefficient
of PS are much smaller.

All estimates in this supply equation are significantly different from zero. All coefficient signs
are correct, and the coefficient values do not differ much from the estimates in part (b).

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 69

EXERCISE 11.29
(a)

The estimates are in column (1) of Table XR 11.29. The coefficients of total wages, W, and

profits, P, have positive signs. There are 17 degrees of freedom for this equation. The 97.5

percentile of the t(17 ) distribution is 2.11. The 99.5 percentile is 2.898. Wages are significant at

the 0.01 level with t = 19.93. Profits are significant at the 0.05 level with t = 2.12.

Table XR 11.29
(1) (2) (3) (4) (5) (6)
CN OLS CN 2SLS W1 RF P RF Hausman Sargan
C 16.2366*** 16.5548*** 43.4356 50.3844 16.5548*** 5.6846
(12.46) (11.28) (1.71) (1.59) (15.51) (0.40)
W 0.7962*** 0.8102*** 0.8102***
(19.93) (18.11) (24.90)
P 0.1929* 0.0173 0.0173
(2.12) (0.13) (0.18)
PLAG 0.0899 0.2162 0.8719 0.8025 0.2162* -0.1885
(0.99) (1.81) (2.09) (1.55) (2.49) (-0.80)
KLAG -0.1230 -0.2161 -0.0432
(-1.28) (-1.81) (-0.80)
W2 -0.4437 -0.0796 -0.2560
(-0.22) (-0.03) (-0.22)
G 0.8662* 0.4390 -0.5044*
(2.76) (1.12) (-2.84)
TX -0.6042 -0.9231 0.1397
(-1.73) (-2.13) (0.71)
TIME 0.7136 0.3194 0.1174
(1.14) (0.41) (0.33)
ELAG 0.0953 0.0220 0.1525
(0.42) (0.08) (1.19)
VHAT1 -0.4537
(-1.89)
VHAT2 0.6899**
(3.23)
N 21 21 21 21 21 21
t statistics in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 70

(b)

The estimated equation is in column (3) of Table XR 11.29. The F-test of joint significance of

the 5 variables KLAG, G, TX, TIME, ELAG yields F = 2.53. The 95th percentile of F( 5,13) is

3.0254. Therefore at the 5% level we do not find these variables jointly significant.

(c)

The estimated equation is in column (4) of Table XR 11.29. The F-test of joint significance of

the 5 variables KLAG, G, TX, TIME, ELAG yields F = 3.35. The 95th percentile of F( 5,13) is

3.0254. Therefore at the 5% level we find these variables jointly significant.

(d)

The regression results are in column (5) of Table XR 11.29. We see that v̂2 is significant at the

1% level. The joint F – test gives F = 5.60. The 95th percentile of F( 2,15) is 3.6823. Therefore

at the 5% level we do find these variables jointly significant. We conclude that either W1t or Pt

is endogenous, or both are. This is what we should find in a system of simultaneous equations.

The endogenous variables are jointly determined.

(e)

The 2SLS estimates are in column (2) of Table XR 11.29. The primary difference is that profits,
P, has a much smaller coefficient estimate and is no longer significant. On the other hand
lagged profits, PLAG, has a larger coefficient estimate and is significant at the 10% level, using
the critical value 1.74 for critical value, which is t( 0.95,17 ) .

(f)

The estimates for this regression are in column (6) of Table XR 11.29. The test statistic is
TR 2 = 21( 0.4177 ) = 8.7715 . The chi-square critical value is 7.8147. Thus we reject the validity

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 71

of the surplus IV. The test does not indicate which IV is the problem, but government spending,
G, is statistically significant in the regression. Valid exogenous variables, and IV, should not
be related to the 2SLS residuals.

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 72

EXERCISE 11.30
(a)

The OLS estimates of the investment function are in column (1) of Table XR11.30. There are

17 degrees of freedom and the 0.975 percentile of the t-distribution is 2.1098, and the 0.95

percentile is 1.7396. We see that profits and lagged profits, P and PLAG are significant at the

5% level with the anticipated positive signs. If profits are higher firms will spend more on

investment goods. The sign of the lagged capital stock, KLAG, is negative and significant.

Having a larger capital stock means a smaller requirement for new investment.

Table XR 11.30
(1) (2) (3) (4) (5) (6)
OLS 2SLS Two steps P RF Hausman Sargan
C 10.1258 20.2782* 20.2782 50.3844 20.2782*** 7.6711
(1.85) (2.42) (2.03) (1.59) (4.31) (0.37)
P 0.4796*** 0.1502 0.1502
(4.94) (0.78) (1.39)
PLAG 0.3330** 0.6159** 0.6159* 0.8025 0.6159*** 0.1899
(3.30) (3.40) (2.86) (1.55) (6.07) (0.56)
KLAG -0.1118*** -0.1578** -0.1578** -0.2161 -0.1578*** -0.0023
(-4.18) (-3.93) (-3.30) (-1.81) (-7.01) (-0.03)
W2 -0.0796 -0.7046
(-0.03) (-0.42)
G 0.4390 0.0343
(1.12) (0.13)
TX -0.9231 -0.0228
(-2.13) (-0.08)
TIME 0.3194 0.2839
(0.41) (0.56)
ELAG 0.0220 -0.1160
(0.08) (-0.63)
PHAT 0.1502
(0.66)
VHAT 0.5745***
(4.03)
N 21 21 21 21 21 21
R2 0.931 0.885 0.837 0.826 0.966 0.086
t statistics in parentheses
*
p < 0.05, ** p < 0.01, *** p < 0.001

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 73

(b)

The estimates are in column (4) of Table XR 11.30. The F-test statistic value is 1.93. The

critical value for a 5% test of significance is F( 0.95,5,13) = 3.0254 . Thus we cannot reject the null

hypothesis that the coefficients of these 5 variables are zero.

(c)

The Hausman test regression is in column (5) of Table XR 11.30. Note that the residuals VHAT
are significant at even the 0.001 level. We conclude that P is endogenous. This is what we
expected from the simultaneous equations model.

(d)

The 2SLS estimates are in column (2) of Table XR 11.30. There are some noticeable
differences. The coefficient on profits, P, is insignificant. The coefficient on lagged profits is
twice as large as the OLS estimate and significant at the 1% level. The lagged capital stock has
a negative coefficient and is significant at the 1% level, the same result as in the OLS
regression.

(e)

The second stage model is reported in column (3) of Table XR 11.30. The coefficient estimates
are identical. However the t-values are different because they are based on an incorrect estimate
of var ( e2t ) which leads to incorrect standard errors. The second stage regression uses residuals

(
eˆ2, wrong = I t − βˆ 1 + βˆ 2 Pˆt + βˆ 3 Pt −1 + βˆ 4 K t −1 )
whereas the correct residuals for estimating the variance are

(
eˆ2,correct = I t − βˆ 1 + βˆ 2 Pt + βˆ 3 Pt −1 + βˆ 4 K t −1 )
See page 496 in POE5 for further discussion.

(f)

Copyright © 2018 Wiley

 Chapter 11, Exercise Solutions, Principles of Econometrics, 5e 74

The Sargan regression is in column (6) of Table XR 11.30. The value of the test statistic is
TR 2 = 21( 0.0864 ) = 1.815 . The 5% critical value for the chi-square distribution with 4 degrees

of freedom is 9.4877. Thus we fail to reject the validity of the surplus IV.

Copyright © 2018 Wiley