Scribd
Upload
15 views

Tutorial 3solutions

Uploaded by

Alireza Kafaei
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
15 views

Tutorial 3solutions

Uploaded by

Alireza Kafaei
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

Tutorial 3 Solutions

1. Details behind those provided in the text:


a. The following four regressions estimate the means for the four sub-samples (Some rows have
been deleted to save space):

Dependent Variable: FTE


Method: Least Squares
Sample: 1 794 IF NJ=1 AND D01=1
Included observations: 319

Variable Coefficient Std. Error t-Statistic Prob.

NJ*D01 21.02743 0.520309 40.41332 0.0000

R-squared 0.000000 Mean dependent var 21.02743


… … … …

Dependent Variable: FTE


Method: Least Squares
Sample: 1 794 IF NJ=0 AND D01=1
Included observations: 77

Variable Coefficient Std. Error t-Statistic Prob.

C 21.16558 0.943221 22.43968 0.0000

R-squared 0.000000 Mean dependent var 21.16558


… … … …

Dependent Variable: FTE


Method: Least Squares
Sample: 1 794 IF NJ=1 AND D01=0
Included observations: 321

Variable Coefficient Std. Error t-Statistic Prob.

C 20.43941 0.508261 40.21442 0.0000

R-squared 0.000000 Mean dependent var 20.43941


… … … …

Dependent Variable: FTE


Method: Least Squares
Sample: 1 794 IF NJ=0 AND D01=0
Included observations: 77

Variable Coefficient Std. Error t-Statistic Prob.

C 23.33117 1.351149 17.26765 0.0000

R-squared 0.000000 Mean dependent var 23.33117


…. … … …
b. From this we obtain 𝛿 = (21.0274 – 21.1656) – (20.4394 – 23.3312)
= 2.7536

c. The appropriate regression based on all 794 observations is given by

Dependent Variable: FTE


Method: Least Squares
Sample: 1 794
Included observations: 794

Variable Coefficient Std. Error t-Statistic Prob.

C 23.33117 1.071870 21.76679 0.0000


NJ -2.891761 1.193524 -2.422877 0.0156
D01 -2.165584 1.515853 -1.428625 0.1535
NJ*D01 2.753606 1.688409 1.630888 0.1033

R-squared 0.007401 Mean dependent var 21.02651


Adjusted R-squared 0.003632 S.D. dependent var 9.422746
S.E. of regression 9.405619 Akaike info criterion 7.325517
Sum squared resid 69887.88 Schwarz criterion 7.349079
Log likelihood -2904.230 Hannan-Quinn criter. 7.334571
F-statistic 1.963536 Durbin-Watson stat 1.841699
Prob(F-statistic) 0.117983

From this regression we find that 𝛿 =2.7536 (with a standard error of


1.6884)

d. We want the test 𝐻 : 𝛿 = 0 vs 𝐻 : 𝛿 ≤ 0 .


𝑏 −0
Our test statistic is T = 𝑠𝑒(𝑏 ) and under the null 𝑇~𝑡 ~𝑁 (0, 1).
Our critical value is -1.645 and we will reject 𝐻 if T < -1.645.
Here, our observed T = 1.63, which is not sufficiently negative to reject
𝐻 . We conclude that the data does not support the business owner’s
claim.

2. Issues that we need to be aware of are whether treatment is random, and


whether the sample is representative of the population of interest
a. Some of the men will choose to have a weight coach while others will
not, and this choice will not be entirely random. In addition, even if
those who had coaches was random, the results would only relate to
overweight men and not to the population at large.
b. Some of the people in the sample would choose to be vaccinated, while
others would not, and this choice would not be random. Vaccination
programs might target older people. Also, some people (eg very young
children) might not have access to vaccinations.
3. Both equations are likely to have measurement errors in the explanatory
variable.
a. Here we are using a last year’s forecast of demand as an explanatory
variable for this year’s production, and forecasts are likely to be
inaccurate. Further, there will be an omitted variable issue, given
that production depends on other factors such as labour and capital.
b. Intelligence is notoriously difficult to measure, so it’s reasonable to
assume that the explanatory variable will be measured with error.
Again, we can expect omitted variable bias, given that study and
other factors will affect exam performance.

4. This is a pair of simultaneous equations


a. C = 𝛽𝑌 + 𝑒 = 𝛽 (𝐶 + 𝐼 ) + 𝑒 → (1 − 𝛽)𝐶 = 𝛽𝐼 + 𝑒
𝛽 𝑒
→𝐶= 𝐼+
1−𝛽 (1 − 𝛽)
b. Y= C + I = 𝐼+ +I= 𝐼+
( ) ( )
c. The expression for Y contains the error term e, which shows that
COV(Y, e) ≠ 0 , which implies that the OLS estimate b for 𝛽 will be
inconsistent.
Since Y = (𝐼 + 𝑒)
d. 𝐸 (𝑌. 𝑒 ) =E (𝐼 + 𝑒). 𝑒 = 𝐸 [(𝐼 + 𝑒). 𝑒] = 𝐸 (𝑒 )

= 𝜎

𝑉𝐴𝑅 (𝑌) = 𝐸 (𝑌 ) = [𝜎 + 𝜎 ]

∑
e. We know that 𝑏 = 𝛽 + ∑
, but we cannot immediately put the
expectation operator through the second term to show that it is non-
zero because of its nonlinear form. We can, however, use the law of
iterated expectations to show this (The L.I.E. says that one can always
write the expectation of a function g(Y) of a RV Y is the expectation
(wrt Y) of the (conditional) expectation of g(Y)|Y).
Applied here we have
P
ye
E(b) = +E P t 2t
yt
P
yt et
= +E E P 2 jY
y
P t
E( yt et jY )
= +E P 2
yt
P
yt E(et jY )
= +E P 2
yt
P
y F (y )
= +E Pt 2 t where F (yt ) is nonzero
yt
6=

Question 5
The two equations are

demand: Qt = a0 + a1 Ptw + a2 Ptb + ut ((1))


supply: Qt = b0 + b1 Ptw + b2 St + vt ((2))
(a) Equating equations (1) and (2) we have
a0 + a1 Ptw + a2 Ptb + ut = b0 + b1 Ptw + b2 St + vt
(a1 b1 )Ptw = (b0 a0 ) + b2 St a2 Ptb + (vt ut )
(b0 a0 ) + b2 St a2 Ptb + (vt ut )
Ptw =
(a1 b1 )

which shows that Ptw is correlated with ut (and also with vt ).


Substituting the expression for Ptw back into (1) leads to
(b0 a0 ) + b2 St a2 Ptb + (vt ut )
Qt = a0 + a1 + a2 Ptb + ut
(a1 b1 )
a0 (a1 b1 ) (b0 a0 ) + b2 St a2 Ptb + (vt ut ) a2 (a1 b1 )Ptb + (a1 b1 )ut
= + a1 +
(a1 b1 ) (a1 b1 ) (a1 b1 )
(a1 b0 a0 b 1 ) a1 b2 St a2 b1 Ptb a1 v t b1 ut
= + + +
(a1 b1 ) (a1 b1 ) (a1 b1 ) (a1 b1 ) (a1 b1 )

We could have substituted the expression for Ptw into (2 )to obtain
(b0 a0 ) + b2 St a2 Ptb + (vt ut )
Qt = b0 + b1 + b2 St + vt
(a1 b1 )
b0 (a1 b1 ) (b0 a0 ) + b2 St a2 Ptb + (vt ut ) b2 (a1 b1 ) (a1 b1 )
= + b1 + St + vt
(a1 b1 ) (a1 b1 ) (a1 b1 ) (a1 b1 )
b
(a1 b0 a0 b 1 ) a1 b 2 a2 b 1 P t a1 b1
= + St + vt ut
(a1 b1 ) (a1 b1 ) (a1 b1 ) (a1 b1 ) (a1 b1 )
which (of course) gives the same result)

(b) The reduced form for Ptw shows that Cov(Ptw ; ut ) 6= 0 and also that Cov(Ptw ; vt ) 6= 0:
Thus, in each equation (1) and (2) an explanatory variable is correlated with the error
term, which implies that OLS will provide bias and inconsistent estimates for (1) and
(2)

You might also like