Professional Documents
Culture Documents
Problems PDF
Problems PDF
• Problems 1–5 are designed to give you insights about basic IV estimation.
where (vt , ηt ) are i.i.d. with E[k(vt , ηt )k2 ] < ∞ and E[(vt , ηt )] = 0.
expectation function, but observes the realized sales which are expected sales plus
noise.
yt = xt β0 + εt .
1
University of Wisconsin–Madison Spring 2018, ECON 710
Department of Economics M. Sølvsten, J. Freyberger
(b) Does the assumptions imply exogeneity of xt−1 or yt−1 ? What about xt+1
or yt+1 ? What about xt−2 or yt−2 ?
y i = xi β 0 + εi (i = 1, . . . , n)
√ σ 2 E[zi2 ]
!
d
n β̂IV − β0 →− N 0,
E[zi xi ]2
(c) Compare the two IV estimators that use respectively zi = xi and zi = x2i
as instruments (asuming that zi = x2i is relevant). Is one estimator always
(weakly) more efficient than the other? What happens if xi is constant
across i?
yi = xi β0 + εi
xi = zi π0 + ui (i = 1, . . . , n) (1)
2
University of Wisconsin–Madison Spring 2018, ECON 710
Department of Economics M. Sølvsten, J. Freyberger
(a) Write
yi = γ0 zi + vi (2)
(c) Let π̂ and γ̂ be OLS estimators applied to (1) and (2). Show that the IV
estimator applied to the simultaneous model is equal to γ̂/π̂.
yi = α0 + xi β0 + εi
with E[εi | zi ] = 0.
yi = xi β0 + wi0 γ0 + εi
with E[εi | zi , wi ] = 0.
P −1 P
Let δ̂ = n
i=1 wi wi0 n
i=1 wi zi be the OLS estimator in the regression model
zi = wi0 δ0 + νi ,
3
University of Wisconsin–Madison Spring 2018, ECON 710
Department of Economics M. Sølvsten, J. Freyberger
Consider the IV regressor that uses (zi , wi0 ) as an instrument for (xi , wi0 ):
−1
1
n
X zi xi zi wi0 1 n
X z y
i i
β̂IV = .
n i=1 w x w w0 n i=1 w y
i i i i i i
(c) (Hard) Show that the first entry of β̂IV is equal to β̃IV .
y i = xi β 0 + εi
xi = zi π0 + ui (i = 1, . . . , n)
√
where β0 = 0, π0 = 1/ n, and
εi 1 .7 0
ui ∼ N 0, .7 1 0 .
zi 0 0 1
π02 nE[zi2 ]
1+ .
σu2
What is this value? What is the probability limit of β̂OLS ?
4
University of Wisconsin–Madison Spring 2018, ECON 710
Department of Economics M. Sølvsten, J. Freyberger
(b) In a statistical software of your choice, generate data from the above model
and calculate the following five statistics (defined in lectures 3 and 4):
√
β̂OLS , β̂IV , F̂ , 1 √nβ̂IV
> 1.96 ^
, 1 AR(0) > 1.96 .
V̂
HC
(c) Repeat (b) 10000 times and report the simulated median of each the first
two statistics and the simulated mean of the last three statistics. Relate
your findings to the model and theory presented in class.
Hint: The Canvas course site contains some incomplete R-code that can be used
to perform the simulations.
y i = xi β 0 + εi
xi = zi0 π0 + ui (i = 1, . . . , n)
√
where ` = 15, β0 = 0, π0 = (1, . . . , 1)0 / n, and
εi 1 .7 0
ui ∼ N 0, .7 1 0.
zi 0 0 I`
nE[(zi0 π0 )2 ]
1+ ,
`σu2
What is this value? What is the probability limit of β̂OLS and β̂2SLS under
many (weak) instruments (lecture 5)?
(b) In a statistical software of your choice, generate data from the above model
and calculate the following five statistics (defined in lecture 5):
√
nβ̂
β̂OLS , β̂2SLS , β̂JIV E , F̂ , 1 √
2SLS > 1.96 .
V̂
HC
5
University of Wisconsin–Madison Spring 2018, ECON 710
Department of Economics M. Sølvsten, J. Freyberger
(c) Repeat (b) 10000 times and report the simulated median of each the first
three statistics and the simulated mean of each of the last two statistics.
Relate your findings to the model and theory presented in class.
0
zi π̂(i) −Pii xi
Hint: In calculating β̂JIV E , you can use that zi0 π̂(i) = 1−Pii
where Pii =
P −1
zi0 n 0
i=1 zi zi zi .
Your written answer to problems 6 and 7 must contain two files: (1) a pdf file
that contains your code and potentially the output, and (2) a pdf file that contains
your findings and discussion (maximum length of the second file is two pages).