Scribd
Upload
22 views3 pages

PS5 Econ320 2024

ECON320,Emory University

Uploaded by

Will Martinez
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
22 views3 pages

PS5 Econ320 2024

ECON320,Emory University

Uploaded by

Will Martinez
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/ 3

ECON 320 Fall 2024

Problem set 5
Due at 11.59pm on November 26th

1. In this problem we derive the expression for including “irrelevant variables” in the
regression and see more formally what it does to the variance. Assume as in the
lecture notes that all variables are demeaned, i.e., that X̄ = Z̄ = Ȳ = 0, where
X̄ = ni=1 Xi . We assume the true model is
P

Yi = βXi + εi , (1)
and we define the corresponding regressor as β̂. The model we actually estimate is
Yi = βXi + γZi + ui , (2)
and we define the corresponding regressor as β̃. Moreover, we assume that Xi and
Zi are deterministic (non-random), so that our Gauß-Markov assumptions on
the regression model look like
E (εi ) = 0 and Var (εi ) = σ 2
as well as E[εi εj ] = 0 for i ̸= j. Throughout, all random variables are sampled iid.
We will solve this problem together step by step.

(a) Set up the optimization problem for the OLS estimators β̃, γ̃ in the model (2).
(b) Show that the first order optimality conditions are
n
X
(Yi Xi − βXi2 − γZi Xi ) = 0
i=1
Xn
(Yi Zi − βXi Zi − γZi2 ) = 0.
i=1

(c) Deduce that P


i−
i (YP βXi )Zi
γ= 2
.
i Zi

(d) From this, by substitution, deduce that


! ! ! !
X X X X
Yi Xi Zj2 − β Xi2 Zj2
i j i j
! ! ! !
X X X X
− Zi Xi Yj Zj +β Zi Xi Xj Zj = 0.
i j i j

(e) Deduce that


( i Zi2 ) ( i Xi Yi ) − ( i Xi Zi ) ( i Zi Yi )
P P P P
β̃ = .
( i Xi2 ) ( i Zi2 ) − ( i Xi Zi )2
P P P

This is the expression in the lecture notes.


(f) Now we want to derive the variance Var(β̃) and want to show that Var(β̃) =
P
Var(β̂) if i Zi Xi = 0. For this, start by showing that

Var (( i Zi2 ) ( i Xi Yi ) − ( i Xi Zi ) ( i Zi Yi ))
P P P P
Var(β̃) = 2 .
( i Xi2 ) ( i Zi2 ) − ( i Xi Zi )2
P P P

(g) Deduce that

Zi2 ) ( i Xi εi ) − ( i Xi Zi ) ( i Zi εi ))
P P P P
Var (( i
Var(β̃) = 2 .
( i Xi2 ) ( i Zi2 ) − ( i Xi Zi )2
P P P

[Hint: Alway plug in the true model!]


(h) Deduce further that
P h P  P i
2
Var ε
i i X i Z
j j − Zi j X Z
j j
Var(β̃) = 2 .
( i Xi ) ( i Zi ) − ( i Xi Zi )2
P 2 P 2 P

(i) Finally deduce that


P h P 2  P i2
i X i Z
j j − Zi j X j Z j
Var(β̃) = σ 2  P 2 .
( i Xi2 ) ( i Zi2 ) − ( i Xi Zi )2
P P

P
(j) This is the variance of β̃. Now assume that i Xi Zi = 0. Show that

σ2
Var(β̃) = P 2 .
i Xi

Confirm that this is Var(β̂).


(k) Without any further computations, argue that it must be that Var(β̃) ≥ Var(β̂)
in general.

2. In the beginning of the lecture notes on IV regression we introduced the Wald IV


estimator and a version of the 2-SLS estimator for the simple model with only one
endogenous variable and one instrument:

Yi = βXi + εi .

As usual, we normalize all variables Ȳ = X̄ = Z̄ = 0 to make formulas and compu-


tation easier. We also have the first stage model

Xi = γZi + ui

with E[εi |Zi ] = E[ui |Zi ] = 0.

(a) Write out the Wald estimator in summation form (NOT in matrix form).
(b) Denote the fitted values of Xi from the first stage as X̂i = γ̂Zi . Write out the
2-SLS estimator in summation form.

(c) Show that β̂ Wald = β̂ 2SLS . [Hint: What can you say about
P
i (Xi − X̂i )Zi ? Use
the normal equations.]

3. On CANVAS, I have posted the famous paper “Experimental Estimates of Education


Production Functions” (1999) by Alan Krueger in the QJE. In it, he basically wants
to find out whether small classes have a positive effect on learning outcomes. In
order to do so, he uses data from a famous experiment in Tennessee, called STAR,
where children in different schools were randomly assigned to small, medium, or large
classes. Students were followed in this experiment for a few years in order to evaluate
“long-term” outcomes—this was measured by real outcomes in tests. The problem
now was that the experiment was not perfect, i.e. students switched schools (lost
observations) and there was an issue with compliance: parents of student who got
randomly assigned to a large class would try a lot in order to get their child into a
small class, so instead of having all children in the same class-sizes over a few years,
there were some “switchers”.
Krueger used an instrumental variables approach for this problem. The goal is to
regress test-scores (Y ) on the dummy variable 1(small) indicating whether the stu-
dent went to a small class and some other exogenous covariates. The important
contribution of this paper is the way Krueger instrumented the (now endogenous)
dummy variable 1(small) by the variable 1(initial small), i.e. the dummy variable
indicating whether a student was initially assigned to attend a small class (even if
she switched classes later on).
Now based on this, please answer the following questions:

(i) Why is the variable 1(small) endogenous now? Explain in no more than 5 sen-
tences.

(ii) Argue in no more than 5 sentences that 1(initial small) is a valid instrument.

4. Suppose you want to estimate the effect of a change in tax laws in the state of
fictitious Georgia, where in 2022, the income tax brackets were lowered. You want to
estimate the effect of this policy change on the average household income in fictitious
Georgia. Assume you have the data of average household income in all 50 states of
the US from January 2010 to November 2024. Tell me in no more than 8 sentences
how you would conceptually estimate this causal effect. In particular, tell me what
assumptions you would make in order to make sure you actually get at the true causal
effect.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy