Semester 1, AY2021-2022
Assignment 1 for Weeks 3/4
EC3304: Econometrics II
Instructor: Dr. Sorawoot Srisuma
E-mail:
[email protected]You will receive full credit for participation if you constructively engage during tutorial discussions
when asked regardless whether or not you give the correct answers.
1. Suppose you have a fair coin. Let Yi indicate the outcome of a coin toss for i = 1; 2; 3.
Speci…cally, Yi takes value 1 if it is heads and 0 if it is tails. Let Y = 13 (Y1 + Y2 + Y3 ).
(a) What is the probability mass function of Y ?
(b) Calculate E Y .
(c) Calculate V ar Y .
(d) What assumptions did you make in answering (a) to (c)?
(f) Grab a coin and toss it three times and record the outcomes. Take the average of your score
and compare it to the answer of part (b). Are they, or should they be, the same?
2. Let Y Bernoulli ( ) such that for some 2 (0; 1),
(
1 w.p.
Y = :
0 w.p. 1
Suppose fYi gni=1 is an i.i.d. sample from Y . Let be the sample average constructed from just the
even observations and b be the sample average constructed from the whole sample. You may assume
n is an even number.
(a) For any estimator e for , show that
MSE(e) = (E(e) ))2 + Var(e):
(b) Compare the mean square errors of and b. Which estimator would you prefer?
(c) Provide one example of a biased estimator of .
3. Suppose X N (0; 1). Let’s Y = X.
(a) Do X and Y have identical distribution?
(b) Are X and Y independent?
1
p
4. bn converges in probability to , which can be succinctly written as bn ! , if for all c > 0,
lim Pr c bn + c = 1:
n!1
We should think of bn as an estimator for when the sample size of the data is n. (E.g., see b in
Question 2 where the estimator’s dependence on n is implicit; omitting n is a common practice in
applied econometrics and statistics.)
p
bn is a consistent estimator if bn ! . The purpose of this question is to give a sense of what
it means for an estimator to be consistent. For this, let’s suppose that bn N ; n1 .
(a) Compute Pr 0:1 bn + 0:1 for n = 1; 10; 100.
(b) Compute Pr 0:01 bn + 0:01 for n = 1; 10; 100.
(c) How large must n have to be for Pr 0:01 bn + 0:01 to be larger than 0:95?
(d) Intuitively, what happens to the distribution of bn as n ! 1? (The same intuition rationalizes
why LLN on page 8 of Week 1’s slides holds!)
5. Suppose you don’t know how likely the coin in Question 1 is to land heads or tails. You can
use the Bernoulli framework of Question 2 to infer whether the coin is fair.
(a) Suppose n = 100, what can you assume about the sampling distribution of b using the central
limit theorem?
(b) Suppose the realized value of b is 0.78. Test the hypothesis that the coin is fair against the
alternative that it is not at 5% signi…cance level. Compute the p-value.
6. Suppose E (ujX) = 0. Show that it implies Cov (X; u) = 0.
7. Consider the regression model
Y = 0 + 1X + 2D + 3X D + u:
where X is a continuous variable, D is a binary variable, and X D is the
interaction of X and D. Suppose that E (ujX; D) = 0.
(a) What is E (Y jX; D)?
(b) What is the partial e¤ect of X on Y ?
(c) What is the partial e¤ect of D on Y ?
2
8. The …le eaef01.dta contains, among others, the following variables,
EARNINGS = monthly salary
MARRIED = 1 if individual is married, 0 otherwise
ETHBLACK = 1 if individual is black, 0 otherwise
S = years of schooling.
(a) Estimate the following model
ln (EARN IN GS) = 0 + 1S + 2 M ARRIED + 3 ET HBLACK + u. (Model 1)
(b) Assuming that E (ujS; M ARRIED; ET HBLACK) = 0, interpret the estimates of ( 0; 1; 2; 3 ).
(c) Test whether the coe¢ cient of MARRIED is zero at the 5% level of signi…cance.
(d) Test whether the coe¢ cients of MARRIED and ETHBLACK are both zero at the 5% level of
signi…cance.
(e) Consider the following model
ln (EARN IN GS) = 0 + 1S + 2 M ARRIED + 3 ET HBLACK (Model 2)
+ 4 M ARRIED ET HBLACK + u.
Explain, intuitively (before estimating it), why Model 2 may be more desirable than Model 1.
(f) Estimate model (2). Do you prefer to use model (1) or model (2)? Explain your reasonings.
