Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
77 views

Final EM2 2021 Answer Key

This document contains the answer key for an econometrics exam with multiple choice and open response questions. It provides the correct answers to exam questions about topics like instrumental variable regression, prediction, time series modeling, and estimating dynamic effects. The open responses show the steps to calculate values, test hypotheses, describe appropriate tests, and directly estimate dynamic multipliers from a single regression.

Uploaded by

Il Mulinaio
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
77 views

Final EM2 2021 Answer Key

This document contains the answer key for an econometrics exam with multiple choice and open response questions. It provides the correct answers to exam questions about topics like instrumental variable regression, prediction, time series modeling, and estimating dynamic effects. The open responses show the steps to calculate values, test hypotheses, describe appropriate tests, and directly estimate dynamic multipliers from a single regression.

Uploaded by

Il Mulinaio
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Econometrics II Final Exam 2021, Permutation 01

[Updated in December 2022]

Answer Key

Professors: Kirill Evdokimov and Geert Mesters

Part I: Multiple Choice Questions

Question number 1 2 3 4 5 6 7 8

Permutation 01 A C A A A C B D

Part II: Open Questions


You must show all your work but your final answer must be inside the box. If the final answer
is missing or the derivation/explanation that leads to the final answer are missing, no credit
will be given. Likewise, illegible and ambiguous answers will receive no credit. No partial credit
will be given. Only entirely correct answers will receive points. Answers that are not entirely
correct will receive 0 points.
1. Consider the following Stata output from the TSLS estimation and from the first stage
regression.

(a) Calculate the missing values q1, q2, and q3.


q1 = 0.2927119/0.1904822 = 1.54
q2 = 2 (1 − Φ (1.54)) = 2 (1 − 0.9382) = 0.124
q3 = 0.2927119 − 1.96 × 0.1904822 = −0.08063
(b) Test the null hypothesis that the coefficient on Education equals 1.0 at the 5%
significance level.
Rejecting H0 : t = (1.402443 − 1) /0.1870591 = 2.1514, |t| > 1.96.
(c) How many excluded exogenous variables are there in this TSLS regression?
1 variable (S1)
(d) Can we use the J-test of overidentifying restrictions to test instrument exogeneity
here? If yes, provide the distribution of the J-statistic under the null hypothesis. If
not, explain why not in 30 words or less.
No. The model is just-identified, i.e., there are no overidentifying restrictions, i.e.,
#of IVs = #of endogenous regressors.
(e) Are the instrument(s) weak? Explain your answer in 30 words or less.
No, instruments are not weak, because the first stage F -statistic is F = 3.572 =
12.7 > 10.

2. Suppose Yi are i.i.d. with mean µ and variance σ 2 = 7. We have a sample of n = 19


observations, and would like to predict the value of an out-of-sample Y OOS . Consider the
predictor Yb = Y1 /4 + Y2 /4 + Y3 /4.

1
(a) Calculate the expected value of this predictor.
h i
3
E Yb = E [Y1 /4 + Y2 /4 + Y3 /4] = µ/4 + µ/4 + µ/4 = 4
µ .

(b) Calculate the MSPE of this predictor Yb .


 2 
OOS
M SP E = E Y −Yb

 2  h  2 
OOS OOS
2 i
= E Y − µ + µ − Yb =E Y −µ + E µ − Yb
 2 " 2 #
1 3 1 2 h i
= V Y OOS + = σ2 +
 
µ +E µ − Yb µ + V Yb
4 4 16
1 2 3σ 2 1 19σ 2
= σ2 + µ + = µ2 + = 1 2
16
µ + 133
16
≈ 0.0625µ2 + 8.31
16 16 16 16

3. In this exercise we are interested in modeling and forecasting the US unemployment


rate. The first step is to search for an appropriate model. For this we estimated AR(p)
models for p = 1, . . . , 4 using data from 1957:Q1 to 2013:Q4. Some estimation results are
summarized in the following table.
T p SSR(p)
220 1 29.39
220 2 16.44
220 3 16.33
220 4 16.04

(a) Compute the BIC criteria for each model. Based on this criteria which model would
you prefer?
We have
T p SSR(p) BIC(p) = ln[SSR(p)/T ] + (p + 1) ln T /T
220 1 29.39 -1.96
220 2 16.44 -2.52
220 3 16.33 -2.50
220 4 16.04 -2.49
We choose AR(2) as it minimizes BIC.
(b) A next step would be to test whether there are any breaks in the coefficients of the
preferred AR(p) model. Discuss which test should be used and how it should be
implemented for the AR(2) model. You need to clearly describe the test statistic
and how it should be computed in practice.
The Quandt likelihood ratio (QLR) statistic (or sup-Wald statistic) should be used.
The test statistic is
max F (τ )
τ0 ≤τ ≤τ1

where τ0 and τ1 are the truncation points of the sample and F (τ ) is the F statistic
for testing H0 : δ0 = δ1 = δ2 = 0 in the regression

Yt = β0 + β1 Yt−1 + β2 Yt−2 + δ1 I(t ≥ τ ) + δ1 Yt−1 I(t ≥ τ ) + δ2 Yt−2 I(t ≥ τ ) + ut .

with I(t ≥ τ ) the indicator function that is equal to 1 when t > τ .

2
4. A researcher wants to estimate the effect of the world oil price, Ot , on quarterly Spanish
exports, EXt . The researcher estimates the following model

\t ) = 0.05 − 0.75 ln Ot − 0.21 ln Ot−1 − 0.12 ln Ot−2 + 0.17 ln Ot−3


ln(EX

(a) What is the estimated value of the one quarter dynamic multiplier? One period (i.e.,
one quarter) dynamic multiplier is −0.75 − 0.21 = −0.96 .
(b) What is the estimated long run effect (in %) of a one time oil price increase of 5%
on the exports?
(−0.75 − 0.21 − 0.12 + 0.17) × 5 = −4.55 %.
(c) How can we estimate all cumulative dynamic multipliers directly using a single re-
gression?
We can estimate the regression

ln(EXt ) = γ0 + γ1 ∆ ln Ot + γ2 ∆ ln Ot−1 + γ3 ∆ ln Ot−2 + γ4 ln Ot−3 + ut .

Then coefficient γj+1 is the j-quarters cumulative dynamic multiplier.

You might also like