E n{Ê Lof{i0oit Scxoot-

oF ECOIIOMICS
PoLITICAL ScIENcE
^flo
¡

Summer 2018 Exam

EC486
Econometric Methods

Suitable for all candidates

lnstructions to candidates
Thís paper contains two parts. Answer TWO questions from Part A, and TWO questions from
Part B. Each question carries equalweight.
Time Allowed - Reading Time: None
Writing Time: 3 hours

You are supplied with: No additional materials

You may also use: No add ition al ateri al s
.m
Calculators: Calculators are allowed in this examination

Pagel of11
o)LSE ST 2018/8C486
 Part A

Answer any TWO of the following four questions.

1 . Consider the problem of evaluating the impact of class size on student performance
in California as discussed in class. There is a district-level dataset where we have
information about average class size in each district as well as test scores, and
income. The researcher estimates the causal effect of class size on test scores by
OLS, using the following model

testscore¿: a¿l- B x si,ze¿Il xincome¿i €¿. (1)

(a) Suppose the researcher decides to estimate the short regression instead:
testscore¿ : ùI þ x s'ize¿ * e¿.

Derive the formuta of the coefficient of size in the short regression and discuss
the relation with the coefficient estimated in the lonE regression (1) above. [4
marksl

Suppose the variabl a s'izeis measured with an error, i.e. : s'iz€¿* ur¿, where
-size¿
cou(size¿,w¿) :0 and cou(w¿,i,ncome¿) : Q.

(b) Show that the regression coefficient, B, suffers from attenuation bias, À. ln
deriving the regression coefficient formula, assume lhat cou(si,ze¿,incornei) :
0. [5 marks]
(c) Now assume lhat cou(si,zeij'incorne¿) * 0 and the long following regression is
estimated:
testscore¡ : a * p * si,ze¿ I 1 x i,ncome¿ t €,i

The formula of the regression coeffibient is B : ÉÀlwith À' :

&,where À is
the attenuation factor as derived above, and Rl" is the R2lrom the population
regression oÍ s¿ze on'incom,e¿. Show lhaT R!, < À. [6 marks].
(d) Assume that cou(s'ize¿,i,ncome¿) l0 and the long regression is estimated:

testscore¿ : a I p * si,zq I 1 x'income¿ * e¿.

ls the coefficient 7 estimated with a bias? Explain your motivations in detail

without explicitly deriving any formula. [5 marks]

@ LSE ST 201 8/EC486

Page2of11
 (e) A colleague suggests estimating the following short regression:

testscore¿ : a * p * si,zq + e¿.

instead of the long regression in (d) to reduce the bias in the coefficient B.
Assuming thal cou(si,ze¿,'incorneo) # 0, do you agree with your colleague?
Explain your motivations without explicitly deriving any formula. [5 marks]

ST 2018/EC486 Page3of11
oLSE
 2. You are working for a government agency and you have been asked to evaluate
the importance of business education on wages. A government agency gives you
access to a data set of 20,000 randomly selected employees with information on
their average wage and whether or not they have an MSc in Management or its
equivalent. This is coded as QUAL: 1 if an individual has such a qualification
and QUAL: 0 if not.

(a) The government has done some preliminary analysis on the data and found
that the average wage (WAGE) of individuals with QUAL: 1 is 80% higher
than those wilh QUAL: 0. Does this mean that obtaíning an MSc is likely to
raise wages by 80%? lf not, what other variables could be collected to obtain
a better estimate? [5 marks]
(b) You run the following regression:

ln(WAGE¿) : aQUAL¿ * X¿þ I e¿,

where ln(WAGE¿) is the log of the wage of individual i, p is a vector of param-

eters to be estimated and Xo includes whether individuali has an undergradu-
ate degree and any of the exogenous observable controls that you suggested
in answer to part (a) above. The e¿ is an error term. The OLS estimate of a
is 0.08. What does this now suggest about the returns to management edu-
cation? Why do you think the estimated value of managerial education has
fallen compared to the case when we assumed þ :0? [5 marks]
(c) A government official suggests that we should use a "fixed effect" estimator
to control for "unobserved ability". Why do you think she is suggesting this?
What data would you need to implement this? Do you believe that this is a
good estimation strategy? (Hint: write down the appropriate formula explicitly
in your answer). [5 marks]

(d) A colleague suggests that you try an instrumental variable approach. The
dataset has information on the geographical location of where the individual
went to school and the distance from this to the nearest university offering an
MSc in business education (DIST). He suggests using DIST "as an instru-
ment for schooling". Why do you think he is suggesting this? Do you think this
is a credible identification strategy? [5 marks]

o)LSE ST2018/EC486 Page4of11

 (e) Using the fixed effects method in part (c), the estimate of a is aFE : 0.02.
Using the instrumentalvariable estimator in part (d), the estimate of a is aIV -
0.15. What could account for these different results? lf you had a second
instrurnental variable that you were sure was a valid instrument would this
help you? [5 marks]

oLsE sT2018/EC486 Page5of11

 3. This question looks at a population of applicants to the Ph.D. in Economics in 1985
who had similarly marginal credentials, which are followed up over time. Suppose
that admission was offered randomly to half of the population in question. Consider
the following econometric framework:

Y: D¿Yt¿ + (1 - D¿)Yo¿

P¿: D¿Pt¿+ (1 - D¿)Po¿

where D¿ is an indicator for person i having been admitted in 1985 to a Ph.D. in

Economics; Y¿ is log annual earnings of person iin2ß10;Yr¿ is log annualearnings
of person ? in 2010 if they had been admitted in 1985 to the Ph.D. in Economics;
Y¡¿ is log annual earnings of person ù in2010 if they had not been admitted in 1985

to the Ph.D. in Economics; P¿ is an indicator for person z working as a

professor in 201 0, Pu is an indicator for person i working as a professor in 2010
if they had been admitted in 1985 to a Ph.D. in Economics; and Ps¿ is an indicator
for person z working as a professor in 2010 if they had not been admitted in 1985
to a Ph.D. in Economics.

(a) Could you estimate the causal effect of admission to the Ph.D. in economics
in 1985 on log annual earnings in 2010 for person z in this population? Explain
your answer briefly. [6 marks]
(b) Could you estimate an (average) causal effect of admission to the Ph.D. in
Economics degree in 1985 on log annual earnings in 2010? Explain your
answer briefly.[6 marks]
(c) Now suppose that we regress Y on Dofor the people for whom P¡: 7. Write
the estimated coefficient on D¿and its relation to the effect EIY¿-Y¡¿)P1: 7l;
Explain your answer. [7 marks]
(d) Explain intuitively (without equations) why conditioning a regression of l¿ on
D¿Íor the people for whom P¿: fundamentally different from conditioning
L is
a regression ol Y on D¿ on an indicator for being born after 1962. Which
approach makes more sense, and why? [6 marks]

oLSE ST2018iEC486 Page6of11

 4. Consider the simple linear model with a single, possibly endogenous, explanatory
variable r, ârìd a single instrument z:

a:þo*þfi1-u,
E (u) : 0, Cou (r,u) :0, Cou (",ù I 0, E (u2lz) : o2.

(a) What is the lV estimato r of B1? Show that the lV estimator is unbiased. t6
marksl
(b) Explain intuitively what happens if Cou (z,r) -+ 0. [5 marks]
(c) Assume now that z is binary. Show that the lV estimato r of Blcan be written
as
at-ao
rt-ro
where where 96 and ro ãtê the sample averages oÍ y¿ and n¿ ovèr the part of
:
the sample with z 0, and 91 and rt altê the sample averages of y¿ and r¿
:
over the part of the sample with z I.17 marksl

(d) Explain in details why different instruments might lead to different lV esti-
mates. [7 marks]

Page7of11
@)LSE ST2o18/EC486
 Part B

Answer any TWO of the following four questions.

5. Suppose you estimate demand using the logit model. The random utility is
given by
u¿¡: x¡0 - aln(p¡) * {¡ * e¿¡,

where X, are observed product characteristics, p¡ is price and (¡ is an unob-

served product characteristic and e¿¡ iS thê error term.

i. What is the utility of the outside option? ls it possible to estimate the

mean utility of all products as well as the outside option? ls it possible to
estimate the standard deviation of the error term? Explain. (6 marks)
ii. What assumptiorì orì e¿¡ generates the logit model? Write the individual
and aggregate demand for product 7. (5 marks)
iii. The main problem with the logit model specified above is that it implies
the lndependence from lrrelevant Alternatives (llA). Explain the problem.
Compute the own and cross price elasticities for product j and illustrate
the implication from ¡lA. (7 marks)
iv. Explain how you would estimated the model using aggregate data and if
there is any endogeneity problem. (7 marks)

€)LSE ST 2018/EC486 Page8of11

 6. Suppose you wish to estimate the parameters {B6, þt, þx} of the following
production function:

log (yir) : þo I þrlog(L") + þxlog(K¿) -f e¿¿ (2)

where Y¿, is the revenue, .L¿¿ is the labor and K¿tis the capital of firm i,inyear
ú, and e¿¿ is ân unobserved component.

i. Explain what are the potentialproblems if you estimate equation (2) using
OLS. (6 marks)
ii. Suppose that the firms in your sample are price-takers in factor markets
(i.e, labor,L and capital K) and in the output market, and that there are
no costs to adjust factors. How could you estimale {lt. and 0x? (6 marks)
iii. Assume now that there are costs to adjust factors and you have a bal-
anced panel of firms. Can you propose an instrumental variable proce-
dure to estimate þt and p67 Whal are the potential problems with this
estimator? (6 marks)
iv. Explain the Olley and Pakes procedure. (7 marks)

oLSE ST2018/EC486 Page9of11

 7. Consider the entry model of Bresnahan and Reiss. You have a sample of
different markets of sizes ,S¿, each with ,AL active firms and you calculate the
ratio s¿: ft.
i. What is the interpretation of the ratio s¿? (8 marks)
ii. How do expect the ratio s¿ to change as the market size ,S¿ increases?
Why? (8 marks)
iii. Assume that firms' profits are

r¿: S¿V (^/,) - F (1V¿) +e

where
N
y (^/,) : (rr Ðo'
n:2
1V

r (¡r,) : ?r * Dr,"
n:2

represent variable profits and fixed costs, respectively. How would you
estimate the parameters a¿ and y7 How are they separately identified?
(9 marks)

oLSE ST2018iEC486 Page10of11

 8.i Consider the model of Varian (1980): l/ firms sell a homogeneous good
and all have common marginal cost c; a unit measure of buyers have
unit demand and they all have the same valuation u > c for the good; a
fraction 0 < I < 1 of buyers sample only one seller at random (i.e., they
are uninformed), whereas the fraction 1 - À sample all sellers and buy
from the lowest-cost seller (i.e., they are informed).
A. Show that there are no pure-strategy equilibria. (5 marks)
B. Consider a symmetric mixed-strategy equilibrium. Derive the equilib-
rium mixed strategy ,F' (p) (i.e., the probability distribution over prices
p) and the lower bound of the support of F (p). (8 marks)
ii Consider the paper "Equilibrium Price Dispersion in Retail Markets for
Prescription Drugs" by A. Sorensen.
A. Describe the data and the empirical model. (8 marks)
B. Describe the results of the paper. (4 marks)

@)LSE ST 2018/EC486 Page11of11