ECON0021: Microeconometrics SUMMER TERM 2021
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
SUMMER TERM 2021
24-HOUR ONLINE EXAMINATION
ECON0021: Microeconometrics
All work must be submitted anonymously. Please ensure that you add your candidate number and the module code to the template answer sheet provided. Note that the candidate number is a combination of four letters plus a number, e.g. ABCD9. You can find your candidate number in your PORTICO account, under “My Studies” then the “Examinations” container. Please, note that the candidate number is NOT the same as your student number (8 digits), which is printed on your UCL ID card. Submitting with your student number will delay marking and when your results might be available.
Page limit: 15 pages
Your answers, excluding the cover sheet, should not exceed this page limit. Please note that a page is one side of an A4 sheet with a minimum margin of 2 cm from the top, bottom, left and right borders of the page. The submission can be handwritten or typed, but the font size should be no smaller than the equivalent to an 11pt font size. This page limit is generous to accommodate students with large handwriting. We expect most of the submissions to be significantly shorter than the set page limit. If you exceed the maximum number of pages, the mark will be reduced by 10 percentage points, but the penalised mark will not be reduced below the pass mark: marks already at or below the pass mark will not be reduced.
The estimated amount of time it should take to complete this examination is 3 hours.
Answer ALL questions from Part A and one question from Part B. Questions in Part A carry 60 per cent of the total marks in total and questions in Part B carry 40 per cent of the total marks each.
In cases where a student answers more questions than requested by the examination rubric, the policy of the Economics Department is that the student’s first set of answers up to the required number will be the ones that count (not the best answers). All remaining answers will be ignored. Allow enough time to submit your work. Waiting until the deadline for submission risks facing technical problems when submitting your work, due to limited network or systems capacity.
Honour pledge
By submitting this assessment, I pledge my honour that I have not violated UCL’s Assessment Reg- ulations which are detailed in https://www.ucl.ac.uk/academic-manual/chapters/chapter-6-student- casework-framework/section- 9-student- academic-misconduct-procedure, which include (but are not lim- ited to) plagiarism, self-plagiarism, unauthorised collaboration between students, sharing my assess- ment with another student or third party, access another student’s assessment, falsification, contract cheating, and falsification of extenuating circumstances.
PART A
Answer ALL questions from this section .
A1 (30 points)
(a) You are interested in evaluating a training programme o↵ered to managers of food retail firms . The programme was designed to increase investment and growth of these firms . You have firm-level panel data from retail firms in the following sectors: food, beverages, clothing . The data covers several time periods before the training programme and two time periods after the programme . How can you evaluate the impact of the training programme? Write down the estimand and explain the assumptions required . Which treatment param- eter can be identified under these assumptions?
(b) A school is testing a new software to allow students to communicate with their teacher
about class assignments . Researchers are interested in measuring the e↵ect of the software on student satisfaction with their university experience . A randomly selected group of 200 students (in a school with 2000 students) are provided with access to the software . Three months later, all 2000 students are asked to fill in a satisfaction survey. Discuss how you would estimate the impact of the software, what data you need and what assumptions you need to make . Spell out the estimating equation and explain what each term captures .
(c) Now consider that some students in the treatment group gave their login details to their friends in the control group; other students in the treatment group did not use the software . Suppose you receive the list of 200 students initially given access to the software, and the list of students who actually used the software . Can you provide an alternative estimator that would allow you to recover a treatment parameter in this context where some treated individuals did not use the treatment and some control individuals obtained access to treatment? Should you exclude from your analysis the students in the control group who received the login details from their friends? If you were able to redo the random assignment of the software, how would you suggest to do it?
A2 (18 points) Judge if each statement given below is correct or incorrect, providing your reasoning .
In statements (a)- (c), (yi,xi) , i = 1 , . . . ,n, are iid observations of a dependent variable yi 2 {0, 1} and a vector of regressors (observable characteristics), xi = (1,x1i , . . . ,xKi )\ . The vector of intercept and coefficients is denoted by β = (β0 , β 1 , . . . , βK )\ , where β0 is the intercept and βk , k = 1 , . . . ,K, is the coefficient of regressor xki .
(a) Conditional maximum likelihood estimation requires the researcher to specify and param- eterize the joint distribution of yi and xi .
(b) The only drawback of the linear probability model is that it requires xβ 2 [0 , 1] .
(c) Consider the following variation of the threshold-crossing model studied in class: y*i = β0 + β1x1i − ✏i , yi = *i*i
where ✏i are iid N(0, 1), and β0 , β1 and a are unknown parameters . Then, the unknown parameters β0 , β1 and a can not be separately identified and estimated by maximum like- lihood .
(d) The multinomial logit model is likely to generate unreasonable substitution patterns if some of the options are close substitutes .
A3 (12 points) Consider a collection of iid random variables {zi} drawn from a distribution with
the probability density function (pdf) fZ (z;λ0 ) for some unknown λ0 > 0 . The pdf fZ (z;λ0 ) is known to belong to the following parametric family (with λ > 0):
fZ (z;λ) =
if z > 1 ,
otherwise .
As usual, λ0 denotes the true value of the parameter of interest, which the researcher wants to estimate .
(a) Write down the log-likelihood function .
(b) Derive the maximum likelihood estimator and argue that it’s consistent for λ0 . (Hint: To show consistency use that fact that Eλ0 [log zi] = λ)
(c) Derive the Fisher information for λ and the asymptotic variance for , i .e . provide an expression for ⌃ such that satisfies ^n( − λ0 ) !d N(0, ⌃) .
(d) Suppose that n = 100 and P log zi = 0 .25 . Use the asymptotic distribution established in (c) to test the hypothesis H0 : λ0 = 5 against the alternative Ha : λ0 5 at the 5% significance level .
PART B
Answer one of the two questions in this section .
B 1 (40 points) Assume you work as an impact evaluation specialist with the government’s tax
audit office . The office aims to target audits at the firms most likely to engage in tax evasion . Every year, government tax analysts calculate a risk score for every firm . The risk score takes into account firm characteristics and prior firm behaviours that are associated with tax evasion propensity. The risk score varies from 0 to 200 . Firms with a risk score above 120 are selected for audit . The director of the tax audit office is concerned that a tax audit might drive firms into informality (i .e . stop filing their tax declaration) . She asks you to evaluate the e↵ect of audits in 2018 on tax filing behaviour in 2019 . (For the purpose of this question, assume that tax declarations for year X are filed in year X) .
(a) Describe how you would conduct this evaluation . What is the objective of interest you will
estimate, and what assumption would you need to make?
(b) How would you test the assumption? What data would you need for this?
(c) Assume that your evaluation (using the methods described in part a) suggests that being subject to an audit increases a firm’s likelihood of filing its tax declaration in the following year by 5 percentage points . The director of the tax administration has long been striving to increase tax filing rates . Non-filing is particularly high among small firms . The director concludes that your evaluation “shows that we can increase tax filing rates among small firms by auditing these firms” . Do you agree with this statement?
(d) After delivering your first evaluation report, you learn that the 2018 audit programme actually worked as follows: The tax audit office used the risk score to select 90 firms for an audit . This group contains all the firms that had a risk score above 120 . Among these 90 firms, 10 firms were deemed “politically sensitive”, and were not audited . Can you still apply the procedure you proposed in part a) to evaluate the e↵ect of the audits on tax filing in 2019? Why, why not, or under which conditions? Please illustrate your answer with examples .
B 2 (40 points) Consider consumers who face the choice between J alternative products 1 , 2 , . . . ,J
(with no “outside option”) . The products have one common observed characteristic xj 2 R and consumer i’s utility from product j is given by
uij = ↵ + β(Ii− pj)+ Vxj + ✏ij ,
where pj is the observed price of product j and Ii is individual’s income, that is not observed . Unobserved utility components ✏ij are iid across both i and j and have extreme value type- I distribution . As usual, let ✓ = (↵ , β ,V)\ denote the vector of unknown parameters . Let Yi 2 {1, 2, . . . ,J} be the choice of consumer i. To sum up, the researcher observes a large random sample of individual choices {yi} and the product characteristics {pj,xj}, which are treated as non-random.
(a) Write down the conditional choice probability P(Y = j|I;✓).
(b) Is that a problem that the researcher does not observe Ii? Explain. Provide the expression for P(Y = j|✓) and write down the likelihood (or log-likelihood) function (note that it should depend on the observables only).
(c) Can ↵ be identified and estimated by maximum likelihood? Explain.
(d) Explain why we need J > 3 in order to be able to identify and estimate the remaining parameters β and V. (Hint: This is a harder question. Since we do not have any observed variation in the choice related characteristics at the individual level, maximum likelihood simply tries to match the observed products’ shares sj by the choice probabilities P(Yi = j|✓) = P(Yi = j|β ,V) predicted by the model. Try to see what goes wrong when J = 2 .)
Now consider a random-coefficient version of the previous model
uij = ↵ + β(Ii− pj)+ Vixj + ✏ij ,
where unobserved Vi is allowed to be consumer-dependent. Suppose that Vi are iid across i and independent of all the other variables. We also suppose that Vi are drawn from the following distribution
Vi = (
with with
probability q,
probability 1 − q,
where V 6 V. As before, ✓ = (↵ , β ,V,V,q) denotes the vector of unknown parameters.
(e) Provide the expression for P(Yi = j|✓) and write down the likelihood (or log-likelihood) function. It might be helpful to start with writing down P(Yi = j|I,V, ✓).
(f) Can ↵ be identified and estimated by maximum likelihood in this model? Explain. How many products do we need in order to be able to estimate the remaining parameters β , V, V, and q? Explain. (Hint: This question is closely related to question (d) . Exactly the same reasoning applies .)
2023-05-19