ECON6004 Quantitative Methods 2020-21
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
SEMESTER 1 TAKE-HOME FINAL ASSESSMENT 2020-21
ECON6004 Quantitative Methods
1 Consider a bivariate regression model:
yi = β0 + β1x1i + ✏i
[10]
(a) Explain first how OLS works in words, and show this diagram- matically. [7]
(b) When running a regression yi = β0 + β1x1i + ✏i, a colleague recommends that you use robust standard errors. With the help of a diagram, explain the issue that robust standard errors help to prevent. Does this issue a↵ect the resulting point estimates of β0 and β1? [3]
2 (a) The weight in a country has long been stable, normally dis- tributed with mean 67kg and standard deviation 8.1kg. Last year, the country allowed fast food restaurants to open chains in the country. Recently the health ministry in the country con- ducted a survey, randomly sampling 250 adults and finding an average weight of 68.4kg. The government is concerned about this and is considering withdrawing the licences of these fast food firms due to the increase in weight suggested in the data. Are they justified in doing so, based solely on the data collected? Be sure to clearly specify the hypothesis being tested, explain your answer and use ↵ = .01 as the level of significance for the test. [8]
(b) The failure rate for an electrical product produced by a company is 7.9%. The company has recently introduced productivity pay for its factory workers, which has markedly increased the out- put of the workers. The company is worried though about the
downside of this change in payment structure, as during a recent audit of 105 products, the failure rate was 12.1%. Is this a sta- tistically significant increase in the failure rate for these goods? Be sure to clearly specify the hypothesis being tested, explain your answer and use ↵ = .05 as the level of significance for the test. [7]
(c) The heads of two school districts are both running to be elected as regional school director. The two school districts have very similar budgets, and pupil intakes. In district 1, the success rate of the 105 pupils taking a national standardised test is 72%. In the second district, the 110 pupils had a 62% success rate on the same test. The head of district 1 wants to use her pupils’ performance as part of her campaign. Is this a wise move based on the data? Be sure to clearly specify the hypothesis being tested, explain your answer and use ↵ = .05 as the level of sig- nificance for the test. [5]
3 We are interested in modelling wages. We write down a regression model for log wages as:
ln(wage)i = xiβ1 + β2educi + ✏i ,
where ln(wage)i is the natural logarithm of wages, x1i is a vector of exogenous characteristics (including a constant) and educi is com- pleted years of education. [35]
(a) Write down the population moment conditions required for the consistent estimation of β1 and β2 using OLS. [5]
(b) We are concerned that education may be endogenous. What does this imply for the moment condition for educi? What does this imply for the OLS estimates of β1 and β2? Give a clear ex- ample of why education may be endogenous, and use equations to detail this. [10]
(c) Building on part (b), suppose we have a continuous instrument for educi. Call this z2i. Explain in words and with equations the two key features a valid instrument will satisfy. Write down both the population moment conditions, and then the corre- sponding sample moments, for the instrumental variables (IV) estimator to consistently estimate the parameters in the model. [5]
(d) In Figure 1 below, you will find OLS and IV estimates (and robust standard errors) of β2 - the returns to education from a random sample of working age males in England and Wales. In this case, we used a change in the minimum school leaving age to instrument for education. The estimates for the vector β1 are not presented for brevity. At first these results may be puzzling, as we expected the OLS estimate of β2 to be upwards biased, and we also expected the IV estimates to be a consistent estimate of the true population parameter, β2. Discuss two di↵erent reasons why this may be the case, using equations and diagrams as part of your answer. [15]
Figure 1: OLS and IV Estimates - Returns to Education
(1)
OLS
(2)
IV
Education 0.081
(.002)
0.142
(.009)
N 23, 724 23, 724
4 We are interested in modeling the labour force participation decision for married women. We specify:
yi = 1 yi = 0
if individual i works
if individual i does not work
[20]
(a) If we assume E[✏i|xi] = 0 and estimate the following regression
by OLS:
yi = x β + ✏i
where xi is a vector of exogenous characteristics, then what are the econometric problems with this approach? Name, and dis- cuss two in detail, including relevant equations. [6]
(b) For a continuous regressor xij, write down the partial e↵ect for the model in part (a), that is when this binary choice model is estimated by OLS. [2]
(c) Now consider a latent variable approach to this binary choice problem:
y = x β + ✏i
7
where
yi = 1 yi = 0
if y > 0
otherwise
Assume ✏i is symmetrically distributed, and let F denote the dis- tribution function of ✏i. Note that this is a general distribution function. You can use f = F0 to denote the density function of ✏i. Write the probability that yi = 1 given xi in terms of F. For a continuous regressor xij, write down the partial e↵ect for this model. How does the partial e↵ect you calculated here di↵er from the one calculated in part (b)? [4]
(d) Discuss in detail how we can derive the probit model from an underlying latent variable model given in part (c). What nor- malisations do we need to make here, and why do we need to make these? Answer in detail. [6]
(e) Below see Figure 2 for output from a probit regression. You are given βˆprobit and x. Use the values in the table to work out P[yi = 1|xi], where you can set all values in xi to sample averages. Be careful regarding the constant in this step. Next work out the marginal e↵ects for both education and age. [11]
(f) Finally assume that we include a new variable cityi, which is a dummy variable that =1 if individual i lives in a city, and =0 otherwise. Let’s say that βˆcity,probit = .1, and that when we in- clude this as an additional explanatory variable, the coefficients on the other variables remain exactly the same as they are in
part (e). Calculate the marginal e↵ect of cityi [6]
Figure 2: Probit Estimates and Variable Means
nwifeinc educ exper expersq age
kidslt6
hushrs/1000
Constant
Observations
Estimated coefficients |
Mean of Variables |
-0.0109 (0.0049) 0.1310 (0.0252) 0.1201 (0.0187) -0.0018 (0.0006) -0.0566 (0.0080) -0.8917 (0.1183) -0.1328 (0.0867) 0.7831 (0.4984) |
20.129 12.287 10.631 178.039 42.538 .238 2.267 |
753
2022-01-13