Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

QUESTIONS:

Answer ALL  TWO questions from Part A and answer ONE question from Part B.

Questions in Part A carry 60 per cent of the total mark and questions in Part B carry 40 per cent of the total mark.   Tables for the normal and F-distribution are at the  end of the examination paper.

PART A

Answer all questions from this section.

A.1 You are working for a major political party called A and wish to better understand which types of voters that are more likely to vote for either A or its main competitor called B. You collect data on n voters where for each individual you observe whether he or she voted for A or B, together with k voter characteristics. For voter i = 1,...,n in the sample, let

8 0,   didn’t vote for A or B

yi = 1,            voted for A      ,

: 2,            voted for B

and xi  = (xi1,...,xik) contains the k household characteristics (such as years of education of each member, total income, etc).

(a) To learn about which voter characteristics explain the relative attractiveness of the two di↵erent parties, you estimate the following regression,

y = β0 + β1x1 + ··· + βkxk+ u,

where we are willing to assume that E[u|x] = 0, using your sample.  Is this regression useful for the purpose of your analysis? Explain.

(b)  Next, you decide to build a model for each party’s share of votes from a given segment of the population of voters.  Formally, among the subpopulation of voters with characteristics x, let 0 πj(x) 100 be the percentage that voted for party j 2 {A, B}.  Your chosen model assumes that πj(x) satisfies

πj(x) = βj0 + βj1x1 + ··· + βjkxk ,

for some unknown coefficients βj0 , βj1 , ..., βjk   for j  2  {A, B}.   Explain  how  you  would estimate the coefficients in these two equations by OLS.

(c) How would you compute standard errors for the OLS estimators in part (b)? Justify your answer.

(d) A colleague claims that more efficient estimators of the coefficients are available.  Is your colleague right? Explain. If your answer is yes then describe such estimators in detail.

(e)  Given estimates of the coefficients βj0 , βj1 , ..., βjk  for j  2 {A, B}, how would you predict the likelihood of a given voter, whose characteristics x you know, will vote for neither A nor B?

(f) How would you estimate the share of voters who would vote for neither A nor B in the whole population? Justify your answer. Provide standard errors for your proposed estimator.

A.2 You collect data on log–wages, lwage, work experience, exper , and education, educ, of 2000 individuals and obtain the following estimated regression function:

with R(¯)2 = 0.26 and e12(=) .(.)2(9) + .(.)0(3)exper + .(.)0(1)exper2 + .(.)0(3)educ,

(a)  Consider an employed male individual with 10 years of education and 5 years of experience. Based on the above regression, provide a prediction of this individual’s wages if he stays in his current position one additional year. Justify your answer and any limitations to it.

(b) Is there an optimal level of years of experience in terms of predicted earnings?  Explain.

(c)  Given the information provided, do you nd that the above model provides a better descrip- tion of data compared to the model that assumes that log-wages are linear in experience and education? Explain.

(d) You estimate the following alternative model:

with R(¯)2 = 0.32 l.(.). .et(xp)t(e)h(r)e(+)io(u)itima(educ)ted coecient.

(e) Which of the above two models would you recommend based on the information provided? (f) How would you formally test the two models against each other?

PART B

Answer ONE question from this section.

B.1 Fajgelbaum, Goldberg, Kennedy, and Khandelwal (Quarterly Journal of Economics, 2020) es- timate the demand for imports from China in the U.S. and the supply of exports in China to analyze the consequences of the trade war between the two countries in 2018. To do so, they leverage the variation in import tari↵s imposed by the U.S. in that year across products. For a large sample of products they specify the import demand equation (where we simplify some details and suppress the intercepts):

qi = ↵d · (pi+ ⌧i)+ di ,                                                                  (1)

where qi  is the quantity of product i imported from China into the U.S. in 2018, ⌧i  is the new import tari↵, pi  is the producer price in China (not including the tari↵), pi+ ⌧i  is the consumer price (inclusive of the tari↵), di  is the error term, and ↵d  < 0 is the slope of the demand curve.

qi = ↵spi+ si ,                                                                              (2)

where si  is the error term and s  > 0.1 The authors assume that the tari↵s are exogenous with respect to demand and supply factors, di  and si.

(a) Write down the reduced-form equations for price and quantity.

(b) Does OLS regression of qi  on the producer price pi  yield a consistent estimate for ↵s? And for ↵d? In each case, explain why or why not and discuss whether the bias is likely towards zero or away from zero.

(c) How can you consistently estimate ↵s? Describe the estimator precisely. What problem with your estimator can arise if export supply is very elastic? Explain the economic intuition for this problem.

(d) How can you consistently estimate ↵d? Discuss what is special about this setting, relative to the model of demand and supply studied in class, that allows you to estimate both parameters.

(e) Suppose now that export supply elasticities vary across industries, such that equation (2) is replaced by

qi = ↵ipi+ si ,                                                                       (20 )

where ↵i  is positive in all industries and independent from the tari↵. Continue to assume that the import demand elasticity ↵d  is homogeneous. Are your estimators from parts (c) and (d) consistent for E[↵i] and ↵d, respectively? Explain why or why not and, in cases when it’s not, discuss the direction of the bias.

B.2 This question concerns the problem of forecasting oil prices, using either autoregressive models or autoregressive distributed lag models with U.S. employment growth.  Figure 1 shows the Brent oil price and total U.S. employment, both in 12-month percent changes. The variables are defined in Table 1, and Table 2 gives regression results. Figures 2 and 3 follow with model diagnostics.

Figure 1: Brent oil price and U.S. Employment 12-month percentage changes

Table 1: Variable Definitions

Unit of Observation: Monthly, July 1987 – November 2014 (T = 329)

Variable name  Variable definition                                        Mean  Std. Dev.

dlbrent

Monthly percentage change in brent oil price (computed as 100∆ ln(Brentt), where Brentt is the Brent oil price in month t)

0.48

8.74

dlemp

Total employment in the U.S., monthly percent change (com- puted as 100∆ ln(Empt), where Empt is the monthly em- ployment)

0.93

0.169


Table 2: Oil price forecasting models, monthly data, 1988-2014

Estimation period:

(1)       (2)       (3)       (4)       (5)       (6)       (7)

1988m1-  1988m1-  1988m1-  1988m1-  2010m1-  1988m1-  2010m1- 2012m12  2012m12  2012m12  2012m12  2012m12  2014m11  2014m11

Regressors

dlbrentt 1

0.265***

0.275***

0.275***

0.262***

0.214

0.270***

0.321**

(0.075)

(0.076)

(0.077)

(0.074)

(0.212)

(0.073)

(0.142)

dlbrentt−2

-0.037

(0.61)

-0.041

(0.64)

dlbrentt−3

0.013 (0.064)

dlempt 1

1.85**

(0.78)

-8.67

(6.12)

1.50**

(0.68)

11.76** (5.35)

Constant

0.46

(0.51)

0.48

(0.51)

0.46

(0.51)

0.30

(0.66)

1.75

(1.05)

0.20

(0.64)

1.48

(0.93)

BIC

4.334

4.351

4.370

Adjusted R2

0.0670

0.0674

0.0641

0.075

0.009

0.074

0.104

RMSFE, 2013m1-

2014m11

3.96

3.96

3.96

4.03

3.98

3.95

3.59

Total no. observations

300

300

300

300

36

323

59

Notes:  All regressions are estimated by  OLS. Standard errors  (in parentheses below coefficients) are heteroskedasticity- robust.   The  RMSFE  is  the  root  mean  squared  forecast  error,  computed  over  the  period  2013m1-2014m11,  using  the reported estimation period to estimate the model parameters (note that the units of the RMSFE are the same as for the dependent variable).  Coefficients are significant  at the ** 5% and *** 1% significance levels.

(a)  Of the three autoregressive models  (1)-(3), choose a preferred model.  Justify your choice.  (b)  Figure 2  plots  the  Quandt  Likelihood  Ratio  statistic,  computed  for  regression  (4)  (note:

the maximum value is 3.67) .

i.  Explain precisely what is plotted in the figure and how you could produce those values.  ii.  In order to draw a conclusion, to what must you compare the values in Figure 2?  Why? iii.  What do you conclude?

(c)  Choose a preferred forecasting model from regressions  (4)-(7) and justify your choice. (d)  Using your  preferred model from question part  (c):

i.  Forecast  the  Brent  oil  price  for  January  2015   (in  $/barrel),  based  on  the  following partial  data  through  December  2014 .    You  can  assume  that  the  predicted  change  is

“small”.  (Hint  1:  first produce  a  forecast of the percentage change) . Variable                                 December  2014 value

Brent price ($/barrel) dlbrentt

dlempt

$62.12

-16.3

0.19

ii.  Give a 67% forecast interval for your forecast.  (Hint 2:  first compute a forecast interval for the percentage change.)

(e)  Suppose  you  were  interested  in  estimating  the  causal  e↵ect  of  changes  in  employment, dlempt, on oil prices, and added dlempt to the regression model in (6).  Would the estimated coefficient on dlempt  be an unbiased estimator of this causal e↵ect?  Why or why not?

(f)  The FRED-MD database is a monthly database consisting of 127 macroeconomic timeseries, with  767  observations  for  each.    Suppose  you  had  th