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

ECON3018J  Advanced Econometrics midterm

PART A  Multiple-choice quesitons (10 pts)

1) Omitted variable bias:

A) will always be present as long as the regression R2 < 1.

B) is always there but is negligible in almost all economic examples.

C) exists if the omitted variable is correlated with the included regressor but is not a determinant of the dependent variable.

D)  exists if the omitted variable is correlated with the included regressor and is a determinant of the dependent variable.

E) can be attenuated by removing independent variables from the model.

2) We wish to estimate the model  = 0  + 11 ∗  +  . Suppose that we observe 1, which    corresponds to the “true” 1 with some noise : 1  = 1 ∗  +  , with  being some classical errors/noise. Which of the following statement is accurate?

)  will be biased towards 0

B) There is no measurement error issue in this case

C) The larger the variance of the noise, the smaller the attenuation.

D) 1 will be biased towards the true population parameter

E) There is an expansion bias.

3) The 95% confidence interval for  is the interval:

A)  ±  .  ()

B)  ± 1.96(̂ )

C) ̂ ± 1.96( )

D)  ± 1.96( )

E) None of the answers above

4) The proof that OLS is BLUE requires all of the following assumptions with the exception of:

A) the errors are homoskedastic.

B) the errors are normally distributed.

C) E(ui .

D) X is full rank.

E) E(′|) =   2

5) Consider the following regression line:  ̂  = 698.9 + 2.28 × education. You are testing     whether education has a significant effect on income and you are told that the corresponding t- statistic on the slope coefficient is +3.92. What is the standard error of the slope coefficient?

A) 0 because that is the value of the parameter when the null hypothesis holds.

B) 1.96

C) 1.72

D) 0.58

E) It cannot be computed from the information provided.

6) The p-value tells us how likely it would be

A) under the alternative hypothesis to randomly obtain a t-statistic that is less extreme than the one we estimated ().

B) under the null hypothesis to randomly obtain a t-statistic that is less extreme than the one we estimated ().

C) under the alternative hypothesis to randomly obtain a t-statistic that is more extreme than the one we estimated ().

D) under the null hypothesis to randomly obtain a t-statistic that is more extreme than the one we estimated ().

E) None of the answers above.

7) Which expression corresponds to the 2 ?

A)  1

B)    2  =

C)      =  ()

D)   2  =

E)   None of the expressions above.

8) Which assumption is NOT needed to be able to characterize the distribution of the OLS estimators in the multiple regression model?

A) Large outliers are unlikely

B) E(ui |Xi) = 0

C) No perfect collinearity between the independent variables

D) {( ,  ):  = 1,  2,  3, . . . . , } are I.I.D.

E) (|) =    

9) In the multiple regression model, the ordinary least squares estimator is derived by:

A) setting the sum of squared errors equal to zero.

B) minimizing the absolute difference of the residuals.

C) forcing the smallest distance between the fitted values and the squared residuals.

D) minimizing the sum of squared prediction mistakes.

E) minimizing the absolute difference of the squared residuals.

10) We are interested in estimating (Y ) on X. We know  = 136.78 and  = 61.63.      When estimating  =   0  +  1  +   , we obtain an estimated value for the intercept equal to 114.26.

What would be the predicted value of the intercept when regressing (Y ) on X?

A)  -22.52

B)   11.26

C)   0.027

D)  251.04

E)   We can’t know with the information provided.

PART B  OLS and related properties (15 pts)

We are interested in estimating the following model:

   =    +    +            =  1, … ,

with Z the dependent variable of interest and W the independent variable.

1 1    

b)  Show that the sample covariance between  and the residuals is always zero. (3 pts)

1 ( )( )

−1

c)   Show that the sample mean of  equals the sample mean of  (2 pts)

PART C  Rescaling and demeaning (15 pts)

a)      We are interested in estimating the following model using OLS:

 =   0  +  1  +  

with u representing the error term and  representing the control variable.               Assume that there is a change in the units of measurement on Y. The new variable is Y*=aY. What effect will this change have on the regression slope 1 ? (4 pts)

b)     What will the above change have on the intercept 0 ? (3 pts)

c)       Instead of estimating the model  =   0  +  1  +    (1),    we decide to estimate the model   =    0  +   1   +    (2) with  =   and  =   .

Show that the predicted value of the intercept of model (2) will be equal to 0. (4 pts)

d)    Suppose now that we are interested in this model

 =   0  +  1  +  

but we consider the linear transformation  =    0  +   1  with  =  0  +   1             Derive the expression of the estimated constant (as a function of , ,  0,  1 ) in the model in which we regress  on   , that is  = 0(∗)  +  1(∗)  +    (4 pts)

PART D  (15 pts) Standard errors and correction

We saw in class that, in order to be able to conduct statistical tests, we need an estimate of  2 .

We saw that an estimator for it was 2  = , which would then allow us to estimate | ) = 2 () 1

a)   What is the critical assumption underlying the expression  | ) = 2 () −1 and what does it mean? (2 pts)

b)   Why would it be problematic if we make the assumption in a) but that assumption is not correct when estimating | )? (3 pts)

c)   Discuss two different ways to deal with a situation in which the assumption in a) doesn’t hold (assume here that the error terms are not correlated with each other) (5 pts)

d)    Following your response in c), why would one approach be preferable over the other in practice? (3 pts)

e)    In the ideal scenario where the two approaches could be reliably and successfully       implemented, which of the two approaches would be more efficient, and why? (2 pts)