Scribd
Upload
14 views

Problem Set 2

Uploaded by

leafko
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
14 views

Problem Set 2

Uploaded by

leafko
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

AEREC 511

Econometrics II
The Pennsylvania State University
Spring 2015

Problem set 2
Due by 11:59pm on February 27, 2015

1. Consider the model


et=ρet-1+vt
a. Suppose ρ=0.9 and σ v2 =1. What is (i) the correlation between et and et-1? (ii) the
correlation between et and et-4? (iii) the variance σ e2 ?
b. Repeat part (a) with ρ=0.4 and σ v2 =1. Comment on the difference between your answers
for parts (a) and (b).
    
2. Show that ( X 'X) −1 X ' y = ( X 'X) −1 X ' y .

3. Consider a simple time series model where the explanatory variable has classical measurement
error:

yt =β 0 + β1 xt* + ut
x=
t xt* + et

where ut has zero mean and is uncorrelated with xt* and et . We observe yt and xt only. Assume
that et has zero mean and is uncorrelated with xt* , and that xt* also has a zero mean.

a. Write xt=*
xt − et and plug this into the model. Show that the error term in the new
equation, vt, is negatively correlated with xt if β1>0. What does this imply about the OLS
estimator of β1 from the regression of yt on xt?
b. In addition to the previous assumptions, assume now that ut and et are uncorrelated with
all past values of xt* and et ; in particular, with xt*−1 and et −1 . Show that E ( xt −1 , vt ) = 0 , where
vt is the error term in the model from part (a).
c. Are xt and xt-1 likely to be correlated? Explain.
d. What do parts (b) and (c) suggest as a useful strategy for consistently estimating β0 and
β1?

4. Consider the simple regression model y =β 0 + β1 x + u , where we think that x and u are
correlated. We are considering z as a possible instrumental variable for x. While IV is
consistent when z and u are uncorrelated, and z and x have any positive or negative correlation,
IV estimates can have large standard errors if z and x are only weakly correlated. Weak

1
correlation between z and x can have even more serious consequences: the IV estimator can
have a large asymptotic bias, even if z and u are only moderately correlated.
Define the plim of the IV estimator and of the OLS estimator, respectively, in terms of
population correlations and standard deviations as:
 Corr ( z , u ) σ u
p lim β= β1 +
Corr ( z , x) σ x
1

σ
p lim β= β1 + Corr ( x, u ) u
1
σx
a. When is IV preferred to OLS, on asymptotic bias grounds? When could the inconsistency
in the IV estimator be large?
b. Assume that σu=σx, Corr(z,u)=0.1 and Corr(z,x)=0.2. What is the asymptotic bias in the
IV estimator?
c. How much correlation would have to exist between x and u before OLS has more
asymptotic bias than IV?

5.
a. Use the data on married working women in the MROZ dataset to estimate (by OLS) the
return to education in the simple model:
log(wage)=β0+ β1educ+u
Comment on the OLS estimate for β1.
b. You suspect that educ may be endogenous, and are considering the use of fatheduc as an
instrumental variable. Based on what we discussed in class, why might fatheduc be a good
instrument? Is this a valid instrument in our sample? What percentage of the variation in
educ in the sample does fatheduc explain? (Note: pay attention to the number of observations
when checking the validity of the instrument)
c. Using fatheduc as instrument for educ, check the endogeneity of educ using a Durbin-Wu-
Hausman test. What do you conclude?
d. Estimate the model in (a) using IV (and fatheduc as an instrument for educ). Compare the
IV estimate of the return to education to the OLS estimate. What do you note? Is this
result consistent with omitted ability bias?
e. Compare the standard errors of the IV and OLS estimates of the return to education.
What do you conclude? Is the difference between the two estimates statistically significant?
f. Compare the estimates from the second step of the endogeneity test in (c) to the estimates
obtained in (d). What do you observe?
g. Re-estimate the log(wage) equation adding exper and expersq as exogenous variables, and
motheduc as instrument for educ. Compare these estimates and standard errors to those in
(d).
h. Use two least squares regressions to obtain the same IV estimates in part (g).
i. Test the overidentifying restriction.
j. Add huseduc as instrumental variable for educ, and re-estimate the log(wage) equation.
Compare the estimates with those in (g): what do you note? Test the two overidentifying
restrictions.

You might also like