Вы находитесь на странице: 1из 2

Subject M300 Econometric Methods

Instructor: Alexei Onatski

E S 2
D  M , N  2,  12:00
1. (Time series regression assumptions) Suppose that, for a province or district, the crime rate,
crimet , is a two-year distributed lag of the clear-up rate (percentage of crimes resulting in
a conviction):
crimet = 0 + 0 clearupt + 1 clearupt1 + 2 clearupt2 + t ,
where t is uncorrelated with clearupt , clearupt1 , clearupt2 , and all other past values of
the arrest rate. Suppose that, through expenditures on law enforcement, the clear-up rate
can be made to react to last years crime rate:
clearupt = 0 + 1 crimet1 + vt
(a) Explain, in behavioral terms, what 1 > 0 means.
(b) If vt in uncorrelated with all past values of clearupt and t , argue that clearupt and
t1 must be correlated. Which Gauss-Markov assumption does this violate?
(c) Assume that the conditional expectation of t given the current and all past values of
clearup and all past values of crime is zero:
E (t |clearupt , clearupt1 , crimet1 , ...) = 0.

(1)

Using the law of iterated expectations, show that in such a case E (t s |xt , xs ) = 0
for any t = s, where
xt = (clearupt , clearupt1 , clearupt2 ) ,
so that there is no serial correlation in the errors.
(d) Assumptions like (1) are sometimes referred to as the dynamic completeness assumptions. Give some reasons why the dynamic completeness may be violated for the crime
and clearup example. What will be consequences of such a violation for the properties
of OLS?
2. (Regression interpretation, spurious regression, FGLS) This problem is based on example
10.4 of Wooldridge. The general fertility rate (gf r) is the number of children born to every
1,000 women of childbearing age. For the years 1913 through 1984, the equation,
gf rt = 0 + 1 pet + 2 ww2t + 3 pillt + ut ,

(2)

explains gf r in terms of the avergae real dollar value of the personall tax exemption (pe)
and two binary variables. The variable ww2 takes on the value unity during the years 1941
through 1945, when the United States was involved in World War II. The variable pill is
unity from 1963 on, when the birth control pill was made available for contraception.
(a) Using the data FERTIL3.dta, available from the course web site, run regression (2).
Save and report the STATA output.
(b) Using Graphics/ Time-series graphs/ Line plots menu, create the time series plots
of gf rt and of pet . You will need to type "tsset year" in the command line before
using the Graphics menue to define time as year variable. Do gf rt and pet appear
stationary? Do they appear weakly dependent? Explain your answers.
1

(c) Run regression


gf rt = 0 + 1 pet + 2 ww2t + 3 pillt + 1 t + 2 tsq + r tcu + ut ,
where t is the time trend t = 1, ..., 72, tsq = t2 , and tcu = t3 . Compare to the results
in a) focusing on the estimate of the coecient on pet . Do the results in a) appear to
be spurious?
(d) Using Statistics/ Linear models and related/ Regression diagnostics Specification test
etc. menu, test the hypothesis that there is no serial correlation in the errors of the
regression in c) against the hypothesis of the first order correlation. Use BreuschGodfrey test, and specify the lag order to be tested as 1. Do you reject the null?
What does this imply regarding the validity of your results in c)?
(e) Using Statistics/ Time series/ Prais-Winsten regression menu, run the FGLS for the
regression equation given in c). Use the "Single-lag OLS of residuals" option in the
drop-down menu. Compare the obtained results with those in c) focusing on the
estimate of the coecient on pet . Which results would you prefer and why?
(f) Gary Becker (1960) An Economic Analysis of Fertility argues that children can be
viewed as a durable good, so equations explaining gf rt can be interpreted as demand
for children. Given such an interpretation, what additional variables you would include
in regression (2)? Will these variables be able to capture the apparent time trend in
gf rt ? Discuss.
3. (Beveridge-Nelson decomposition) Assume that log of real aggregate consumption Ct in the
UK is an I(1) process. Further, let its first dierence log Ct satisfy the following AR(1)
model
log Ct = ( log Ct1 ) + t ,
where t is i.i.d. shock with mean zero and variance 2 , || < 1, and = E ( log Ct ) .

(a) Suppose that you know the values of and . What would be your forecast of
log Ct+h in terms of log Ct ?
(b) What would be your forecast of log Ct+h h in terms of log Ct and log Ct ?
(Hint: log Ct+h = log Ct + log Ct+1 + ... + log Ct+h ).
(c) What would be the limit of your forecast from b) as h ?
(d) Now suppose that the Beveridge-Nelson decomposition of log Ct is given by
log Ct = log C0 + t + (1 + ... + t ) + et ,
where et is a zero mean weakly dependent stationary process. Show that the forecast
of log Ct+h h made at time t must converge to
log C0 + t + (1 + ... + t )
as h . For this, assume that t , t1 , ... are observable at time t. Recalling c),
express the trend log C0 + t + (1 + ... + t ) in terms of log Ct and log Ct .
(e) File BN.dta contains the values of log Ct for 1955-2014 period. By running AR(1)
in first dierences, estimate and . Given your answer in d), estimate the trend
log C0 + t + (1 + ... + t ) component of log Ct . Obtain an estimate of the cycle
component et , and plot it against time.
4. You are given a MATLAB code unitroot1.m that 1) simulates 1000 observations of a random
walk series yt with t = 1, ..., 500; 2) for each of the simulations, run regression of yt =
3) Draws a histogram of the stored
and ;
+ yt1 + t and stores the OLS estimates

values of . Please, change the code so that the value of the t-statistic for testing the
hypothesis H0 : = 1 is stored, and its hystogram is drawn. What do your results suggest
about the asymptotic distribution of the t-statistic for testing the unit root hypothesis?

Вам также может понравиться