Shanghai Jiaotong University

Shanghai Advanced Institution of Finance

Financial Econometrics I

Assignment 1 - due 12/18/2017

For additional help on Matlab, you can get additional materials from the website of “www.math.ufl.edu/
help/matlab-tutorial/”. For problems that use Matlab, you need to hand in your compute code along with
your discussion of the results for full credit. Please do not just hand in your computer printouts.

1. (Statistics Review)
(a) Suppose a random variable x follow a normal distribution of N [1, 22]. Let z = 
3x2 + 2x + 1,
compute μz = E(z) and 2
σ = V ar(z). [Hint: 2
for f (x) = N (μ, σ ), we have xf (x)dx = μ,
  z 
(x − μ)2 f (x)dx = σ 2 , (x − μ)3 f (x)dx = 0, (x − μ)4 f (x)dx = 3σ 4 . For x2 you can rewrite it
as (x − μ)2 + 2μx − μ2 ]
(b) Suppose {x1 , x2 , · · · , xn } is a sample with n observations drawn from the distribution of x in
(a). For each observation xi , you can compute the corresponding zi according to the relation in
 
(a). With {z1 , z2 , · · · , zn }, you can then calculate (1/n) ni=1 zi and (1/(n − 1)) ni=1 (zi − z̄n )2
as the for the mean and variance estimators of z, respectively. For a sample size of n = 5,
compute and denote your first estimates z̄(5),1 and σ̄z(5),1 2 . Discard your sample of xs and
redraw a new 5-observation sample for x. Compute and denote your second round estimates as
2
z̄(5),2 and σ̄z(5),2 . Repeat the exercise for 10 times.1 Plot z̄(5),1 , · · · , z̄(5),10 against the theoretical
value μz found in (a), what pattern do you observe? Discuss your variance estimates in the
same way.
(c) Set the sample size to n = 50, and Repeat (b). What do you observe?
2. (Possible Bias in OLS) Matlab can generate two independently distributed random variables with
standard normal distribution, i.e., η1 ∼ N (0, 1), η2 ∼ N (0, 1), where Cov(η1 , η2 ) = 0. Using η1 and
η2 , we can generate two variables x1 and x2 in the following two cases:
Case I (independent xs): Let x1 = 1 + 2η1 and x2 = 2 + η2

x1 = 1 + 1.6η1 + 1.2η2
Case II: (dependent xs): Let
x2 = 2 + 0.923η1 − 0.385η2
Generate a vector of 30 independent η1 = [η1,1 , η1,2, · · · , η1,30] and η2 = [η2,1 , η2,2, · · ·, η2,30] sepa-
rately, we can construct two vectors X1 and X2 for each of the above cases. Use these two Xs as
the independent variables, we can generate the corresponding dependent variable Y according to the
true model for each case
Y = 2 × X1 + 1 × X2 +  (1)
where the vector  also has a independent normal distribution (N (0, I)).
1
If you write a Matlab program, you cannot run it for ten times because the same initial value for your random number
generator will result in the same answers. Instead, you need to use loop.

1
 (a) Under Case I, run one OLS regression of Y on a Constant and X1 only, and one OLS regression
of Y on a Constant, X1 , and X2 . We record the coefficient estimates and their standard errors
from each set of regression. Repeat this process for 20 times. Each time, we use the same Xs
but redraw s to generate new Y . Construct a similar table as the one on slide 27 of Lecture
Note 2. You only need to report the first 10 replication and last two rows of summary statistics.
Are your estimates biased, and why?
(b) Under Case II, repeat the same exercise as in (a). Are your estimates biased? Does correlation
between x1 and x2 matter in affecting the biasness of the coefficient estimates, and why? [Hint:
consider model misspecification in Lecture 4.]
(c) Now repeat (b) but run an OLS regression of Y on X1 and X2 only (without the constant
term). Does the coefficient estimates biased, and why? If you are not sure whether the true
model include a constant or not, should you include the constant term in your regression?

3. (Multivariate Regression) In this problem, we study the effect of change in hourly wage (WPH) on
the demand of labor (hours worked, HRS). In particular, we have data on the 38 demographic groups
in data file p1 wage.txt). There are other factors that will affect the demand of labor, which include
number of years of schooling (SCHOOL) and the non-earned income (NEIN). To estimate the supply
equation, you should include a constant term.

(a) Estimate a univariate regression of HRS on WPH. In order to appreciate the beauty of matrix
option, run the regression first use the summation formula found on slide 22 of Lecture Note 1,
and then apply matrix formula for the OLS estimate. What is the impact of a dollar change
in wage on the demand, and does the sign of the coefficient makes sense? Is your estimate
significant?
(b) Now estimate the following multivariate regression model,

HRS = β0 + β1 W P H + β2 SCHOOL + u.

Interpret your coefficient estimates, and test their significance.

HRS = β0 + β1 W P H + β2 SCHOOL + β3 ln(N EIN ) + u.

Are coefficient estimates significant? If you consider N EIN as measuring ”skill”, interpret your
results. Comparing with estimates from (b), why do you see the difference?
(d) What if you delete the 18th observations, and rerun the multivariate regression in (c), what do
you observe when comparing with those of (c), and why?

[Note: there is an endogeneity problem in our estimation, which will be discussed in Lecture 5]

4. (Statistical Inference) [Hint: It’s easy to use Matlab] Given the following information on two sub-
samples with 100 and 120 observations, respectively, (Note, the first X is the constant term)
⎡ ⎤ ⎡ ⎤
120 240 100 60
⎢ ⎥ ⎢ ⎥
X1 X1 = ⎣ 240 550 140 ⎦ , X1 Y1 = ⎣ 150 ⎦ , Y1 Y1 = 1300
 

100 140 800 300

2
 ⎡ ⎤ ⎡ ⎤
100 180 100 50
⎢ ⎥ ⎢ ⎥
X2 X2 = ⎣ 180 600 200 ⎦ , X2 Y2 = ⎣ 250 ⎦ , Y2 Y2 = 1200
 

100 200 700 320

(a) Using OLS formula to compute the coefficient estimates, β̂1 = [β̂1,1 β̂1,2 β̂1,3] , and β̂2 =
[β̂2,1 β̂2,2 β̂2,3] for the two subsamples, respectively
(b) What are the residual variance, the variance-covariance matrix of estimates, and the R2 using
each subsample? [Hint: e e = Y  Y − β̂  X  Y and Ỹ  Ỹ = Y  Y − n(1 Y /n)2 ]
(c) For the first subsample, first test whether the coefficient estimates β1,2 and β1,3 are significant,
if so, at what significance level (10% or 5%)? At a 5% significance level, test the hypothesis of
1) β3 = 0.15 and 2)β2 = 2β3 separately.
(d) For the first subsample test the two hypotheses of 1) β3 = 0.15 and 2)β2 = 2β3 simultaneously
(e) For the second subsample, imposing the constraints of β2 = 2β3 , what is your coefficient esti-
mates? Pleae use the substitution method in Lecture Note 3.
(f) Using the whole sample (that is without break) to estimate the same model coefficients, β̂c =
[β̂c,1 β̂c,2 β̂c,3 ] , and ec ec . Test the structure break using Chow’s test. [hint: for unconstrained,
eu eu = e1 e1 + e2 e2 ]