Università della Svizzera italiana Faculty of Economics
Prof. Loriano Mancini & Mr. Hao Ma Fall Semester
FINANCIAL ECONOMETRICS – Final Exam
22 January 2021
Instructions: For Exercise 1, you can write your answers in a Word file (or equivalent) or you
can write by hand on a sheet of paper and then take a picture. If you write by hand, please write
as clearly as possible . If we cannot read your handwriting, we cannot grade the exercise. For
Exercise 2, please copy and paste your code and code output in the Word file. Provide comments
about the code and code output in the Word file, explaining the meaning of the code and result
output. Please upload one single PDF. You can convert the Word file in a PDF. If you wrote
the answers by hand, you can for example copy and paste the picture in the Word file and then
convert it to a PDF.
Exercise 1: Theory (50 Points)
Consider the linear model
yi = α + x0i β + εi , for i = 1, . . . , n (1)
xi,1 β1
where yi , α, εi are scalars, xi =
xi,2 , β = β2 .
xi,3 β3
Let’s assume the error terms are exogenous to the independent variables
E[ε|x1 , . . . , xn ] = 0 (2)
1. Let’s write the model in Equation (1) in matrix form
y = Xγ + ε
α
where γ = . Give y, X, ε and their dimensions.
β
2. Let’s define the column vector
x1,k
.
xk = .
.
xn,k
Show that
E[ε] = 0, (3)
0
E[xk ε] = 0, k = 1, 2, 3 (4)
0
where xk is the transpose of xk .
Hint: Use the Law of Iterated Expectations.
3. In Statistics, an estimator γ̂ of a parameter γ is said to be unbiased if E[γ̂] = γ. Prove that
the OLS estimator is unbiased.
Hint: Use the assumption in Equation (2) and the Law of Iterated Expectations.
4. Use Equations (3) and (4), and the Law of Large Numbers to show that
1 0
lim Xε=0
n→∞ n
5. Consider the null hypothesis: β1 = β2 and β2 = β3 . Write the null hypothesis in the form
Rγ = q. Give explicitly form and dimension of R and q. Explain what needs to be estimated
to test the null hypothesis Rγ = q.
6. Suppose that β3 = 0 in Equation (1). That is, the data generating process is
yi = α + xi,1 β1 + xi,2 β2 + εi , for i = 1, . . . , n.
The econometrician instead estimates the model in Equation (1) which includes xi,3 β3 . Is
the OLS estimator of α, β1 , β2 and β3 biased or unbiased? Explain your answer in words
(no need to provide a formal proof).
Exercise 2: Coding (50 Points)
Consider the univariate linear model
yi = β0 + β1 xi + εi , for i = 1, . . . , n (5)
where yi , β0 , β1 , xi and εi are all scalars.
1. Generate your sample data
The data generating process is as follows: xi follows a normal distribution with mean 10
and standard deviation 5; yi is a linear function of xi as in Equation (5) with β0 = 30 and
β1 = 3; the error term εi follows a uniform distribution between −10 and 10.
(a) Use the data generating process above to generate n = 30 observations of data pairs
(yi , xi ).
(b) Store your observations of yi , xi and εi into three n × 1 vectors, and name them as y,
x and ε (for the latter, please use the full name epsilon). Report mean, median and
standard deviation of each variable.
2. Run OLS regression
Let’s write the linear model in Equation (5) in matrix form
y = Xβ + ε (6)
where X = [1 x] is an n × 2 matrix (1 represents a vector of ones), β = [β0 β1 ]0 is a 2 × 1
vector, y, x and ε are all n × 1 vectors.
(a) Construct the matrix X, then obtain your estimated parameter b = [b0 b1 ]0 by using
the OLS estimator b = (X 0 X)−1 X 0 y. Report the estimates.
(b) Plot in the same graph: 1) the values of yi versus xi as scattered points, 2) the estimated
linear model ŷi = b0 + b1 xi , for i = 1, . . . , n, as a straight line with xi on the horizontal
axis, and 3) the true conditional expectation of yi given xi , i.e., yi = β0 + β1 xi , for
i = 1, . . . , n, as a straight line with xi on the horizontal axis.
Provide some comments about the above graph (accuracy of the estimates, distribution
of scattered points, etc.).
(c) Estimate the variance of the error term εi using the unbiased estimator s2 = e0 e/(n−K).
Recall, e = Y − Xb is the vector of residuals, n is the size of the simulated sample and
K is the number of estimated parameters.
Compare the estimated variance of εi (namely s2 ) and the sample variance of the
simulated εi . Are these two variances expected to be “close” to each other? Is it
relevant that εi follows a uniform distribution, not a normal distribution?
(d) Estimate the standard error of the two estimated parameters, and report their values.
Recall that the (conditional) variance of the OLS estimator is V [b|X] = σ 2 (X 0 X)−1 ,
where σ 2 = V [εi ]. Therefore, an unbiased estimator of V [b] is s2 (X 0 X)−1 .
For each parameter, compute the ratio of the estimated standard error over the esti-
mated parameter value, e.g., (estimated standard error of b0 )/b0 . Which parameter is
estimated more accurately (i.e., the ratio is lower)?
