Researcher in Computational Finance

Maths Test Questions

You have one-and-a-half hours to submit answers to two questions only. Marks for each
section are provided in bold, square brackets on the right hand side. Please show all working and
start each question on a separate sheet of paper.
Question 1 Suppose X, Y are zero mean random variables satisfying yt = xTt β + εt , for in-
dependent and normally-distributed random variables {εt } with variance σ 2 . Let y be an N × 1
vector of observations of Y and x an N × n matrix of observations of X.
1. Find β̂, the value of β which minimises the variance of the {εt }. [2]
2. Show that the estimator β̂ is unbiased. [1]
3. Find the covariance matrix of β̂. [2]
4. Find the likelihood function L of unknown parameters given observations y and x. The
likelihood function is the conditional probability density function of the data given the
unknown parameters, and is treated as a function of the parameters. [2]
5. Estimate the parameters β and σ from the observations by maximising the likelihood func-
tion. [3]

2
Question 2 Let Xt be a stochastic process defined as

Xt = Xt−1 + σUt ,

where Ut ∼ N (0, 1) are i.i.d. With regular sample period ∆ ≥ 1, we make observations Zk of Xk∆ ,
i.e. Z0 = X0 , Z1 = X∆ ,..., ZN = XN ∆ .

1. Find an estimator σ̂ 2 of σ 2 using all the observations Zk for a given ∆ assuming they are
noiseless. [2]
2. Find the mean and variance of the estimator. [2]
3. What is the optimal sample period for the estimator? [2]

Now assume observations are corrupted by Gaussian noise Vk such that Zk = Xk∆ + aVk . We can
assume Vk ∼ N (0, 1) are i.i.d. and independent of each of the Ui .

4. What is the mean and variance of your estimator? [2]

 13
 r ! 13 r ! 31 
2a4 T 2a4 2a4

∆∗ =  1+ 1− 4 2 + 1− 1− 4 2  (2.1)
σ4 σ T σ T

3
Question 3 Let X, Y be independent random variables and f, g : R → R.

1. Given that fX,Y (x, y) = fX (x)fY (y), show that E(f (X)g(Y )) = E(f (X))E(g(Y )) for con-
tinuous random variables X, Y . [1]
2. Show that independent random variables are uncorrelated. [1]
Let {Xi } be n independent identically distributed random variables with mean µ and finite variance
def
σ 2 . Let X̄n = n1
P
Xi .
3. Show that for any i

Cor Xi − X̄n , X̄n = 0. [2]
4. Let X, Y be independent random variables taking values 0 and 1 with probability 1/2. Let
U = X+Y2 , V = X − U . Find the complete joint probability distribution pU,V . [2]
5. Use this to deduce that uncorrelated random variables are not necessarily independent. [1]
6. For  > 0, show that
lim P (|X̄n − µ| ≥ ) = 0 .
n→∞

You may find it useful to know that for any non-decreasing, non-negative function g
E(g(X))
P (X ≥ t) ≤ .
g(t)
[3]

4
Question 4 Consider three univariate time-series xt , yt and zt in discrete time t = 1, 2, ....
Denote the column-vectors of observations as x = (x1 , ..., xT ), y = (y1 , ..., yT ) and z = (z1 , ..., zT ).
Let ρxy = corr(xt , yt ), ρxz = corr(xt , zt ) and ρyz = corr(yt , zt ) be the pairwise contemporaneous
correlations between the pair of time-series for all t. Suppose ρxy = ρxz = 0.5.

1. Find all possible values of the correlation between y and z, ρyz . [2]
2. What is the correlation matrix C of x, y, z with the smallest nonnegative correlation ρyz ?
For this case, compute the largest eigenvalue of C and the corresponding eigenvector, p.[2]
3. Let q be the projection of the observation matrix A = (x, y, z) onto the largest eigenvector
p. Derive an expression for q. [2]
4. Find the variance of q and explain what it represents. [2]
5. Suppose xt , yt and zt are financial time-series. Provide an example in which the answers in
3. and 4. might be used. In which circumstances is the eigenvector of the correlation matrix
more appropriate than that of the covariance matrix? [2]