STAD70 Statistics & Finance II: Assignment 2
STAD70 Statistics & Finance II: Assignment 2
Assignment 2
Due date: March 9, 2021 (by 11:59pm). Late submissions will be penalized.
1. (Open ended question) On Feb 3, 2022, the stock price of Meta Platforms Inc (symbol FB)
dropped by more than 25%.
(a) Given reason(s) to explain the price drop. (no less than 150 words)
(b) Discuss the event in relation to the efficient market hypothesis. Does it support or
oppose the EMH? (no less than 200 words)
In both parts, cite all references (news articles, research papers, price charts, etc) that you
use.
rt = φrt−1 + εt ,
where |φ| < 1 and (εt ) is an i.i.d. white noise process. Show that the theoretical
variace ratio is equal to
φ 2 φ(1 − φq )
VR(q) = 1 + 2 − , q = 1, 2, . . . .
1 − φ q (1 − φ)2
(b) In Theorem 5.5 of the lecture notes we cited a result about the variance ratio:
√ suppose
rt are i.i.d. (and satisfy suitable technical conditions). Then, as T → ∞, T (VR(q)−
d
1) converges in distribution to N (0, 2(q−1)). [In the earlier version, the p there should
be a q, where q is the lag.] In this problem, we consider a simulation study to observe
this convergence.
i.i.d. i.i.d.
Consider two cases: (i) rt ∼ N (0, 1) and (ii) rt ∼ Laplace(0, 1). Here we recall
that the Laplace distribution Laplace(µ, b) has density
1 |x − µ|
f (x) = exp − , x ∈ R.
2b b
Consider q = 1, 2, 3, 4, 5 and T = 10, 100, 1000, 10000. For each case and each q-T
pair, simulate N = 5000 batches of data
√ points. The N batches provide an empirical
distribution for the distribution of T (VR(q)
d − 1). Plot a density estimate and
compare with that of the asymptotic distribution. Your final output should contain
25 = 5 × 5 graphs (arrange them in a matrix of graphs). Comment on the results.
3. (Quadratic programming)
1
(a) Prove Proposition 6.10 of the lecture notes (shape of the unconstrained efficient fron-
tier).
(b) A quadratic programming problem is an optimization problem of the form
1
min b> Db − d> b subject to A> b ≥ b0 .
2
b∈Rn
Consider daily (arithmetic) return of these stocks from 2017-01-01 to 2022-1-31. We use
the data from 2017-01-01 to 2019-12-31 for fitting, and the data from 2020-01-01-2022-
01-31 for testing. For the purposes of this problem, we ignore the time dependency of
the returns. In this problem, we are interested in finding the global minimum variance
portfolio.
(a) Randomly assign the fitting data into two parts. The first part (70%) will be used
as training data, and the rest (30%) will be use for validation. When missing data
exists, use the maximal subset of dates such that the return exists for all stocks.
2
(b) For α ∈ {0, 0.05, 0.1, . . . , 0.95, 1}, consider the skrinkage estimator
Σ̃ = (1 − α)Σ̂ + (1 − α)D̂,
where Σ̂ is the sample covariance matrix, and D̂ is the diagonal matrix of Σ̂.
For each α, estimate Σ̃ using the training data. Compute the global minimum variance
portfolio under the no short selling constraint (see Problem 3).
(c) Using the validation data, pick the α∗ which minimizes the standard deviation of the
portfolio.
(d) Using the α∗ from (c) and the corresponding portfolio, compute the realized standard
deviation of the returns in the testing period. Comment on the performance of your
estimator. [This last part is non-trivial and a good answer can grant you extra credits
(at most 5%, and your total for this assignment cannot exceed 100%).]