STAT0011 - Exam 2021
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
Question 1 [14 marks]
Suppose that a parameter θ can take only three values, θ = 0, θ = 1 or θ = 2. Your prior on θ before observing any data is p(θ = 0) = 0.3, p(θ = 1) = 0.5 and p(θ = 2) = 0.2. The distribution of Y is as follows:
p(Y lθ = 0) = Gamma(2, 3)
p(Y lθ = 1) = Gamma(1, 2)
p(Y lθ = 2) = Gamma(3, 3)
Your task is to estimate the value of θ . Let action a0 corresponds to claiming that θ = 0, a1 corresponds to claiming that θ = 1, and a2 corresponds to claiming that θ = 2. The losses corresponding to each action and values of θ are represented by the following loss matrix :
θ = 0 θ = 1 θ = 2
a0
a1
a2
For example, the loss associated with action a0 when the true value is θ = 1 is L(θ = 1, a0 ) = 1. You obtain one observation Y = 2.
(a) Compute the posterior distribution for θ . [7]
(b) Compute the Bayesian expected loss associated with all three actions given this observa-
tion, and decide which action to take. [7]
Question 2 [24 marks]
Consider the following historical record for the daily log-returns of a financial stock: Y1 = 0.4, Y2 = _0.61, Y3 = _0.53, Y4 = 0.22, Y5 = _0.018, Y6 = _0.39
The log-returns are assumed to be independent and identically distributed draws from a Normal distribution with known mean 0 and unknown variance σ 2 .
(a) Using the Inverse-Gamma(α, β) distribution as a conjugate prior for σ 2 , derive its
posterior distribution given the data. You do not need to evaluate any integrals or normalising constants. [5]
(b) Let denote the log-return on a particular day in the future. Show that the posterior
predictive distribution is of the following form:
1 β˜ Γ( + )
^2π Γ() (β˜ + )+
for some values of and β˜ . Using the numeric values of Y1 , . . . , Y6 provided above, give expressions for and β˜ in terms of α and β . State the range of possible values
that can take. [7]
(c) [Type] State the Pickands_Balkema_de Haan theorem. Explain using your own words how this theorem can be used in risk analysis. Use no more than 100 words. [4]
(d) Find the probability of log-return, Y7 , on the next day being between _0.75 and _0.7 using a Generalised Pareto Distribution (GPD) with threshold u = _0.38. Use the method of moments to estimate the GPD parameters. [8]
Question 3 [21 marks]
A local meteorological office located on an island in the Pacific Ocean would like to model the number of hurricanes that occur in a year. Let Xt be a discrete random variable which represents the number of hurricanes that occurred in year t. The company considers a Poisson model to estimate the frequency of hurricanes, i.e. X Poisson(λ). The following historical record shows the number of hurricanes over the past 5 years:
Year, t Number of hurricanes
1
2
3
2
2
The meteorological office suspects that during this period there might have been a struc- tural change in the hurricane frequency due to climate change. In particular, the meteoro- logical office believes that the change point has occurred in year t = 3. Thus, observations X4 and X5 still come from a Poisson distribution but with a different parameter λ, i.e.:
Xt ~
Define the two models:
M0 : There has been no change point in the hurricane frequency.
M1 : There has been a change point at t = 3.
(a) Compute the marginal likelihood for Model M0 without a change point. Note that
you should use the Gamma(1, 1) prior in Model M0 . [6]
(b) Compute the marginal likelihood for Model M1 with a single change point in year
t = 3. Note that you should use the Gamma(1, 1) prior for both segments in Model M1 . [11]
(c) Compute the posterior distribution for both models, and decide whether there has been a structural change in the hurricane frequency in year t = 3. Both models M0 and M1 are equally likely a priori. [4]
Question 4 [6 marks]
Let Yt = ln(Pt ) _ ln(Pt一1 ) denote the daily log-return on a financial asset, where Pt is the opening daily price at time t. The following model has been proposed to describe the behaviour of log-returns:
Yt = ut
ut = σt εt
ε N(0, 1)
σt(2) = ω + αut(2)一1 + βσt(2)一1
where ω , α , β > 0 and α + β < 1.
(a) Derive the unconditional variance of Yt , Var(Yt ). Clearly show all the steps leading to your answer. [6]
Question 5 [16 marks]
(a) Suppose that you are faced with taking one of the following three actions: a1 , a2 , or
a3 . However, there is a loss associated with each action which depends on the state of nature. The losses corresponding to each action ai , i = 1, 2, 3, and the state of nature θj , j = 1, 2, 3, are represented by the following loss matrix :
|
1 |
2 |
3 |
1 |
_10 |
_08 |
_ 8 |
2 |
_3 |
_3 |
_2 |
3 |
_6 |
_5 |
_6 |
For example, if you take action a1 , and the state of nature is θ 1 , then the incurred loss is 10.
(i) State for each action a1 , a2 and a3 , whether it is admissible or inadmissible. [3]
(ii) Find the minimax nonrandomized action. Clearly show all the steps leading to your answer. [5]
(b) Let Y Poisson(λ) where λ > 0 is unknown. It is of interest to estimate λ using
the following loss function:
L(λ,) = log ╱ 、 + ←λλ + log ╱ 、
Assuming that no data are available, compute the Bayesian point estimate of λ under the loss function L(λ,). [8]
Question 6 [19 marks]
(a) Consider the following information on the hypothetical portfolio of 入7,000 in-
vested in two assets. The information on each daily asset return is provided in the table below. It is assumed that these returns are jointly normally distributed.
Asset 1 Asset 2
Mean Standard deviation Portfolio weights |
0.008 0.3 0 2 |
0.02 0.1 0 8 |
|
Portfolio value Correlation coefficient |
入7,000 0.2 |
Standard normal distribution table.
巫 -2.326 -2.054 -1.881 -1.751 -1.645 -1.555
Φ(巫) 0.01 0.02 0.03 0.04 0.05 0.06
Compute the 99% 1-day Value-at-Risk (VaR) of the portfolio in value terms. Interpret your findings. [5]
(b) [Type] Suppose that you have been hired as a risk analyst by a financial com-
pany to assess its current internal approach to computing the 99% 10-day Value- at-Risk (VaR) for a particular financial asset. Let X芒 denote the daily log-return on that asset which evolves over time as follows:
X芒 = µ芒 + u芒
where
P
µ芒 = c + δiX芒一i
i=1
u芒 = σ芒 ε芒 ε芒 i. . N (0, 1)
s o Q
σ芒(2) = ω + α5u芒(2)一5 + γou芒(2)一o I[at −o<0] + βg σ芒(2)一g
5=1 o=1 g=1
where P = 3, S = 2, O = 1, and Q = 2. After careful inspection of internal models, you learn that the company computes the 99% 10-day VaR using the following formula:
VaRa,h = 2.326 . . ^h _ . h
where α = 0.99, h = 10, and are the sample mean and the sample standard
t 1
一& γ
where F (x) is the distribution function of a random variable X denoting log- returns, δ > 0 is the order of the moment of F (x), and t is some target value. Show that, for certain values of t, γ and δ, this modified risk measure corresponds to the Expected Shortfall at the 99% confidence level. [8]
2023-04-27