Title | Gaussian Distribution Introductory Level-quiz |
---|---|
Author | Srijan Chaudhury |
Course | Stochastic Processes |
Institution | Национальный исследовательский университет Высшая школа экономики |
Pages | 5 |
File Size | 103.9 KB |
File Type | |
Total Downloads | 60 |
Total Views | 126 |
Gaussian Distribution assignment.Questions to test understanding of Normal Distibutions.Intermediate level maths knowledge required...
Quiz-4 answers and solutions Coursera. Stochastic Processes December 30, 2020 1. Choose the functions K (t, s), which can be covariance functions of some stochastic processes defined for t ∈ [0, 1]. Options: a) K (t, s) = 1 − max{t, s} c) K (t, s) = 1 + (t − s)3
b) K (t, s) = max{t, s} − min{t, s} d) c > 0
Answer: a) K (t, s) = 1 − max{t, s}, d) c > 0 Solution: a) Clearly, this function is symmetric, since K (t, s) = 1 − max{t, s} = 1 − max{s, t} = K (s, t). Let us show that it is also positive semi-definite. Indeed, define gt (x) := I{x ∈ [t, 1]}. Then Z∞ 1 − max{t, s} = gs (x)gt (x) dx, 0
from which n n X X
uj uk K (tj , tk ) =
n n X X
uj uk
j=1 k=1
j=1 k=1
=
Z∞X n X n 0
Z∞
gtj (x)gtk (x) dx
0
uj uk gtj (x)gtk (x) dx
j=1 k=1
! Z∞ X n n X = uj gtj (x) uj gtk (x) dx j=1
0
k=1
2 Z∞ X n = uj gtj (x) dx ≥ 0. j=1
0
Thus, K (t, s) = 1−max{t, s} can be a covariance function of some process defined for t ∈ [0, 1].
b) This function cannot be a covariance function of any process. Assume the converse: let there exist some process Xt such that its covariance function K (t, s) = max{t, s} − min{t, s}. Then Var Xt = K (t, t) = max{t, t} − min{t, t} = t − t = 0, 1
i.e., Xt = const for all t ∈ [0, 1]. However, cov(Xt , Xs ) = max{t, s} − min{t, s} = 6 const
∀s 6= t,
(s, t) ∈ [0, 1]2 ,
which leads to the contradiction. c) Since
K (t, s) = 1 + (t − s)3 6= 1 + (s − t)3 = K (s, t),
the symmetry property is violated, and this function cannot be a covariance function of any process. d) It can be seen that K (t, s) = c = K (s, t) and n n X X
uj uk K (tj , tk ) =
n n X X j=1 k=1
j=1 k=1
uj uk c = c
n X j=1
2
uj ≥ 0,
meaning that this function is symmetric and positive semi-definite. Thus, it can be a covariance function of some process. A corresponding example is given by Xt = ξ, where ξ is a random variable such that Var ξ = 1. 2. Which of the following processes are Gaussian? (In the answers below Wt is a Brownian motion). Options: a) Xt = Wt2 b) Xt = aξt−1 + ξt , c) Xt = W1−t − W1 , d) Xt = Wt + ηξ,
ξi ∼ i.i.d. N (0, 1) ∀i = 0, 1, . . . , t,
t = 1, 2, . . .
t ∈ [0, 1]
( −1, 1/2 ξ ∼ N (0, 1), η = , 1, 1/2
Wt , ξ, η are indepen-
dent for all t ≥ 0 Answer: b), c), d) Options: The process Xt is Gaussian if and only if for any t1 , . . . , tn : n P λk Xtk ∼ N , ∀~λ. (Xt1 , . . . , Xtn ) is a Gaussian vector, that is, k=1
a) Since P{W t2 < 0} = 0, Wt2 is not normally distributed, and this process is not Gaussian. b) A finite dimensional distribution for this process has the form (Xt1 , Xt2 , . . . , Xtn ) =
(aξt1 −1 + ξt1 , aξt2 −1 + ξt2 , . . . , aξtn −1 + ξtn ).
Clearly, any linear combination of finite dimensional distributions of Xt can be represented as a sum of i.i.d. normally distributed random variables: for instance, if ti are such that ti − 1 = ti−1 ∀i = 1, 2, . . . , n n X
λk Xtk
=
λ1 (aξt1 −1 + ξt1 ) + λ2 (aξt2 −1 + ξt2 ) + · · · + λn (aξtn −1 + ξtn )
=
λ1 (aξt1 −1 + ξt1 ) + λ2 (aξt1 + ξt2 ) + · · · + λn (aξtn−1 + ξtn )
k=1
=
λ1 aξt1 −1 + (λ1 + aλ2 )ξt1 + (λ2 + aλ3 )ξt1 + · · · + λn ξtn ∼ N . 2
c) This process is Gaussian as n X
λk Xtk
=
k=1
=
λ1 W1−t1 − λ1 W1 + · · · + λn W1−tn − λn W1 n X k=1
λk W1−tk − W1
n X k=1
λk ∼ N
as a linear combination of the components of the finite dimensional distribution (W1−t1 , W1−t2 , . . . W1−tn , W1 ) of a Brownian motion which is a Gaussian process. d) Let us show that ηξ ∼ N (0, 1). Indeed, P{ηξ ≤ x}
P{−ξ ≤ x|η = −1}P{η = −1} + P{ξ ≤ x|η = 1}P{η = 1} 1 1 (1 − P{ξ > x} + P{ξ ≤ x}) = (1 − 1 + P{ξ ≤ x} + P{ξ ≤ x}) = 2 2 = P{ξ ≤ x}.
=
Thus n X
λk Xtk =
k=1
n X
λk (Wtk + ηξ) =
k=1
n X k=1
˜λk (Wt − Wt ) + ηξ k k−1
n X k=1
λk ∼ N
as a sum of i.i.d. normally distributed random variables. 3. Which of the following processes is a Brownian motion? (In the answers below Wt is a Brownian motion). Options: a) Xt = Wt2 b) Xt = aξt−1 + ξt , c) Xt = W1−t − W1 , d) Xt = Wt + ηξ, dent for all t ≥ 0
ξi ∼ i.i.d. N (0, 1) ∀i = 0, 1, . . . , t,
t = 1, 2, . . .
t ∈ [0, 1]
( −1, 1/2 ξ ∼ N (0, 1), η = , 1, 1/2
Answer: c) Xt = W1−t − W1 ,
Wt , ξ, η are indepen-
t ∈ [0, 1]
Solution: Since Brownian motion is a Gaussian process Wt such that EWt = 0 and cov(Wt , Ws ) = min{t, s}, we should only check the last two properties for the processes b)-d). For b) E[Xt ] = E[aξt−1 + ξt ] = aE[ξt−1 ] + E[ξt ] = 0, but cov(Xt , Xs ) = cov(aξt−1 +ξt , aξs−1 +ξs ) = (1+a2 )I{t = s}+aI{|t−s| = 1}, thus, it is not a Brownian motion. For c) E[W1−t − W1 ] = E[W1−t ] − E[W1 ] = 0 3
and cov(W1−t − W1 , W1−s − W1 ) =
cov(W1−t , W1−s ) − cov(W1 , W1−t )
− cov(W1 , W1−s ) + cov(W1 , W1 )
min{1 − t, 1 − s} − (1 − t) − (1 − s) + 1
=
t + s − max{t, s} = min{t, s}.
= Therefore, Xt is Brownian motion. For d)
E[Wt + ηξ] = E[Wt ] + E[η]E [ξ] = 0, but cov(Wt + ηξ , Ws + ηξ)
=
cov(Wt , Ws ) + cov(Wt , ηξ) + cov(Ws , ηξ ) + cov(ηξ, ηξ )
=
min{t, s} + 0 + 0 + E[(ηξ)2 ] − (E[ηξ ])2
ηξ∼N (0,1)
=
min{t, s} + 1 − 0
min{t, s} + 1 6= min{t, s}.
= So, Xt is not a Brownian motion.
4. Find the covariance function of the process Xt = Wt − tW1 , where Wt is a Brownian motion.
t ∈ [0, 1],
Answer: min{t, s} − ts
Solution:
cov(Xt , Xs ) = = =
cov(Wt − tW1 , Ws − sW1 )
cov(Wt , Ws ) − t cov(W1 , Ws ) − s cov(Wt , W1 ) + ts cov(W1 , W1 )
min{t, s} − ts − st + ts = min{t, s} − ts.
5. Find the covariance function of the process Xt = ξWt , independent of Wt ∀t, t ≥ 0.
ξ ∼ N (1, 1) is
Answer: 2 min{t, s}.
Solution:
cov(ξWt , ξWs )
= = t>s
=
= =
E[ξ 2 Wt Ws ] − E[ξWt ]E[ξWs ]
Eξ 2 E[Wt Ws ] − (E[ξ])2 EWt EWs
(1 + 12 )E[(Wt − Ws + Ws )Ws ] − 02 · 0 · 0
2(E[(Wt − Ws )(Ws − W0 )] + E[Ws2 ]) 2s = 2 min{t, s}.
6. Harry is playing some game. At each given time moment t the number of points he gets is determined by the value of the process Xt = σWt , that is, he earns points if Xt > 0 and loses if Xt < 0. Every hour (at time moments t = 1, 2, . . .) the results are automatically recorded, and if the sum of 3 results in a row is greater than 10, Harry gets a special prize. Find the probability that Harry gets this prize after 3 hours of playing (in 4
the answers below Φ is the distribution function of the standard normal distribution). 10 Answer: 1 − Φ √ σ 14
Solution: We need to find the probability that P{X1 + X2 + X3 ≥ 10}
= =
P{σ(W1 + W2 + W3 ) ≥ 10}
1 − P{W1 + W2 + W3 ≤ 10/σ }.
As E[W1 + W2 + W3 ] = E[W1 ] + E[W2 ] + E[W3 ] = 0 and Var(W1 + W2 + W3 ) =
Var W1 + Var W2 + Var W3 + 2 cov(W1 , W2 ) +2 cov(W2 , W3 ) + 2 cov(W1 , W3 )
=
1 + 2 + 3 + 1 · 2 + 2 · 2 + 1 · 2 = 14,
1 − P{W1 + W2 + W3 ≤ 10/σ} = 1 − Φ
10 √ σ 14
.
7. Choose the correct statements about the process Xt = σWt , σ > 0: Options: a) cov(Xt , Xs ) = min{t, s }
b) the Kolmogorov continuity theorem E[|Xt − Xs |α ] ≤ c|t − s|1+β holds with c = 3, α = 4, β = 1 c) Xt is not stationary d) Xt is a Gaussian process Answer: Xt is not stationary and is a Gaussian process. Solution: Since Xt is a Brownian motion multiplied by a positive constant, it is also a Gaussian process. Now, cov(Xt , Xs ) = σ 2 cov(Wt , Ws ) = σ 2 min{t, s} and since the covariance function depends not on the difference, but on the time moments t and s, Xt is not stationary. Finally, E[|Xt − Xs |α ] = E[|σ(Wt − Ws )|α ] = σ α E[|Wt − Ws |α ], therefore, the condition of the Kolmogorov continuity theorem E[|Xt − Xs |α ] ≤ c|t − s|1+β holds with α = 4, β = 1 and c = 3σ 4 .
5...