MTH5540 Tutorial 6 Problem Set
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
Tutorial 6 Problem Set
1. Consider that we have the following state transition model and observation model:
Xk+1 = Xk(2) + Vk ,
Yk = Xk + Uk .
where Uk ∼ N(0, 1) and Vk follows the exponential distribution with pdf λ exp(−λx). Derive the pdf of the state transition probability f (xk+1|xk) and the likelihood function g(yk+1|xk+1).
2. At time tk, let Yk = {y1 , . . . , yk} denote all the observations made at and before tk . Suppose we have the posterior distribution of the state Xk as p(xk|Yk). Prove that the predictive distribution takes the form of
∫
p(xk+1|Yk) = f (xk+1|xk)p(xk|Yk)dxk ,
and the posterior given the new data yk+1 has the form of
p(xk+1|Yk+1) = g(yk+1|xk+1)p(xk+1|Yk).
3. Consider the stochastic volatility model
Xn = αXn − 1 + σVn ,
Yn = β exp(Xn/2)Wn ,
where X1 ∼ N(x; 0, ), and Vn and Wn are iid N(0, 1).
(a) Simulate (Xn, Yn) for n = 1, . . . , 1000, with α = 0.91, σ = 1.0 and β = 0.5.
(b) Use the bootstrap filter to generate N = 10000 estimates of X1 , . . . , X1000 given the observations
Y1 , . . . , Y1000 . Plot the filter mean ±2 SD on the same axes as the true hidden states X1 , . . . , X1000 .
4. Consider we have a Gaussian random variable X ∼ N(0, 1), a Gaussian random variable Y that has the law depending on X in the form of X | Y ∼ N(X, a), and another Gaussian random variable Z that has the law depending on X in the form of Z | X ∼ N(−X, a), for some constant a > 0. First derive the joint density of (X, Y, Z), and then derive the marginal density of (Y, Z). Are Y and Z independent? Are Y and Z conditionally independent given X? Justify your answer. You can also generate random variables (X, Y, Z) to verify your answer.
2023-06-01