I'd like a check in this exercise, taken from Beichelt's book:

Let $X(t) = sin(\Phi t)$, where $\Phi$ is uniformly distributed over the interval $[0,2\pi]$.


(1) The discrete-time stochastic process $\{X(t), t= 1,2, \dots\}$ is weakly but not strongly stationary.

(2) The continuous-time $\{X(t), t \in (-\infty,+\infty)\}$ is neither weakly nor strongly stationary.


First, I need to check that $X(t) \in L^2(\Omega)$, with $\Omega=[0,2\pi]$, and this is shown by the computation of $\int_{0}^{2\pi} sin^2(\Phi t)d\Phi= \pi$.

$\mathbb{E}[X(t)]= \frac{1}{2\pi}\int_{0}^{2 \pi}sin(\Phi t) d \Phi=\frac{\cos(2\pi t) - 1}{2\pi}$. If $T \subset N$, then $\mathbb{E}[X(t)]=0 \quad \forall t$.

Conversely, if $T \in \mathbb{R}$, then the trend function is not constant and thus the process is not stationary (dns this answers secondo question).

So from now I suppose $T \subset \mathbb{N}$.

$Cov(X(t),X(s))=\mathbb{E}[\sin(\Phi t) \sin(\Phi s)] = \frac{1}{2}\cdot \mathbb{E}[\cos(\Phi(t-s)) + \cos(\Phi(t+s))]$

The computation of this integral gives $\frac{1}{4\pi} [\frac{\sin(2\pi(s-t))}{s-t} + \frac{\sin(2\pi(s+t))}{s+t}]$. Since $t,s \in \mathbb{N}$, then $s+t $ and $ s-t \in \mathbb{N}$ and the sum summand is identically zero, i. e., $Cov(X(t),X(s))=0$.

In order to show that $X(t)$ is not strongly stationary I need to check if $F_t(x)=F(x), t \in T$. So, for $x \in [0,1]$:

$F_t(x)=\mathbb{P}(sin(\Phi t)\leq x)=\mathbb{P}(\Phi \leq \frac{arcsin(x)}{t}) = \int_{0}^{\frac{\arcsin(x)}{t}} \frac{1}{2\pi} d \Phi=\frac{\arcsin(x)}{2\pi t }$.

Since $F_t(x)$ depends on $t$, this process is weakly but non strong stationary.

But I have a big doubt: is it correct to state that the covariance function depends on the time splitting $\tau=t-s$, in order to state that the process is weakly stationary?

I hope to have explained myself well


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.