4
$\begingroup$

If we consider $x_t$ an Ornstein-Uhlenbeck process (with $W_t$ the Wiener process), does anyone know what would be the variance of the convolution of $x_t$ with a given filter $A$ i.e. $V(x_t \star A)$ ?

I've been trying to derive the solution starting from the analytical solution such as

\begin{align} x_t = {} & \mu_x \int_0^\infty A(p-t) (1-e^{-p/T}) \, dp \\ & + \left(\frac{2\sigma} T \right)^{1/2} \int_0^\infty \left(A(p-t) \int_0^p e^{-(p-s)/T} \, dW_s \right) dp \end{align}

and using properties of stochastic integrals and Itô isometry but I can't get the end of it...

So far, my "most successful" unsuccessful developments are the following. Since the integration of a stochastic process is a stochastic process such that

$$\int_0^p e^{-(p-s)/T} \, dW_s = X_p = \mathcal{N} \left(0,\int_0^p e^{-2(p-s)/T} \, ds \right) = \mathcal{N}\left(0,\frac{T}{2}(1-e^{-2p/T}) \right)$$

and that one can consider (with $W_p$ a Wiener process)

$$X_p = \left(\frac{\frac{T}{2}(1-e^{-2p/T})} p \right)^{1/2} \, W_p $$

then

\begin{align} x_t = {} & \mu_x \int_0^\infty A(p-t) (1-e^{-p/T}) \, dp \\ & + \left( \frac{2\sigma}{T} \right)^{1/2} \int_0^\infty A(p-t) \left(\frac{\frac{T}{2}(1-e^{-2p/T})} p \right)^{1/2} W_p \, dp \end{align}

and given that

$$ \int_0^a g'(p) W_p \, dp \sim \mathcal{N}\left(0, \int_0^a \left(g(s)-g(p)\right)^2 \, ds\right)$$

one therefore gets with $a=\infty$

$$\sigma_x^2 = \sigma^2 \int_0^\infty \left( \int_p^\infty A(u-t) \left(\frac{1-e^{-2u/T}}{u}\right)^{1/2} \, du \right)^2 \, dp$$

But I can't go anywhere from the last equation. Even taking $A$ as a dirac, it's not easy...any help would be greatly appreciated !

Thanks!

$\endgroup$

1 Answer 1

2
$\begingroup$

Welcome to MO! This is not quite a research level question, but here is the answer anyway.

Represent the Orenstein-Uhlenbeck process as white noise passing through a low-pass (this is really the representation in the equation for $x_t$ you wrote). Call the transfer function of the low-pass $H$. Then your process is white noise passing through the filter $AH$. Now the variance you ask about is simply the square of the $L^2$ norm of $AH$, i.e. $\int |AH(\omega)|^2 d\omega$.

$\endgroup$
2
  • $\begingroup$ Thank you for your anwser ! And sorry, I couldn't get any anwser on mathstackexchange so I thought my question was maybe a complicated one...although it was not. Indeed, going spectral makes the reasoning much easier...Although I don't find it to be straight forward to derive the filter H...I'll think of it. Thanks ! $\endgroup$
    – Romain
    May 8, 2018 at 11:30
  • $\begingroup$ So H can be calculated using the link between the autocorrelation of $x_t$ and the spectral density. Here an even more general derivation math.stackexchange.com/questions/1647793/… $\endgroup$
    – Romain
    May 8, 2018 at 13:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.