Say I have $X$ that follows an Ornstein-Uhlenbeck process:

$$ dX_t = \phi (\mu - x_t) d_t + \sigma d W_t $$

Let $Y_t = \exp(X_t)$. Is there anything that helps me compute $\lim_{t\to\infty}E[Y_t^\gamma]$, $\gamma \neq 1$?

Here is my approach:

We know that the stationary solution for $X_t$ is Gaussian with mean $\mu$ and variance $\sigma^2/(2\phi)$. Hence, the expected value of $\lim_{t\to\infty} Y_t$ is that of the log-normal distribution,

$$\lim_{t\to\infty} E[Y_t] = \exp\left(\mu + \frac{\sigma^2}{2\phi}\right)$$


$$Y_t^\gamma = \exp(\gamma X_t)$$

I would intuitively guess that the process $\gamma X_t$ follows

$$ d\gamma X_t = \phi \gamma (\mu - x_t) d_t + \gamma \sigma dW_t$$

and therefore compute

$$\lim_{t\to\infty} E[Y_t^\gamma] = \exp\left(\mu + \frac{\gamma^2\sigma^2}{2\gamma\phi}\right)$$

But clearly, I'm out of my reach here.

  • $\begingroup$ If my answer bellow is satisfactory, please accept it so that others asking the same question can find the answer more quickly. Thanks! $\endgroup$
    – byouness
    Commented Jun 5, 2018 at 16:09

1 Answer 1


It can be shown that (by using Itô's lemma on $X_t e^{\phi t}$): $$X_t = X_0e^{-\phi t} + \mu (1 - e^{-\phi t}) + \int_0^t \sigma e^{\phi (t-s)} dW_s $$

Which leads to: $$\gamma X_t = \gamma X_0e^{-\phi t} + \gamma \mu (1 - e^{-\phi t}) + \int_0^t \gamma \sigma e^{\phi (t-s)} dW_s $$

This process has:

  • mean $E[\gamma X_t] = \gamma X_0e^{-\phi t} + \gamma \mu (1 - e^{-\phi t})$
  • variance $Var[\gamma X_t] = \int_0^t \gamma^2 \sigma^2 e^{2 \phi(t-s)} ds = \frac{\gamma ^2 \sigma ^2}{2 \phi} (1 - e^{-2\phi t}) $

At $t \rightarrow \infty$, the process $\gamma X_\infty$ is gaussian with:

  • mean $E[\gamma X_\infty] = \gamma \mu$
  • variance $Var [\gamma X_\infty] = \frac{\gamma^2 \sigma^2}{2\phi}$

So its exponent is lognormal with mean: $$E[Y_\infty ^\gamma] = E[e^{\gamma X_\infty}] = e^{E[\gamma X_\infty] + \frac{1}{2} Var[\gamma X_\infty]} = e^{\gamma \mu + \frac{\gamma^2\sigma^2}{4\phi}}$$

You were almost there but were a bit too quick.

PS: one can also apply Itô's lemma to get the expression of $Y_t^\gamma$ then get its mean and take the limit. I chose to do $\gamma X_t$ and take the exponent at the end to follow your approach.


Your Answer

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

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