(Edited) I met a univariate Ornstein-Uhlenbeck type process but with self soft-thresholding: $$ dX(t) = - c\ \mbox{sgn}(X(t))\big[|X(t)|-c_1 t^{\mu}\big]_+ dt + \sigma dB(t), \quad X(0)=0, $$ where $B(t)$ is the standard Brownian motion, and $c,c_1,\sigma>0$ are constants, and $[\cdot]_+$ is a soft-thresholding operator with $[a]_+=a$ if $a> 0$ and $[a]_+=0$ if $a\leq 0$.

My Question: does the stationary distribution exist for the SDE above, for different $\mu\in[0,1]$? I suspect that there is a regime switching phenomenon at around $\mu=1/2$...

Any comment will be much appreciated!

  • $\begingroup$ I think $X(t)$ will be oscillating between $(k+o(1))\sqrt t$ and $(-k+o(1))\sqrt t$ as $t\to\infty$, for some real $k>0$. So, it seems that no stationary distribution exists. Also, in view of the law of the iterated logarithm for the Brownian motion, with probability $1$ the paths of $X$ will be substantially different from those of the Brownian motion. Also, there seems to be no reason to believe that there exists a closed-form solution to this SDE. $\endgroup$ – Iosif Pinelis May 20 '18 at 21:32
  • $\begingroup$ Thank you. Suppose I consider the following process instead:$$ dX(t) = - c\ \mbox{sgn}(X(t))\big[|X(t)|-c_1 t^{\gamma}\big]_+ dt + \sigma dB(t), \quad X(0)=0, $$ where $c_1>0$, I wonder if anything can be said about the behavior of $X(t)$ with respect to different value of $0 < \gamma\leq 1$? $\endgroup$ – Nick May 21 '18 at 1:49
  • $\begingroup$ I think, again in view of the law of the iterated logarithm, it will depend on whether $\gamma\le1/2$. If so, I think $X(t)$ will be oscillating between $(k+o(1))t^\gamma$ and $(−k+o(1))t^\gamma$ as $t\to\infty$. If $\gamma>1/2$, then I think $X$ will be similar to $\sigma B$. $\endgroup$ – Iosif Pinelis May 23 '18 at 20:44
  • $\begingroup$ Thank you! I ran some simulations. For $\gamma>1/2$, the behavior of $X$ does look like $\sigma B$. For $\gamma<1/2$, $X$ explodes. However, rather than oscillating between positive and negative values, $X$ explodes toward one of the directions. Some simulation paths explode toward $+\infty$, some toward $-\infty$, but they do not oscillate. $\endgroup$ – Nick May 24 '18 at 3:11
  • $\begingroup$ I think you should recheck your simulation at least for $\gamma\le1/2$. If $X$ takes a very large positive (say) value, then there will be a strong negative trend. $\endgroup$ – Iosif Pinelis May 24 '18 at 11:05

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.